Datasets:

Hendrik-a

/

arxiv

like 0

we study the evolution of the observed photospheric magnetic field and the modeled global coronal magnetic field during the past 3 1/2 solar activity cycles observed since the mid-1970s . we use synoptic magnetograms and extrapolated potential - field models based on longitudinal full - disk photospheric magnetograms from the nso s three magnetographs at kitt peak , the synoptic optical long - term investigations of the sun ( solis ) vector spectro - magnetograph ( vsm ) , the spectro - magnetograph and the 512-channel magnetograph instruments , and from the u. stanford s wilcox solar observatory . the associated multipole field components are used to study the dominant length scales and symmetries of the coronal field . polar field changes are found to be well correlated with active fields over most of the period studied , except between 2003 - 6 when the active fields did not produce significant polar field changes . of the axisymmetric multipoles , only the dipole and octupole follow the poles whereas the higher orders follow the activity cycle . all non - axisymmetric multipole strengths are well correlated with the activity cycle . the tilt of the solar dipole is therefore almost entirely due to active - region fields . the axial dipole and octupole are the largest contributors to the global field except while the polar fields are reversing . this influence of the polar fields extends to modulating eruption rates . according to the computer aided cme tracking ( cactus ) , solar eruptive event detection system ( seeds ) , and nobeyama radioheliograph prominence eruption catalogs , the rate of solar eruptions is found to be systematically higher for active years between 2003 - 2012 than for those between 1997 - 2002 . this behavior appears to be connected with the weakness of the late - cycle 23 polar fields as suggested by luhmann . we see evidence that the process of cycle 24 field reversal is well advanced at both poles . = 1

1303.1218

since synoptic photospheric magnetogram observations began in the mid-1970s , three full solar activity cycles have been observed and a fourth has begun . as has been widely reported , the recent behavior of the solar magnetic field has stood out as unusual compared to the previous three decades or so of regular synoptic measurements . since the cycle 23 polar field reversal , the polar fields have been about 40% weaker than the two previous cycles ( wang et al . 2009 , gopalswamy et al . 2012 ) and the activity level of cycle 24 has been unusually low , following a very long activity minimum . the weak polar fields have been linked by wang et al . ( 2009 ) to a 20% shrinkage in polar coronal - hole areas and a reduction in the solar - wind mass flux over the poles . low - latitude coronal holes were larger and more numerous than during the previous minimum ( lee et al . 2009 , de toma 2011 ) . large , low - latitude coronal holes were present as late as 2008 , finally disappearing in 2009 ( de toma 2011 ) . meanwhile , the coronal streamer structure and the heliospheric current sheet only became axisymmetric in the equatorial plane after sunspot numbers fell to unusually low values in mid-2008 ( wang et al . 2009 , thompson et al . 2011 ) when the heliospheric flux was at its weakest level since measurements began in 1967 ( sheeley 2010 ) . the activity level of cycle 24 looks unlikely to reach even the modest heights of cycle 23 . at the time of writing , the polar fields are on the point of swapping polarities , consistent with cycle 24 having reached its maximum activity level , and yet the sunspot number is currently little more than half the the typical value recorded during the cycle 23 maximum . in the past , polar field polarity reversals have coincided with activity cycle maxima . according to observations of the `` rush to the poles '' in the fe xiv corona 1.15 solar radii ( altrock 2013 ) and prominence eruption activity in microwave brightness observations ( gopalswamy et al . 2012 ) , solar maximum conditions have arrived at the northern hemisphere but not in the south . on the other hand , there seems to be no reason why the sunspot number can not increase further during this cycle . some recent work has been directed towards relating these observed changes in solar magnetism to standard babcock - leighton models of the global solar dynamo . babcock ( 1959 , 1961 ) produced a powerful phenomenological description of the solar cycle based on his magnetograph observations . this model was then developed by leighton ( 1964 , 1969 ) into a kinematic model for the transport of photospheric magnetic flux , in which photospheric turbulent diffusion played a key role in decaying active - region fields , causing their leading polarities to interact across the equator and spreading their lagging polarities pole - ward , providing a link between active - region and polar fields . this model was later modified by wang and sheeley ( 1991 ) and wang et al . ( 1991 ) to relate the polar flux distribution to the observed poleward transport of decayed active - region flux by meridional flows . since then the active - region and polar fields have been coupled in babcock - leighton models via a combination of surface diffusion and meridional flow . the recent unusual behavior of the solar field has been modeled by introducing unusual changes in the meridional flow speed in babcock - leighton models ( e.g. , schrijver and liu 2008 , wang et al . 2009 ; nandy et al . , 2011 ) . other work has applied observed solar magnetograms to relate ideas from kinematic dynamo theory to patterns in photospheric magnetic fields and extrapolated coronal models . using more than 37 years of nso kitt peak and mt . wilson 150-foot tower full - disk longitudinal magnetograms , petrie ( 2012 ) found strong correlation in each hemisphere between poloidal active region fields , high - latitude poleward field surges , and polar field changes , consistent with the relationship between poloidal active region fields and polar field changes in the babcock - leighton model . the results also showed that the weak polar fields observed since the maximum of cycle 23 may have been caused by the hemispheric bias of the active region poloidal field component in each hemisphere effectively disappearing after the cycle 23 polar field reversal . derosa et al . ( 2012 ) decomposed wilcox solar observatory ( wso ) and michelson doppler imager ( mdi ) field measurements into multipole components and studied the wso multipoles evolution over three solar cycles to characterize the dominant large - scale structure of the corona over this time ( see also hoeksema 1984 ) . derosa et al . focused on axisymmetric multipole components , and found a coupling between those antisymmetric and symmetric about the equator . they found a corresponding coupling between antisymmetric and symmetric fields in axisymmetric babcock - leighton kinematic dynamo solutions that pointed to the necessity of both classes of field for cyclical dynamo behavior to be maintained . in this paper we seek to extend the above results . we will characterize the relationship between the axisymmetric and non - axisymmetric components of the fields in some detail and with reference to the sunspot number and polar field measurements from nso and wso . separating the fields into classes symmetric and asymmetric about the rotation axis and symmetric and anti - symmetric about the equator , we will determine which classes follow the sunspot activity cycle and which are dominated by the polar fields . the evolving relationship between the active and polar fields over time will be investigated and unusual behavior reported . we will also determine whether the field evolution can shed light on recent unexpected changes in activity indices , in particular the coronal mass ejection ( cme ) rate and the prominence eruption ( pe ) rate . the relationship between cmes and pes has been extensively studied in the past . studying the association rate , relative timing , and spatial correspondence between pes and cmes , gopalswamy et al . ( 2003 ) found that most ( 72% ) pes were clearly associated with cmes and that during the solar minimum , the central position angle of the cmes tends to be offset closer to the equator compared to the pes but not during solar maximum ( see also gopalswamy et al . shimojo et al . ( 2006 ) found that the prominence eruption rate follows the sunspot number , with differences between their peak times , and describes a butterfly pattern but with some hemispheric asymmetry . luhmann et al . ( 2012 ) have noted that the cycle 24 cme rate is comparable to the cycle 23 rate even though the cycle 24 sunspot number is significantly weaker . they suggested that this phenomenon might be related to the weakness of the polar fields since the cycle 23 polarity reversal . we will explore this suggestion using the magnetic field data and solar eruption statistics from three well - known public catalogs . the paper is organized as follows . section [ sect : data ] describes the various data sets analyzed in the paper . in section [ sect : activepolar ] the evolution of the active and polar fields , and the axisymmetric and non - axisymmetric multipoles , are described in detail . the relationship between these fields and the cme and pe rates is explored in section [ sect : cmes ] . we conclude in section [ sect : conclusion ] . we analyze synoptic maps for the radial photospheric magnetic field component from three nso kitt peak magnetographs that have been observing the full - disk longitudinal photospheric field daily , weather permitting , since 1974 : the synoptic optical long - term investigations of the sun ( solis ) vector - spectromagnetograph ( vsm ) , that has been observing the 630.2 nm fe i spectral line since august 2003 ( keller et al . , 2003 ) ; the spectro - magnetograph ( spmg ) , that observed the 868.8 nm fe i spectral line from november 1992 until september 2003 ( jones et al . , 1992 ) ; and the 512-channel instrument , that observed the 868.8 nm fe i spectral line from february 1974 until april 1993 ( livingston et al . , 1976 ) . the earliest synoptic map covers carrington rotation 1626 in march 1975 . together the data set covers cycles 21 - 23 in their entirety , as well as the end of cycle 20 , and cycle 24 up to the present . the sky images were taken at @xmath0 pixel@xmath1 spatial resolution and @xmath2-pixel full - surface synoptic maps for the radial field were constructed in longitude - sin(latitude ) coordinates by software pipelines running at nso . we also analyze time series of synoptic photospheric radial magnetic field data derived from measurements of the line - of - sight field taken at wso using the fe i spectral line at 525.02 nm ( scherrer et al . 1977 ) . the wso magnetograph scans the solar image with a @xmath3 square aperture . the wso synoptic maps cover about 36 1/2 years from may 1976 ( carrington rotation 1642 ) until the present . the data are analyzed here in the form of spherical harmonic coefficients calculated rotation by rotation at wso and made available online , as well as estimates of the polar field strengths taken every 10-days over a 30-day window . the nso data clearly have much higher spatial resolution than the wso data but a major advantage of the wso data is that the wso telescope has been operating continuously since 1976 with no instrument changes . the nso / kp data set , deriving from three different instruments , has to be cross - calibrated . the analysis is complicated by the fact that the nso sensitivity and saturation levels vary significantly among the three nso instruments , and so the results must be interpreted with care . the wso data can be analyzed without the complication of cross - calibration issues . however , there are known saturation errors in the wso data generally . following derosa et al . ( 2012 ) we do not correct the wso data for known saturation effects that cause the fields to be underestimated by a factor of about 1.8 ( svalgaard et al . 1978 ) . in a later section we will investigate the connection between the active and polar fields on the one hand and the rate of solar eruptions on the other . the computer aided cme tracking project ( cactus , robbrecht et al . 2009 ) automatically detects coronal mass ejections ( cmes ) in image sequences from nasa s large angle and spectrometric coronagraph ( lasco ) without human intervention . compared to catalogs assembled manually by human operators , these automatic cme detections might be more objective as the detection criterion is written explicitly in a program but the lack of manual filtering also means that the catalog needs to be treated with some caution . the detection method and the catalog are described and analyzed in detail by robbrecht et al . ( 2009 ) , who found the statistical properties of the derived cme rates to be encouraging . until 2010 the algorithm ran on level 0 images from the lasco c2 and c3 instruments and since 2010 quick - look images from these instruments have been used . the solar eruptive event detection system ( seeds ) project at george mason university s space weather laboratory also runs an automated algorithm to detect , classify and characterize eruptive events including cmes ( olmedo et al . 2008 ) . a catalog of cmes covering the lasco data set has been collected . we will determine whether the cactus and seeds statistics give consistent results . we also use the catalog of limb prominence eruption ( pe ) events derived from microwave imaging observations at 17 ghz made by the nobeyama radioheliograph ( nakajima et al . the nobeyama radioheliograph is a radio interferometer offering full - disk coverage with @xmath4 resolution at 17ghz and temporal resolution of 1 second . figure [ fig : butterfly ] shows the nso / kp magnetic butterfly diagram , updated from figure 1 of petrie ( 2012 ) to early 2013 and smoothed using a 27-day boxcar filter . this plot was derived using the full - disk sky images collected at nso / kp since february 1974 . the diagram clearly shows the cyclical behavior of the active and polar fields . features of the babcock - leighton model of the solar cycle are clearly evident . the active fields begin each cycle emerging at latitudes around @xmath5 and subsequently emerge at progressively lower latitudes on average , creating the wings of the distinctive butterfly patterns . the diagram also shows the change of polarity of the polar fields once each cycle , coinciding with activity maximum ( babcock 1959 ) . most of the flux that emerges in active regions cancels with flux of opposite polarity but a proportion survives as weak flux that is carried poleward by the meridional flow . this poleward drift of the weak , decayed magnetic flux appears as plumes of one dominant polarity at high latitudes , between about @xmath6 and @xmath7 , corresponding to changes in the polar field . in each hemisphere during each cycle the plumes generally have a strong tendency to take the polarity of the lagging polarity of the active regions . these patterns reflect the well - known facts that magnetic bipoles are typically asymmetric and tilted so that the leading polarity is stronger , more compact and slightly closer to the equator than the following polarity ( hale et al . 1919 ) , and the vast majority of bipoles in each hemisphere have the same leading polarity , the leading polarities are opposite in the two hemispheres during any given cycle , and all polarity patterns reverse each cycle ( hale and nicholson 1925 ) . thus the active and polar fields appear to be linked together in a single magnetic cycle ( babcock 1961 ) . leighton ( 1964 , 1969 ) showed that the active - region and polar fields could be coupled by the effects of photospheric diffusion , and wang & sheeley ( 1991 ) and wang et al . ( 1991 ) demonstrated that systematic pole - ward meridional flows likely play a key role in coupling the two classes of field in a global activity cycle . figure [ fig : butterfly ] clearly shows that the recent cycle 23 minimum was unusually long and quiet . figure [ fig : butterfly ] also shows that cycle 23 s polar fields also stand out as being weaker than the polar fields for any other cycle , and the plumes of decayed active - region flux were generally weak and of unusually mixed polarity following the cycle 23 polar reversal . the relationship between these two latter observations was explored by petrie ( 2012 ) , as is the hemispheric asymmetry of the activity . figure [ fig : ssn ] shows the variation of the sunspot number over the same period . the sunspot maxima / minima coincide with the activity maxima / minima in figure [ fig : butterfly ] , including the unusually long minimum of cycle 23 and the unusually low activity level of cycle 24 . the sunspot number does not represent all active - region fields , especially during a weak cycle such as the present one , but it is a standard index of solar activity and is generally found to correlate well with other activity indices . figure [ fig : poles ] shows estimates of the average field strengths at polar latitudes based on measurements from the two observatories . at wso the north and south polar line - of - sight field strength is measured daily in the @xmath8 apertures nearest the poles , north and south . every 10 days an average is derived for each pole from measurements in a 30-day centered window . the nso / kp estimates derive from the butterfly map shown in figure [ fig : butterfly ] and represent the radial field component . this is part of the reason why the nso measurements are so much stronger than the wso measurements . there are also well documented calibration issues with the wso data ( svalgaard et al . 1978 ) causing the fields to be underestimated by a factor of 1.8 . setting these differences aside , the time series shown in the two panels of figure [ fig : poles ] are clearly well correlated with each other . they agree , for example , that the polar fields were only about 60% as strong during the cycle 23 minimum compared to the cycle 22 minimum . the two plots also agree that the south polar field steadily decreased in strength beginning early in 2010 reversed polarity around mid-2012 , and that the recent change in the north polar field has been more abrupt . the difference in latitude range between the nso data ( between about @xmath9 to about @xmath10 ) and the wso data ( between about @xmath11 and the poles ) in figure [ fig : poles ] cause the change in the north polar field to appear more gradual and to begin earlier in the wso plot than in the nso plot . however , both the nso and wso north pole curves agree that the north polar field is rapidly declining towards zero and figure [ fig : butterfly ] shows that the last remains of a positive - polarity polar cap survive south of @xmath12 . the response of the coronal magnetic field to the photospheric activity patterns can be diagnosed in a simple way by calculating extrapolated potential - field source - surface ( pfss , altschuler and newkirk 1969 , schatten et al . 1969 ) models using the nso and wso synoptic maps as lower boundary conditions . low in the corona , the magnetic field is sufficiently dominant over the plasma forces that a force - free field approximation is generally applicable . moreover , for large - scale coronal structure the effects of force - free electric currents , which are inversely proportional to length scale , may be neglected . a potential - field model for the corona must satisfy , @xmath13 the solution of equation ( [ eq : currentfree ] ) can be represented as the gradient of a scalar potential , @xmath14 . equation ( [ eq : solenoidal ] ) therefore becomes @xmath15 we identify the lower boundary of the model with the solar photosphere at @xmath16 where @xmath17 is the solar radius . we use the photospheric radial field maps from nso to fix @xmath18 at the lower boundary @xmath16 . the use of radial field data , derived from longitudinal field measurements assuming the magnetic vector to be approximately radial in general , has been found to be more successful in reconstructing coronal magnetic structure than the direct application of longitudinal measurements as boundary data ( wang and sheeley 1992 ) . a synoptic map construction method that derives the radial component from longitudinal measurements without assuming the boundary measurements to be approximately radial has been developed and applied to chromospheric data by jin et al . ( 2013 ) with promising results . here we adopt the standard photospheric synoptic maps because measured photospheric fields are found to be approximately radial in general ( svalgaard et al . 1978 , petrie and patrikeeva 2009 , gosain and pevtsov 2012 ) and the resulting models remain competitive for reasons explained by wang and sheeley ( 1992 ) . above some height in the corona , usually estimated to be between 1.5 and 3.5 solar radii , the magnetic field is dominated by the thermal pressure and inertial force of the expanding solar wind . to mimic the effects of the solar wind expansion on the field , we introduce an upper boundary at @xmath19 , and force the field to be radial on this boundary by setting @xmath20 there , following altschuler and newkirk ( 1969 ) , schatten et al . ( 1969 ) and many subsequent authors . the usual value for @xmath21 is @xmath22 although different choices of @xmath21 lead to more successful reconstructions of coronal structure during different phases of the solar cycle ( lee et al . 2011 ) . for simplicity we adopt the standard value @xmath22 in this work . with these two boundary conditions , the potential field model can be fully determined in the domain @xmath23 . we use the national center for atmospheric research s ( ncar ) mudpack package to solve equation ( [ eq : currentfree ] ) numerically subject to the above boundary conditions . although equation ( [ eq : currentfree ] ) can be solved analytically , and has been so treated for several decades , we adopt a finite - difference approach in this subsection to avoid some problems associated with the usual approach based on spherical harmonics ( tth et al . mudpack ( adams 1989 ) is a package for efficiently solving linear elliptic partial differential equations ( pdes ) , both separable and non - separable , using multigrid iteration with subgrid refinement procedures . by iterating and transferring approximations and corrections at subgrid levels , a good initial guess and rapid convergence at the fine grid level can be achieved . multigrid iteration requires less storage and computation than direct methods for non - separable elliptic pdes and is competitive with direct methods such as cyclic reduction for separable equations . in particular , three - dimensional problems can often be handled at reasonable computational cost . we use the mudpack package here to solve equation ( [ eq : laplace ] ) in its non - separable form in spherical coordinates , subject to the boundary conditions described above . jiang and feng ( 2012 ) have recently presented a high - speed combined spectral / finite - difference pfss solution method using the related ncar fishpack package . figures [ fig : pfss1 ] and [ fig : pfss2 ] show hairy - ball plots of 24 example models representing the turns of the 24 years since the beginning of 1990 . the foot - points of open fields in these models correspond to coronal holes observed in , e.g. , extreme ultraviolet or helium images while the set of tallest closed field lines represent streamer belts seen in coronagraph images . in 1990 , corresponding to the beginning of the time series in figure [ fig : pfss1 ] , the solar field was close to cycle 22 maximum as figure [ fig : butterfly ] also indicates . in the first model both polar coronal holes are well defined although the northern hole is significantly displaced from the heliographic north pole . in 1991 the polar coronal holes were still present but weaker . by 1992 the polar holes had reversed polarity , and they were already well developed and centered at the heliographic poles . in 1996 activity cycle 22 was over and the coronal field was approximately axisymmetric . by 1998 cycle 23 was well under way and the coronal field was again highly complex and non - axisymmetric . the cycle 23 polar field reversal was different from the cycle 22 reversal . during cycle 23 the north polar coronal hole disappeared about a year earlier than the southern polar hole . this one - year lag can also be seen in figure [ fig : butterfly ] . as figure [ fig : pfss2 ] shows , the southern hole was not fully formed until about 2004 . cycle 23 did not produce a nearly axisymmetric , dipole - like global coronal field structure until 2009 when the sunspot number became especially low according to figure [ fig : ssn ] . as has been widely discussed in the literature ( see section [ sect : intro ] ) , this is believed to be due to the weakness of the polar fields during cycle 23 . by 2011 cycle 24 had begun and the global coronal field was again complex and non - axisymmetric . the northern polar coronal hole was disappearing by 2012 . at the beginning of 2013 the northern polar hole had effectively disappeared , but had not yet reappeared with reversed polarity , while the southern polar hole was weakening steadily . to investigate the evolution of the coronal field structure in more detail , we also calculate analytical solutions of equation ( [ eq : laplace ] ) that are constructed as sums of spherical harmonics . decomposing global solar magnetic fields into spherical harmonic components reveals which spatial scales and symmetries are dominant during any phase of the field s evolution . these components therefore allow us to summarize the main features of the global field in a useful and instructive way . the wso group distributes pfss models and the associated spherical harmonic coefficients on their website . we compute spherical harmonic coefficients for the nso data using the same method , that described by altschuler and newkirk ( 1969 ) and hoeksema ( 1984 ) . we briefly describe this method below . solutions of equation ( [ eq : laplace ] ) in the domain @xmath23 can be written as , @xmath24 ( g_n^m\cos m\phi + h_n^m\sin m\phi ) p_n^m ( \theta ) \right ] .\ ] ] here the functions @xmath25 are the usual legendre polynomials with the schmidt normalization ( chapman and bartels 1940 , altschuler and newkirk 1969 , hoeksema 1984 ) , and the legendre coefficients @xmath26 and @xmath27 are in units of magnetic field strength and are determined from the known photospheric field distribution @xmath28 by , @xmath29 with the schmidt normalization used here for the legendre polynomials , the coefficients @xmath26 and @xmath27 refer to the magnitudes of the multipole components of the magnetic field multiplied by @xmath30 ( hoeksema 1984 ) . because multipoles of different principal index @xmath31 have different @xmath32-dependence , the coefficients must also be adjusted by an @xmath32-dependent factor before multipole strengths can be compared at any chosen height . for example , to measure the comparative influence of two multipoles on the global field structure , their strengths may be compared at @xmath33 we must apply the factor @xmath34 ( hoeksema 1984 ) . the coefficient @xmath35 corresponds to the monopole term , set to zero in the computation . the @xmath36 is the axial polar dipole ( the dipole parallel to the rotation axis ) and the terms with coefficients @xmath37 and @xmath38 represent the two orthogonal equatorial dipoles . the principal index , @xmath31 , is the total number of circles of nodes on the photosphere and the secondary index , @xmath39 , is the number of those nodal circles passing through the pole . therefore terms with @xmath40 are axisymmetric . the lowest - degree multipoles correspond to the largest length scales , so the comparative amplitudes of the dipole , quadrupole and other low - degree multipoles are useful diagnostics of the global coronal field structure over multiple solar cycles . we describe the temporal evolution of the dipole components in subsection [ s : dipoles ] and the evolution of the higher - order components in subsection [ s : multipoles ] . there are three independent dipole components of the global magnetic field : an axial ( axisymmetric ) component aligned with the solar rotation axis ( @xmath36 ) and two independent components spanning the equatorial plane ( @xmath37 and @xmath38 ) . these dipole components can be added vectorially to give the full strength and direction of the dipole field at any time . the strength of the the equatorial dipole , the vector sum of the equatorial components , can be compared to the axial dipole as shown in figure [ fig : dipoles ] . in this figure the axial dipole points north when positive and south when negative . the axial dipole is observed to be at its strongest during solar minimum , after the previous cycle s decayed field has accumulated at the poles but before the new cycle has begun to send opposite - polarity field poleward weakening the polar fields towards polarity reversal . the equatorial dipole , on the other hand , follows the activity cycle , being largest during activity maximum and smallest during activity minimum . figure [ fig : dipoles ] shows that the axial dipole has been becoming progressively weaker from cycle to cycle since cycle 21 . this evolution follows the pattern of the polar field strengths plotted in figure [ fig : poles ] . the equatorial dipole has also been weaker during cycles 23 and 24 than during cycles 21 and 22 , following the behavior of the sunspot number shown in figure [ fig : ssn ] . the equatorial dipole is formed mostly by the azimuthal ( east - west ) component of active region fields , whenever these form a net dipole moment in the equatorial plane . because of the joy s law tilt bias in each hemisphere , the active regions would tend to produce a small axial dipole moment of the same sign in each hemisphere but this moment is small , and is insignificant compared to the moment of the polar fields because of the generally small joy s law tilt angle . the influence of the active regions on the equatorial dipole is much larger . the effect of the polar fields on the equatorial dipole , through non - axisymmetry in the polar fields , is insignificant as figure [ fig : dipoles ] indicates . the dipole tilt angle , the angle in the poloidal plane between the dipole and the equator , is also plotted in figure [ fig : dipoles ] . this is a useful diagnostic of the global coronal field because it offers a comparison of the polar and low - latitude ( active region ) contributions to the global field . the tilt is close to @xmath41 latitude during solar minimum and passes through @xmath42 at the height of maximum activity when the polar fields reverse polarity . this process can take a few months to a year at each pole , and the two poles do not always reverse polarity around the same time . the axial dipole reversals have taken between one and two years with the cycle 22 reversal around 1990 the quickest and simplest of the sample , corresponding to the comparatively quick and simple cycle 22 reversal seen in figures [ fig : butterfly ] and [ fig : pfss1 ] . the dipole strength is at its weakest during polar reversal but it does not disappear at any time . figure [ fig : dipoles ] also shows the azimuthal angle of the horizontal dipole . this angle indicates which carrington longitude the horizontal dipole is pointing towards . for example , the dipole azimuthal angle is zero when the dipole points towards @xmath42 carrington longitude , the boundary between two carrington rotations . figure [ fig : dipoles ] shows that the dipole azimuth angle has no significant dependence on the activity cycle . the angle often changes abruptly , on time - scales much less than a year , as different active regions emerge and decay at different carrington longitudes . there are , however , periods of time such as during 1984 , 1991 , 2008 and 2009 , for example , when the dipole maintains a nearly stable azimuthal direction over timescales of about a year . this is because during these periods of time most of the activity appears preferentially within a small range of carrington longitudes . there are numerous episodes evident in figure [ fig : dipoles ] when the activity appears to have been dominated by such `` active longitudes '' but such behavior shows no dependence on the activity cycle . one remarkable feature of figure [ fig : dipoles ] is that the active field , represented by the equatorial dipole component , continued to appear with significant strength during the declining phase of cycle 23 long after the polar fields ceased to strengthen . the polar dipole component became stronger between the reversal of the poles in 1998 and 2001 but after 2002 the polar dipole did not become stronger . on the other hand , the active fields continued to appear long after 2002 , until 2006 , without having any significant net effect on the polar dipole . this pattern can also be seen in the dipole tilt angle . after 2002 , after the poles reversed and the dipole tilt swung decisively southward , the trend of the tilt angle halted and even reversed for a few years . during this time the polar dipole did not strengthen while the equatorial dipole , representing the active fields , continued to be strong . indeed , it was arguably stronger in 2003 and 2004 , after the polar dipole stopped developing , than before , as figure [ fig : dipoles ] shows . between about 2001 and 2004 , after the cycle 23 polar field reversal , the equatorial dipole remained comparable in strength to the polar dipole . an essential part of the babcock - leighton mechanism is the change in the polar field strength caused by the poleward drift of decayed active - region flux . between 2002 and 2006 active - region flux continued to appear in large quantities but the polar fields did not change accordingly . we will return to this issue in the following subsection . more recently the polar dipole has steadily weakened as the cycle 24 polar field reversal approaches . meanwhile the equatorial dipole has increased , reflecting the ascent of cycle 24 . the dipole tilt angle is quickly decreasing , principally because of the weakening polar dipole . according to the latest synoptic data from both solis and wilcox the dipole tilt has recently turned positive and the polar dipole has recently pointed north for the first time in this cycle . the simplest way to understand the cyclic behavior of the higher - order multipoles is to plot the axisymmetric ( @xmath40 ) and non - axisymmetric ( @xmath43 ) fields separately . figure [ fig : mpoles ] shows plots of the full field strength , the combined strength of all multipoles , as well as the strength of the axisymmetric and non - axisymmetric components , measured by the two observatories . figure [ fig : mpoles ] shows that the axisymmetric fields dominate during activity minima and the non - axisymmetric fields dominate during maxima . the pearson linear correlation coefficient ( @xmath44 ) between the time series for all wso fields and the sunspot number ( figure [ fig : ssn ] ) is low , about 0.10 . the @xmath44 between all wso fields and the polar field strength , the combined strengths of the north and south polar fields plotted in figure [ fig : poles ] , is about 0.75 . the @xmath44 between the non - axisymmetric wso fields and the sunspot number is much higher , about 0.97 . the axisymmetric wso fields correlate equally well with the total polar field strength with @xmath44 about 0.97 . the axisymmetric fields began to decline slowly after 2007 . figure [ fig : poles ] shows that this may be because the north polar field began to weaken steadily after 2007 . the correlations for the nso / kp data are less impressive because of the difficulty of cross - calibration between the three nso / kp magnetographs ( see section [ sect : data ] ) but the patterns of correlation are qualitatively the same . these correlations illustrate the enormous influence of the polar fields on the global coronal magnetic field , and suggests an association between sunspot / active region fields and non - axisymmetric global coronal structure on the one hand , and polar fields and axisymmetric global coronal structure on the other . this association is not simple , however , as the following discussion will show . in figures [ fig : mpolesarraykp ] and [ fig : mpolesarraywilcox ] the individual poloidal orders are plotted separately . higher - order multipoles differ in behavior from the dipole ( @xmath45 ) components in general . for example , the full quadrupolar fields ( @xmath46 , figures [ fig : mpolesarraykp ] and [ fig : mpolesarraywilcox ] , top left plots , red curves ) differ from the dipolar fields ( @xmath45 , figures [ fig : mpolesarraykp ] and [ fig : mpolesarraywilcox ] , bottom left plots , red curves ) in being strongest during activity maxima and weakest during activity minima . the axisymmetric and non - axisymmetric quadrupolar fields follow the activity cycle in this way . unlike the axial dipole ( top middle plots , red curves ) , the axisymmetric quadrupole ( bottom middle plots , red curves ) is of even order and is symmetric about the equator , and therefore its behavior is not dominated by the polar fields . all quadrupolar components , axisymmetric and non - axisymmetric , are perturbed mostly by the active region fields and their strengths wax and wane with the activity cycle . the octupole fields are of odd order ( @xmath47 ) and so the axisymmetric octupole , like the axial dipole , follows the behavior of the polar fields while the non - axisymmetric octupolar fields follow the activity cycle . the qualitative patterns described here apply to both nso / kp and wso data sets but the correlation coefficients quoted below derive from the wso data only . the high - order fields are generally well correlated with the sunspot number . the full even- and odd - order multipoles correlate equally well ( @xmath48 ) with the sunspot number for orders higher than hexadecapole ( @xmath49 ) . the low - order even multipoles also correlate well ( @xmath50 ) with the sunspot number whereas the low - order odd multipoles , the dipole ( @xmath45 ) and octupole ( @xmath47 ) , instead follow the polar fields . the axisymmetric dipole ( @xmath51 ) and octupole ( @xmath52 ) components correlate well with the polar fields but the axisymmetric multipoles of higher order do not ( @xmath53 ) . the low even axisymmetric orders are not correlated with the sunspot number or with the polar fields . the high even axisymmetric orders show some correlation with the sunspot number ( @xmath54 in some cases ) , possibly perturbed by joy s law tilt of active region fields , but these components are weaker than their non - axisymmetric counterparts . therefore the excellent overall correlation between the axisymmetric fields and the polar fields is exclusively due to the axial dipole and octupole fields . because of their large spatial scale and impressive strength these fields have a major influence on the global coronal field over most of the cycle . the non - axisymmetric multipole fields correlate increasingly well with the sunspot number as the order @xmath31 increases , from @xmath55 for the equatorial dipole to @xmath51 for the ninth - order fields . this behavior corresponds to the patterns in the right plots in figures [ fig : mpolesarraykp ] and [ fig : mpolesarraywilcox ] , where in particular the response of the equatorial dipole to the activity cycle is clearly not as steady as the response of the higher - order non - axisymmetric multipoles . since low multipole orders correspond to large spatial scales , it may frequently occur that multiple active regions perturb high - order multipoles without the lowest - order multipoles responding significantly . for example , two identical active regions at antipodal points would give zero equatorial dipole but non - zero higher multipole components . there is no evidence that the non - axisymmetric multipoles respond to the polar field cycle , including during polar reversal . figures [ fig : pfss1 ] and [ fig : pfss2 ] clearly show that the lowest - order dipole - like structure of the global coronal field is often tilted with respect to the rotation axis . the centers of the polar coronal holes are sometimes significantly displaced from the heliographic poles . the absence of any significant polar field signature in the non - axisymmetric fields in figures [ fig : mpolesarraykp ] and [ fig : mpolesarraywilcox ] shows that the polar field reversal is not caused by a real physical dipole rotation but by the disappearance of the polar caps and their reappearance with opposite polarity , all taking place at the heliographic poles , and that the tilt of the solar dipole is therefore almost entirely due to the active - region fields . in summary , the active regions drive practically all non - axisymmetric fields and high - order fields in the corona . only the lowest odd - order axisymmetric components , the dipole ( @xmath45 ) and the octupole ( @xmath47 ) , follow the polar fields but these components are very influential . because of their large spatial scale they are very dominant during solar activity minima and are influential during maxima except during polar field reversal . the sunspot number does not represent all active - region fields but the above results indicate a strong correlation between the sunspot number and the strength of high - order ( @xmath56 ) and non - axisymmetric multipoles . various explanations for the weakness of the cycle 23 polar fields have been suggested . via simple numerical estimates and detailed kinematic dynamo modeling , dikpati ( 2011 ) showed that even a quite modest decrease in active region field strength from one cycle to the next , such as between cycles 22 and 23 , could produce a large decrease of polar field strength . generally speaking , the presence of active region fields is correlated with changes in the polar fields as figure [ fig : mpoles ] shows . however , there are intervals of time when significant quantities of active region field are present in the photosphere but the polar fields do not change significantly . in particular , between 2002 and 2006 there were significant non - axisymmetric ( active ) fields in the photosphere while the large - scale , odd - order axisymmetric ( polar ) fields remained remarkably constant . because there are periods of time when there are active region fields on the sun that produce no detectable effect on the polar fields , dikpati s analysis does not fully explain the weakness of the polar fields . in the babcock - leighton dynamo model there are two possible explanations for unchanging polar fields in the presence of active regions . according to both explanations , decaying active fields still reach polar latitudes during these time intervals but these decayed active region fields are of such mixed polarity that their net effect on the polar fields is approximately zero . figure [ fig : butterfly ] shows that the plumes of decayed active - region flux moving poleward have been of more mixed polarity since the cycle 23 polar reversal than before . one explanation is that the meridional flows are so fast that the leading polarities in the two hemispheres do not have time to interconnect and interact with each other before being swept poleward ( e.g. , schrijver and liu , 2008 ; wang et al . , 2009 , nandy et al . , 2011 ) . to affect the polar fields significantly , these fast meridional flows would have to occur at active latitudes , as dikpati ( 2011 ) has emphasized . ulrich s ( 2010 ) , basu and antia s ( 2010 ) and hathaway and rightmire s ( 2010 ) meridional flow speed measurements do not show evidence of significantly faster flows at active latitudes during cycle 23 than during previous cycles . an alternative explanation is that the active region joy s law tilts lost their hemispheric bias during cycle 23 ( e.g. jiang et al . 2011 , petrie 2012 ) . in this scenario approximately equal quantities of each polarity would be sent poleward with approximately zero net effect on the polar fields even for slow meridional flow speeds . petrie ( 2012 ) found that the latitude centroids of the positive and negative active region fields converged in each hemisphere around 2003 , implying a disappearing active - region poloidal field around this time . at the same time the high - latitude poleward surges of field were observed to lose their polarity bias in each hemisphere and the polar fields stopped strengthening , consistent with the babcock - leighton model . schrijver and liu ( 2008 ) and stenflo and kosovichev ( 2012 ) have analyzed the the joy s law tilt angle of selected magnetic bipoles over time and found no statistical change in average joy s law tilts , and li and ulrich ( 2012 ) found from a long - term study that tilt angles of spots appear largely invariant with respect to time at a given latitude , but they decrease slowly during each cycle following the butterfly diagram . on the other hand , tlatov et al . ( 2010 ) found distinctly different bipole tilt angle distributions for different classes of bipoles during cycles 21 - 23 , and lefvre and clette ( 2011 ) discovered that the size and complexity distributions of sunspots changed significantly during cycle 23 . some change in the sunspot or bipole size or complexity distributions may be related to the change in the effects of the active region fields on the poles , and the weakness of the polar fields , during cycle 23 . we now investigate consequences of the behavior of the solar field for coronal mass ejection ( cme ) and prominence eruption ( pe ) rates . there are two long time intervals over which continuous cme rate data are available . the 15-year period between 1975 and 1989 was studied by webb and howard ( 1994 ) who carefully cross - calibrated cme rate information from numerous observational sources . the royal observatory of belgium s computer aided cme tracking ( cactus ) project ( robbrecht et al . 2009 ) has produced estimates for cme rates covering the large angle and spectrometric coronagraph ( lasco ) data archive from 1997 until the present , as has the solar eruptive event detection system ( seeds ) project of george mason university . robbrecht et al . ( 2009 ) presented a detailed study of the cactus results for cycle 23 , comparing these results with those of webb and howard ( 1994 ) for cycle 21 and the beginning of cycle 22 ( 1975 - 89 ) . they found broad similarities in the behavior of the two data sets , including good correlation with the sunspot number . here we compare and contrast the data from 1975 - 1989 compiled by webb and howard and the cactus and seeds lasco data sets covering from 1997 to the end of 2012 , with reference to the magnetic results of the previous sections . luhmann et al . ( 2011 ) has noted that the cycle 24 cme rate has so far been higher than expected in view of the relatively low sunspot number of this cycle . they suggested that this may be related to the weakness of the polar fields during cycle 24 , enabling the cme rate to remain comparable to the rates during the two previous stronger field cycles by letting more modest active regions have greater coronal influence . the tallest closed field lines in the pfss models shown in figures [ fig : pfss1 ] and [ fig : pfss2 ] indicate the boundary between the closed active - region and open coronal - hole fields in the models and they generally correspond to the streamer belts that form above active regions . if the polar fields are strong then these streamer structures would tend to be magnetically stronger and more extensive , and they would therefore more effectively inhibit the eruption of the active fields beneath than if the polar fields were weak . this is why the weakness of polar fields may be related to increased eruption rates . figures [ fig : cme ] and [ fig : ssncme ] show that there is excellent correlation between the international sunspot number and the cme rate ( webb and howard ) for the years 1975 - 1989 and almost as good correlation for 1997 - 2012 . this result confirms past findings that the cme rate follows the solar cycle ( webb and howard 1994 , robbrecht et al . 2009 ) . as robbrech et al . note , sunspots only represent a subset of the source regions of cmes . several studies have found that the majority of cmes for which on - disk source regions could be identified are related to filament / prominence eruptions ( e.g. , munro et al . 1979 ; webb and hundhausen 1987 ) . however the correlation of these activity indices shows that they are likely to be consequences of a common magnetic activity cycle . webb and howard quoted a pearson correlation coefficient of 0.94 for 1975 - 1989 but we calculated a coefficient of 0.97 based on a time - weighted average of the various instrument - specific cme rates shown in webb and howard s table 1 . the correlation between the sunspot number and the soho / lasco cme rate covering the period 1997 - 2012 is less impressive , with a pearson linear correlation coefficient of 0.91 for the cactus data and 0.85 for the seeds data . in particular , the cactus and seeds data sets both tell us that the cme rate for the active years of cycle 24 is unexpectedly high relative to the sunspot number , in line with the comments by luhmann et al . ( 2011 ) . this behavior is more pronounced in the seeds data than in the cactus data . there is no corresponding behavior in the webb - howard data for cycle 21 . clearly there are significant differences between the cme rates plotted in figures [ fig : cme ] and [ fig : ssncme ] . besides instrumental differences between the 1975 - 1989 data and the lasco data , the manual identification techniques of webb and howard differ from the automated algorithms of cactus and seeds , which in turn differ from each other . the seeds statistics are two or three times as large as the cactus statistics for the same lasco data . we can not therefore draw conclusions from the amplitudes of the curves in figure [ fig : cme ] . we can , however , compare and contrast the patterns of behavior that these curves represent . it appears that the weaker correlation for 1997 - 2012 compared to 1975 - 1989 signifies a fundamental change in the global solar magnetic field than with problems in the cme statistics . robbrecht et al . ( 2009 ) found that the cactus statistics for lasco can be more consistent and more objective than manually compiled statistics - see their figure 2 . the 1997 - 2012 cme rate statistics follow a simple pattern according to our analysis of two independent catalogs , cactus and seeds , suggesting that a real change of behavior took place . the middle and bottom scatter plots in figure [ fig : ssncme ] show that the active years 2003 - 06 , after the strengthening trend of the polar fields stopped , all lie above the linear regression fit . the two active years of cycle 24 , 2011 - 12 , lie further above the regression fit . the cme rates of these two years are comparable with those of the cycle 23 maximum even though the active magnetic fields have been significantly weaker during cycle 24 compared to cycle 23 . this change seems to correspond to the polar fields becoming weaker during cycle 23 ( section [ sect : activepolar ] ) . the points below the regression fit consist of the minimum years 2007 - 09 and the maximum years before the polar fields became weak , 1997 - 02 , with the exception of a year of polar field reversal , 2000 , which appears above the regression fit . thus the active years generally fall below the regression fit before 2003 and above the regression fit after 2003 . in the top scatter plot of figure [ fig : ssncme ] for webb and howard s 1975 - 89 data there is no such separation between data points . the weaker correlation between cme rate and sunspot number in the lasco era may therefore be related to the change in the behavior of the polar fields that occurred around 2002 or 2003 . there are more subtle phase differences between the cme cycles and sunspot cycles . robbrecht et al . ( 2009 ) reported that during cycle 23 the cme cycle lagged the sunspot cycle by between 6 months and a year , and that the cme cycle rose faster than it declined . they also noted that this behavior may be specific to cycle 23 because webb and howard s ( 1994 ) data for cycle 21 did not show it . the middle and bottom panels of figure [ fig : cme ] show that during cycle 23 the cme rate rose , peaked and declined up to a year after the sunspot cycle and that the rise phase was much faster than the declining phase . this pattern agrees with robbrecht et al.s ( 2009 ) findings , and is particularly clear in the seeds data but is also evident in the cactus data . the ascent of the cme curves for cycle 24 appear to lead the sunspot number curves but this is because the cactus and seeds cme rates for cycle 24 are so strong compared to the sunspot number . the top panel of figure [ fig : cme ] indicates a quite different pattern : the cycle 21 cme cycle rose more slowly and peaked later than the sunspot cycle , but declined more quickly . the phase differences between the sunspot and cme cycles seem to be complex and variable from cycle to cycle . the change in the rate of eruptions relative to sunspot number that occurred during cycle 23 seems to have been a much simpler change of pattern . further evidence of a real change in the rate of solar eruptions comes in the form of a corresponding pattern in prominence eruption rates ( pes ) . figure [ fig : nobeyama ] shows the nobeyama radioheliograph pe rate compared to the sunspot number and the cactus and seeds lasco cme rates over the period 1997 - 2012 . comparing the top panel of figure [ fig : nobeyama ] and the middle and bottom panels of figure [ fig : cme ] , the pe rate has similar behavior to the cme rates : the active years after 2003 tend to lie above the sunspot number curve whereas those before 2003 tend to ( but do not all ) lie below , while the minimum years lie below . this pattern is not as strong for pes as for cmes but it is present in both sets of statistics . the right panel of figure [ fig : nobeyama ] shows that the cme rate has increased relative to the pe rate since the cycle 23 minimum . the pe rate kept pace with the cme rate during the declining phase of cycle 23 when many polar crown prominences were erupting . the cme and pe rates may have evolved slightly differently over time because pes do not occur as high in the atmosphere and therefore may be less inhibited by the overarching streamer structures than cmes are . evidence in support of this argument comes from gopalswamy et al.s ( 2003 ) result that cme central position angles tend to be offset towards the equator compared to their corresponding pe s , indicating a greater influence of overarching streamer arcade structure on cmes than on pes . alternatively the differences may be related to differences in prominence creation rates over time . prominences are structures with significant magnetic shear that has either emerged into the atmosphere or been built up there . a relative shortage of prominences may therefore imply a lack of emerged or evolved magnetic shear low in the atmosphere . for example , the presence of overlying closed fields may allow much shear to build up below before an eruption takes place . cmes , especially the largest ones , are also often associated with shear but cmes come from a variety of sources . the difference between the cme and pe statistics may therefore also be caused by the weaker polar fields and overarching arcades making it more difficult for atmospheric fields to be sheared into prominence configurations before eruptions take place . the overall pattern is that eruption rates per sunspot number have generally been higher since the cycle 23 polar field reversal than they were before . this is consistent with luhmann et al.s ( 2012 ) interpretation of the polar fields as a restraining influence on eruptions , whose effect has been reduced by the weakening of the polar fields during cycle 23 . in section [ s : multipoles ] we found that the polar fields have a dominant influence on the global coronal magnetic field structure over most of the activity cycle via the axial dipole and octupole fields . the statistics from three independent eruption catalogs indicate that the polar fields influence extends to the eruption rates themselves . the pfss modeling applied in this work represents the simplest approximation to the real global coronal magnetic structure . the pfss models neglect the effects of forces and electric currents in the low corona and they only crudely model the inertial effects of the expanding solar wind at their outer boundaries . also these techniques do not fully exploit the full - disk vector spectro - polarimetric magnetic field measurements from the nso s synoptic optical long - term investigations of the sun ( solis ) and nasa s helioseismic and magnetic imager ( hmi ) telescopes that have become available in recent years . on the other hand , these standard pfss techniques make use of the archived scalar magnetic data sets that extend back to the 1970s . they are also well understood and are known to be able to reconstruct the basic global structure of the coronal field cheaply and effectively for reasons explained by wang and sheeley ( 1992 ) . such models usually closely approximate much more expensive magnetohydrodynamics ( mhd ) models in practice ( neugebauer et al . 1998 , riley et al . 2006 ) . for these reasons the pfss model remains the basic tool for routinely modeling the global coronal field at noaa s space weather prediction center and nasa s community coordinated modeling center . the decomposition of potential fields into spherical harmonics and multipole components also offers much valuable insight into the evolution of global coronal structure in terms of dominant length scales and symmetries . using 3 1/2 solar cycles of nso / kp and wso photospheric synoptic data and extrapolated coronal potential field models , we characterized the cyclical evolution of the global photospheric and coronal magnetic fields . the global coronal morphology follows the progress of the magnetic activity cycle , taking a simple dipole - like form during each activity minimum and evolving through series of complex configurations with more structure of smaller active - region scales at the height of each activity maximum . at all phases of the cycle except times of polar field reversal the large - scale , low - order fields , particularly the axisymmetric dipole and octupole , generally dominate the global coronal morphology . the higher - order multipole components of the coronal field , corresponding to the smaller spatial scales , are strongly correlated with the sunspot / active region cycle , as are the non - axisymmetric components of the coronal field . only the axisymmetric dipole and octupole components correlate well with the polar fields but since these components are generally the largest contributors to the global field , except when the polar field reverses , the polar fields play a determinative role in structuring the global corona . there is no significant signature of polar field reversal in the non - axisymmetric multipoles . the apparent dipole tilt observed over the solar cycle is almost entirely due to the active - region fields . polar field changes are generally well correlated with active fields over most of the period studied , except between 2003 - 6 when the active fields did not correspond to significant polar field changes . this change in behavior seems to be related to the well - known fact that , the polar fields created during cycle 23 only became about 60% as strong as those created during the two previous cycles . we see evidence that the correlation between active - region fields and polar fields changes has returned since the cycle 23 minimum and the process of cycle 24 field reversal is well advanced at both poles . the temporal patterns of the different multipole components reveal the behavior of the global solar field in terms of symmetry and length scale . this kind of analysis does not reveal all properties of the active - region fields , and how these properties compare between this cycle and the previous cycles . there may be different active region size distributions and different distributions of sunspot / active region complexity at different times . these changes are beyond the scope of a multipole analysis like this one but , e.g. , tlatov et al . ( 2010 ) and lefvre and clette ( 2011 ) have found that significant changes in such distributions occurred during cycle 23 . also analyses based on longitudinal magnetograms and potential fields can not inform us of the twist and shear of the fields over activity cycles . but zhang et al . ( 2010 ) found from more than 20 years of vector magnetograms of active regions taken at observatories in mees , huairou and mitaka that helicity patterns propagate equatorwards following the sunspot cycle but , unlike sunspot polarity , helicity in each solar hemisphere does not change sign from cycle to cycle . what the potential fields did allow us to do was characterize the influence of different classes of photospheric field on the global magnetic field structure , emphasizing the importance of the polar fields . statistics from three independent solar eruption catalogs then allowed us to compare the eruption rates to the behavior of the solar magnetic field . here the influence of the polar fields was again emphasized . the annual average cactus and seeds cme rates based on lasco data have changed significantly since the cycle 23 polar reversal : they are both systematically higher for active years between 2003 - 2012 than for those between 1997 - 2002 . the nobeyama pe rate follows a similar pattern . these results appear to be connected with the weakness of the late - cycle 23 polar fields as suggested by luhmann , and may be explained by the dominant influence of the polar fields on the global coronal field structure . it would be interesting to investigate the heliospheric consequences of these enhanced eruption rates for the interplanetary magnetic fields solar wind behavior , etc . this work can be extended using more sophisticated nonlinear force - free and mhd modeling techniques and vector synoptic magnetograms . such modeling would yield more detailed information on the physical properties of the global coronal field , such as the magnetic helicity and the free magnetic energy , whose spatial and temporal patterns may provide useful insights into the occurrence , size and geo - effectiveness of flares and cmes and the global behavior of the solar dynamo . i thank an anonymous referee , sanjay gosain and giuliana de toma for helpful comments on the manuscript . solis data used here are produced cooperatively by nsf / nso and nasa / lws . nso / kitt peak 512-channel and spmg data used here were produced cooperatively by nsf / noao , nasa / gsfc , and noaa / sel . wilcox solar observatory data used in this study was obtained via the web site http://wso.stanford.edu courtesy of j.t . this paper uses data from the cactus cme catalog , generated and maintained by the sidc at the royal observatory of belgium . the paper also uses cme statistics from the seeds project at george mason university s space weather laboratory , that has been supported by nasa s living with a star program and nasa s applied information systems research program . this paper uses data from the nobeyama radioheliograph at the national astronomical observatory of japan s nobeyama radio observatory . adams , j. 1989 , applied math . and comput . 34 , 113 altrock , r. 2013 , solar phys . , in press , doi 10.1007/s11207 - 012 - 0216 - 1 altschuler , m. , newkirk , g. 1969 , solar phys . 9 , 131 babcock , h.d . 1959 , apj 130 , 364 babcock , h.w . 1961 , apj133 , 572 basu , s. , antia , h.m . 2010 , apj 717 , 488 chapman , s. & bartels , j. 1940 , geomagnetism , oxford university press , london de toma , g. 2011 , solar phys . 274 , 195 derosa , m.l . , brun , a.s . & hoeksema , j.t . 2012 , apj 757 , 96 dikpati , m. , 2011 , apj 733 , 90 gopalswamy , n. , shimojo , m. , lu , w. , yashiro , s. , shibasaki , k. & howard , r.a . 2003 , apj 586 , 562 gopalswamy , n. , yashiro , s. , mkel , p. , michalek , g. , shibasaki , k. , hathaway , d. 2012 , apjl 750 , l42 gosain , s. & pevtsov , a.a . 2012 , solar phys . in press , doi 10.1007/s11207 - 012 - 0135 - 1 hale , g.e . , nicholson , s.b . 1925 , apj 62 , 270 hale , g.e . , ellerman , f. , nicholson , s.b . , joy , a.h . 1919 , apj 49 , 153 hathaway , d.h . , rightmire , l. 2010 , science 327 , 1350 . hoeksema , j.t . 1984 , phd thesis , stanford univ . jiang , j. , cameron , r. , schmitt , d. , schssler , m. 2011 , space sci . , doi 10.1007/s11214 - 011 - 9783-y jiang , c. , feng , x. 2012 , solar phys . 281 , 621 jin , c.l . , harvey , j.w . & pietarila , a. 2013 , solar phys . , submitted jones , h.p . , duvall , t.l . , jr . , harvey , j.w . , mahaffey , c.t . , schwitters , j.d . , simmons , j.e . 1992 , solar phys . 139 , 211 keller , c.u . , harvey , j.w . , giampapa , m.s . 2003 , in : keil , s.l . , avakyan , s.v . ( eds . ) , innovative telescopes and instrumentation for solar astrophysics , proc . spie 4853 , 194 lee , c.o . , luhmann , j.g . , hoeksema , j.t . , sun , x. , arge , c.n . & de pater , i. 2011 , solar phys . 269 , 367 lee , c.o . , luhmann , j.g . , zhao , x.p . , liu , y. , riley , p. , arge , c.n . , russell , c.t . , de pater , i. 2009 , solar phys . 256 , 345 lefvre l. , clette , f. 2011 , a&a 536 , l11 leighton , r.b . 1964 , apj 140 , 1547 leighton , r.b . 1969 , apj 156 , 1 li , jing & ulrich , roger k. 2012 , apj 758 , 115 livingston , w.c . , harvey , j. , slaughter , c. , trumbo , d. 1976 , appl . 15 , 40 luhmann , j.g . , li , y. , liu , y. , jian , l. , russell , c.t . , kilpua , e. , petrie , g.j.d , hoeksema , j.t . , american geophysical union , fall meeting 2011 , abstract # sh31d-04 munro , r.h . , gosling , j.t . , hildner , e. , macqueen , r.m . , poland , a.i . & ross , c.l . 1979 , sol . 61 , 201 nakajima , h. , nishio , m. , & enome , s. et al . 1994 , proc . ieee , 82 , 705 nandy , d. , muoz - jaramillo , a. , martens , p. c. h. 2011 , nature 471 , 80 neugebauer , m. , forsyth , r. , galvin , a. , harvey , k. , hoeksema , j. , lazarus , a. , lepping , r. , linker , j. , miki , z. , steinberg , j. , von steiger , r. , wang , y.m . , wimmer - schweingruber , r. 1998 , jgr 103 , 14587 olmedo , o. , zhang , j. , wechsler , h. , poland , a. & borne , k. 2008 , solar physics 248 , 485 petrie , g.j.d . 2012 , sol . phys . 281 , 577 petrie , g.j.d . & patrikeeva , i. 2009 , apj 699 , 871 riley , p. , linker , j. , miki , z. , lionello , r. , ledvina , s. & luhmann , j. 2006 , apj 653 , 1510 robbrecht , e. , berghmans , d. & van der linden , r.a.m . 2009 , apj 691 , 1222 schatten , k. , wilcox , j. , ness , n. 1969 , solar phys . 6 , 442 scherrer , p.h . , wilcox , j.m . , & svalgaard , l. et al . 1977 , sol . 54 , 353 schrijver c.j . , liu y. 2008 , solar phys . 252 , 19 sheeley , n.r . , jr . 2010 , in cranmer , s.r . , hoeksema , j.t . , kohl , j.l . ( eds . ) , soho-23 : understanding a peculiar solar minimum , asp conf . 428 , 3 svalgaard , l. , duvall , t. l. , and scherrer , p. h. , 1978 , solar phys . 58 , 225 shimojo , m. , yokoyama , t. , asai , a. , nakajima , h. & shibasaki , k. 2006 , pasj 58 , 85 stenflo j.o . & kosovichev a.g . 2012 , apj 745 , 129 thompson , b.j . , gibson , s.e . , schroeder , p.c . , webb , d.f . , arge , c.n . , bisi , m.m . , et al . 2011 , solar phys . 274 , 29 tlatov , a.g . , vasileva , v.v . , pevtsov , a.a . 2010 , apj 717 , 357 tth , g. , van der holst , b. , huang , z. 2011 , apj 732 , 102 ulrich , r.k . 2010 , apj 725 , 658 wang , y .- m . , robbrecht , e. , sheeley , n.r . jr . 2009 , apj 707 , 1372 wang , y .- m . , sheeley , n.r . jr . 1991 , apj 375 , 761 wang , y .- m . , sheeley , n.r nash , a.g . 1991 , apj 383 , 431 wang , y .- m . , sheeley , n.r . 1992 , apj 392 , 310 webb , d.f . , & howard , r.a . 1994 , jgr 99 , 4201 webb , d.f . & hundhausen , a.j . 1987 , sol . , 108 , 383 zhang , hongqi , sakurai , t. , pevtsov , a. , gao , yu , xu , haiqing , sokoloff , d.d . & kuzanyan , k. 2010 , mnras 402 , l30

we introduce a model of fracture which includes the out - of - plane degrees of freedom necessary to describe buckling in a thin - sheet material . the model is a regular square lattice of elastic beams , rigidly connected at the nodes so as to preserve rotational invariance . fracture is initiated by displacement control , applying a uniaxial force couple at the top and bottom rows of the lattice in mode - i type loading . the approach lends itself naturally to the inclusion of disorder and enables a wide variety of fracture behaviours to be studied , ranging from systems with a simple geometrical discontinuity to more complex crack geometries and random cracking . breakdown can be initiated from a pre - cracked sheet or from an intact sheet where the first damage appears at random , and buckling sets in when a displacement vector containing out - of - place components becomes energetically favourable over one which does not . in this paper we only consider center - cracked sheets with no disorder and include some results relevant to the force- and displacement - fields , and the buckling response ratio . rather than carry out a comprehensive study of such systems , the emphasis presently is on the development of the model itself .

cond-mat0505667

understanding how , and when , materials break are important in many engineering applications . this is so for a number of reasons the motivation to study fracture may , for instance , be related to safety issues , such as determining when cracks form in concrete structures , or it may be one of economical gain , as in the case when the runnability of a printing press in the paper industry is considered . few materials , be they natural or manufactured , are perfect , however . hence , the disorder in the micro - structure needs to be accounted for in order to obtain a realistic description . over the past fifteen years methods have emerged within the statistical physics community to successfully tackle just such problems @xcite . these methods are intermediate between the microscopic , or first principles , approach and the mean - field type of approach . in the former case fracture properties are derived from inter - molecular or inter - atomic forces , representing a problem which is both theoretically demanding and heavy on numerical resources . in the latter case disorder can not be included in a satisfactory way . this is a big drawback since the presence of disorder in a material is crucial to the way it fractures . disorder affects the stress field in such a way as to enhance the already existing heterogeneities . this interplay , between a constantly evolving non - uniform stress field and local variations in material properties , can nevertheless be handled in a numerically tractable way using lattice models . the most common lattice models used in engineering applications are finite element methods ( fem ) , the implementation of which is usually based on commercially available computer codes . the lattice models currently used in statistical physics differ somewhat from the fem - approach in that the grid used is regular , i.e. , the same everywhere , rather than one which adjusts the mesh size according to where the stress field is most intense . although fem modeling is certainly more suitable in describing homogeneous materials , the requirement that the stress field should vary slowly over each element makes the approach cumbersome in the presence of heterogeneities . in the stochastic lattice model , however , the nodes are thought of as being connected by objects such as elastic beams or current carrying elements . while in some respects being less sophisticated than fem methods , the interpretation of the algorithm is much more transparent and the approach also has the advantage of allowing disorder to be included quite generally . in the stochastic models , the local equilibrium in force and moment is considered on a mesoscopic scale , i.e. , on a scale much smaller than the external dimension of the lattice but still sufficiently large for the forces to be governed by well known physical laws . in this sense it is also a very good alternative to the far more complicated approach of including disorder on the microscopic level . since only the nearest neighbours on the lattice are included the calculation of the displacement field reduces to the inversion a sparse matrix , enabling reasonably large systems to be handled computationally . one feature which is of a phenomenological nature , however , is the breaking rule the choice here is guided by intuition rather than by the inner workings of the model itself . in other words , breaking does not arise as a natural consequence of the calculations . this is actually an advantage in that the mechanism by which the system ruptures can be tailored to suit different engineering requirements . if we regard thin planar materials , for instance , the energy required to propagate a crack across a given area is usually much lower in mode - iii fracture , i.e. , tearing , than in the pure tensile loading of mode - i fracture . familiar examples of disordered materials which behave this way are textiles and paper . most of the research done so far in stochastic lattice modeling aims to identify the underlying general principles of the fracture process rather than to address traditional problems in fracture mechanics . in this paper the plane beam model @xcite , which has been used previously to study scaling laws in fracture , is extended to include a specific , and practical , aspect of fracture which is very important for thin sheet materials , i.e. , buckling . as is well known , buckling can profoundly influence the residual strength of such materials @xcite . but before devoting our attention to this problem in full , we briefly mention part of the background which has inspired the use of lattice modeling as a tool in statistical physics . in modeling experiments of random media , the feature which by far has received the most attention is the morphology of crack surfaces . many surfaces in nature are found to be self - affine , i.e. , statistically invariant with respect to anisotropic scale transformations . the morphology of such surfaces can be described by simple scaling laws , behaving very much like fractal objects @xcite . these scaling laws provide a theoretical framework whereby much information can be summed up in a few parameters . certain features have been found to share a common basis with other , seemingly unrelated , problems such as deposition and growth processes , or transport properties in random media @xcite . in the case of fracture it has been established that crack surfaces scale as @xmath0 , where @xmath1 is the system size , @xmath2 is the roughness and @xmath3 is the roughness exponent . other scaling laws have been studied , e.g. , in connection with the distribution of stresses , or for the total amount of damage found at various stages in the breakdown process . by far the most popular tool in such studies has been the random fuse model @xcite . in the fuse model , the nodes on the lattice are connected by current - carrying elements , i.e. , fuses . the threshold for the amount of current which may flow through each fuse is chosen from a random distribution . hence , in the breakdown process a fuse is irreversibly removed from the lattice once its threshold is exceeded . a new distribution of currents is then calculated before the next fuse is removed , and so on , until an uninterrupted path can be traced across the system . although it really describes electrical breakdown , the fuse model is often referred to as a scalar model of fracture due to the similarity in form between ohm s law and hooke s law of linear elasticity . results obtained for @xmath3 with the fuse model are found to be different in two and three dimensions , however . the results are @xmath4 in two dimensions @xcite and @xmath5 in three dimensions @xcite . although the former seems to agree with experimental findings , the latter does not . furthermore , the type of forces involved on the meso - scale also seem to make a difference , i.e. , the results obtained with a scalar model differ from those obtained with a vectorial model . specifically , in calculations with the elastic beam model @xmath6 is obtained in two dimensions @xcite . the difference between the results of the two and three dimensional fuse model indicates that the additional degrees of freedom afforded by the ( three - dimensional ) buckling beam model should provide a lower estimate for the roughness exponent in the vectorial problem as well . since the observed value in real materials , i.e. , @xmath7 @xcite , in fact does lie below the two dimensional beam lattice result , it would be interesting to see if the buckling beam lattice reproduces the universal value observed in nature . however , although such fundamental aspects of the fracture process are certainly interesting , the subject of how buckling affects the scaling laws are left for future study . the focus in this paper is instead on the development of a lattice model that realistically includes the buckling behaviour observed in thin sheet materials . the characteristic out - of - plane deflection known as buckling is perhaps most frequently associated with thin plates or beams under compressive loading . presently , however , we concern ourselves with the special case of a thin planar structure under tensile loading . the interaction of buckling with fracture in such circumstances is a well known phenomenon , although it has often been neglected in fracture mechanics analyses due to the extra complications involved . one of the characteristic features of buckling in a thin tension - loaded sheet is that a stable out - of - plane configuration is obtained after buckling has set in . this is in stark contrast with the case of compressive loading , where loss of stability usually signals complete breakdown . practically all previous work considers the effect buckling has on the strength properties of an already cracked plate or a plate with a geometrical discontinuity such as a circular hole or a rectangular cut - out . if the physical parameters of the plate are such that buckling can be expected before the crack begins to grow , the residual strength of the plate will be significantly lower than what would otherwise be expected , based on an analysis which does not take account of buckling . the present study of fracture and buckling will be more general in scope . in other words , we also regard sheets which , in their initial state , have no cracks or other discontinuities . instead , cracks form by a complex process which depends on the evolving distribution of stresses and its interaction with a disordered meso - structure . the onset of buckling in this scenario , and the effect buckling has on the fracture properties , will vary according to the type of disorder used , i.e. , weak or strong . nonetheless , even for weak disorders the final crack which breaks the system will only rarely appear at the exact center of the sheet , and even then the situation will usually be complicated by additional cracks in the vicinity cracks which interact with the main crack so as to alter the distribution of stresses and hence also the exact shape or mode of buckling . the emphasis here , however , is on the development of the model itself . for illustration purposes , a few results are included on uniform systems with a center - crack . in section [ planb ] the plane beam model is briefly reviewed , before the equations describing the out - of - plane behaviour are derived in section [ bbm ] . typical stress and displacement fields are shown in sections [ sdf ] and [ fff ] , respectively , before the initialization of buckling is discussed in section [ vrn ] and a fracture criterion defined in section [ fracrit ] , where results for the buckling response ratio of a center - cracked sheet are included . where a force couple has been applied uniformly on opposite edges . the strain imposed is consistent with mode - i type fracture and corresponds to a displacement @xmath8 in the @xmath9-direction . the enumeration scheme of the neighbouring beams is shown on the right - hand side , where a rotation at node @xmath10 ( center dot ) induces shearing forces and bending moments in the neighbouring beams . [ beamlatt ] ] the beam model may be defined as a regular square lattice of size @xmath11 , where the spacing is one unit length , and each node in the horizontal and vertical in - plane directions is connected to its nearest neighbours by elastic beams . a beam is then fastened to other beams in such a way that , upon subsequent displacement of neighbouring nodes , the angle between beams remains the same as in the original underlying square lattice , see fig . [ beamlatt ] . furthermore , all beams are imagined as having a certain thickness , providing finite shear elasticity . beginning with the simple two dimensional beam model , there are three possible degrees of freedom , i.e. , translations in the horizontal ( @xmath12 ) and vertical ( @xmath13 ) directions , and rotations about the axis perpendicular to the plane ( @xmath14 ) . as shown in fig . [ beamlatt ] , this allows for both bending moments and transverse shearing forces , in addition to the axially tensile , or compressive , forces . for any node @xmath10 , the nearest neighbours @xmath15 are numbered in an anti - clockwise manner , beginning with @xmath16 to the right of @xmath10 . defining @xmath17 , where @xmath18 , the forces on @xmath10 due to @xmath16 are @xmath19 , \label{mi}\\ _ { y}t_{i}^{(1)}\hspace{-2mm}&=&\hspace{-2 mm } \frac{1}{\beta+\frac{\gamma}{12 } } \bigl[\delta y-\frac{1}{2}\bigl(w_{i}+w_{j } \bigr)\bigr ] , \label{si}\\ _ { x}a_{i}^{(1)}\hspace{-2mm}&=&\hspace{-2 mm } \frac{1}{\alpha}\delta x , \label{fi}\end{aligned}\ ] ] for the moment due to angular displacements @xmath14 , shear and transverse force due to displacements @xmath13 , and axial strain due to displacements @xmath12 , respectively . expressions for @xmath20 are analogous . prefactors characteristic of the material and its dimensions are @xmath21 where @xmath22 is young s modulus , @xmath23 and @xmath24 the area of the beam section and its moment of inertia about the centroidal axis , respectively , and @xmath25 the shear modulus @xcite . the conjugate gradient method @xcite is used to obtain the displacement field from @xmath26=\lambda \left[\begin{array}{l } w_{i}\\ x_{i}\\ y_{i}\\ \end{array } \right ] , \label{ma2x}\end{aligned}\ ] ] where @xmath27 are the components of force and moment . the material is assumed to be brittle , i.e. , each beam is linearly elastic up to the breaking threshold . using @xmath28 and @xmath29 for the thresholds in axial strain and bending moment , respectively , a good breaking criterion , inspired from tresca s formula , is @xmath30 where @xmath31 is the largest of the moments at the two beam ends @xmath10 and @xmath15 . the fracture process is initiated by imposing an external vertical displacement which at the top row corresponds to one beam - length , i.e. , @xmath8 , see fig . [ beamlatt ] . the lattice now consists of horizontally undeformed beams and beams which in the vertical direction are stretched lengthwise . the first beam to break is that for which the ratio @xmath32 is largest , this being the vertically oriented beam which has the lowest value of @xmath28 . if all threshold values are the same , the next beam to break will be one of the nearest lateral neighbours since these now carry a larger load than other beams on the lattice . the case of no disorder is thus one in which the crack propagates horizontally from the initial damage , taking the shortest possible path to break the lattice apart . introducing disorder in the breaking thresholds , material strength varies across the lattice and consequently the crack will not necessarily develop from the initial damage point . instead microcracks and voids form wherever the stress concentration most exceeds the local strength , i.e. , wherever eq . ( [ tresca ] ) dictates that the next beam should be broken . towards the end of the breakdown process smaller cracks merge into a macroscopic crack , forming a sinuous path which ultimately traverses the width of the lattice and thus breaks it apart , see , for instance , fig . [ warp ] . in this scenario the quenched disorder on the thresholds and the non - uniform stress distribution combine to determine where the next break will occur . the stress distribution itself also continually changes as the damage spreads . throughout the process , the equilibrium stress field is re - calculated by use of eq . ( [ ma2x ] ) each time a beam is removed . the stress field therefore relaxes at a rate much faster than the process by which the crack grows . hence , the model describes quasi - static fracture . the displacements of a real material , even if its geometry is essentially confined to a plane , will generally occupy three dimensions . for instance , when opposite forces are applied uniformly along the top and bottom edges of a sheet of paper , with the object of straining it to failure , significant displacements will be observed in the direction perpendicular to the sheet , see fig . this becomes especially evident wherever sizable cracks appear . reasons for this behaviour are deviations in the symmetry of the material itself , or its properties , about the plane through which the externally applied forces act . in some cases such deviations may simply correspond to an uneven thickness , or they may be caused by local variations in density , a gradient in the orientation of the micro structure , and so forth . to include this behaviour , the plane beam model must incorporate at least two additional features . one is the random variation of the material in the out - of - plane direction . since lattice modeling reduces the material to a set of points corresponding to the nodes on a mathematically precise two - dimensional lattice , the most convenient approach is to impose a very small randomly chosen vertical displacement on each node . this is discussed in more detail in section [ vrn ] . the other feature to be included , and the topic of the present section , is the physics of the forces which create , and maintain , the out - of - plane displacement field . in the buckling beam model we have one translational and one rotational displacement relevant to each of the principal axes , i.e. , six degrees of freedom , with the matrix system @xmath33=\lambda \left[\begin{array}{l } u_{i}\\ v_{i}\\ w_{i}\\ x_{i}\\ y_{i}\\ z_{i}\\ \end{array } \right ] \label{ma3x}\end{aligned}\ ] ] replacing eq . ( [ ma2x ] ) . presently the forces are projected onto the @xmath34- , @xmath35 and @xmath36-planes , and hence @xmath37 of eq . ( [ ma3x ] ) , that is , @xmath38 can be stated as @xmath39 where the term @xmath40 , for instance , is the @xmath12-component of the buckling ( b ) force due to @xmath16 , as projected onto the @xmath35-plane . axial and transverse contributions are denoted ( a ) and ( t ) , respectively . the rotational displacements about the @xmath9- and @xmath41-axes are denoted @xmath42 and @xmath43 , respectively , and @xmath44 is used for vertical displacements along the @xmath45-axis . a coordinate system is placed on each node , whereupon forces and moments are expressed as functions of the displacements . to this end , an elastic beam with no end restraints @xcite is considered , as in the case of the plane model . in the buckling model , however , the coordinate system is additionally rotated about the relevant angle within the @xmath35- , @xmath36- or @xmath34-plane , i.e. , @xmath42 , @xmath43 or @xmath14 . with the exception of the signs on the terms of eq . ( [ xf ] ) , contributions from neighbours @xmath16 and @xmath46 are similar , as are those from @xmath47 and @xmath48 . consequently , if we define @xmath49,\end{aligned}\ ] ] with @xmath50 , and @xmath51 with @xmath52 , for notational convenience , then the total force on @xmath10 along the @xmath41-axis , with the contributions from all four of the neighbouring beams having been included , reads , at node @xmath10 due to @xmath16 in the case of an axially compressive load , showing the angular displacements , @xmath53 and @xmath54 , the bending angle , @xmath55 , the axial force , @xmath56 , and the component @xmath57 , of @xmath56 , which is parallel to the beam axis at node @xmath10 . also shown is the original @xmath35-system and the x@xmath58z@xmath58-system . [ bucfig ] ] @xmath59s_{j } \nonumber\\ & + & \frac{1}{\beta+\frac{\gamma_{\rm z}}{12 } } \sum_{j=1}^{4 } \left[\bigl(p_{j}\delta x+q_{j}\delta y\bigr ) \sin{w_{i}}+\bigl(-1\bigr)^{j}\bigl(p_{j}\delta y+q_{j } \delta x\bigr)\cos{w_{i}}-s_{j}\bigl ( \frac{\delta w}{2}+\sin{w_{i}}\bigr)\right ] \nonumber\\ & \times & \bigl(p_{j}\sin{w_{i}}+q_{j}\cos{w_{i}}\bigr ) \nonumber\\ & + & \frac{1}{\beta+\frac{\gamma_{\rm y}}{12 } } \sum_{j=1}^{4 } \left[\delta x\sin{u_{i}}+\bigl(-1\bigr)^{j } \delta z\cos{u_{i}}+r_{j}\bigl(\frac{\delta u}{2}+ \sin{u_{i}}\bigr ) \right]p_{j}\sin{u_{i}}. \nonumber\end{aligned}\ ] ] in eq . ( [ xif ] ) , moreover , @xmath60^{-1 } + q_{j}\frac{\delta v}{2}\bigl[sin(\frac{\delta v}{2 } ) \bigr]^{-1 } \label{angc}\end{aligned}\ ] ] is an angular correction to @xmath61^{2 } } \hspace{1mm}\right\ } , \label{fiforce}\end{aligned}\ ] ] the latter being the projection onto the @xmath35-plane of the force along the axis of the beam . if we consider the @xmath16 component in eq . ( [ xf ] ) , @xmath62 is the contribution due to elongation or compression along the axis of the beam , @xmath63\sin{w_{i } } \label{trxy}\end{aligned}\ ] ] is due to forces which are transverse to the axis of the beam , and @xmath64 is the contribution due to buckling . the latter arises when a beam in a bent configuration is under compressive or tensile axial loading , see , e.g , fig . [ bucfig ] where the term @xmath65 has been shown . out - of - plane contributions with components along the @xmath41-axis are @xmath66\sin{u_{i } } \label{trxz}\end{aligned}\ ] ] from transverse forces , and @xmath67 from buckling . the buckling term , as obtained in the lowest order approximation , is essentially the product of a bending angle , @xmath55 , and an axial force component , @xmath57 , the latter being parallel to the axis at the opposite end of the beam . the component of the buckling reaction in the x@xmath58z@xmath58-system which lies along the @xmath41-axis in the @xmath35-system is then @xmath40 , i.e. , eq . ( [ bucxz ] ) . ( [ trxz ] ) , moreover , corresponds to that part of the transverse force ( including shear ) which does not include buckling and is similarly obtained , i.e. , by rotating the axes in eq . ( [ si ] ) . finally , the axial term becomes @xmath68 when the out - of - plane displacements are set to zero . hence , in this case , only when the rotation of the @xmath34-system onto the x@xmath58y@xmath58-system is neglected does eq . ( [ xf ] ) reduce to eq . ( [ fi ] ) , of the plane beam model . although forces and displacements on a beam under simultaneous axial and transverse loading can not , in general , be obtained by superposition , combinations such as @xmath69 in eq . ( [ xf ] ) result when only the leading terms , in @xmath57 , are retained after inverting the expressions of ref . this also causes the buckling term in tensile loading to be the same as that in compressive loading , a change of sign being the only difference . the expression for the @xmath9-component is similar to that of the @xmath41-component , and is obtained by changing around the directions in eq . ( [ xf ] ) , i.e. , @xmath70 the @xmath45-component , furthermore , is @xmath71 i.e. , also similar in form to eq . ( [ xf ] ) but with lines number two and five omitted . the full expressions are then @xmath72r_{j } \nonumber\\ & + & \frac{1}{\beta+\frac{\gamma_{\rm z}}{12 } } \sum_{j=1}^{4 } \left[\bigl(-1\bigr)^{j}\bigl(p_{j}\delta x+q_{j}\delta y\bigr ) \sin{w_{i}}+\bigl(p_{j}\delta y+q_{j } \delta x\bigr)\cos{w_{i}}-r_{j}\bigl ( \frac{\delta w}{2}+\sin{w_{i}}\bigr)\right ] \nonumber\\ & \times & \bigl(p_{j}\cos{w_{i}}+q_{j}\sin{w_{i}}\bigr ) \nonumber\\ & + & \frac{1}{\beta+\frac{\gamma_{\rm x}}{12 } } \sum_{j=1}^{4 } \left[\bigl(-1\bigr)^{j}\delta y\sin{v_{i}}- \delta z\cos{v_{i}}-s_{j}\bigl ( \frac{\delta v}{2}+\sin{v_{i}}\bigr ) \right]q_{j}\sin{v_{i } } \nonumber\end{aligned}\ ] ] for the force in the horizontal direction , and @xmath73p_{j}\cos{u_{i } } \nonumber \\ & -&\frac{1}{\beta+\frac{\gamma_{\rm x}}{12 } } \sum_{j=1}^{4}\left[\bigl(1-r_{j}\delta y\bigr ) \sin{v_{i}}+r_{j}\delta z\cos{v_{i}}+\frac{\delta v}{2 } \right]q_{j}\cos{v_{i } } \nonumber\end{aligned}\ ] ] for the force in the direction perpendicular to the rest plane . considering next the rotational contributions , a beam under axial loading , which is simultaneously bent , is shown in fig . [ moment ] . in this case a buckling term arises which is again the product of a bending angle , @xmath55 , and an axial force component . for rotations about the @xmath45-axis , this gives @xmath74 for each beam in eq . ( [ wf ] ) there are two terms , one analogous to eq . ( [ mi ] ) and denoted ( m ) , and one extra term , such as @xmath75 , which is the contribution due to the beam being simultaneously bent while under axial loading . similarly , we have to the in - plane moment at node @xmath10 from @xmath47 , due to buckling , in the case of a tensile axial load . shown are the angular displacements , @xmath76 and @xmath77 , the bending angle , @xmath55 , the axial force , @xmath56 , and the component @xmath57 , of @xmath56 , which is parallel to the axis of the beam at the opposite end , i.e. , at node @xmath47 . [ moment ] ] @xmath78 for rotations about the @xmath9-axis , and @xmath79 for rotations about the @xmath41-axis , where ( q ) denotes the torque . in eq . ( [ uf ] ) , the torque is simply @xmath80 where , assuming @xmath81 , the material constant is @xmath82 when @xmath14 denotes the width of the beam cross section and @xmath83 its thickness . the buckling term reads @xmath84 and that part of the bending moment which does not involve buckling becomes @xmath85 \label{trmomxz}\end{aligned}\ ] ] when the axes are rotated . ( [ uf ] ) , when written out in full , is now @xmath86\right\}- \sum_{j=1}^{4}\left[f_{i}^{(j)}p_{j } \frac{\delta u}{4}\cos{\delta u}-q_{j}\xi\delta u\right ] \label{uif}\end{aligned}\ ] ] and eq . ( [ vf ] ) is analogous , i.e. , @xmath87\right\}- \sum_{j=1}^{4}\left[f_{i}^{(j)}q_{j } \frac{\delta v}{4}\cos{\delta v}-p_{j}\xi\delta v\right ] . \label{vif}\end{aligned}\ ] ] finally , for rotations within the @xmath34-plane , eq . ( [ wf ] ) becomes @xmath88 \label{wif } \\ & + & \frac{\beta}{\gamma_{\rm z}(\beta+\frac{\gamma_{\rm z}}{12 } ) } \sum_{j=1}^{4}\delta w- \frac{1}{4}\sum_{j=1}^{4}f_{i}^{(j)}\delta w\cos{w_{i}}. \nonumber\end{aligned}\ ] ] and its nearest neighbours @xmath16@xmath89 , shown when the lattice is in an advanced state of buckling . the plane passing through @xmath10 is uniquely defined by any @xmath15 and @xmath90 neighbours of @xmath10 , as shown by the broken lines , and is no longer parallel to the @xmath34-plane . the out - of - plane reaction ( a ) is normal to the x@xmath58y@xmath58-plane and ( b ) is a bending moment about the y@xmath58-axis . [ 4beam ] ] in the six components of eq . ( [ ma3x ] ) , derived above , prefactors characteristic of the beam and its dimensions vary according to the principal axis of bending . hence , in eq . ( [ mate ] ) , we use @xmath91 for bending within the @xmath34-plane , and @xmath92 for bending within the @xmath36- and @xmath35-planes . we have then assumed beams with a rectangular cross - section , as already noted in connection with eq . ( [ eqxi ] ) . this is convenient in the study of how thin sheets behave during fracture , since one may then simply visualize beams with a flat profile , see fig . [ 4beam ] . in the present calculations the chosen width - to - thickness ratio is 10:1 , so that resistance towards bending within the plane is much larger than that which governs out of plane bending . in the following , results are displayed for non - disordered systems with a central crack . to illustrate the nature of the forces , moments and displacements involved , sections of the lattice parallel with the crack are referred to as @xmath93 , 2 , ... , @xmath94 . hence , on the bottom part of the lattice in fig . [ warp ] , the set of nodes @xmath95 to @xmath96 , located on the same row parallel with the @xmath41-axis , is referred to as @xmath93 . with a total of @xmath94 rows @xmath97 parallel with the @xmath9-axis , the `` near '' edge of the crack coincides with @xmath98 while the `` far '' edge coincides with @xmath98 . likewise , the set of nodes @xmath95 to @xmath99 parallel with the @xmath9-axis is referred to as , from left to right , @xmath100 to @xmath101 . the equations governing force and moment in a buckling beam lattice were derived in the previous section . at present we have not taken account of the poisson contraction which is observed in elastic systems at least not at the level of the individual beam . such an effect does show up , however , on length scales spanning several beams . of course , in the macroscopic behaviour of the lattice , an example of this is precisely the buckling behaviour we intend to study . the bulging of the crack edges shown in fig . [ warp ] , for instance , comes about as a result of transverse compressive stresses which develop in the neighbourhood of the crack . [ 3comp ] shows the in - plane displacements @xmath102 and @xmath103 in a lattice of size @xmath104 at various stages of crack advancement . the out - of - plane deflection @xmath105 is also shown on the right - hand side . displacements refer to the initial coordinate system on each node . in the specific example shown the crack grows towards the left - hand side of the lattice , with the out - of - plane deflection increasing with the extent of the crack opening . as previously mentioned , fracture is initialized by displacing the top row a unit distance . in the absence of geometrical discontinuities , each horizontal row @xmath97 is then incrementally displaced by an amount @xmath106 with respect to the previous row @xmath107 . in the absence of cracks , the displacement field @xmath108 then consists of a set of equidistant lines between zero and one . with a crack present , this is altered into the pattern shown in fig . [ 3comp ] , e.g. , for @xmath109 . as expected , transverse displacements @xmath110 are largest close to the face of the crack . if we consider @xmath111 , and move from left to right along the edge of the crack , beams are seen to be stretched wherever the slope is positive and compressed wherever it is negative . as the crack grows the net effect , however , is to cause the lattice to contract in the transverse direction , e.g. , with the edge on the left - hand side moving inward by about 1.5 beam lengths in the case of @xmath112 . the rotation of axes mentioned above is necessary to obtain the correct feedback between the force components in the system , such as mutual consistency between @xmath34-forces and the @xmath45-forces . to illustrate this , regard the lattice before it begins to buckle , i.e. , when the stress field is confined to the @xmath34-plane . when a crack grows beyond a certain critical size , interaction between the stress field and random variations in the @xmath45-direction initiates out - of - plane displacements which ultimately result in a buckled lattice . the driving forces are terms normal to the @xmath34-plane , i.e. , terms such as @xmath65 and @xmath113 , which belong to @xmath114 and @xmath115 , respectively . these terms are not large . in the flat lattice , for instance , the last two terms on the right - hand side of @xmath116 are identically zero while the terms @xmath117 and @xmath118 are very small in comparison with the leading axial term . the terms @xmath65 and @xmath113 , however , although being small in comparison with either @xmath119 or @xmath120 , are non - negligible . this owes to the fact that there is no physical obstruction in the lattice to inhibit displacements in the @xmath45-direction , i.e. , there is no leading @xmath121 term . moreover , as can be seen from fig . [ 4beam ] , when the lattice is in an advanced state of buckling there will be regions where the out - of - plane buckling reaction is inclined with respect to the @xmath45-axis , resulting in contributions of the type @xmath122 . these are smaller than @xmath65 , but are also assumed to be non - negligible . in order to include such terms the axes are rotated to the local deflection of the lattice , whereupon the components along the @xmath41- , @xmath9- and @xmath45-axes are obtained . an example of the effect this has is when a large crack , perpendicular to the force couple , opens up in the center of the lattice , as in fig . [ warp ] or fig . although the initial displacement of the crack edges is normal to the rest plane , an in - plane component appears as the crack grows , i.e. , the near edge of the crack is pulled slightly back along the negative @xmath9-axis while the far edge is pulled forward in the opposite direction . this can be seen clearly in fig . [ 3comp ] , where in @xmath123 the rows nearest to the crack edges are displaced below or above the fixed values of the top and bottom rows . in the case of the near edge this means that displacements are negative , i.e. , they have moved slightly backwards with respect to their equilibrium positions in the unrestrained - strained lattice . these displacements become more pronounced as the crack grows , as can be seen from @xmath124 and @xmath125 . it is particularly instructive to compare the displacements of a buckling lattice with those obtained for the same lattice when the out - of - plane degrees of freedom are suppressed . in fig . [ w4 ] a lattice of size @xmath126 is shown in four stages of crack advancement . the corresponding in - plane displacements of the crack - edges are shown as thick lines in fig . [ p8 ] , with thin lines representing the same lattice in a non - buckling fracture mode . in the latter case , the @xmath103 displacements are seen to be confined between the fixed values of the top and bottom row , the physical structure of the lattice itself effectively acting as an obstruction to displacements outside this range . a further feature that can be noted concerns the aforementioned `` poisson '' contraction . this effect is seen to be present in a non - buckling lattice as well , although to a much lesser degree . as for the angular displacements @xmath76 ( not shown ) , these are seen to be somewhat larger in the buckling case , except for the peak values obtained at the crack tips , which are more or less the same . these peak values increase only as the crack nears the outer boundaries of the lattice , where @xmath102 is large . in general , the expressions derived for force and moment from ref . @xcite are accurate for small displacements only . this assumption we simply extend to all displacements . a second reason for rotating the axes , then , is to conserve the consistency of the approximations used . to illustrate the point one may , for instance , regard a straight beam which , in its rest state , lies along the @xmath41-axis . the term @xmath127 then expresses the transverse force as a function of @xmath128 . when @xmath53 is large , however , and the coordinate axes are fixed , @xmath128 , in addition to the transverse force , also implies some measure of axial strain in the beam even though this effect has already been included via eq . ( [ fiforce ] ) . hence , rotating the axes precludes the introduction of systematic errors , e.g. , due to angular deflections of the lattice . it also improves the quality of the approximations used since angular displacements are rendered less severe in a rotated coordinate system . large angular deflections usually involve a number of nodes on the lattice and hence by rotating the axes we avoid that too many errors accumulate . it should also be pointed out , moreover , that with three rotational degrees of freedom there will in principle be several displacement combinations which correspond to a given space orientation of the beam axis . hence the projection of forces into the @xmath34- @xmath35- and @xmath36-planes is an approximation based on the assumption that large deflections about more than one axis simultaneously are rare . otherwise the exact orientation of the beam would be history - dependent , i.e. , it would depend on the sequence in which the ( final ) angular displacements @xmath53 , @xmath129 and @xmath76 were incremented . the out - of - plane force components are small , but their collective effect has a significant impact on the stress field . in the presence of significant cracks there is a feedback from the @xmath45-displacements which allows the @xmath34-displacements of the buckled lattice to relax with respect to the @xmath34-displacements of the flat lattice . one example of this is the transverse compressive stress stored in the region in front of and behind the crack in the non - buckling lattice . buckling releases this stress , as can be seen in fig . [ twox ] , where the @xmath16 component of eq . ( [ sumx ] ) is shown in the non - buckling ( a ) and the buckling ( b ) cases . in ( a ) a region of compressive stress confined between the crack tips is seen to extend for a distance of about 6 - 8 rows away from the crack edge . in ( b ) only a vestige of this is left , and then only in the immediate vicinity of the crack . the tensile stress at the crack tips increases slightly in the buckled configuration . in the non - buckling beam model , the extra non - linear terms which arise when the beam is simultaneously bent while under axial compression , or tension , are of lesser importance . all forces now act within the structure which defines the plane so that , in calculating the in - plane displacement field , corrections such as those due to @xmath76 in eq . ( [ limxx ] ) may be neglected . the out - of - plane displacement field , on the other hand , is obtained from an equilibrium state in force and moment between a number of terms which are individually small . for instance , the axial , transverse and buckling components which make up the @xmath16 contribution to eq . ( [ zf ] ) are shown in fig . [ compz1 ] . here the vertical scales on the three subplots have been adjusted to the relative sizes of the components . the magnitudes , furthermore , refer to the scale of fig . [ twox ] and thus gives an idea of the `` smallness '' of the out - of - plane force components . in agreement with fig . [ twox ] , the forces in fig . [ compz1 ] are seen to be most significant within a region nearest to the crack edge , extending about 6 - 8 rows to either side of the crack . [ t ] [ b ] at the onset of buckling the axial and transverse terms , @xmath130 and @xmath127 , are identically zero while the buckling term , @xmath65 shown in fig . [ bucfig ] , is non - zero . in this - situation the sum of contributions from @xmath131 is non - zero . as the out - of - plane deflection increases , an equilibrium is approached where the sum of forces is zero . at this equilibrium the buckling terms , e.g. , terms such as @xmath65 in fig . [ compz1 ] , remain non - zero . this is also the case with the out - of - plane moments , shown in fig . the three subplots are not mutually to scale in this case , but the magnitudes again refer to the scale of fig . [ twox ] . out - of - plane moments are seen to be somewhat larger than the axial and transverse buckling forces of fig . [ compz1 ] . hence , at the equilibrium , the most significant of the non - linear terms are those relevant to the momentum , @xmath132 being about five times larger than @xmath65 . it is instructive to see what happens when buckling terms such as @xmath65 and @xmath132 are removed . shown in fig . [ 41cgr ] , at the onset of buckling , is the movement of the crack - edge as a function of time , the time - steps being defined by the iteration procedure which locates the equilibrium of force and moment . just before the point at which equilibrium is reached , all buckling terms are `` switched off '' whereupon the remaining forces set about to locate a new minimum of elastic energy . this new minimum , of course , is none other than the flat configuration . as would be expected , not only do terms such as @xmath65 and @xmath132 cause buckling , they also sustain it once it has been established . finally some remarks on the angular correction in eq ( [ fiforce ] ) , which is included to allow for the possibility that axial force may increase or decrease as a consequence of bending . in eq . ( [ fiforce ] ) it is assumed that the additional elongation due to bending can be obtained from a multiplicative factor . this factor is based on the ratio of a circular arc @xcite to a straight line , the former being the semi - circle defined by the angular difference @xmath133 at the end - points and the latter the line which connects these . on the level of the individual beam , the presence of inflection points are neglected in this approximation . in other words , up - down curvatures may only occur in combinations of two or more beams in an end - to - end alignment . furthermore , as can be seen from eq . ( [ fiforce ] ) , the effect of in - plane bending moments , or transverse displacements perpendicular to the rest axis of the beam , are neglected as contributions which would otherwise add to the axial length of a beam . an important feature to be included in the model is the random variation of the material in the out - of - plane direction , as was remarked in section [ bbm ] . in thin materials such as paper , cloth , membranes and so forth , the most important factor influencing the behaviour during fracture is not the three - dimensional structure of the material itself . rather it is the out - of - plane deflection of this structure which makes a difference . nevertheless , random variations in the thickness direction provide an important part of the mechanism which initiates buckling . this is because such variations combine with the externally applied force and the emerging cracks to create local forces and moments which are not perfectly aligned within the plane . once a buckled configuration has been established , however , the variation in the thickness direction is far less important . since we presently regard the out - of - plane deflection of a structure which has no vertical extent , buckling must be initiated by other means . specifically , in modeling the fracture process , the equilibrium stress field is re - calculated by use of eq . ( [ ma3x ] ) after a beam has been removed . at each step of this process , i.e. , for each beam removed , a sample - specific random noise in the form of a small vertical displacement is imposed on all nodes of the lattice . presently , we use a random number uniformly distributed on the interval @xmath134 $ ] . in the early stages of the fracture process , the stress field is calculated in the presence of these variations until buckling commences . before sizable cracks appear , forces combine to flatten out the vertical displacements . that is , a flat configuration is energetically preferred to begin with , and fracture propagates according to the non - buckling simulation . as significant cracks begin to appear , however , the conditions at some point become favourable for the out - of - plane components of the stress field to be realized , and buckling sets in . from here on the random noise is discarded , and the next displacement configuration is simply calculated from the previous coordinates . when the lattice has been broken , a new set of vertical displacements is generated for the next sample , i.e. , a sample - specific random noise is used . lattice buckling modes in the presence of a center - crack of size @xmath135 are shown in fig . the only disorder present here is that due to the out - of - plane initialization , but evidently a number of buckling modes may appear . in the following , cases where the deflection of the edges of the crack is to the same side , i.e. , up - up or down - down , are referred to as symmetric buckling , and cases where the deflection is to opposite sides is referred to as anti - symmetric buckling . in fig . [ disp ] , ( d ) and ( e ) are examples of the former , and ( f ) and ( h ) are examples of the latter . another buckling mode which frequently appears is that shown in ( a ) , ( g ) or ( j ) , where the main bulge at one of the crack edges is made up of four , rather than three , half - waves . based on the few dozen samples observed , ( a ) and ( j ) evolve into type ( d ) after a few more beams have been broken , and ( g ) into type ( e ) , i.e. , the symmetric buckling mode prevails in each case . however , due to the randomness introduced , examples such as ( h ) are not completely anti - symmetric about the neutral plane . hence , even for this simple crack configuration , the exact shape of the out - of - plane deflection can vary considerably . the overall shape , however , tends to fall within the main categories , i.e. , one of the two symmetric or anti - symmetric buckling modes . this way of initializing the out - of - plane deflection is suitable for studying disordered systems , where a lattice without any initial geometrical discontinuity is strained until random cracks begin to appear . as the cracks grow buckling sets in at some point , depending on the configuration , position and size of the initial cracks . another way of initializing the out - of - plane deflection is to impose a small vertical deflection on a very few nodes in strategic positions . this is most practical when studying an `` ideal '' buckling scenario , such as the fracture of a non - disordered plate with a perfect center - crack . in order to study how buckling affects the fracture properties of a two - dimensional structure an appropriate breaking criterion should be chosen . as previously mentioned , this can be done to suit a range of engineering requirements . often the mode of rupture in the out - of - plane direction is radically different from that which takes place within the plane . in paper or cloth , for instance , the phenomenon which first springs to mind is tearing . the energy required to propagate a crack across a given area in tear mode is much less than that which causes the same area to fracture in pure tensile loading . this is especially the case with paper . out - of - plane contributions to the breaking criterion must be included by some other mechanism than that provided by eq . ( [ tresca ] ) , since the latter is relevant to regions which are comparable in size to a beam . the stress intensification due to buckling , on the other hand , is due to much smaller regions , i.e. , comparable in extent to the sharp crack tip . one way of enhancing the stress due to buckling is to combine torsion with axial stress . the larger the load , the more sensitive the beam will be to the presence of a given amount of torque . compressive loads are assumed to alleviate the torsional moment , but only to a very small degree . hence , the breaking criterion can be stated as @xmath136 where @xmath137 is the effective stress , @xmath138 the torque , and @xmath139 the combined bending moment . with @xmath14 and @xmath83 denoting the width and thickness , respectively , of the beam , @xmath140 is the aspect ratio of the cross section , and @xmath141 is the enhancement factor in eq . ( [ ft ] ) . considering fig . [ 2x50 ] , the breaking stress is increased in case ( b ) and also in case ( a ) provided the deflections in front of and behind the crack are not congruent . when the bulges are completely symmetric , however , ( a ) does not intensify the breaking stress and thus contradicts experimental findings . another possibility is to assume a crack - tip stress enhancement which depends on the out - of - plane bending moment . here the in - plane displacement component @xmath103 observed in fig . [ p8 ] , i.e. , the backward and forward movement of the crack edge , creates an angular displacement about the @xmath41-axis . for a sufficiently thin plate the resistance towards bending will not be sufficient to halt the out - of - plane deflection once it has commenced , since the forces involved act over a region much larger than the immediate neighbourhood of the crack . due to the short distance which separates the top and bottom surfaces , the resulting `` lever - arm '' effect creates an asymmetric stress - gradient across the crack front in the direction of the thickness . whereas tensile force on the concave side is then reduced , it increases on the convex side . this increase comes in addition to the stress already concentrated along the crack front , i.e. , the very presence of a crack creates a screening effect which re - distributes the in - plane stresses so as to cause a build - up in the load at the crack tips . for a crack that has grown to an extent which allows buckling to occur , this in - plane stress is significant . the crack - tip opening angle also plays an important role . buckling in brittle materials , for instance , is known to have a profound effect on the maximum load the system can tolerate before breaking . in a fem study by seshadri and newman @xcite a hypothetical very large critical crack - tip opening angle was used to model buckling in a ductile material . strength reduction in this case was found to be significantly smaller than for brittle materials . in the beam model , the crack tip is never sharper than exactly one beam length . to emulate the above stress enhancement due to out - of - plane bending we instead imagine a sharp crack to be embedded within that beam which on the lattice defines the tip of the crack , and consider a combination of axial stress and moment . out - of - plane bending modes are shown in figs . [ uu ] and [ du ] , where the displacements of the schematic lattice at the top have been exaggerated somewhat to illustrate the point . specifically , the @xmath45-displacements of contour b have been scaled up 100 times with respect to those of contour a , which itself is scaled up with respect to the horizontal extent of the lattice . the in - plane @xmath9-displacements have also been adjusted accordingly . experimental evidence indicates that the stress enhancement at the crack tips is more or less similar in the symmetric and anti - symmetric buckling modes . to incorporate this we distinguish between the two cases . hence , retaining eq . ( [ bcri ] ) , we introduce @xmath142 where @xmath143 replaces @xmath144 in eq . ( [ ft ] ) . in this prescription , @xmath145 denotes symmetric ( @xmath146 ) or anti - symmetric ( @xmath147 ) buckling , respectively , with the signs referring to the direction of the moment at the two beam ends . , is ( a ) the @xmath148 contour , passing through the middle of the lattice , and ( b ) the @xmath149 contour , passing through the left - hand side crack tip . also shown is the bending mode of the beam which defines the crack tip at the junction between @xmath149 and @xmath150 . at the bottom are shown the out - of - plane angular displacements @xmath129 about the @xmath41-axis , in the case of ( b ) above . also shown are the neighbouring contours , ( c ) @xmath151 and ( d ) @xmath152 , where ( d ) is discontinuous due to the intersecting crack . [ uu ] ] for the effective stress in the beam , we now use @xmath153 in the symmetric case , i.e. , when @xmath154 and @xmath155 in the anti - symmetric case , i.e. , when @xmath156 the enhancement factor in eqs . ( [ symmfc ] ) and ( [ antifc ] ) is @xmath157 where @xmath158 is a discontinuity operator . the choice made above causes the breaking stress of the beams at the crack tips to increase by a comparable amount in symmetric and anti - symmetric buckling . the expressions for @xmath159 and @xmath160 in eqs . ( [ tsens ] ) and ( [ bsens ] ) , respectively , have been chosen , very generally , to incorporate some overall effects related to size , material and relative dimensions . hence , it is reasonable to assume that , for a given size , `` tearability '' , or the `` lever - arm '' effect , increases with decreasing sheet , or plate , thickness . , is ( a ) the @xmath148 contour , passing through the middle of the lattice , and ( b ) the @xmath149 contour , passing through the left - hand side crack tip . also shown schematically is the the beam which defines the crack tip at the junction between @xmath149 and @xmath150 . although the bending mode is correct , the actual angles at the ends of the beam are negative and not positive as shown . out - of - plane angular displacements @xmath129 about the @xmath41-axis are included at the bottom , with the notation being the same as in fig . [ uu ] . [ du ] ] in conjunction with this , the crack - tip opening angle , which decreases with increasing resistance towards in - plane bending , also enters the picture . both effects are presently included via the ratio of the in - plane to the out - of - plane inertial moment for bending , i.e. , @xmath161 . the length of the arm with which the out - of - plane forces act is assumed to be proportional to the vertical extent of the buckling zone . since this , in turn , is proportional to system size , a factor @xmath162 is also included . as noted previously , fracture is initiated by displacing the top row of the lattice a fixed distance , usually corresponding to one beam length . to avoid scale effects associated with this , a further factor @xmath163 is included , where @xmath163 is the size of the reference system for which the top row displacement is exactly one beam length . the introduction of a reference system allows for the possibility of comparing systems of varying size where the physical behaviour involved requires the same relative external boundary conditions . for instance , referring to the intact lattice , mode - i loading then imposes the same initial strain of @xmath164 on each beam . although computational time increases when @xmath165 , features such as how various buckling modes appear with respect to system size will depend on the external loading . otherwise , @xmath166 might probably be used in cases where we are interested in features which depend on the internal processes of the fracture mechanism , such as the roughness exponent of crack interfaces . to guard against unphysical breaks , we introduce @xmath167 which contributes when one or both nearest lateral neighbours are intact , and @xmath168 which contributes when a certain number of neighbours have been broken . for any node @xmath10 , the array @xmath169 now keeps track of the status of the beam which extends away from @xmath10 in the direction of the @xmath9-axis , i.e. , it remembers whether this is broken or intact , respectively . the combined expression , @xmath170 has the property @xmath171 as has @xmath172 governing cracks in the normal direction . in other words , eq . ( [ disco ] ) ensures that the stress enhancement mechanism is activated only in cases where the lateral neighbour on one side is intact while simultaneously a certain number of beams , defining a minimum crack length @xmath173 , are broken on the other side . in most cases the operator @xmath174 is not necessary . it has been included to avoid cracking being induced near the top and bottom rows of the lattice . for very large systems , and especially in cases where @xmath1 is significantly larger than @xmath163 , breaks sometimes occur due to the large angular gradients in beams extending up from @xmath93 or down from @xmath175 , see , for instance , figs . [ uu ] and [ du ] . in the present formalism the properties of beams with inflection points are not considered , the smallest crack that can cause buckling , i.e. , a bulge consisting of at least three half - waves , is therefore approximately @xmath176 . this is confirmed in numerical runs for systems with small @xmath1 , but where @xmath177 . a problem with using such a large value of @xmath173 is that it excludes cracks inclined at an angle with respect to the horizontal . over a wide range of system parameters and external boundary conditions , however , @xmath178 was found to be adequate . [ t ] in the limit of no buckling , i.e. , @xmath179 or @xmath180 , eq . ( [ bcri ] ) reduces to eq . ( [ tresca ] ) . in other words , if buckling is not activated , then neither is the stress enhancement mechanism in the fracture criterion . finally , in order to illustrate how eq . ( [ bcri ] ) , with eqs . ( [ momdef ] ) to ( [ lresu ] ) defining the stress enhancement mechanism , works within our model of buckling , we consider the buckling response ratio of the residual strength of the system . that is , @xmath181 , where @xmath182 and @xmath183 represent the maximum applied external force a restrained or buckled plate , respectively , can tolerate before breaking apart . early experimental results show that the decrease in strength due to buckling increases as the ratio of the crack - length @xmath184 to the thickness @xmath83 is increased . a linear relationship was proposed by kuhn and figge @xcite which , in the case of brittle materials , has been shown to agree well with more recent fem calculations @xcite . in fig . [ kuhn ] , results obtained with the beam model are compared with the kuhn - figge relationship . a small correction to the size of the central crack has been made to account for the finite size of the beams . the effect is very small , shifting the values of the smallest systems slightly to the left , thus improving the agreement with the kuhn - figge relationship from very good to excellent . to summarize , we have included the additional degrees of freedom necessary to describe the interaction of cracks with buckling in the elastic beam model . this model is stochastic in nature , so that sheets with random cracking at any level of meso - structural disorder can be studied , including systems with no disorder . in addition to important issues of practical relevance in traditional fracture mechanics , such as strength properties and stability , the present model also enables fundamental aspects of fracture in random media to be explored . 99 see , e.g. , h. j. herrmann and s. roux , _ statistical models for the fracture of disordered media _ , ( north - holland , amsterdam , 1990 ) . s. roux and e. guyon , j. phys . ( paris ) lett . * 46 * , l999 ( 1985 ) . r. g. forman , `` experimental program to determine effect of crack buckling and specimen dimensions on fracture toughness of thin sheet materials '' , technical report affdl - tr-65 - 146 , air force flight dynamics laboratory , dayton , ohio ( 1966 ) . j. r. dixon and j. s. strannigan , `` stress distributions and buckling in thin sheets with central slits '' , in _ proc . second int . conf . on fracture 105 - 118 , edited by p. l. pratt _ et al . _ , ( chapman and hall , 1969 ) . g. f. zielsdorff and r. l. carlson , eng . mech . * 4 * , 939 ( 1972 ) . b. b. mandelbrot , d. e. passoja and a. j. paullay , nature * 308 * , 721 ( 1984 ) . a. l. barabasi and h. e. stanley , _ fractal concept in surface growth _ , ( cambridge university press , cambridge , england , 1995 ) . l. de arcangelis , s. redner and h. j. herrmann , j. phys . ( paris ) lett . * 46 * , l585 ( 1985 ) . a. hansen , e. l. hinrichsen and s. roux , phys . lett . * 66 * , 2476 ( 1991 ) . g. g. batrouni and a. hansen , phys . lett . * 80 * , 325 ( 1998 ) . b. skjetne , t. helle and a. hansen , phys . lett . * 87 * , 125503 ( 2001 ) . e. bouchaud , j. phys . condens . matter * 9 * , 4319 ( 1997 ) . h. j. herrmann , a. hansen and s. roux , phys . rev . b * 39 * , 637 ( 1989 ) . m. r. hestenes and e. stiefel , nat . j. res . * 49 * , 409 ( 1952 ) . r. j. roark and w. c. young , _ formulas for stress and strain _ , ( mcgraw - hill book company , new york , 1975 ) . for the parameters chosen in our model the difference in angular displacement at the beam ends is always small . in the buckling behaviour of fabrics , however , the local curvature can be very large . with an exact expression it has been shown that , even in the case of very large bending angles , the equilibrium shape remains close to that of a circular arc . see , e.g. , k. j. choi and h. s. ko , acm t. graphic . * 21 * , 604 ( 2002 ) . b. r. seshadri and j. c. newman jr . , `` analyses of buckling and stable tearing in thin - sheet materials '' , technical memorandum nasa - tm-1998 - 208428 , langley research center , hampton , virginia ( 1998 ) . p. kuhn and i. e. figge , `` unified notch - strength analysis for wrought aluminum '' , technical note nasa - tn - d-1259 , langley research center , hampton , virginia ( 1962 ) .

transverse stratification is a common intrinsic feature of astrophysical jets . there is growing evidence that jets in radio galaxies consist of a fast low density outflow at the jet axis , surrounded by a slower , denser , extended jet . the inner and outer jet components then have a different origin and launching mechanism , making their effective inertia , magnetization , associated energy flux and angular momentum content different as well . their interface will develop differential rotation , where disruptions may occur . we here investigate the stability of rotating , two - component relativistic outflows typical for jets in radio galaxies . for this purpose , we parametrically explore the long term evolution of a transverse cross - section of radially stratified jets numerically , extending our previous study where a single , purely hydrodynamic evolution was considered . we include cases with poloidally magnetized jet components , covering hydro and magnetohydrodynamic models . with grid - adaptive relativistic magnetohydrodynamic simulations , augmented with approximate linear stability analysis , we revisit the interaction between the two jet components . we study the influence of dynamically important poloidal magnetic fields , with varying contributions of the inner component jet to the total kinetic energy flux of the jet , on their non - linear azimuthal stability . we demonstrate that two - component jets with high kinetic energy flux , and an inner jet effective inertia which is higher than the outer jet effective inertia are subject to the development of a relativistically enhanced , rotation - induced rayleigh - taylor type instability . this instability plays a major role in decelerating the inner jet and the overall jet decollimation . this novel deceleration scenario can partly explain the radio source dichotomy , relating it directly to the efficiency of the central engine in launching the inner jet component . the frii / fri transition could then occur when the relative kinetic energy flux of the inner to the outer jet grows beyond a certain treshold .

0910.0332

there is strong observational and theoretical evidence that magnetic fields play a crucial role in the acceleration and the collimation of extragalactic jets . most active galactic nuclei ( agn ) jet formation scenarios involve magnetic fields threading a rotating black hole ( in the ergosphere ) and its accretion disk , thereby removing from them angular momentum , allowing the central black hole to accrete . at least close to the jet launching region , the jet rotation profile persists , reflecting its ( general relativistic and/or magneto - rotational ) origin . thus both ingredients , magnetic fields and rotation , are very important in jet formation , as well as in jet propagation and stability . moreover , detailed astrophysical jet observations point out that relativistic jets are structured , in a direction perpendicular to the jet axis , typically consisting of a fast spine and slower outer flow . in the case of agns , this jet structuring plays an important role in explaining the morphology of the jet high energy radiation @xcite , with sometimes clear evidence for a very fast , light inner jet and a heavy slow outer outflow @xcite . furthermore , observations of the tev bl lacertae objects show brightenings and rapid variability in their tev emission . this variation in their high energy emission implies high lorentz factor flows occuring at smaller scale , suggesting ultra - relativistic bulk motion of the ( inner ) jet . at the same time , complementary ( radio ) observations with vlbi of the pc - scale jet structure indicate a broad , ` slowly ' ( albeit relativistic ) moving outflow . in combination , this clearly suggests the presence of a two component jet morphology @xcite . two - component jet structure has also been proposed in more theoretical work , addressing the physics of jet launching , collimation and propagation mechanisms @xcite . while our jet dynamics computations will be representative for agn jet conditions , radially structured jet flows are now known to exist in virtually all astrophysical jet outflows . transversely structured , ultra - relativistic jet - like outflow has been proposed in the context of gamma ray bursts @xcite , to explain the break observed in their afterglow light curve . in the case of stellar outflows , recent observations of some t tauri jets @xcite also suggest a fast inner outflow bounded by a slow outer outflow . in these young stellar objects , a clear signature of jet rotation around the symmetry axis was detected @xcite , fully supporting scenarios of magneto - centrifugal jet launch and acceleration . theoretical models of two - component jets in classical t tauri @xcite then postulate that the inner outflow is turbulent and pressure driven , associated with the young star wind . the inner jet then has a small opening angle , as it is collimated by the outer jet , which is in turn magneto - centrifugally driven from the surrounding disk . the outer disk wind then carries most of the mass loss in the jet . various authors @xcite have demonstrated using axisymmetric mhd simulations that the outer outflow is self collimated by its intrinsic magnetic field , and that the turbulent inner outflow gets collimated by the outer jet . furthermore , @xcite investigated the topological stability of two - component outflows for young stellar objects , performing extensive numerical simulations to determine whether analytic self - similar models demonstrate robustness in axisymmetric conditions . also for relativistic jet simulations , axisymmetry assumptions are often adopted , excluding the development of all non - axisymmetric perturbations . these can address details of how helical field configurations ( naturally expected from magneto - centrifugal launch mechanisms ) effectively may transport their helicity down the jet beam @xcite , with magnetically aided reacceleration by field compression across internal cross - shocks . while @xcite concentrated on kinetic energy dominated jets , initially poynting flux dominated jets were simulated by @xcite in axisymmetric relativistic mhd , finding that the transition to a matter - dominated jet regime occurs very close to the central engine ( within @xmath0 ) . our model computations will therefore assume kinetic energy dominated jets . as far as the magnetic field topology is concerned , we will restrict ourselves in this paper to purely poloidally magnetized jet components . our two - component jet model determining our initial conditions can actually allow for helical fields , as explained in sect . [ s - model ] ( we include this more general case here , for future reference in follow - up studies ) . as indicated before , during the first acceleration phase of agn jets , magneto - centrifugal mechanisms play an important role and a helical or even strongly toroidal magnetic field is likely produced . present 3d general relativistic mhd simulations for rapidly rotating black holes producing jets with strong toroidal fields . they find a prominent role of the accreted magnetic field geometry for achieving ` stable ' jets . helical or strongly toroidal field topologies can be subject to current - driven kink instabilities @xcite , with @xmath1 toroidal modes that helically displace the jet axis . this requires full 3d numerical simulations , such as performed by @xcite in non - relativistic mhd , or addressed by @xcite in relativistic mhd for a static force - free equilibrium . dispersion relations for non - axisymmetric modes and @xmath1 kinks in particular for relativistic mhd were analysed by @xcite for purely toroidal fields , and electromagnetically dominated , force - free jets were analysed spectrally by and more recently by @xcite . in our work , we will restrict attention to 2.5d scenarios with the somewhat unusual assumption of translational invariance along the jet axis . the overall configuration is schematically indicated in fig . [ fig1 ] , and we simulate a transverse cross - section of the jet at sufficient distance from the two - component jet source , where all three velocity components ( axial , azimuthal and radial ) are included , but their variation along the jet axis is ignored . our aim is to investigate all non - axisymmetric instabilities , primarily induced by the ( sheared ) rotation . this approximation is valid because in the poloidal direction , the flow is supersonic with high lorentz factor , and then the growth rate of poloidal instabilities is expected to be low . on the other hand , the rotation is subsonic , facilitating the growth of toroidal instabilities . we then address jet stability in a cross section of a rotating two - component jet , initially collimated by thermal pressure and/or poloidal magnetic field . consecutive snapshots of the cross - sectional evolution can be interpreted as mimicking the jet flow conditions at increasing distance from the source . note that this particular assumption allows us to follow both axisymmetric and all non - axisymmetric ( including @xmath1 ) mode development , but does exclude helical mode axis displacement typical for kink modes . our model therefore mimics jet evolution adequately as long as the radial axis displacement is smaller than the axial wavelength associated with possible @xmath1 kinks . our assumption does also neglect conical jet expansion , assuming cylindrical propagation . this is justified given the low observed values for jet opening angles . since we defer the study of toroidal and helical magnetic field configurations in 2.5d and 3d to later work , we start off with numerically investigating the influence of a purely poloidal magnetic field on the stability of rotating , two - component relativistic jets . our work complements the studies looking into kink development , by putting the emphasis on the azimuthal variation and on the effect of the two - component jet stratification . as far as a purely poloidal magnetic field topology is concerned , the work by @xcite suggests that jets should be collimated by poloidal magnetic field pressure , rather than by toroidal magnetic field , as toroidal jet magnetic fields can introduce kink instability ( but only slow mode growth was found for force - free jets by @xcite ) . to justify purely toroidal fields , we can argue that the toroidal magnetic field in the jet acceleration phase can be gradually dissipated involving reconnection , a mechanism which in turn contributes to ( axial ) jet acceleration @xcite . indeed , during the acceleration phase a fraction of the poynting flux ( angular momentum ) carried by the magnetic field is converted to kinetic energy by internal dissipation / reconnection of the toroidal magnetic field @xcite . in accord with this mechanism , we model the region where the eventual jet rotation is low . in this paper , we analyse five cases in detail , to determine the effects of differing poloidal magnetic field configurations in the two - component structure on its long - term stability , and on the overall deceleration of the jet as it propagates away from the central source regions . the role of poloidal magnetic field is in these cases most prominent in its added effect on total pressure and effective fluid inertia , and this is shown to play a prominent role in the two - component jet stability . we set up a relativistic magnetohydrodynamic model of a two - component jet , which elaborates on the earlier model described in . the jet model uses as its basic characterising parameter a total jet kinetic luminosity flux typical for powerful radio loud galaxies , namely @xmath2 . secondly , the outer radius of the two - component jet is taken to be @xmath3 . this is a value directly inferred from observed values for m87 @xcite , which has a known jet opening angle of @xmath4 at @xmath5 from the jet source . for the less constrained inner jet radius , we adopt the initial value @xmath6 . the remainder of the @xmath7 condition is then characterized by the velocity profiles in both inner and outer jet components and by the magnetic configuration in the two - component jet . for the initial flow condition , we adopt a uniform outflow velocity @xmath8 along the jet axis in each jet component , with @xmath9 ( hence @xmath10 , since speeds are normalised to the light speed ) for the outer , slow jet located within @xmath11 . this represents a typical outflow value for relativistic jets at parsec scale from the source . the inner jet has a much faster outflow speed , which we set to @xmath12 , with corresponding lorentz factor @xmath13 . the difference in lorentz factor by an order of magnitude in the inner versus outer jet layers is representative ( though somewhat exaggerated ) for differences inferred for the bl lac object markarian 501 between its central spine and surrounding ( shear ) layer jet , as described by @xcite . however , it remains lower than the lorentz factor @xmath14 suggested for tev blazar pks 2155 - 304 . two - component jet models also commonly assume that the spin of the inner beam is higher than the spin of the outer jet . the physical argument is that the inner beam is extracting angular momentum from the inner parts of the accretion disk and from the black hole itself , both implying fast rotation . therefore , we adopt different initial rotation profiles for inner versus outer jet , both radially self - similar . although we will in our numerical results below only adopt purely poloidal magnetic field configurations , i.e. we will have @xmath15 , one can generalize our initial equilibrium configuration to allow for both toroidal speeds and toroidal magnetic fields , sharing the same @xmath16-dependence . in that general case , one takes in practice @xmath17 @xmath18 where the @xmath19 is the 3-velocity , and similarly for the magnetic field . at @xmath7 , we then have a discontinuity in the toroidal velocity at the boundary between the two jets when fixing @xmath20 and @xmath21 . a faster rotating inner component is consistent with a small expansion of the inner streamlines , which are confined by the outer outflow . the fast rotating inner jet effectively extracts the angular momentum from the central region and carries it away with a very small mass flux . in fact , our models typically have the inner jet extracting order @xmath22 of the total angular momentum associated with the two - component jet . we fix the exponent @xmath23 for the inner jet and @xmath24 for the outer jet . we then choose for the outer jet a rotation profile implying a radially constant angular momentum . this is motivated by the fact that this component is believed to be launched from an accretion disk , where the outer streamlines in the jet expand faster than the inner streamlines . the rotation profile of the inner jet then obeys the relativistic equivalent of the rayleigh criterion for stability , namely that the angular momentum flux given by @xmath25 increases with @xmath16 @xcite . this expression uses proper density and pressure @xmath26 , @xmath27 , assumes a polytropic index @xmath28 appearing in a simple polytropic equation of state ( our actual numerical simulations will in fact relax this assumption and use a full synge - type equation of state ) , and writes @xmath29 , @xmath30 for poloidal three - velocity and magnetic field strengths . for the chosen radial prescriptions for @xmath31 given by eq . ( [ q - vphi ] ) , we have for the inner jet @xmath32 making the inner jet centrifugally stable . the outer outflow on the other hand is marginally stable , following the same argument . in the numerical simulations further in this paper , we neglect toroidal field @xmath33 , and then the interface between the two components does not verify the rayleigh criterion and this shear flow interface is unstable from the start , since the angular momentum at the interface decreases when @xmath34 this means that in all cases we will simulate , the initial shear flow interface at @xmath35 will develop small - scale instabilities , and we intend to address the ultimate non - linear stability of such initial two - component structure . the central question investigated then is whether the dynamics will lead to complete destruction of the initial two - component nature of the outflow at further distances from the source . related to this question , we will quantify the potential deceleration of the central fast jet by mixing processes , induced by the nonlinear evolution of the two - component jet . the densities and the poloidal magnetic field @xmath36 are assumed to be constant within the three different regions , namely throughout inner jet beam , outer jet , and external region . the external medium is hot and rarified and has a small reference number density . the typical ism number density value of @xmath37 is in the computation used as a scaling value . we arbitrarily set @xmath38 with @xmath39 , and adopt @xmath40 and a unit of length of @xmath41 . this sets the unit of pressure , and a corresponding scaling value for the magnetic field is then @xmath42 ( as mass is normalized to proton mass @xmath43 ) . the jet environment represents a relativistically hot , rarified , static external medium . this is consistent with the fact that at the parsec scale , the jet head has previously shocked the jet surroundings , so that the jet itself gets embedded in a dilute , hot medium . assuming a number density for this external medium of @xmath39 is appropriate , since this is near values obtained in numerical simulations of jet propagation and are then typical for the jet cocoon surroundings @xcite . values for the jet component proper densities are estimated from the kinetic luminosity flux and its division over inner and outer jet components . requiring that the outer jet carries a case - specific fraction of the total kinetic luminosity flux by @xmath44 we can deduce the density @xmath45 , if we in addition prescribe the ( outer ) jet lorentz factor and make the initial approximation that the thermal energy in the outer jet is negligible compared to its mass energy ( as valid in all cases studied ) . in eq . ( [ kinjet ] ) , @xmath46 is the specific enthalpy in the outer jet , and its expression depends on the equation of state . when we consider that the inner component carries only a small fraction @xmath47 of the total kinetic luminosity flux , such that the outer jet carries the remaining @xmath48 , we find that the resulting density is @xmath49 . a similar argument for the inner , light jet , works out to fix @xmath50 . while in all cases we investigate here numerically , we assume these initial , piecewise constant , densities for the inner and outer jet , the actual kinetic luminosity carried by each component will change from one case to another . this relates to our prescription of the pressure and magnetic variation in the jet components , and we will simulate five cases with varying thermal pressure contribution at the jet axis . in four cases ( a ) , ( b2 ) , ( c ) and ( d ) , the total pressure at the jet axis ends up similar , while case ( b1 ) has a lower total pressure value . by varying the relative contribution of thermal to magnetic pressure in the jet components , we aim to analyze the effect on overall jet stability of varying magnetic field and kinetic energy contributions to the kinetic luminosity in each component . in all five cases studied , we set the pressure of the external medium to ensure total pressure balance with the outer jet . also , in the four cases ( a ) , ( b1 ) , ( c ) , ( d ) , we assume that the thermal energy in the inner ( hot ) component is higher than its mass energy . only in the case ( b2 ) , the thermal energy is lower than the mass energy . they differ in the following sense : in the first case ( a ) , both components are non - magnetized and the pressure at the jet axis is @xmath51 in our simulation units . this purely hydrodynamic case differs from the hydro case already studied in , as now the inner jet has an even higher lorentz factor , a slightly different inner rotation profile , and an on axis pressure value which makes the entire inner jet relativistically hot . in the cases ( b1 ) and ( b2 ) , only the inner jet is magnetized and we set it to a constant value @xmath52 ( making the jet parameter @xmath53 expressing magnetic to rest - mass energy of order @xmath54 , as for kinetic energy dominated jets ) in the case ( b1 ) and @xmath55 in the case ( b2 ) . in the case ( b1 ) , the thermal pressure on axis is set to @xmath56 and in the case ( b2 ) the thermal pressure on axis is @xmath57 . in the case ( c ) , only the outer jet is magnetized with @xmath58 , and we adopt the same order on axis inner pressure as in ( a ) , namely @xmath59 . in case ( d ) , both inner and outer jet are magnetized , with @xmath60 and @xmath61 . we then take the pressure at the jet axis @xmath62 . in all magnetized cases , the magnetic field strenght is set according to the observations at parsec - scale . clearly , in all four cases ( a ) , ( b1 ) , ( c ) and ( d ) , the thermal energy in the inner component dominates , only in the case ( b2 ) the thermal energy is lower than the mass energy . the pressure profile through inner and outer jet is taken from transverse equilibrium conditions , discussed next . since we assume an initial near steady - state axisymmetric two component jet , the actual pressure variation is deduced from transverse equilibrium between pressure gradient , centrifugal force , and lorentz force . this is expressed by the steady - state equation @xmath63 here @xmath26 is the proper density of particles , @xmath27 and @xmath64 are the pressure and enthalpy per particle , respectively , @xmath65 is the fluid three - velocity , @xmath66 the lorentz factor , and ( @xmath67 ) denote the electromagnetic fields . we have @xmath68 and for steady - state conditions @xmath69 as the associated charge and current densities , while @xmath70 in ideal relativistic mhd . if we introduce the total pressure in the fluid frame , @xmath71 the radial component of eq . ( [ eqeuler ] ) writes as @xmath72 this has assumed a polytropic equation of state . we can then determine the radial pressure profile @xmath73 by solving the first order differential equation ( [ difpseudopressure ] ) and using ( [ pseudopressure ] ) , together with the self - similar prescriptions from eq . ( [ q - vphi])-([q - bphi ] ) to get @xmath74 in this expression , the constant @xmath75 is a self - similarity exponent , which we use to set the inner jet exponents @xmath76 and the outer @xmath77 , and the constant @xmath78 , which is different for inner versus outer jet @xmath79 , @xmath80 . also , @xmath81 is an integration constant , which is to be deduced from boundary conditions . for the inner jet , we use it to set the pressure on axis @xmath82 from @xmath83 to fix the constants for the outer jet , we have to match conditions at the interface between the two components . at the contact between the two components , rankine - hugoniot conditions should hold . in terms of effective total pressure ( [ pseudopressure ] ) obeying eq . ( [ difpseudopressure ] ) , we need to match at the interface @xmath84 this yields for the constant pressure integration in eq . ( [ pressure ] ) for the outer jet @xmath85 \left(1-\alpha_{\rm out}\right)^{\frac{\gamma_{\rm out}}{a_{\rm out}\left(\gamma_{\rm out}-1\right)}}\nonumber\\ & + & \left[-b_{\varphi , \rm out}^2+\left(b_{\varphi , \rm out } v_{z , \rm out } - v_{\varphi , \rm out } b_{z , \rm out}\right)^2\right]{\left(1-\alpha_{\rm out}\right)^{\frac{\gamma_{\rm out}}{a_{\rm out}\left(\gamma_{\rm out}-1\right)}}}\nonumber\\ & \times&\left(\frac{1}{2 } + \frac{\left(1-\alpha_{\rm out}\right)\left(\gamma_{\rm out}-1\right)\left(a_{\rm out}+2\right)}{2\,\alpha_{\rm out}\left[a_{\rm out}\left(\gamma_{\rm out}-1\right)+\gamma_{\rm out}\right]}\right)\,.\end{aligned}\ ] ] this fixes the pressure variation throughout both components . in the external medium , the thermal pressure is constant and set equal to the total pressure in the outer jet component . this is actually a slightly different value for cases ( a ) , ( b1 ) and ( c ) , ( d ) ; albeit of the same order . the governing equation of state is taken as a synge - type relation , also used in @xcite . as a result of the above profile prescriptions , the outer jet is relativistically cold with an effective polytropic index @xmath86 and the inner jet has ultra - relativistic state @xmath87 . case ( b2 ) is an mhd variant to the hydro case with more stratified inner effective polytropic index already simulated in . the matter state between the two components is different ( relativistically hot inner and cold outer jet ) and we will see that during the time evolution , a shear region with intermediate matter state forms . this variation in the matter state between various jet regions makes it vital to use a synge - eos to model growing instabilities and resulting turbulence . however , in the initial conditions of all cases , the polytropic index is relatively constant throughout each component . this is why we could use the polytropic eos assumption to deduce a _ near_-equilibrium solution of the initial two - component jet . as a result of the initializations described above , the local fast magnetosonic speed in the inner jet for case ( d ) and the local sound speed in the inner jet for cases ( a ) and ( c ) are of the order of @xmath88 ( light speed units ) , while cases ( b1 ) and ( b2 ) have a higher local fast magnetosonic speed . in the outer jet , the local fast magnetosonic speed of cases ( c ) , ( d ) and the local sound speed in case ( a ) and ( b2 ) are of the order of @xmath89 , while it is @xmath90 in the case ( b1 ) . both inner and outer jets are kinetically dominated . the inner jet has an effective relativistic mach number @xmath91 up to @xmath92 for cases ( a ) , ( c ) , ( d ) , while the outer jet has @xmath93 in cases ( a ) , ( b2 ) , ( c ) , ( d ) and @xmath94 in case ( b1 ) . however , both components are subsonically rotating , and during a complete rotation of the inner jet , the fast magnetosonic or sound wave will propagate about @xmath95 times from the edge of the inner jet to the axis and back . such overall configuration can easily develop non - axisymmetric instabilities , with growth times of order of the radial sound - crossing time @xcite . the computational domain of this simulation is a 2d box of size @xmath96 and @xmath97 . the simulation is performed in cartesian coordinates using an hllc flux formula . hllc is a two state extension of the harten , lax and van leer flux formulation ( hll ) , which includes a proper representation for the contact wave . we use a piecewise parabolic method ( ppm ) limiter @xcite . the combination of ppm reconstruction ( third order accurate ) and hllc flux computations is extremely robust and handles both sharp discontinuities and turbulence development accurately . the lateral boundaries assume open boundary conditions , with a clipping of any inwardly directed momentum as soon as turbulent flow features start crossing the boundaries . the simulation is run till time @xmath98 , which due to our normalization translates to 163.2 year . the corresponding distance of jet propagation of the jet beam during this time is about @xmath99 . the simulation is done using the amrvac code @xcite with synge - type equation of state . we take a base resolution of @xmath100 , allow for 5 grid levels , reaching an effective resolution of @xmath101 . shorter timescale runs at even higher resolutions were done to confirm that the dominant initial large scale structure development is adequately resolved , although more fine scale features inevitably turn up . the simulations are performed typically using @xmath102 processors for about 2 days per case . we add some white noise both at the initial time and at time @xmath103 when various waves have already developed ( this latter addition may not be essential to the evolution ) . the fact that we use cartesian coordinates could give preference to instabilities with mode number proportional to @xmath104 character ( the case studied in was therefore confirmed separately in cylindrical coordinates , with the overall mode number dominance of @xmath105 recovered ) . however , in all cases we study here , the instabilities dominating the dynamics and evolution of the jet have a clear mixture of many mode numbers . in the actual simulations performed in this work , the magnetic field in the jet is taken purely poloidal . as explained before , our five simulations differ in their magnetic configuration : ( a ) is hydrodynamic ; ( b1 ) and ( b2 ) has only the inner jet magnetized ; ( c ) has only the outer jet magnetized ; while in ( d ) , both outflows are magnetized . this represents the main difference between the five cases , and while all cases will develop fairly complex nonlinear evolutions governed by multiple , interacting instabilities , our main aim is to determine which configuration can result in a clearly sustained two - component jet flow over a sufficiently long distance ( time ) . it will turn out that cases ( a ) , ( c ) , and ( d ) all result in deceleration due to mixing between the two components , leading to decollimation of the jet , while only cases ( b1 ) and ( b2 ) convincingly maintain their two - component character . for later reference , important models parameters are listed in table [ table:1 ] . we now continue to discuss the complex nonlinear evolution of the five cases in some detail . [ cols="^,^,^,^,^ " , ] ( one full rotation of the inner jet corresponds to @xmath106).[figa ] according to the rayleigh criteria mentioned , the interface between the two components is always unstable . also , the effective inertia of the inner jet is higher than in the outer jet . in fact , at the interface @xmath35 , we have initially @xmath107 . in table [ table:1 ] , a @xmath7 ratio is given using a value midway the outer jet divided by the axial value . the overall effective inertia contrast is very different from the case we investigated in , where the ratio between ( mean value of ) effective inertia of the inner component jet to the effective inertia of the outer component jet was such that @xmath108 . initially , a linear surface mode develops at this interface @xmath35 . as a direct consequence , a small , radially extended shear region with low effective inertia @xmath109 forms at this location . at both bounding interfaces of this small shear layer , one with the outer and one with the inner jet , small wavelength instabilities develop meanwhile , acoustic waves from the inner interface propagate inwards , reflect on the jet axis , and then give rise to a kelvin - helmholtz - like body mode with dominant azimuthal mode number @xmath110 in the inner jet . meanwhile , the surface instability at the interface between inner jet and shear shell reaches a clearly non - linear phase . the overall interaction then gives rise to the growth of a relativistically enhanced rayleigh - taylor - type instability , having a body mode character propagating into the inner jet ( fig . [ figa ] , top ) . this instability regime shows about four arms at first , that collapse to three arms which progressively propagate inwards . as will be detailed later on , this instability can be explained from the fact that in this case ( a ) , fluid elements in the shear layer have a lower effective inertia @xmath109 and higher pressure than pre - existing material in the inner jet . we could qualitatively describe this relativistic rayleigh - taylor instability as follows . a fluid element in the shear layer is at first typically rotating with a speed of order of @xmath111 , the initial rotation speed at the inbound interface . the lower inertia of the shear shell relative to the inner jet makes that transfer of angular momentum from the inner wind / jet region to the shear layer is rather efficient , causing a fast rotational speedup of the shear layer matter . at the same time , the centrifugal force acting on a fluid element moving inwards varies with @xmath112 . because the fluid inertia in the shear region ends up at about a twelve times lower value than the inertia in the inner jet , the centrifugal force acting on this fluid element is much weaker than in the inner wind . at the same time , the pressure in the shear layer fluid element is higher than in the inner wind . therefore , the element will expand inwards until the pressure gradient gets balanced by the centrifugal force . in this particular simulation , three prominent fingers form and start propagating inwards . however , as the initial pressure profile decreases towards the jet axis , the centrifugal force in the inner jet decreases inward . with such a configuration , we find that in the phase where the intruding , rotating fingers of the rayleigh - taylor mode propagate inward , they will nearly reach the jet axis . very close to the axis , the centrifugal force acting on them gets compensated by the pressure force . in the phase that follows , the rotating fingers get deflected sideways in the direction of rotation . after less than half a full rotation of the inner jet ( at @xmath113 , see fig . [ figa ] , top ) , this relativistic rayleigh - taylor mode then dominates the body mode instability in the inner jet . the rayleigh - taylor fingers propagating inwards also compress the inner jet . since angular momentum is conserved , the centrifugal force increases and the inner jet fluid moves outwards towards the shear shell . this means that we get enhanced angular momentum transfer from the inner jet to the shear layer shell , which accelerates the rotational speed of the shear shell . we find that after less than one total rotation of the inner jet ( at @xmath114 , see fig . [ figa ] , center ) , the three rayleigh - taylor fingers nearly merge , making the inner jet appearance totally dominated by the ( growing ) shear shell . an analysis of this relativistic rayleigh - taylor - type instability is described seperately in section [ lninst ] . the low effective inertia of fluid in the shear region makes the interface between the shear flow and the inner jet unstable to this instability , where centrifugal forces act as an effective ( radially outwards pointing ) gravity . once initiated , these grow faster than the previously formed kelvin - helmholtz body mode . a number of ` arms ' develop from the shear region , having lower angular momentum than the fluid from the inner region , these propagate inwards , while inner jet fluid having higher angular momentum propagates outwards . the end result is transfer of energy and angular momentum from the inner jet to the growing shear region . as a consequence of this complex mixing of jet and shear layer matter , we find significant deceleration of the inner jet , while the density and the radius of the inner jet plus shear region increases ( this is quantified later in fig . [ inradius_evolution ] ) . the external boundary of the growing shear region at @xmath115 which borders the outer cold jet is also kelvin - helmholtz unstable . for the case at hand , in a first phase , a larger scale kelvin - helmholtz type mode develops with 8 undulations initially , eventually converging to form about 5 large scale structures ( see fig . [ figa ] , center ) . this is different from the case studied before in , where the most prominent feature in the non - linear evolution was the development of a kelvin - helmholtz type mode with @xmath116 arms . this mode transfers angular momentum from the shear region to the outer jet and decollimates the jet . at smaller spatial scale , but also at the interface between the shear region and the outer jet , other instabilities develop as well . these are again more centrifugally driven , as angular momentum and pressure increase in the shear region . thus also at this interface , the centrifugal force becomes locally higher than in the outer outflow . small bubble - like protrusions from the shear shell enter the outer outflow . however , since the pressure of the outer jet slowly increases outward , the growth of this smaller scale instability is stopped when the centrifugal force of the protruding bubbles gets balanced by the pressure of the outer jet . therefore , this instability remains weak and the large - scale dynamics of the interaction between the shear layer and the outer jet remains dominated by the kelvin - helmholtz instability . the small scale instabilities do influence the dynamics , by increasing the efficiency of the angular momentum transfer from the kelvin - helmholtz structures to the outer jet . finally , also the outer interface at @xmath117 between outer jet and the hot , dilute medium is rayleigh - taylor unstable , and is dominated by small wavelength perturbations . as the overall outcome of all the interacting instabilities , the two - component jet decollimates , such that the jet radius increases to @xmath118 ( see fig . [ radius_evolution ] ) . the inner component jet also spreads and its radius reaches @xmath119 ( see fig . [ inradius_evolution ] ) . also during the simulated @xmath120 years in physical time ( fig . [ figa ] , down ) , the rotation speed of both components decreases and the interaction between the shear region and the outer component is dominated by kelvin - helmholtz induced behaviour , together with the relativistically enhanced rayleigh - taylor instability of the inner rotating fluid . the main result of this interaction between the two components is the dramatic deceleration of the inner jet . in fact , the relativistic rayleigh - taylor - type instability leads to a deceleration of the inner component where the lorentz factor of the inner component drops to @xmath121 ( quantified later for all models in fig . [ lorentzfactor_evolution ] ) . we now describe the two - component jet evolution with a magnetized inner component . the effective inertia in the inner jet is lower than in the outer jet with initial contrasts @xmath122 in case ( b1 ) and @xmath123 in the case ( b2 ) , due to the initial lower pressure at the jet axis and the initial different distribution over thermal and magnetic energy in them . these cases are then such that the inner interface is now stable against the dominating rayleigh - taylor type instability described above and explained analytically in section [ lninst ] . still , the interface at @xmath115 between the inner and outer jet is subject to kelvin - helmholtz instability because of the differential rotation . this modifies the shape of the interface surface , leading to non - axisymmetric reflection of magnetohydrodynamic waves propagating through the inner jet . these waves in turn disturb locally the initial equilibrium between total pressure gradient and centrifugal force . during the evolution , some spikes develop at the interface , where the outer jet locally interchanges with the inner jet . these spikes are accelerated in the toroidal direction by the faster rotating inner jet . the centrifugal force acting on these slows down their inward expansion , and they then mainly propagate in the toroidal direction . their interaction with the inner jet material induces kelvin - helmholtz body mode instability , with a spiral pattern forming , this time having 3 arms in the two cases ( see fig . [ figb ] , top and fig . [ figb2 ] , top ) . they extract some angular momentum from the outer regions of the inner jet , slowing its rotation . again , a shear shell then forms in this region , continuously extracting angular momentum from the inner jet / wind . the spikes then slowly expand to the jet axis . the inner wind gets compressed by the shear layer . however , these mainly kelvin - helmholtz type instabilities turn out to have lower efficiency in extracting angular momentum than the relativistic rayleigh - taylor instability encountered in case ( a ) . the compression of the inner jet by the shear region is followed by a modest expansion . in the case ( b2 ) , the extraction of the angular momentum is weaker than in the case ( b1 ) , because in case ( b2 ) the inner jet has lower inertia . meanwhile , the angular momentum extracted from the inner jet leads to an outwards extension of the shear shell that formed at @xmath124 . at the interface between this growing shear shell and the outer jet , a kelvin - helmholtz instability surface mode with azimuthal mode number @xmath125 character then develops . also at this interface , smaller scale rayleigh - taylor instability proceeds . the shear shell has an effective inertia which is lower than the outer jet inertia . as locally centrifugal forces dominate , low inertia ` bubbles ' radially extend toward the outer jet . the growth rate of these bubbles is rather slow , since the pressure of the outer jet increases mildly outwards and the slowly rotating outer jet breaks their rotation . at the outer surface @xmath126 of the outer jet in the case ( b1 ) ( see fig . [ figb ] ) , we also witness rayleigh - taylor instability , as in case ( a ) . on the other hand , case ( b2 ) shows little evidence for such instability at the outer jet surface ( see fig . [ figb2 ] ) , which is due to the different external pressure conditions there . throughout the simulation , the jet in ( b1 ) and ( b2 ) remains collimated by the outer component , which compresses the inner component and shear region . the inner component jet radius in case ( b1 ) at @xmath120 years remains lower than @xmath127 and in case ( b2 ) even below about @xmath128 ( see fig . [ inradius_evolution ] ) . however , in case ( b1 ) the rayleigh - taylor instability at the surface of the outer region forms an extended sheat , which gives an increase to the apparant total jet radius @xmath129 ( see fig . [ radius_evolution ] ) , whereas the jet in case ( b2 ) does not show any sign of decollimation . both the inner jet and the shear region end up magnetized . the main difference between these cases and all others studied here , is that despite all small - scale instability development , the inner jet decelerates little , dropping to an average lorentz factor of about @xmath130 ( see fig . [ lorentzfactor_evolution ] ) . the two - component jets for case ( b1 ) and ( b2 ) remain clearly separable in inner and outer jet components ( see figs . [ figb ] , down-[figb2 ] , down ) , which is not the case for all other evolutions shown here . in both cases ( b1 ) and ( b2 ) , we do find that the inner jet component is deplaced from its on - axis position due do significant non - axisymmetric mode development . the jet stratification converges to a structure with an inner fast , magnetized spine having a lorentz factor of about @xmath130 . the spine is surrounded by a shear shell being @xmath131 denser than the spine and with lower lorentz factor and low magnetisation . the difference between the two cases is that the high magnetic pressure and low thermal pressure in the inner jet for case ( b2 ) increases its non - linear stability as compared to case ( b1 ) . in this case , only the outer jet is magnetized with a purely poloidal magnetic field . as mentioned before , the rotation of the jet is subsonic , such that acoustic waves which start to propagate from the inner - outer jet interface at different stages in the evolution propagate inwards and are reflected about @xmath95 times during one rotation period of the inner jet . our simulations follow about 2.5 full inner jet rotations . as the interface between the two components is unstable to various mode types as already encountered in the previous cases , its shape changes and the acoustic wave propagation becomes strongly non - axisymmetric . these waves play a clear role in disturbing the initial axisymmetry of the inner jet , and mitigate the development of non - axisymmetric body mode instabilities in the inner jet . in this case ( c ) , in a first stage , linear small - wavelength ( compared to case ( a ) ) kelvin - helmholtz instabilities develop at the interface @xmath115 between the two components . these reach a non - linear phase in less than a tenth of a full rotation . spikes of the outer jet , which are this time magnetized , arise in the inner jet . the large ratio of the fluid inertia between the outer and inner jet makes these spikes once more unstable to the centrifugally mediated , relativistically enhanced rayleigh - taylor instability seen previously in case ( a ) . this gives rise to the formation of rotating ` bubbles ' that protrude inward . in the case at hand , about ten such bubbles form ( see fig . [ figc ] , top ) , having smaller size compared to those seen to emerge in case ( a ) . this difference results due to the difference in the kelvin - helmholtz surface mode development mentioned earlier . the initial smaller size of the protruding bubbles makes their growth rate slower than in the purely hydrodynamical case ( a ) , until they reach the merging phase where in this case about three larger size bubbles form and converge to the jet axis . meanwhile , these bubbles compress and push fluid elements from the inner jet to a shear layer ( see fig . [ figc]center ) . thus once again , the larger - scale mode structures act to transfer angular momentum from the inner jet to the shear region . since the bubbles which move radially inwards form at the shear shell , their ratio of magnetic pressure to thermal pressure @xmath132 is lower than in the outer jet . in fact , the shear region is made up of fluid from the inner and the outer jet , and its thermal pressure dominates total pressure . the external surface of the shear shell is kelvin - helmholtz unstable and about 6 arms form ( see fig . [ figc ] , center ) . from each one , a centrifugally driven spike propagates outward . in fact , angular momentum extracted from the inner jet in the end gets transferred to the external part of the shear region . then , locally the centrifugal force at the interface between the shear shell and the outer jet acts to develop spikes which decollimate the jet ( see fig . [ figc ] , down ) . the total jet radius increases from @xmath133 at the initial stage to @xmath134 ( fig . [ radius_evolution ] ) , after the jet has propagated for a distance of @xmath99 . in the same time , the inner jet spreads as well , decelerates and its lorentz factor drops to about @xmath135 . at the endstate , the jet is constituted of an inner turbulent component with a radius of @xmath136 ( see fig . [ inradius_evolution ] ) and a small outer component . the final rotation speed profile increases from the axis to @xmath137 at @xmath128 , further decreasing outwards . in this last case , both inner and outer jet are magnetized . initially , from the discontinuity between the two components , fast magnetosonic waves propagate both outwards into the outer jet , and inwards into the inner jet . the end result of the initial readjustment is that a small shear region with lower magnetization than both component develops at the interface @xmath115 . also this configuration is kelvin - helmholtz unstable . at the surface from the shear region roughly 13 bubbles arise , moving inwards into the inner jet ( see fig . [ figd ] , top ) . they merge to form 3 large arms expanding towards the jet axis ( see fig . [ figd ] , center ) . fluid from the inner jet is pushed outwards , thereby extracting angular momentum from the inner jet . since the structures form at the shear region , the magnetic pressure inside the bubbles is lower than in the inner and outer component . however , the thermal pressure is increased in the shear region . during the evolution , dominant forces are magnetic pressure and thermal pressure that balance each other , while the contribution of the centrifugal force is low however , the centrifugal force remains responsible for triggering instabilities . as in case ( c ) , the shear region grows with eventual almost complete destruction of the inner jet component ( see fig . [ figd ] , center ) . in the last phase simulated , the dominating inner shear region becomes totally turbulent , and strong vortices form , changing the full jet shape to elliptic ( fig . [ figd ] , down ) . at this phase the average lorentz factor of the inner jet has dropped down to @xmath138 ( see fig . [ lorentzfactor_evolution ] ) . in cases ( a ) , ( c ) , and ( d ) , a relativistic , rayleigh - taylor type instability developed . here , we first describe in a qualitative way how this instability ensues , and then continue to explain it using an approximate linearization . at the interface between the two outflow components , we mentioned that ( linear and then non - linear ) surface modes develop , due to velocity shear hence of kelvin - helmholtz type . fluid of lower inertia ( low @xmath139 ) and higher pressure then penetrates the inner jet . since the inertia in the penetrating spikes is lower than in the surrounding inner jet material , the centrifugal force acting on these is weaker . the spikes then expand inwards until the pressure is balanced by the centrifugal force . they expand sideways as well , again on the basis of pressure balance and ram pressure arguments . due to the big difference in inertia , transfer of angular momentum from the inner jet to these inward penetrating spikes is very efficient . rather quickly then , their rotation speed reaches the rotation speed of the inner jet at the initial interface , set by @xmath140 . during its further inward expansion , this rotation speed remains constant . however , the low inertia of the spikes keeps their angular momentum lower than the inner jet matter . to explain this phenomenon semi - analytically , we confine our attention to a perturbation depending only on the radial direction @xmath16 . let @xmath141 represent the radial displacement of the contact surface between the two components during @xmath142 , and @xmath143 the perturbation speed . the momentum equation governing the fluid near our initial equilibrium writes as @xmath144\vec{v } + \nabla p_{\rm total}+\vec{v}\frac{\partial p_{\rm total}}{\partial t}+\ldots = 0\,,\ ] ] with @xmath145 . in this equation we already ignored the lab frame contribution of the charge separation , which is justified for rotational flows far within the light cylinder , as is the case in our model here at a distance of @xmath5 for m87 . the estimated light cylinder radius for m87 is smaller than @xmath133 if a large fraction of the angular momentum ( then poynting flux ) extracted from the accretion by the magnetic field is transferred to the stream lines during the acceleration phase by the ideal rmhd mechanisms . but in our model we assume that the jet is accelerated also by non - ideal rmhd mechanisms with decay of the toroidal magnetic field by dissipation / reconnection , leading to a fast central spine jet , and overall low rotation profiles ( weak transfer of angular momentum from magnetic fields to stream lines ) . ] moreover , the magnetic topology taken here does not involve magnetic pinching , hence magnetic effects only add to total pressure and to the effective inertia , where also @xmath146 is used . then the radial component of eq . ( [ euler ] ) can be further approximated , if we neglect temporal variation of the total pressure and use @xmath147 , to get @xmath148+\frac{\partial p_{\rm total}}{\partial r}=0\,.\ ] ] according to the initial condition , the inertia @xmath149 of the fluid varies slowly in the inner and outer jet . we can thus argue that both inertia and toroidal speed in each fluid element varies slowly when it undergoes a small radial displacement @xmath150 from the initial interface position @xmath124 , with @xmath151 . using the main equilibrium balance between centrifugal force and total pressure gradient , the variation of the total pressure can be argued to lead to @xmath152\,.\ ] ] to get an approximate dispersion relation , we assume that to first order , the perturbation speed is potential , i.e. @xmath153 and that we have @xmath154 . noting that the displacement @xmath155 must be identical for inner / outer regions , and using total pressure and displacement continuity arguments , we then get the essential proportionality relation between the instability growth rate and wave number @xmath156 \,.\ ] ] this approximative dispersion relation indicates that for two - component stability , we need @xmath157 requiring that @xmath158 . this confirms that the interface between the two rotating components is stable against a centrifugally driven , relativistic rayleigh - taylor instability , when the effective inertia of the outer component is higher than the effective inertia of the inner component . this important result correlates well with the results of all simulations in section [ simulations ] as well as in our previous publication . in fact , this equation explains why cases ( a ) , ( c ) , ( d ) are relativistically rayleigh - taylor unstable since we find typical contrasts during the evolution in ( a ) @xmath159 , in ( c ) and ( d ) @xmath160 , whereas cases ( b1 ) and ( b2 ) are relativistically rayleigh - taylor stable since @xmath161 in ( b1 ) and @xmath162 in ( b2 ) . initial effective inertia ratios derived from the initial conditions are mentioned in table [ table:1 ] , and it is clear that we can set up stable versus unstable cases by varying the contrast of effective inertia between inner versus outer jet . this also works in pure hydro , and indeed the pure hydro case from does not suffer from this newly discovered instability . for the purely poloidal field configurations studied here , magnetic field effects are prominent in total pressure and effective inertia alone . note finally that it is a truly relativistic effect , since the same argument in classical mhd just involves the density difference between outer versus inner jet . under the conditions of a light inner jet with heavy outer jet taken here , a classical variant of relation ( [ disprel ] ) involves only @xmath163 and would predict stability . we examined five configurations of magnetized two - component jets . all share the same density ratio between inner and outer component and identical rotation profiles . the magnetic and thermal pressure configuration in each model differs , though . this in turn translates to different distributions of the ( total , fixed ) kinetic energy flux over the inner and outer jet component . in fact , in case ( a ) the kinetic energy flux in the inner component is about @xmath22 , while it is @xmath164 for case ( b1 ) and @xmath165 in case ( b2 ) , and around @xmath166 for cases ( c ) and ( d ) . for the outer component , we thus have @xmath167 in case ( a ) , @xmath168 in case ( b1 ) and @xmath169 in case ( b2 ) , and only @xmath170 for cases ( c ) and ( d ) . the most important difference between the two model categories is then : cases ( a , c , d ) have an inner jet component with higher inertia @xmath171 than their outer jet component , while cases ( b1 ) and ( b2 ) have an inner jet component inertia which ends up lower than in the outer jet component . from the detailed analysis of the simulations , as well as from the approximate stability analysis , this criterion distinguishes between cases where relativistically enhanced rayleigh - taylor modes ultimately dominate the evolution , leading to complete mixing of both components and inner jet deceleration . this is quantified most clearly by showing the time evolution of the mean lorentz factor over the inner jet region for all cases in fig . [ lorentzfactor_evolution ] . this requires a clear criterion to distinguish inner versus outer jets in the turbulent evolutions . in cases ( a ) , ( c ) and ( d ) , we locate the outer jet component as having a lorentz factor @xmath172 and effective polytropic index @xmath173 . the inner component is the region defined by a lorentz factor @xmath174 and effective polytropic index @xmath175 . in the cases ( b1 ) and ( b2 ) , the effective polytropic index in the inner and outer jet can be locally of the same order there , hence we use that the inner jet is magnetized . in the cases ( b1 ) and ( b2 ) , the inner component jet and shear region are not totally mixed during the evolution , since the inner component is compressed and has higher magnetization and lorentz factor . thus to distinguish the inner component jet we put a condition on magnetic field strength @xmath176 , where @xmath177 is the magnetic strength assumed initially . under these precise quantifications of inner / outer jet regions , fig . [ lorentzfactor_evolution ] demonstrates clearly how stable cases ( for the relativistically enhanced rayleigh - taylor modes ) remain at high speed , while unstable cases decelerate . using the same means to distinguish inner versus outer jet regions at all times , we can quantify the inner jet radius for all cases , as well as the total jet radius for all cases . these are shown in figs . [ inradius_evolution]-[radius_evolution ] , and quantify the decollimation effects discussed in section [ simulations ] . during the entire evolutions , the toroidal and radial speeds remain weak as we have typically a maximal @xmath178 and @xmath179 . this means that the contribution of the laboratory frame charge separation force @xmath180 to the lorentz force is negligible at all times . under these conditions , in both radial and toroidal directions , the contribution of magnetic energy to fluid inertia is @xmath181 weaker than the contribution of the magnetic pressure to the total pressure . this explains why cases ( b1 ) and ( b2 ) are then more stable than the other cases . despite the fact that case ( b2 ) has similar axial total pressure than other cases ( a , c , d ) , the low contribution of thermal energy to total pressure makes the effective inertia ratio between the inner and outer jet low . this two - component jet is then stable against the relativistically enhanced rayleigh - taylor instability . extraction of angular momentum and energy from the inner jet to the shear shell is less efficient in cases ( b1 ) and ( b2 ) than in all other cases . as a result , in cases ( b1 ) and ( b2 ) the inner relativistic jet with high lorentz factor @xmath182 persists . in the other cases , the kinetic energy flux in the inner jet was initially relatively high , making them unstable and leading to deceleration to lorentz factors around @xmath135 we investigated the stability of two - component jets beyond the launching region , where both components are rotating differently and a clear two - component structure exists . we initialized this model in accord with magneto - centrifugal models for jet generation , also using the observed analogy between radio source jets and two - component jets in young stellar objects , where the rotation within the jet can actually be observed . we performed five very high resolution simulations of magnetized two - component jets with various magnetization and kinetic energy flux stratifications . the two - component jets with a low inner kinetic energy flux contribution are more stable and remain relativistic for long distances , whereas jets with a highly contributing inner jet to the total jet kinetic energy flux , are subject to a relativistic rayleigh - taylor type instability . this instability turns out to be very efficient to decollimate and decelerate the inner jet . jets that are subject to this instability become turbulent after propagating for a distance of about @xmath183 . this new result on two - component jet models is important because it can explain the classification of radio sources in fanaroff - riley i / ii categories according the energy stratification of the inner jet . this ultimately relates to the jet launch region and the properties of the inner accretion disk . in fact , by analogy between the fri / frii classification and the results of our model , an fri jet would correspond to a two - component jet with a high energy flux contribution from the inner jet , whereas the frii jet corresponds to relatively low energy fluxes in the inner jet . the model we propose here to explain the fri / frii dichotomy is different from the model we proposed earlier @xcite where the transition occurs due to external density stratification . that model explains the group of peculiar hybrid morphology radio source " ( hymors ) which appear to have an fr ii type on one side and an fr i type diffuse radio lobe on the other side of the active nucleus . since the launch conditions on each side are presumably similar in these kind of radio sources , the different fanaroff - riley morphologies on either side must be attributed to the change in the properties of the ambient media , as shown convincingly in @xcite . the results of the present paper nicely complement these earlier findings with a quantifiable role of the central engine contribution . we currently continue this study of the interaction between two component jets in full 3d . we thereby intend to explore the relative influence of azimuthal versus longitudinal instabilities for realistic multi - component jets . another extension is to allow for aximuthal magnetic fields , in accord with the initial profiles as given generally in this paper . it then rremains to be shown that ( 1 ) the newly discovered instability persists in 3d hydro and magnetohydrodynamic configurations , where the potential role of axial mode development is incorporated , and introduces helical jet axis displacements ; ( 2 ) how the instability gets modified ( stabilized or destabilized ) by the inclusion of toroidal field components , first in 2.5d neglecting helical axis displacements , and consecutively in 3d , where also current - driven kinks may occur . we acknowledge financial support from the fwo , grant g.027708 , and from the k.u.leuven goa/09/009 . part of the computations made use of the high performance computing vic cluster at k.u.leuven . z. meliani acknowledge financial support from hpc europa ( project number : 228398 ) . visualization was performed using paraview , see www.paraview.org . 99 appl , s. , & camenzind , m. , 1993 , a&a , 270 , 71 bacciotti , f. , mundt , r. , ray , t.p . , eiskffel , j. , solf , j. , camenzind , m. , 2000 , apj , 537 , l49 bacciotti , f. , ray , t.p . , mundt , r. , eiskffel , j. , solf , j. , 2002 , apj , 576 , 222 baty , h. , & keppens , r. , 2002 , apj , 580 , 800 begelman , m. c. , li . z. y. , 1994 , 426 , 269 begelman , m.c . , 1998 , apj , 493 , 291 biretta , j.a . , junor , w. , livio , m. , 2002 , newar , 46 , 239 bogovalov , s. , & tsinganos , k. , 2001 , mnras , 325 , 249 coffey , d. , bacciotti , f. , woitas , j. , ray , t.p . , eislffel , j. , 2004 , apj , 604 , 758 coffey , d. , bacciotti , f. , ray , t.p . , eislffel , j. , woitas , j. , 2007 , apj , 663 , 350 cranmer , s.r . , 2008 , apj , 689 , 316 fendt , c. , 1997 , a&a , 319 , 1025 fendt , c. , 2009 , apj , 692 , 346 gopal - krishna , wiita , p. j. , 2000 , a&a , 363 , 507 ghisellini , g. , tavecchio , f. , chiaberge , m. , 2005 , a&a , 432 , 401 giroletti , m. , giovannini , g. , feretti , l. , cotton , w.d . , edwards , p.g . , lara , l. , marscher , a.p . , mattox , j.r . , piner , b.g . , venturi , t. , 2004 , apj , 600 , 127 ghisellini , g. , tavecchio , f. , 2008 , mnras , 386 , l28 gnther , h.m . , matt , s.p . , li , z .- y . , 2009 , a&a , 493 , 579 hardcastle , m.j . , 2006 , mnras , 366 , 1465 hardee , p.e . , 2004 , ap&ass , 293 , 117 heyvaerts , j. , & norman , c. 2003 , apj , 596 , 1240 istomin , ya.n . , pariev , v.i . , 1996 , mnras , 281 , 1 jester , s. , harris , d.e . , marshall , h.l . , meisenheimer , k. , 2006 , apj , 648 , 900 jester , s. , meisenheimer , k. , martel , a.r . , perlman , e.s . , sparks , w.b . , 2007 , mnras , 380 , 828 kataoka j. , stawarz , l. , harris , d.e . , siemiginowska , a. , ostrowski , m. , swain , m.r . , hardcastle , m.j . , goodger , j.l . , iwasawa , k. , edwards , p.g . , 2008 , apj , 685 , 839 keppens , r. , meliani , z. , van der holst , b. , casse , f. , 2008 , a&a , 486 , 663 komissarov , s.s . , barkov , m.v . , vlahakis , n. , knigl , a. , 2007 , mnras , 380 , 51 matsakos , t. , tsinganos , k. , vlahakis , n. , massaglia , s. , mignone , a. , trussoni , e. , 2008 , a&a , 477 , 521 mckinney , j.c . , blandford , r.d . , 2009 , mnras , 394 , l126 mignone , a. , plewa , t. , bodo , g. , 2005 , apjs,160 , 199 mignone , a. , bodo , g. , 2005 , mnras , 364 , 126 mizuno , y. , lyubarsky , y. , nishikawa , k .- hardee , p.e . , 2009 , apj , 700 , 684 meliani , z. , sauty , c. , tsinganos , k. , vlahakis , n. , 2004 , a&a , 425 , 773 meliani , z. , sauty , c. , vlahakis , n. , tsinganos , k. , trussoni , e. , 2006a , a&a , 447 , 797 meliani , z. , casse , f. , sauty , c. , 2006b , a&a , 460 , 1 meliani , z. , keppens , r , casse , f. , giannios , d. , 2007 , mnras , 376 , 1189 meliani , z. , & keppens , r. , 2007 , a&a , 475 , 785 meliani , z. , keppens , r. , giacomazzo , b. , 2008 , a&a , 491 , 321 meier , d.l . , 2003 , newar , 47 , 667 narayan , r. , li , j. , tchekhovskoy , a. , 2009 , apj , 697 , 1681 osullivan , s. p. , gabuzda , d. c. , 2009 , arxiv:0907.5211 racusin , j. l. , et al . , 2008 , nature , 455 , 183 rawlings , s. , & saunders , r. , 1991 , nature , 349 , 138 siemiginowska , a. , stawarz , l. , cheung , c.c . , harris , d.e . , sikora , m. , aldcroft , t.l . , bechtold , j. , 2007 , apj , 657 , 145 sol , h. , pelletier , g. , assero , e. , 1989 , mnras , 237 , 411 spruit , h.c . , foglizzo , t. , stehle , r. , 1997 , mnras , 288 , 333 spruit , h.c . , 2008 , arxiv , 0804.3096 tavecchio , f. , maraschi , l. , sambruna , r.m . , urry , c.m . , cheung , c.c . , gambill , j.k . , scarpa , r. , 2004 , apj , 614 , 64 van der holst , b. , & keppens , r. , 2007 , jcp , 226 , 925 van der holst , b. , keppens , r. , meliani , z. , 2008 , cophc , 179 , 617 woitas , j. , bacciotti , f. , ray , t.p . , marconi , a. , coffey , d. , eislffel , j. , 2005 , a&a , 432 , 149

we present a study that addresses both the stationary properties of the energy current and quantum correlations in a three - mode chain subjected to ohmic and super - ohmic dissipations . an extensive numerical analysis shows that the mean value and the fluctuations of the energy current remain insensitive to the emergence of a rich variety of quantum correlations : such as , two - mode discord and entanglement , bipartite three - mode and genuine tripartite entanglement . the discussion of the numerical results is based on the derived expressions for the stationary properties in terms of the two - time correlation functions of the oscillator operators , which carry quantum correlations . interestingly , we show that quantum discord can be enhanced by both considering initially squeezed thermal bath states and imposing temperature gradients .

1503.09051

entanglement is one of the most striking phenomena in quantum physics . composite systems exhibiting genuine quantum correlations defies our intuition , in the sense that they are not interpreted by classical or semiclassical means @xcite . quantum correlations are at the heart of many quantum information tasks , such as quantum teleportation and quantum communication , as well as at the core of a variety of many - body physics phenomena @xcite . _ quantum thermodynamics _ seeks the understanding of the emergence of thermodynamics laws from those of quantum dynamics @xcite . in this sense , one may wonder about the role of quantum correlations and coherence in different phenomena of interest in quantum thermodynamics , for instance in the thermodynamics of quantum thermal machines @xcite or , more generally , in thermal non - equilibrium systems @xcite . quantum correlations have different fates depending on the environmental influence @xcite . although entanglement is fragile with respect to thermal fluctuations and decoherence , stationary entanglement still could remain in a system subjected to dissipation @xcite . quantum discord @xcite on the other hand seems more stable to environmental noise @xcite . recently , efforts have been made to study how dissipation may precisely drive the system onto preferred states , e.g. onto a genuine entangled state @xcite , by engineering the interaction with environments @xcite . according to the non - equilibrium theory , the analysis of the ( _ linear _ ) response of many - body systems to macroscopic thermodynamic forces , such as those induced by temperature or chemical potential gradients , and to ( weak ) external fields provides an opportunity to test some predictions from condensed matter theory and statistical physics . as an illustration , multipartite entanglement in spin chains has been explored through precise measurements of the magnetic susceptibility @xcite and the heat capacity @xcite . also theoretical studies of these two magnitudes seem to provide observable signatures of entanglement in spin chains at thermal equilibrium @xcite . nowadays , the energy transport through systems involving spatial continuous variables , such as chains of trapped ions , can be experimentally measured @xcite . given the increasing interest in systems under non - equilibrium thermal conditions in the quantum regime @xcite , one may naturally ask whether the stationary response of a system to a temperature gradient may be influenced by the presence of pure quantum correlations , and in particular by _ genuine _ multipartite entanglement . the present work tries to elucidate whether the _ average _ properties of the stationary energy current across a harmonic chain , such the mean values and fluctuations , are sensitive to the presence of two - mode and _ genuine _ tripartite entanglement in the system , and more generically to quantum correlations as measured by discord . significant advances in the context of quantum spin networks indicate that the presence of bipartite entanglement does not play an important role on excitation transport @xcite , whereas a strong correlation between quantum coherence and transport efficiency can be present @xcite . although quantum correlations tend to disappear in systems subjected to a temperature gradient , it has been shown that entanglement and discord can still survive in systems under such conditions @xcite . much less is known about the influence of genuine multipartite entanglement and the structure of discord on the stationary energy current in strongly dissipated harmonic chains at low temperatures . this work focuses on stationary quantum correlations in a continuous - variable system within such domain , and analyzes their possible relation to non - equilibrium conditions . we consider an open system model composed of a linear arrange of three harmonic oscillators , each of them interacting with its own independent heat bath . we assume that the heat baths are in an initial squeezed thermal state @xcite . this set - up is particularly interesting in the study of the generation of entanglement between distant modes of a quantum network @xcite , and also a convenient model to analyze many issues concerning the quantum thermodynamics of continuous variable systems @xcite . we employ the open - system formalism based on the generalized langevin equation ( gle ) @xcite to carry out an extensive numerical study of the stationary properties . we focus on the two- and three - mode entanglement and the discord in the presence of an energy current through the harmonic chain , for a large range of system parameters . we will analyze whether the average and the fluctuations of the energy current exhibit any evidence of the quantum correlations emerging under non - equilibrium thermal conditions . the paper is organized as follows . in section [ sec:1 ] we describe the model of the system and introduce the covariance matrix , which fully characterizes the stationary state of the system . section [ sla ] reviews the generalized langevin equation approach considered to obtain this state . in section [ sec:3 ] we derive the expressions giving the average and the fluctuations of the energy current in terms of two - time correlation functions , and introduce the quatum correlations , characterized by means of the two- and three - mode entanglement and the ( right ) discord . the numerical results are presented in section [ sres ] , the corresponding discussion is given in section [ sdis ] , and the main conclusions are put together in section [ scon ] . we consider an open one - dimensional chain composed of three harmonic oscillators , see figure [ system ] , labeled as @xmath0 ( left ) , @xmath1 ( center ) , and @xmath2 ( right ) , with identical mass @xmath3 , natural frequencies @xmath4 , and position and momentum operators @xmath5 . we assume bilinear interactions between first - neighbor oscillators , @xmath6 and @xmath7 , with strength given by a single parameter @xmath8 . each @xmath9 oscillator is coupled with an independent heat bath composed of @xmath10 independent harmonic oscillators , with masses @xmath11 , frequencies @xmath12 and position and momentum operators @xmath13 . eventually we will consider the quasi - continuum limit @xmath14 . the hamiltonian of the global system can be written as @xmath15 where @xmath16 with @xmath17 corresponds to the isolated chain , and @xmath18 describes the three independent baths and their interactions with the oscillators , which are assumed bilinear with coupling constants @xmath19 . , @xmath20 and @xmath21 . @xmath22 indicates the energy current from the heat bath to the @xmath9 oscillator , and @xmath23 the energy current from the @xmath24 to the @xmath9 oscillator , in the case of @xmath25 . @xmath8 is the coupling constant between first - neighbor oscillators , and @xmath19 the coupling constant between the @xmath9 chain oscillator and the @xmath26 oscillator of the bath . ] the interaction term in the microscopic model given by @xmath27 includes the renormalization terms @xmath28 which ensures that the frequency @xmath29 is maintained as the bare frequency of the @xmath9 oscillator @xcite , and the complete positivity of the total hamiltonian ( [ mm ] ) . in general , a system under the influence of dissipative effects will evolve in the long time limit toward a stationary state in which any trace of its initial state has been wiped out . the initial condition is only relevant in determining the transient dynamics previous to this asymptotic state . we fix the initial state at @xmath30 , and assume a barely chance of interaction between the system and the environment at to this point . then it is reasonable to consider that the system and the environment are initially uncorrelated . as our analysis is based on quantum properties in the asymptotic stationary state , without loss of generality we will assume an initial product state given by @xmath31 @xcite , where @xmath32 is the initial state of the isolated chain and @xmath33 initial gaussian quantum states corresponding to the baths , which are not necessarily at thermal equilibrium states . assuming that initially the baths are in squeezed thermal states with zero first - moments @xcite , the following averages over the initial state @xmath34 are satisfied @xmath35\,\right ] , \nonumber \\ \frac{1}{2}\left\langle \left\lbrace { \hat p}_{i\nu}(t_{0 } ) , { \hat p}_{i\mu}(t_{0})\right\rbrace\right\rangle_{\hat{\rho}_{0 } } & = & \delta_{\nu\mu}\,\frac{\hbar m_{i\mu}\omega_{i\mu}}{2}\left[\,1 + 2\,n\left(\omega_{i\mu}\right)\,-\,2\,\text{re}\left[\,m\left(\omega_{i\mu}\right)\,\right]\,\right]\,,\nonumber \\ \frac{1}{2}\left\langle \left\lbrace { \hat q}_{i\nu}(t_{0 } ) , { \hat p}_{i\mu}(t_{0})\right\rbrace\right\rangle_{\hat{\rho}_{0 } } & = & \delta_{\nu\mu}\,\hbar\,\text{im}\left[\,m\left(\omega_{i\mu } \right)\,\right ] , \label{cmhb}\end{aligned}\ ] ] where @xmath36 $ ] and @xmath37 $ ] denotes the real and imaginary part of @xmath38 , and @xmath39 which satisfy the relation @xmath40 . we have considered the same squeeze @xmath41 for all the oscillators of the @xmath9 bath . @xmath42 is a global arbitrary rotation of the bath state @xmath43 , and @xmath44 is the average occupation number of the @xmath45 oscillator in the @xmath9 bath in a thermal equilibrium state . to induce an energy current across the chain , see figure [ system ] , we fix the left and right heat baths at different temperatures , @xmath46 and @xmath47 respectively , with @xmath48 low enough to ensure that the system remains within the quantum regime . then we modify the temperature of the central bath , @xmath49 , by considering different values of @xmath50 . this setup is particularly interesting as it makes possible to establish a quasiclassical regime in the central oscillator while maintaining the lateral oscillators in the quantum regime . below we will show that the asymptotic stationary state derived from this manipulation of the central oscillator can exhibit a rich variety of quantum correlations , such as two - mode and bipartite three - mode and genuine tripartite entanglement . since the total hamiltonian ( [ mm ] ) is quadratic in both positions and momenta , and we have considered gaussian initial bath states , the asymptotic stationary state , denoted by @xmath51 , will be gaussian for any initial state of the oscillators @xcite . then , the stationary quantum properties will be determined by just the first and second moments of the positions @xmath52 and the momenta @xmath53 . the former can be made arbitrarily close to zero by unitary local transformations that do not affect the non - local properties such as entanglement . whereas the second moments determining all the correlation properties required in our analysis are given in terms of the covariance matrix @xmath54 with @xmath55 and @xmath56 , and the two - point ( symmetrical ) correlation functions @xmath57 the second moments of the energy currents also involve the imaginary part of the two - point correlation @xmath58 $ ] , given by @xmath59 \,\right)\ , . \label{cv2}\\\ ] ] the covariance matrix of the state @xmath60 corresponding to the subsystem defined by the @xmath9 and @xmath24 oscillators can be obtained from ( [ cvmatrix ] ) by just taking the elements associated with these two oscillators . it should be emphasized that at stationary conditions the correlation functions ( [ cv ] ) and ( [ cv2 ] ) only depend on the time difference @xmath61 , and a particular initial time @xmath62 is irrelevant to obtain them . this will be important in what follows , when computing these stationary correlations by using the generalized langevin equation approach . within the langevin approach , the equations of motion that govern the evolution of the stationary correlations are derived from the microscopic model ( [ mm ] ) by writing the heisenberg equations for the oscillator positions and tracing out the degrees of freedom of the heat baths . this leads to the so - called generalized langevin equation , @xmath63 where we have introduced the potential @xmath64 the susceptibilities @xmath65 and the fluctuating forces @xmath66 with @xmath67 the heaviside step function , and the spectral density of the environment given by @xmath68 as long as the stationary solution of the generalized langevin equation is guaranteed , one may take the limit @xmath30 in eq . ( [ leq ] ) , and then , use the fourier transform @xmath69 to obtain the stationary solution of the position and momentum operators @xcite . by replacing these solutions into the correlation elements ( [ cv ] ) and averaging over the initial state @xmath34 , it follows @xmath70 with @xmath71 and @xmath72 where @xmath73 . the two - points correlation functions @xmath74 ( [ cv2 ] ) satisfy an expression identical to ( [ sleq1 ] ) , but replacing in @xmath75 the anticommutator of the fluctuating force by the commutator . expression ( [ greenf ] ) is nothing but the fourier transform of the ( matrix ) green function @xcite for the generalized langevin equation ( [ leq ] ) . the real and imaginary parts of the fourier transform @xmath76 of the susceptibility ( [ sus ] ) are given by ( see appendix [ asusc ] for further details ) @xmath77&=&\hbar\left[\,\theta(\omega)\,j_{i}(\omega)-\theta(-\omega)\,j_{i}(-\omega)\,\right ] , \label{imx } \\ \text{re}\left[\,\tilde{\chi}_{i}(\omega)\,\right]&=&\frac{1}{\pi}\,p\int\frac{\text{im}\left[\,\tilde{\chi}_{i}(\omega')\,\right]}{\omega'-\omega}\,d\omega ' , \label{rex}\end{aligned}\ ] ] where @xmath78 denotes the cauchy principal value . the second expression is the well - known kramers - kronig relation @xcite arising from the causal nature of the susceptibility . in order the three - mode system can reach a stationary state , the function @xmath79 must approach a combination of decaying exponentials in the long time limit @xcite . according to a previous study of the equations of motion of the system - plus - environment complex in terms of normal modes @xcite , the existence of a well defined stationary solution entails that @xmath80 has no any real root @xmath81 corresponding to the frequency of a bound normal mode , which implies that @xmath82\neq 0 $ ] . from eq . ( [ imx ] ) , the latter condition means that @xmath81 must be contained within the domain of the bath spectral density @xmath83 @xcite . in general , the heat baths can be considered as composed of a large number of degrees of freedom with finite broad band spectrum , in which the most energetic environmental - degree is roughly determined by a cut - off frequency @xmath84 . this ensures that the natural frequencies of the system are well embedded in the environmental spectrum , and consequently it makes possible an irreversible energy transfer from the system to the environment , at least in a finite time much larger than the natural time scale of the system . we shall impose @xmath85 in order to ensure an irreversible evolution of the three - oscillator chain toward a well defined asymptotic stationary state . the covariance matrix of this stationary state is completely determined by the correlation functions ( [ sleq1 ] ) evaluated at equal time , once the correlation functions of the fluctuating forces ( [ ff ] ) have been obtained . these correlations depend only on the initial environmental state . below we show the relation between the fluctuating forces and the initial covariance matrix ( [ cmhb ] ) of the heat baths . our choice of the initial environmental state implies the statistical independence of the fluctuating forces corresponding to different heat baths , _ i.e. _ @xmath86 for all @xmath87 . whereas , according to eqs . ( [ cmhb ] ) , the symmetrical two - time correlation function of the fluctuating forces associated with a given @xmath88 bath is given by ( see appendix [ aff ] for further details ) @xmath89\cos\left(\omega_{l\mu}(t+t'-2t_{0})\right)+\text{im}\left[\,m_{l}(\omega_{l\mu})\,\right]\sin\left(\omega_{l\mu}(t+t'-2t_{0})\right)\bigg]\ , . \label{ffe}\end{aligned}\ ] ] the average of the corresponding commutator can be expressed as @xmath90\right\rangle_{\hat{\rho}_{0}}= i\,\delta_{lm}\,\frac{\hbar}{2}\sum_{\mu=1}^n \frac{g_{l\mu}^2}{m_{l\mu}\omega_{l\mu}}\,\sin\left(\omega_{l\mu}(t'-t)\right)\ , . \label{aff2a}\end{aligned}\ ] ] the dependence of the symmetrical two - time correlation functions on the initial time @xmath91 is eliminated in the case of an initial thermal equilibrium state of the @xmath88 bath , in which @xmath92 and @xmath93 for all @xmath94 values . although in the previous section we have already fixed the time limit @xmath95 in order to obtain the stationary solution , we shall maintain the notation @xmath96 for convenience in order to make more clear the following discussion . as shown in appendix [ aff ] , the non - stationary terms in eq.([ffe ] ) come from the average of factors involving @xmath97 and @xmath98 , with @xmath99 the annihilation ( creation ) operator of the @xmath45 mode in the @xmath100th reservoir . these terms describe non - conservative energy processes that take place in the heat bath at the initial time @xmath96 , and therefore , they may influence the transient dynamics of the three - oscillator chain . however , they become highly oscillatory in the long time limit @xmath101 and their contribution to the stationary properties may be disregarded @xcite . when taking the quasi - continuum limit @xmath102 in the environment spectral density , only the stationary term in eq.([ffe ] ) remains . this assertion holds for an environment with a broad spectrum limited by @xmath103 , and a finite interaction between the reservoir modes and the system oscillators . mathematically , the latter translates into that the spectral densities @xmath104 are finite continuous functions , and the corresponding coupling strengths should decay at least as @xmath105 at high frequencies . under these conditions , the long time limit of the symmetrical two - time correlation function ( [ ffe ] ) reduces to the following expression in the frequency domain ( see appendix [ aff ] for details ) @xmath106\coth\left(\frac{\hbar\,\omega}{2\,k_{b}\,t_{l}}\right)\cosh\left(2r_{l}\right)\ , . \label{cffdal}\ ] ] the average of the corresponding commutator reduces to @xmath107\right\rangle_{\hat{\rho}_{0}}= 2\pi\,\delta(\omega+\omega')\,\text{im}\left[\,\chi_{l}(\omega)\,\right]\ , . \label{cffdal2}\ ] ] similar results have been previously obtained within the path integral formalism @xcite , see also @xcite . we point out that an initially squeezed state of the environment makes the reduced system to _ notice _ an effective temperature above the temperature @xmath108 of the heat bath at thermal equilibrium . this effect has interesting consequences in the efficiency of thermal machines within the quantum regimen @xcite . now we can replace the autocorrelations ( [ cffdal ] ) into the expressions ( [ gij ] ) , and perform the integral in the frequency @xmath109 to obtain a closed - form expression for the two - time correlation functions @xmath110 . notice that @xmath111 due to the stationary condition of the fluctuating force correlation . a similar procedure is followed for the functions @xmath74 ( [ cv2 ] ) . in general , there are not analytic expressions giving the integrals involved in the correlation functions in terms of the system parameters , such as the bath temperatures , the oscillator frequencies and the coupling strengths . we will compute them by means of numerical methods . the non equilibrium conditions imposed by the different bath temperatures drive an energy current through the system , see figure [ system ] . a discrete definition of the energy currents associated with each chain oscillator can be derived from its local energy @xcite @xmath112 with @xmath113 for @xmath114 , and @xmath115/4 $ ] . the time derivative of @xmath116 leads to the discrete continuity equations @xmath117 with @xmath118 for @xmath114 , and @xmath119 . the term @xmath120 \label{j1}\end{aligned}\ ] ] can be identified as the energy current from the @xmath24 oscillator to the @xmath9 oscillator , whereas @xmath121 corresponds to the energy current from the @xmath9 heat bath into the @xmath9 oscillator . at stationary conditions the total current coming from the baths into the system becomes zero . here we will focus on the analysis of the total current flowing from the @xmath0- to the @xmath2-oscillator , defined as @xmath122 our study is based on the stationary properties of the total energy current , which are basically determined by its first- and second - moments , or equivalently , by its average and fluctuations . the steady state average of the total energy current @xmath123 can be obtained by tracing ( [ j1 ] ) over the the stationary state and using the stationary solutions of the two - time correlation functions ( [ sleq1 ] ) , which leads to @xmath124\ , \label{ef1 } \end{aligned}\ ] ] for the local currents @xmath125 and @xmath126 . since the quantum correlations shared by the oscillators , in particular entanglement , are partially encoded on the correlation terms indicated in ( [ j1 ] ) , one might expect that the energy current could be sensitive to these correlations . notice that the total current involves the correlations between the central and the side oscillators , while it does not depend on the crossed correlation function between the two side oscillators . we shall further analyze this issue in section ( [ sdis ] ) . to have a better understanding of the system behavior under non - equilibrium conditions we also study the two - time correlation functions of the energy currents ( [ j1 ] ) . the fluctuations can be obtained from the symmetrical version of the classical two - time correlations , and expressed as @xmath127 theoretically , the response of a system under external perturbations can be studied in terms of these correlations functions @xcite . notice that we evaluate the fluctuations in the non - equilibrium stationary state , so that we might expect that eq.([cj1 ] ) can elucidate some properties of stationary non - equilibrium , rather than the equilibrium quantum correlations @xcite . furthermore , it has been shown that eq.([cj1 ] ) is related to the fluctuations of the stationary energy current across the chain @xcite . as the stationary state obeys a gaussian distribution , the four - time correlation terms implicit in @xmath128 can be decomposed into terms involving the product of two - time correlations , in the form @xmath129\right\rangle \left\langle \left [ \hat p_{j}(\tau),\hat p_{m } ( 0)\right]\right\rangle\,\right ] \nonumber \\ & + & 2\,\left[\left\langle \left\lbrace \hat x_{i}(\tau),\hat p_{m}(0 ) \right\rbrace\right\rangle \left\langle \left\lbrace \hat x_{l } ( 0),\hat p_{j}(\tau)\right\rbrace\right\rangle - \left\langle \left [ \hat x_{i}(\tau),\hat p_{m}(0 ) \right]\right\rangle \left\langle \left [ \hat x_{l } ( 0 ) , \hat p_{j}(\tau)\right]\right\rangle\right ] \ , . \label{ftc}\end{aligned}\ ] ] then the current - current response function ( [ cj1 ] ) can be expressed as @xmath130\right . \\ \nonumber \left.\,+\,s_{\alpha,\beta'}\left[\,c_{x_\alpha p_\beta}(\tau,0)\,c_{x_{\beta'}p_{\alpha'}}(-\tau,0)- y_{x_\alpha p_\beta}(\tau,0)\,y_{x_{\beta'}p_{\alpha'}}(-\tau,0)\right ] \,\right]\,,\hspace*{1cm}\end{aligned}\ ] ] where @xmath131 is the sign of the cofactor of the element @xmath132 in the @xmath133 array defined by the indexes @xmath134 . finally , according to eq.([gef1 ] ) , the autocorrelation function of the total current flowing from the @xmath0- to the @xmath2-oscillator is given by @xmath135 in contrast to the average energy current , the correlation function @xmath136 involves crossed terms between the @xmath2- and the @xmath137- oscillators . in addition , while @xmath138 ( [ tec ] ) is given by a linear combination of two - time correlation terms , @xmath136 has a nonlinear dependence on such terms . these two aspects will be useful in the subsequent discussion . alternatively , the behavior of @xmath138 and @xmath136 will help us to gauge whether the average properties of the energy current are sensitive to quantum correlations , such as genuine tripartite entanglement . we shall analyze the two - mode quantum correlations between the @xmath9 and @xmath24 oscillators by means of the discord measure on the right @xcite , denoted by @xmath139 . the entanglement between both modes can be quantified by the well - known logarithmic negativity @xmath140 @xcite . in particular we devote special attention to the entanglement @xmath141 and the discord @xmath142 between the side oscillators . we use a recent criterion in the realm of continuous - variable systems @xcite to study tripartite entanglement , which is a good estimator for @xmath143-partite entanglement @xcite in @xmath144-mode gaussian as well as non - gaussian states . a tripartite harmonic system may develop _ bipartite _ three - mode entanglement , which means that there is at least a bipartition of the three - mode system that is entangled , or _ genuine tripartite _ entanglement , which corresponds to the case in which all the bipartitions are entangled and the state @xmath51 can not be written as a convex combination of bipartite separable states . here , the criterion reduces to evaluate a figure of merit @xmath145 , such as a positive value of @xmath146 ( @xmath147 ) indicates that the state @xmath51 is genuine tripartite entangled ( bipartite three - mode entangled ) @xcite . as we are dealing with stationary gaussian states , all the previously mentioned indicators of quantum correlations can be directly computed from the covariance matrix @xmath148 given by eq.([cvmatrix ] ) . the logarithmic negativity can be expressed as @xcite , @xmath149 where @xmath150 stands for the lowest symplectic eigenvalue of the partial transpose covariance matrix @xmath151 , corresponding to the reduced density matrix @xmath60 . the ( right ) discord is given by @xcite , @xmath152 with the total correlations @xmath153 and the classical correlations @xmath154 which are given in terms of the von newman entropy @xmath155 . closed form expressions for the quantum discord as a function of the covariance matrix @xmath156 have been derived in @xcite . it is important to realize that these indicators of quantum correlations involve a non - linear dependence on the density operator and the two - time correlation functions . see appendix [ asc ] for further details of the logarithmic negativity , the quantum discord and the separability criteria @xmath145 . we now investigate the average properties of the total current @xmath157 when the three oscillators share two - mode and tripartite entanglement . in many realistic situations , e.g. quantum brownian motion , the interaction with the environment leads to an ohmic dissipation . in a first approach , for a nanomechanical setup , one may think that the thermal relaxation is mainly due to the coupling with the acoustic phonons of the substrate , which may lead to linear spectral density at low oscillator frequencies . however , in some cases , the dimensionality of the environment may induce super - ohmic dissipation . here we analyze both the ohmic and super - ohmic dissipations , which are characterized by the spectral densities @xmath158 and @xmath159 respectively , with @xmath160 the dissipative rate for the @xmath9 oscillator and @xmath84 the frequency cut - off of the environmental spectrum . as argued in section [ sla ] , the stationary state is reached in a time scale larger than any of the natural time scales implicit in the dynamics of the open chain ; namely @xmath161 . from now on we set the environmental parameters @xmath162 , @xmath163 , @xmath164 , and the typical values for nanomechanical oscillators @xmath165 and @xmath166 . with this configuration the system begins to exhibit quantumness at temperatures in the range of mili - kelvin . we also assume off resonance oscillators with frequencies , @xmath167 , @xmath168 , and @xmath169 , given in terms of a detuning parameter @xmath170 . we start by analyzing the behavior of two - mode entanglements and the total energy current with the temperature gradient @xmath50 . as figure [ fig1 ] ) shows , the three oscillators become two - mode entangled at low temperature gradients , with this entanglement exhibiting a plateau for negative gradients . interestingly the total current flowing through the oscillator chain presents a similar plateau . this can be related to the proximity of the central oscillator to its ground state at very low temperatures @xmath171 , which is effectively reached for @xmath172 . a similar result for entanglement has been obtained in the study of the influence of heat transport on the two - mode entanglement between oscillators that are embedded in a disordered harmonic chain connected to heat baths at both ends @xcite . it has been shown that a plateau emerges when the energy spectrum is bounded from below since each site of the chain suffers an harmonic potential . as the central oscillator gets closer to the ground state for negative values of @xmath50 , the energy flowing across this oscillator becomes bounded as the temperature gradient decreases . in the absence of the harmonic confinement the logarithmic negativity would continue growing up to a maximum value , as the heat transport decreases @xcite . moreover , the plateau in the entanglement remains even when the average energy current across the chain becomes zero , though the temperature gradient @xmath50 is not zero . this occurs when @xmath173 and the left and right oscillators are identical @xmath174 , and therefore @xmath175 . this last point also underlines that the appearance of entanglement is mainly attributed to proximity of the system to its ground state , rather than to the presence of an energy flow induced by non - equilibrium conditions . the two - mode entanglement rapidly decreases for positive @xmath50 , while the energy current grows monotonically . this is expected as the temperature of the whole three - mode system is increased on average , which is generally harmful for entanglement . previous results have suggested this behavior , in fact it has been shown that in a harmonic chain an increasing @xmath176 is detrimental to build up bipartite or tripartite entanglement due to the rise in the thermal noise @xcite . in addition , it can be shown from ( [ cffdal ] ) that an initially squeezed environmental state effectively increases the temperature . hence , in the present setting an initial squeezed bath state does not favor the appearance of stationary entanglement . moreover , one may expect that non - markovian effects , which are more relevant for super - ohmic dissipation , would substantially degrade the entanglement with respect to the ohmic situation . according to figure ( [ fig1 ] ) , the two - mode entanglement is essentially the same for the chain suffering ohmic or super - ohmic dissipation ; namely , @xmath141 , @xmath177 and @xmath178 practically coincide for both situations . this result is in contrast with the observed transient evolution of the two - mode entanglements under different environmental spectral densities , in which the super - ohmic dissipation induces stronger disentanglement effects @xcite . the coincidence of the stationary two - mode entanglements also differs from the emergence of entanglement in a situation in which the oscillators are affected by the same bath @xcite . in the case of the energy current , we observe that it is strongly affected by the interaction with the heat baths , determined by the fixed spectral density . we have performed an extensive analysis of the two - mode entanglements and the energy currents involving the central oscillator , in terms of both the temperature gradient @xmath50 and the coupling strength @xmath8 . as figure ( [ fig2 ] ) shows , a similar behavior to that illustrated in the figure ( [ fig1 ] ) is reproduced for different values of @xmath8 . the entanglements @xmath177 and @xmath179 increase for lower temperature gradients and stronger couplings . whereas the energy currents @xmath180 and @xmath181 exhibit a relatively weak dependence on the coupling strength , and the expected increase with the temperature gradient . also the plateau of small energy currents arising in the proximity of the ground state of the central oscillator can be clearly observed . hence , our results indicate that the energy currents across the system are insensitive to the emerge of two - mode entanglements between the oscillators , both under ohmic and super - ohmic dissipation , and irrespective of the coupling strength with the heat baths . the two - mode entanglement @xmath141 and total energy current @xmath182 remain nearly unchanged provided that the central oscillator is close enough to the ground state , at temperatures between @xmath183 and @xmath184 . an increase in the temperature of this oscillator results in a deterioration of the entanglement , and an increase in the energy current . and @xmath185 as a function of the temperature gradient for ohmic dissipation . right : the time evolution of the fluctuations of the total current under ohmic dissipation , in a system that is genuine entangled @xmath186 ( black solid line ) , bipartite three - mode entangled @xmath187 ( blue dashed line ) , and likely separable in the three possible bipartitions @xmath188 ( red dot - dashed line ) . the remaining parameters are @xmath189 , @xmath190 , @xmath191 , and @xmath192.,title="fig : " ] and @xmath185 as a function of the temperature gradient for ohmic dissipation . right : the time evolution of the fluctuations of the total current under ohmic dissipation , in a system that is genuine entangled @xmath186 ( black solid line ) , bipartite three - mode entangled @xmath187 ( blue dashed line ) , and likely separable in the three possible bipartitions @xmath188 ( red dot - dashed line ) . the remaining parameters are @xmath189 , @xmath190 , @xmath191 , and @xmath192.,title="fig : " ] in this section we analyze the three - mode entanglement and the energy current correlations between the left and right oscillators , which includes correlation terms involving the three oscillators , see eq.([ccr ] ) . figure ( [ fig3 ] ) shows the bipartite three - mode ( @xmath193 ) and the genuine tripartite ( @xmath194 ) entanglements measured by the corresponding criteria @xmath195 @xcite . the results for both the ohmic and super - ohmic dissipations are quite similar . in the low temperature and strong coupling regime , the three - mode system exhibits genuine tripartite entanglement , though this feature rapidly disappears for positive values of @xmath50 , such as occurs with the two - mode entanglement . strikingly , the system still remains bipartite three - mode entangled at relatively high temperature gradients ( @xmath196 ) . hence the tripartite entanglement is more robust to temperature changes than the two - mode entanglement between the side oscillators . figure ( [ fig3 ] ) also shows the initial time evolution of the energy current correlations for three different three - mode entanglement configurations . as expected , the fluctuations of the energy current exhibit an oscillatory behavior , which should be progressively attenuated at larger time intervals . according to a previously reported exponential time decay of the two - time correlation functions ( [ sleq1 ] ) in a damped harmonic oscillator at low temperature @xmath48 , such oscillations should be effectively suppressed at time @xmath197 @xcite . as evidenced by figure ( [ fig3 ] ) , the energy current correlations exhibit similar oscillations as the system evolves from genuine tripartite to bipartite three - mode entanglement . the most significant discrepancy between these two configurations is an increase in the oscillation amplitude , which can be attributed to the thermal fluctuations that arise with increasing the temperature gradient . indeed , a similar oscillating behavior in the fluctuations is still observed at relatively large temperature gradients @xmath198 , when the system is expected to be separable in the three possible bipartitions . the results we have obtained from an analysis considering an extensive set of parameters @xmath199 corresponding to different multipartite entanglement configurations , for both ohmic and super - ohmic dissipations , also indicate that the energy currents correlations across the harmonic chain are insensitive to the emerge of tripartite genuine or bipartite three - mode entanglement . to conclude this section we focus on the energy current correlations evaluated at equal time , see figure ( [ fig31 ] ) , which will be useful in the subsequent discussion . both the stationary fluctuations and the average of the energy current , see figure ( [ fig2 ] ) , grow with increasing the temperature gradient . but , in contrast to the average energy current , the plateau at small values of the fluctuations is observed above a given value of the coupling strength , which is larger in the case of ohmic dissipation . once again , the fluctuations of the energy current are insensitive to whether the system experiences bipartite three - mode or genuine tripartite entanglement . similar results are obtained for the current - current response involving the central and the side oscillators . one might expect that a scenario similar to the one previously described for the two- and three - mode entanglement would be repeated in the presence of other non - classical correlations , such as discord . in this section we analyze a possible connection between the energy current and the quantum correlations measured by the right - discord @xmath200 . although not shown in this work , similar results are obtained from the analysis of the discord @xmath201 measured from the left . we also point out that the two - mode discord contains the contribution of the two - mode entanglement studied in previous sections . we have found that discord exhibits a strong dependence on both the temperature gradient and the frequencies of the oscillators . as shown in figure ( [ fig4 ] ) , its presents a sharp peak centered at the resonance condition @xmath202 , and with an amplitude that grows with increasing the temperature gradient . interestingly , the discord appears strongly correlated with the mean interaction energy between the oscillators , @xmath203\ , , \label{mei}\ ] ] which has maximum strength also at resonance , see figure ( [ fig4 ] ) . this resonant interaction becomes stronger for higher temperature gradients , which also increase the discord . as all the negative contribution to @xmath204 comes from the crossed correlation terms , it becomes evident that the maximum interaction strength occurs when these correlations take the highest values , which also turns into the optimal conditions for discord . this result evidences an underlying connection between the interaction energy and the discord , as could be anticipated considering that the discord would be zero in the absence of interaction . figure ( [ fig4 ] ) also shows that the discord in the resonant system begins to grow almost linearly with the temperature gradient , and then approaches a constant value at higher gradients . an initially squeezed central reservoir enhances the creation of discord , both for ohmic and super - ohmic dissipation . this is in agreement with the foregoing results , as the squeeze of the initial bath state effectively increases the temperature perceived by the oscillators , see eq.([cffdal ] ) . then an increase of the stationary discord between the side oscillators may be induced either by initially squeezing the central reservoir or increasing its temperature . it has been shown that discord may be additionally created by local noisy operations , such as dissipation @xcite . hence , it may happen that discord would be generated by an energy current induced by a temperature gradient , as this current would make each oscillator to dissipate . considering that the discord contains all the quantum correlations , it would be interesting to analyze whether the entanglement available in the system contributes to its increase with the temperature gradient . at this respect , since entanglement can be only created by non - local manipulations and it becomes zero at high temperature gradients , see section ( [ steec ] ) , we may conclude that such increase of the discord must be mainly due to local operations . the paramount role of the local manipulations in the creation of quantum correlations at resonance conditions is correlated with a maximum average interaction strength between oscillators of similar frequencies , see figure ( [ fig4 ] ) and eq . ( [ mei ] ) . notice that the discord grows with @xmath50 even in the absence of an energy current between the side oscillators ( @xmath173 ) . the plateau of maximum discord at high temperature gradients can be attributed to the very low temperatures of the side oscillators @xmath205 , which guarantees the `` coherence '' of the local manipulations . indeed , the increase in the discord gradually disappear as the mean temperature @xmath48 increases , and therefore the three - mode system approaches to a classic state . we have observed that the discord @xmath200 has almost disappeared at temperature @xmath206 . as expected , the two - mode discord between the central and the side oscillators is enhanced by increasing the interaction strength , see figure ( [ fig5 ] ) . whereas , the trend of the central oscillator towards a classical state , by increasing its temperature through higher values of @xmath50 , causes a progressive deterioration of the discord . in the case of the energy currents between the central and the side oscillators , they exhibit an almost linear increase with the temperature gradient , which is barely modified by the strength of the coupling interaction . once again , the energy currents do not seem to be related to the significant non - classical correlations shared by the oscillators . ( left panel ) and the energy current @xmath207 ( right panel ) between the central and right oscillators in the resonant system under ohmic dissipation . similar results are obtained for the discord and the energy current between the @xmath1- and @xmath0- oscillators , both under ohmic and super - ohmic dissipations . the parameters are the same as figure ( [ fig4]).[fig5],title="fig : " ] ( left panel ) and the energy current @xmath207 ( right panel ) between the central and right oscillators in the resonant system under ohmic dissipation . similar results are obtained for the discord and the energy current between the @xmath1- and @xmath0- oscillators , both under ohmic and super - ohmic dissipations . the parameters are the same as figure ( [ fig4]).[fig5],title="fig : " ] considering that the energy current between two oscillators has an explicit dependence on crossed correlations between them , one could expect that the emerge of entanglement in the system should have detectable effects on such current . therefore , it would be interesting to determine whether a formal connection between entanglement and the average properties of the energy current can be formally established . the results presented in section [ steec ] indicate that the behavior of the energy current is not modified by the presence of two - mode entanglement . a direct comparison between the two - mode entanglement between the side oscillators @xmath208 and the total energy current @xmath182 suggests an elusive correlation between them , see figure [ fig1 ] ) . although the most important contribution to the local currents @xmath209 ( [ ef1 ] ) comes from crossed correlation terms that encode part of the quantum correlations shared by the oscillators , the total current @xmath182 ( [ tec ] ) does not have an explicit dependence on the correlation terms involving the @xmath0- and @xmath2- oscillators . then the two - mode entanglement @xmath208 should not necessarily affect the total current . this reasoning does not exclude , however , the possibility that the total current could be sensitive to the two - mode entanglement shared by the central and the side oscillators . the conjecture that entanglement and energy current are intimately related would lead to the natural question whether the current @xmath210 could serve as an useful witness of the entanglement between the @xmath9 and @xmath24 oscillators . according to the theory of entanglement , an entanglement witness @xmath211 based on a ( bounded ) hermitian operator @xmath212 may be constructed as @xmath213 , where the last term is the infimum value of @xmath212 among all the product states @xmath214 @xcite . for @xmath215 ( [ j1 ] ) it follows @xmath216 where @xmath217 is obtained from ( [ sleq1 ] ) by considering the average over product states . @xmath218 is a good candidate to unveil the two - mode entanglement @xmath140 provide that @xmath219 takes a negative value for at least one entangled state . unfortunately such a rigorous proof requires a closed form expression for @xmath217 , which is currently out of scope as we are dealing with a system under the non - equilibrium conditions induced by two different temperature gradients . though we can not guarantee whether eq.([ewj ] ) constitutes a good estimator of entanglement , the results of section [ steec ] evidence the difficulty of assessing entanglement through @xmath218 , mainly due to the apparent insensitivity of the energy current @xmath220 to the two - mode entanglement @xmath140 , see figure ( [ fig2 ] ) . in addition , though the correlation terms in eq . ( [ j1 ] ) partially carry the quantum correlations shared by the chain oscillators , they themselves do not necessarily manifest entanglement . indeed , the so - called peres - hodorecki - simon inequality @xcite @xmath221 which provides a criterion to detect the two - mode entanglement between the @xmath9 and @xmath24 oscillators , already displays a non - linear relation between entanglement and the elements of the covariance matrix . a common feature of the entanglement measures ( or entanglement monotones ) is their non - linear functional dependence on the density operator @xcite , as occurs in the case of the logarithmic negativity @xcite . in some sense , the reliable observation of entanglement relies on the ability to measure non - linear properties of the quantum state @xcite . according to the expressions for the total energy current @xmath138 ( [ ef1 ] ) and the criterion for entanglement ( [ sc ] ) , the first - moment of the energy current depends linearly on the crossed correlation terms , whereas the entanglement exhibits a non - linear dependence on such terms . hence , the energy current between the @xmath9 and @xmath24 is not expected to manifest the emergence of two - mode entanglement . following the previous argument , the fluctuations of the total current ( [ ccr ] ) could manifest the emergence of entanglement , as it involves correlation terms between all the oscillators . in section [ sccr ] we focused on the tripartite entanglement , and showed that @xmath222 seems to be insensitive to the inseparability properties of the three - mode chain . a similar conclusion was drawn from the comparison of the time evolution of the fluctuations @xmath136 at different entanglement configurations of the stationary state , see figure ( [ fig3 ] ) . we remark that this result is not in contradiction with the previous argument based on criterion ( [ sc ] ) , as it provides a necessary , but not sufficient , condition for the existence of entanglement . considering that , in contrast to entanglement , almost any quantum state has a non - negative discord @xcite , a distinct behavior of these two quantum correlations might be expected @xcite . the results of section ( [ snc ] ) indicate that in the proximity of a resonance condition , a finite energy current induced by the temperature gradient @xmath50 may generate non - classical correlations between the left and right oscillators , even when the total energy current between them becomes zero . this behavior has been correlated with a maximum strength of the average interaction between the harmonic oscillators , see figure ( [ fig4 ] ) . the same results also show that the energy current is not modified by the emerge of discord , see figure ( [ fig5 ] ) . the average properties ( mean value and fluctuations ) of the energy current as a whole exhibit a ` linear ' behavior ruled by the temperature gradients , irrespective of the significant two - mode quantum correlations that may be present in the system . as a measure of such correlations we have analyzed the logarithmic negativity , the tripartite entanglement , characterized by the criteria @xmath223 and @xmath224 , and the quantum discord . finally , although we have focused on a specific system configuration in which each chain oscillator is connected to an independent heat bath , similar results might be expected for other arrangements , since the emergence of quantum correlations in linear harmonic chains or lattices essentially stems from the proximity of the system to the ground state . the study of the energy current and quantum correlations in systems exhibiting non - linearities deserves further attention . we have performed an extensive analysis of the quantum correlations , and the mean value and fluctuations of the energy current across a three - oscillator linear chain at the stationary state , both under ohmic and super - ohmic dissipation . we have considered initially squeezed reservoir thermal states , and applied the gle formalism to determine the correlation functions between the position and momentum operators of the chain oscillators , which completely characterize the stationary properties . interestingly , the results obtained for the quantum correlations are quite similar for both ohmic and super - ohmic dissipation . this suggests that the non - markovian effects do not significantly modify the markovian results for the _ stationary _ properties of the quantum correlations in a system of oscillators in contact with independent heat baths . moreover , the initial squeezing of a heat bath effectively increases the temperature that the oscillator chain perceives from this bath , and eventually becomes detrimental for the build - up of stationary entanglement . a different behavior is observed in the case of discord , which can be created by local noisy operations . these quantum correlations highly depend on both the interaction strength between the oscillators and the detuning of their natural frequencies . in particular , the two - mode discord between the side oscillators in the presence of temperature gradients is enhanced at resonance , which indicates that quantum coherence may be favored by thermal non - equilibrium conditions . according to our results , both the average and the fluctuations of the stationary energy current across the oscillator chain are mainly determined by the two imposed temperature gradients , and do not seem to be related to the appearance of a rich variety of quantum correlations in the system , comprising two - mode discord and entanglement , bipartite three - mode and genuine tripartite entanglement . the absence of quantum correlation effects in the average energy current can be partially understood in terms of its linear dependence on the correlation terms between the oscillators . in the case of the fluctuations , the more intricate dependence on such terms makes more complex to elucidate their possible connection with quantum correlations . nowadays the quantum correlations under thermal non - equilibrium conditions have become a topic of great interest in the fields of quantum information , quantum thermodynamics and the theory of open quantum system . generally the non - classical correlations , such as entanglement , exhibit a non - linear dependence on the density operator that make difficult to establish any formal connection between them and the response of the system to non - equilibrium constrains . we hope that this work may contribute to stimulate further research in this direction . the authors acknowledge fruitful discussions with l.a . correa , j.p . palao and b. witt . this project was funded by the spanish micinn ( grant nos . fis2010 - 19998 and fis2013 - 41352-p ) and by the european union ( feder ) . a.a.v . acknowledges financial support by the government of the canary islands through an aciisi fellowship ( 85% co - financed by the european social fund ) . in this appendix we derive the real and imaginary parts of the fourier transform of the susceptibility @xmath225 ( [ sus ] ) . considering the spectral density ( [ jden ] ) , the imaginary part of the fourier transform @xmath226 can be expressed as @xmath227=\frac{\pi \hbar}{2}\sum_{\mu=1}^{n}\frac{g_{i\mu}^2}{m_{i\mu } \omega_{i\mu}}\,\left[\,\delta\left(\omega-\omega_{i\mu}\right)-\delta\left(\omega+\omega_{i\mu}\right)\,\right]=\hbar\left[\,\theta\left(\omega\right)j_{i}\left(\omega\right)-\theta\left(-\omega\right)j_{i}\left(-\omega\right)\,\right]\ , . \label{appa1}\ ] ] then the real part can be obtained from the causality of @xmath225 , according to @xmath228=\mathcal{h}\left[\,\text{im}\left(\chi_{i}(\omega')\right)\,\right](\omega)\doteqdot \frac{1}{\pi}\,p\int\frac{\text{im}\left[\,\chi_{i}(\omega')\,\right]}{\omega'-\omega}\,d\omega'\,,\ ] ] where @xmath229(\omega)$ ] denotes the hilbert transform of @xmath38 . using eq . ( [ appa1 ] ) it follows that @xmath230=\mathcal{h}\left[\,\theta(\omega')j_{i}(\omega')\,\right](\omega)+\mathcal{h}\left[\,\theta(\omega')j_{i}(\omega')\,\right](-\omega)\ , . \label{appa2}\ ] ] below are given the expressions of the susceptibility for ohmic and super - ohmic spectral densities . assuming the ohmic spectral density ( [ sd1 ] ) , and considering the well - known properties of the hilbert transform , the expression ( [ appa2 ] ) leads to @xmath231&=&\frac{\hbar}{2}\,\pi m \gamma_{i}\left[\,\mathcal{h}\left[\,\theta(\omega')\,\omega'e^{-\frac{\omega'}{\omega_{c}}}\,\right](\omega)+\mathcal{h}\left[\,\theta(\omega')\,\omega'e^{-\frac{\omega'}{\omega_{c}}}\,\right](-\omega)\,\right]\nonumber \\ & = & \frac{\hbar}{2}\,\pi m \gamma_{i}\,\omega\left[\,\mathcal{h}\left [ \theta(\omega')\,e^{-\frac{\omega'}{\omega_{c}}}\,\right](\omega)-\mathcal{h}\left[\,\theta(\omega')\,e^{-\frac{\omega'}{\omega_{c}}}\,\right ] ( -\omega)\,\right ] + \hbar\,m \gamma_{i}\,\omega_{c } \nonumber \\ & = & \frac{\hbar}{2}\,m \gamma_{i}\,\omega\left[\,e^{-\frac{\omega}{\omega_{c}}}\,\gamma(0,-\omega/\omega_{c } ) -e^{\frac{\omega}{\omega_{c}}}\,\gamma(0,\omega/\omega_{c})\,\right ] + \hbar\,m\gamma_{i}\,\omega_{c}\,,\end{aligned}\ ] ] where @xmath232 denotes the incomplete gamma function . following the same procedure as in the previous section for the super - ohmic spectral density ( [ sd2 ] ) , one obtains @xmath231&=&\frac{\hbar}{2\omega_{c}}\,\pi m \gamma_{i}\left[\,\mathcal{h}\left[\,\theta(\omega')\,(\omega')^2\,e^{-\frac{\omega'}{\omega_{c}}}\,\right](\omega)+\mathcal{h}\left[\,\theta(\omega')\,(\omega')^2\,e^{-\frac{\omega'}{\omega_{c}}}\,\right ] ( -\omega)\,\right]\nonumber \\ & = & \frac{\hbar}{2\omega_{c}}\,\pi m\gamma_{i}\,\omega^2\left[\,\mathcal{h}\left[\,\theta(\omega')\,e^{-\frac{\omega'}{\omega_{c}}}\,\right ] ( \omega)+\mathcal{h}\left[\,\theta(\omega')\,e^{-\frac{\omega'}{\omega_{c}}}\,\right](-\omega)\,\right ] + \hbar m \gamma_{i}\,\omega_{c } \nonumber \\ & = & \frac{\hbar}{2\omega{c}}\,m \gamma_{i}\,\omega^2\left[\,e^{-\frac{\omega}{\omega_{c}}}\gamma(0,-\omega/\omega_{c})+e^{\frac{\omega}{\omega_{c}}}\gamma(0,\omega/\omega_{c})\,\right ] + \hbar m \gamma_{i}\,\omega_{c}\,.\end{aligned}\ ] ] in this appendix we derivate the correlation function of the fluctuation forces associated with the heat baths given in eq.([cffdal ] ) . considering the time dependence of these forces , it follows @xmath233 replacing the identities ( [ cmhb ] ) in eq.([aff1 ] ) , and applying the fourier transform , one obtains @xmath234 \int\!\!\int dt\ , dt ' e^{i\omega t}\,e^{i\omega ' t ' } \cos(\omega_{i\mu}(t+t'-2t_{0 } ) ) \nonumber \\ & + & \text{im}\left[m(\omega_{i\mu})\right]\int\!\!\int dt\,dt ' e^{i\omega t}\,e^{i\omega ' t'}\sin(\omega_{i\mu}(t+t'-2t_{0}))\big)\ , . \nonumber \end{aligned}\ ] ] to compute the previous integrals we express the trigonometric functions as complex exponentials , introduce the change of variable @xmath235 in the first integral , and @xmath236 in the second and third ones . secondly , we use the definition of the delta function @xmath237 and the identity @xmath238 , which lead to @xmath239\coth\left ( \frac{\hbar \omega}{2k_{b } t_{l}}\right)\cosh(2r_{l } ) \nonumber \\ & -&2\pi\hbar\delta(\omega-\omega')\coth\left ( \frac{\hbar \omega}{2k_{b } t_{l}}\right)\sinh(2r_{l})\text{re}\left[e^{i \theta_{l}}\right]\bigg(\int_{0}^{\infty}d{\bar\omega } { \bar j}_l(\omega-{\bar\omega})e^{2i{\bar\omega } t_{0 } } - \int_{0}^{\infty}d{\bar\omega } { \bar j_l}(\omega+{\bar\omega})e^{-2i{\bar\omega } t_{0 } } \bigg ) \nonumber \\ & + & 2i \pi\hbar\delta(\omega-\omega ' ) \coth\left ( \frac{\hbar \omega}{2k_{b } t_{l}}\right)\sinh(2r_{l})\text{im}\left[e^{i \theta_{l}}\right ] \bigg(\int_{0}^{\infty}d{\bar\omega } { \bar j}_l(\omega-{\bar\omega})e^{2i{\bar\omega } t_{0 } } + \int_{0}^{\infty}d{\bar\omega } { \bar j}_l(\omega+{\bar\omega})e^{-2i{\bar\omega } t_{0}}\bigg)\ , , \nonumber \\ \label{cffd}\end{aligned}\ ] ] with @xmath240 . before to continue , we pay attention to the four integrals having an explicit dependence on the initial time @xmath96 . for both the ohmic and super - ohmic spectral densities , ( [ sd1 ] ) and ( [ sd2 ] ) , the corresponding @xmath241 decays more rapidly than @xmath242 at high frequencies . this allows us to use the riemann - lebesgue lemma which states @xcite @xmath243 for @xmath244 an absolutely integrable function in the interval @xmath245 . as a consequence , only the first term in eq.([cffd ] ) survives after taking the long - time limit @xmath30 . the riemann - lebesgue lemma has been successfully employed in the study of the properties of the stationary state for the damped harmonic oscillator @xcite . in the following we briefly describe the computation of the logarithmic negativity , quantum discord , and the separability criteria @xmath145 . the evaluation of the logarithmic negativity ( [ logneg ] ) requires the symplectic eigenvalues of the partial transpose @xmath246 . these are given by the positive square roots of the eigenvalues of @xmath247 @xcite , which are given in terms of the so - called symplectic matrix @xmath248 where @xmath249 is the @xmath144-dimensional unit matrix . the corresponding partial transpose matrix is obtained from @xmath250 , with @xmath251 the evaluation of the quantum discord ( [ disc ] ) on the state @xmath60 involves an optimization procedure over all positive operator - valued measurements ( povms ) on the @xmath252-mode , denoted by @xmath253 . in eq . ( [ disc2 ] ) , @xmath254 is the probability associated with the @xmath88 measurement outcome , and @xmath255 is the corresponding post - measurement reduced state of the @xmath100-mode . an explicit formula providing @xmath256 for any input gaussian state @xmath257 is given in @xcite . @xmath258 can be obtained from a similar optimization procedure on the povmss in the subsystem @xmath100 . in general , both evaluations of discord may return different values ( i.e. @xmath259 ) . now we describe a hierarchy of separability criteria recently proposed to characterize from genuine multipartite to bipartite entanglement @xcite . according to this proposal , the state @xmath260 of a @xmath144-partite system is @xmath143-partite entangled , _ i.e. _ there is at least a entangled subsystem composed of @xmath143 parties , provide that a given function @xmath261 takes a positive value . the evaluation of the function @xmath262 involves the selection of a set of @xmath263 pure states that allows to assess multipartite entanglement . the important point is that a reliable characterization of entanglement requires an appropriate choice of such probe states . however , a priori there is no information about the ` optimal ' probe states enable to unveil the entanglement encapsulated by an arbitrary density operator @xmath260 . one may circumvent this difficulty by carrying out an optimization procedure over all possible selections in order to obtain the maximum of @xmath262 , whose positive value would reveal the entanglement in the system . we denote @xmath145 such maximum . in continuous - variable states a gaussian selection of the probe states provides a readable expression of @xmath262 @xcite , which can be optimized with standard procedures , and which is strong enough to detect entanglement for a broad class of gaussian and non - gaussian states . in the gaussian case this expression reads @xcite , @xmath264 where @xmath265 are constant values @xcite , @xmath266 is a real @xmath263-vector , @xmath267 is the standard form of the so - called symplectic matrix , and @xmath268 and @xmath269 are @xmath270 real matrices . the objects @xmath266 and @xmath269 denote a compact form of the first and second moments of the probe states @xcite . hence , the detection of entanglement consists basically in optimizing eq.([stau ] ) over the variables @xmath266 and @xmath269 , _ i.e. _ @xmath271 69 ifxundefined [ 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

we present an all - sky catalog of m dwarf stars with apparent infrared magnitude j . the 8,889 stars are selected from the ongoing superblink survey of stars with proper motion mas yr , supplemented on the bright end with the tycho-2 catalog . completeness tests which account for kinematic ( proper motion ) bias suggest that our catalog represents of the estimated m dwarfs with j expected to populate the entire sky . our catalog is , however , significantly more complete for the northern sky (% ) than it is for the south (% ) . stars are identified as cool , red m dwarfs from a combination of optical and infrared color cuts , and are distinguished from background m giants and highly - reddened stars using either existing parallax measurements or , if such measurements are lacking , on their location in an optical - to - infrared reduced proper motion diagram . these bright m dwarfs are all prime targets for exoplanet surveys using the doppler radial velocity or transit methods ; the combination of low - mass and bright apparent magnitude should make possible the detection of earth - size planets on short - period orbits using currently available techniques . parallax measurements , when available , and photometric distance estimates are provided for all stars , and these place most systems within 60 parsecs of the sun . spectral type estimated from v - j color shows that most of the stars range from k7 to m4 , with only a few late m dwarfs , all within 20 pc . proximity to the sun also makes these stars good targets for high - resolution exoplanet imaging searches , especially if younger objects can be identified on the basis of x - ray or uv excess . for that purpose , we include x - ray flux from rosat and fuv / nuv ultraviolet magnitudes from galex for all stars for which a counterpart can be identified in those catalogs . additional photometric data include optical magnitudes from digitized sky survey plates , and infrared magnitudes from 2mass .

1108.2719

m dwarfs are main sequence stars whose spectra display bands of tio and other molecules such as cah , vo , feh , and crh @xcite . the high molecular opacities are the result of a cool atmosphere with effective temperatures in the range 2400k-3700k @xcite . observed orbital motion of astrometric binaries @xcite show that m dwarfs have masses ranging from @xmath50.5 m@xmath6 all the way down to the hydrogen - burning limit ( @xmath50.075 m@xmath6 ) . m dwarfs have relatively small sizes , with radii roughly proportional to their masses . recent measurements of eclipsing systems @xcite indicate a roughly linear relationship between masses and radii , with a 0.4 m@xmath6 ( 0.1 m@xmath6 ) m dwarf having a radius of 0.4 r@xmath6 ( 0.15 r@xmath6 ) . by number , m dwarfs form the bulk of the stars in our galaxy . the census of stars within 33 parsecs of the sun @xcite shows m dwarfs outnumbering all other hydrogen - burning objects by a factor of three to one . because of their low mass - to - light ratio , m dwarfs contribute only a small fraction of the light emission in galaxies , but they may be the dominant baryonic component in at least some galaxies @xcite . in recent years there has been an increasing interest in m dwarfs due to the discovery that they too host exoplanets . as of january 2011 , 25% of the 41 doppler - confirmed planets with @xmath730m@xmath8 are orbiting m dwarfs . the geneva / harps planet search team reports that @xmath930% of low - mass stars harbor planets with masses @xmath030@xmath10 on orbits shorter than 50 d @xcite and preliminary _ kepler _ results suggest that m dwarfs host just as many , if not more , super - earths and neptunes as solar - mass stars @xcite . exoplanet candidates have been identified by _ kepler _ in stars having colors consistent with m dwarfs @xcite , despite the fact that very few m dwarfs are being targeted by the mission @xcite . large exoplanet surveys have now started to monitor sizable numbers of m dwarfs , such as the m2k program which is targeting some 1,600 m dwarfs for radial velocity monitoring @xcite , and the mearth project @xcite which is designed to detect exoplanet transits in nearby late - type m dwarfs . the two principal methods of exoplanet detection , doppler measurement of radial velocity ( rv ) and photometric detection of transits , are more sensitive to planets around stars of lower mass . for a given planet mass and orbital period , rv amplitude scales as @xmath11 . in contrast to late - type m stars , the vast majority of early m dwarfs are not rapid rotators or chromospherically active ( w@xmath12 ) @xcite and are thus amenable to these search techniques . in fact , inactive k and early m stars have weaker p - mode oscillations and lower doppler noise ( `` jitter '' ) than their f or g counterparts @xcite . photometric transit depth scales approximately as @xmath13 for planets of similar size . because of their significantly smaller radii , m dwarfs display deeper transits which do not require high precision photometry for detection , as they do for g stars @xcite . another method of planet detection is high - contrast imaging at infrared wavelengths where a ( giant ) planet is self - luminous @xcite . m dwarf systems present a much more favorable planet / star contrast ratio compared to solar - mass stars , due to the significantly lower luminosity of m dwarfs . young systems are of particular interest because the luminosity of giant planets is predicted to be orders of magnitude higher at ages of @xmath14 yr @xcite providing a more favorable contrast , even as the host star itself is moderately more luminous at this younger age . young solar mass stars can be identified by their enhanced rotation , dynamo activity , and chromospheric emission @xcite , although the trend for late m stars for which the mechanism of dynamo operation is thought to be different is less clear @xcite . another challenge for this approach is the apparent rarity of giant planets on long - period orbits around solar - mass stars @xcite combined with the paucity of giant planets around m dwarfs compared to g dwarfs @xcite . it is also possible that young giant planets may be less luminous and more difficult to detect than previously predicted @xcite . low luminosity m dwarfs are also attractive targets because planets within their diminutive habitable zones are easier to detect @xcite . the correlation between high metallicity and the presence of giant planets found among solar - mass stars @xcite has been recently extended to m dwarf stars @xcite , and metal - rich m dwarfs , if they can be identified @xcite , would be especially propitious targets for planet searches . despite the attractiveness of m dwarfs for exoplanet surveys , there still is no systematic , all - sky census of these objects to a useful limiting magnitude . part of the problem is that the low - mass m dwarfs are stars of relatively low luminosities ( m@xmath15 ) . all but the closest m dwarfs are too faint ( v@xmath912 ) to be part of the _ hipparcos _ catalog @xcite , which means that existing m dwarf samples are selected from proper motion catalogs with fainter magnitude limits . these include the tycho-2 catalogue @xcite , which extends the hipparcos catalog a few magnitudes fainter , and also the older and deeper luyten half second ( lhs ) and new luyten two - tenths ( nltt ) catalogs @xcite which were the primary source material for the well - known _ catalog of nearby stars _ @xcite . newer and more complete proper motion catalogs , such as the lspm - north @xcite have the potential to expand the list of available m dwarf targets . in this paper , we present the first all - sky catalog of bright ( j@xmath010 ) m dwarfs . the sample is extracted from the all - sky superblink proper motion catalog of stars with @xmath140 mas yr@xmath2 , an extension of lspm - north currently in development . the selection method is detailed in 2 . the catalog itself is presented in 3 . the completeness of the catalog is evaluated in 4 . conclusions follow in 5 . the superblink proper motion survey is an all - sky search for stars with large proper motions based on a re - analysis of images from the digitized sky surveys ( dss ) using a specialized image - differencing algorithm and software . the basic search method is described in @xcite . quality control procedures , including cross - correlation with other catalogs and the compilation of astrometric and photometric results , is discussed at length in @xcite . stringent quality control procedures , including visual confirmation of most objects , guarantees a very low level of false detections . the survey has a fixed , low proper motion threshold of @xmath16 mas yr@xmath2 . new stars with very large proper motions ( @xmath17 mas yr@xmath2 ) discovered in the survey have been published in @xcite , @xcite , @xcite , and @xcite . a complete list of northern stars with proper motions @xmath18 mas yr@xmath2 was published in @xcite ; a similar list covering the southern sky has been completed and is available from the authors . those complete lists incorporate data from the hipparcos and tycho-2 catalogs at the bright end , and the catalogs are estimated to be @xmath19 complete to a visual magnitude @xmath20 , and extend to @xmath21 . the completeness is significantly higher at high galactic latitudes , where field crowding is less of an issue . dense fields at low galactic latitudes have a lower completeness due to faint stars often being lost to non - linearity in the crowded areas of photographic plates ; saturation also prevents the detection of proper motion sources in the extended glare ( up to several arc minutes ) of extremely bright stars ( e.g. vega , sirius , canopus ) . as of july 2011 , the full superblink catalog ( stars with proper motions @xmath16 mas yr@xmath2 ) comprised 2,270,481 objects , with the most extensive sky coverage north of decl.=-20@xmath22 . the catalog was complete for the northern sky ( decl.@xmath90@xmath22 ) , but the survey was still in progress for large parts of the southern sky . the southern survey was however complete for stars with proper motions @xmath18 mas yr@xmath2 . this version of the superblink catalog is the one used here . the use of a proper motion catalog is a convenient way to identify nearby stars in general , and m dwarfs in particular , because proper motion - selected samples discriminate against distant background sources including m giants that would otherwise significantly contaminate a sample of cool stars selected by color alone . however , proper motion selection carries its own inherent bias : stars with small velocities in the plane of the sky tend not to be detected in proper motion surveys , depending on the distance to the source . nearby , slow - moving stars can be included by lowering the proper motion limit of the sample , but this has the effect of increasing the contamination from background sources . there is a fine balance to achieve in adjusting the proper motion threshold , which would ideally have the sample incorporate the largest possible fraction of stars up to a specified distance , minimizing the kinematic bias , while at the same time limiting the contamination from background objects . the superblink catalog , with a proper motion limit @xmath16 mas yr@xmath2 , achieves such a balance for stars located within 100 parsecs of the sun . this can be demonstrated with an analysis of proper motion in the _ hipparcos _ catalog . figure 1 shows the fraction of hipparcos stars at a given distance from the sun which have proper motions @xmath18 mas yr@xmath2 and @xmath16 mas yr@xmath2 . the former value is the proper motion limit of the lspm - north and south catalogs ( which is also the approximate proper motion limit of the nltt catalog and , by extension , the catalog of nearby stars ) , while the latter is the proper motion limit of the extended superblink catalog . the diagram shows that operating with a proper motion limit of @xmath18 mas yr@xmath2 will only detect half the stars at 40pc and very few stars ( only those with very large components of motion ) at 100pc . however a sample with a proper motion limit @xmath16 mas yr@xmath2 will include @xmath595% of the stars at 40pc and @xmath570% of the stars at 100pc . background giant stars , on the other hand , tend to have small proper motions , which exclude them from the superblink catalog . figure 2 shows the distribution of proper motions for red giant stars in the hipparcos catalog , selected with the color and absolute magnitude cuts @xmath231.0 , @xmath243.0 . the bulk of the giants have proper motions below the superblink threshold . this particular subset of giants , because it is from the hipparcos catalog , also represents the nearest red giants to the sun , since they all have magnitudes @xmath25 . more distant giants would have even smaller proper motions , and would also be excluded from our sample . figure 3 shows the distribution of optical - to - infrared @xmath26 colors for a subset of m dwarfs with spectral classification from the sloan digital sky survey database @xcite . infrared @xmath27 magnitudes are from 2mass , while @xmath28 magnitudes are the superblink visual magnitudes , estimated from palomar sky survey photographic plate measurements as described in @xcite . the subset includes all stars in sdss which have counterparts in the superblink proper motion catalog . for this subset of nearby m dwarfs , @xmath26 color ranges from @xmath29 at subtype m0 , to @xmath30 at subtype m9 . we adopt an m dwarf color - cut of @xmath31 , which is expected to include most stars of subtype m0 and later , at the expense of some contamination from late - k stars . in addition , we select stars that are bright enough to be monitored by current and planned doppler radial velocity instruments . stars as faint as @xmath32 are routinely monitored using the high resolution spectrometer ( hires ) on the 10 m keck i telescope , e.g. @xcite , and fainter but highly meritorious stars with transiting planets , e.g. gj 1214 @xcite , have been observed . m dwarf stars are brighter in the near - infrared , and future high - resolution spectrometers should be able to detect planets around stars as faint as @xmath33 . anticipating such developments , we select stars with @xmath34 , which also includes all m dwarfs with @xmath35 . to summarize , we select bright m dwarfs among high proper motion stars ( @xmath16 mas yr@xmath2 ) in the superblink catalog using these color / magnitude cuts : @xmath36 figure 4 shows the full magnitude - color distribution of superblink stars , and the region targeted by our selection . out of 2,270,481 stars in the superblink catalog , our color and magnitude cuts select 10,659 bright red stars . this we use as our initial sample to identify bright m dwarfs . the sky distribution of the color - selected stars is shown in figure 5 . separate plots show the stars listed in the tycho-2 and hipparcos catalogs ( top ) , stars from the superblink catalog which are not listed in tycho-2 or hipparcos _ and _ have proper motions @xmath1 150 mas yr@xmath2 ( center ) , and finally stars from the superblink catalog not listed in tycho-2 or hipparcos _ and _ with proper motions 40 mas yr@xmath37 150 mas yr@xmath2 ( bottom ) . the subsample from the tycho-2 and hipparcos catalogs includes 3,319 of the brighter stars , spread over the entire sky . their distribution shows a slight overdensity in the direction of the galactic bulge , which is likely due to the larger number of red giants found in that part of the sky . the very high proper motion ( @xmath1 150 mas yr@xmath2 ) subsample from superblink includes 4,398 stars , and extends uniformly over the entire sky . the moderately high proper motion ( 40 mas yr@xmath37 150 mas yr@xmath2 ) subsample from superblink includes 2,941 stars , and clearly suffers from incompleteness in the southern hemisphere ; the subsample also displays an overdensity around ( @xmath38,@xmath39)@xmath5(80,+20 ) which is associated with the nearby hyades cluster . giant stars constitute the main source of contaminant in the sample . indeed , an excess of red sources in the apparent magnitude range @xmath40 ( see figure 4 ) points to a modest population of very red m giants in the superblink catalog ( probably including agb stars ) . while these giants are only a tiny fraction of the full superblink catalog , they are systematically selected in our initial color / magnitude cut . to eliminate these giants from our sample we apply three selection criteria . the first is based on evaluation of absolute magnitudes , and is used on stars with existing parallax data . the second method identifies giants from their location in a reduced proper motion diagram , and is used for stars for which there are no existing parallax measurements . a third method selects for m dwarfs based on @xmath41 and @xmath42 colors , further eliminating giants and interlopers with bad optical and/or infrared photometry . we searched the literature to find astrometric parallaxes for stars in our subsample of m dwarf candidates . parallaxes for 1,622 stars were recovered from the updated version of the hipparcos catalog @xcite . ground - based astrometric parallaxes were additionally recovered for 10 stars in @xcite , 46 stars in @xcite , 15 stars in @xcite , 6 stars in @xcite , 3 stars in @xcite , 1 star in @xcite , 2 stars in @xcite , 18 stars in @xcite , 4 stars in @xcite , 14 stars in @xcite , 15 stars in @xcite , 3 stars in @xcite , and 3 stars in @xcite . whenever parallaxes were found for one star in more than one bibliographical source , we adopted the most recent measurement , following the order above . the numbers quoted above refer to incremental additions to the list of parallaxes . in addition , we found parallaxes for 3 stars in the sky2000 master catalog , version 4 @xcite , and for 315 more from the nstars database . in all cases we only retained parallax measurements with errors @xmath020% . the final tally comprises a total of 2,080 stars with astrometric parallaxes . from the parallax @xmath43 we calculate absolute visual magnitudes using : @xmath44 the resulting hertzsprung - russel ( h - r ) diagram is displayed in figure 6 . the stars populate two very distinct loci . the more luminous group includes all m giants and the less luminous group consists of all the m dwarfs . for stars with @xmath29 , the giants are typically 7 - 8 magnitudes more luminous than the dwarfs . the difference in luminosity only increases at redder colors , with the reddest giants almost 20 magnitudes more luminous than dwarfs of similar @xmath26 color . from the absolute magnitude @xmath45 and optical - to - infrared color @xmath26 , we define the following criterion for a star to be considered an m dwarf : @xmath46 this cut identifies 561 stars as most likely to be red giants . the remaining 1,519 objects are identified as m dwarfs . we emphasize that this selection works only for the @xmath47 stars in our initial sample for which astrometric parallax measurements are available . note that the color - magnitude relationship for the m dwarfs in our sample closely follows : @xmath48 which implies that we are eliminating stars that are more than 4.5 magnitudes overluminous compared to the average m dwarf of the same color . this cut could possibly eliminate some very young m dwarfs , which tend to be overluminous in an optical - to - infrared color - magnitude diagram @xcite . indeed a number of stars with parallaxes appear to be a few magnitudes over - luminous compared to average field stars . however a close examination of figure 6 indicates that our magnitude cut would , in the worst case , eliminate only a handful of objects that could have been overluminous dwarfs . for stars without parallax measurements , it is possible to identify giants based on their location in a reduced proper motion diagram . the reduced proper motion is defined as : @xmath49 a quantity which is analogous to the absolute magnitude @xmath45 ( eq.3 ) except that the proper motion @xmath50 is substituted for the parallax @xmath43 . the two quantities are mathematically connected , with : @xmath51 where @xmath52 is the projected motion in the plane of the sky , i.e. the transverse velocity , here expressed in km s@xmath2 . this means that the reduced proper motion diagram is basically an h - r diagram where the ordinates are modulated by the individual components of transverse velocities . just as in the h - r diagram , m giants and m dwarfs tend to occupy distinct loci , with the giants all having relatively low values of @xmath53 and clustering in the upper part of the diagram . we demonstrate this in the upper panel of figure 7 , which shows the reduced proper motion diagram of the stars in our sample for which there are parallax measurements . stars identified as giants based on their parallax are plotted in red , dwarfs are plotted in black . we define a demarcation line which separates stars of the two groups quite well . from this , we postulate that m dwarfs can be separated from giants following this criterion : @xmath54 because the reduced proper motion depends on both absolute magnitude _ and _ transverse velocity , our selection may eliminate some main sequence m dwarfs if their transverse motion is small . based on equations 5 and 7 , we see that stars on the mean main sequence will be eliminated if their transverse velocity @xmath55 km s@xmath2 . given that the velocity dispersion for stars in the solar neighborhood is @xmath540 km s@xmath2 @xcite , our adopted criterion will exclude relatively few m dwarfs . conversely , giants with large transverse velocities will also have large values of @xmath53 which may place them beyond the limit defined in equation 8 . among the parallax - confirmed giants , we find 4 stars which fall within our defined reduced proper motion cut for m dwarfs . one of the stars has a inferred transverse velocity @xmath56 km s@xmath2 based on its hipparcos parallax and proper motion . however the other three stars have suspiciously faint absolute magnitudes ( [email protected] ) for m giants , and may instead be m dwarfs with erroneous ( underestimated ) parallaxes measurements . the reduced proper motion diagram of stars with no parallax measurements is shown in the bottom panel of figure 7 . of the 8,580 stars with proper motions but no parallax measurements , we identify 626 as probable m giants , and those are eliminated from the sample . the remaining 7,954 stars are retained as m dwarf candidates . this leaves 9,473 stars , with or without parallax , on the list of m bright m dwarf candidates . a significant number of sources are found to have very red colors ( @xmath585 ) and reduced proper motion in the range typical for dwarf stars ( @xmath5912 ) and yet fall above the line we defined as the giant - dwarf boundary . a close examination of their infrared colors shows that most of these stars fall outside the normal range for m dwarfs ( see below ) and are likely either stars with erroneous ( overestimated ) v magnitudes , or distant f - g - k stars with large components of reddening . most of these stars are eliminated after the infrared cuts described below . giant stars are notable for having infrared colors different from the m dwarfs , most apparent in a [ ( @xmath41),(@xmath60 ) ] color - color diagram . this has been known since @xcite . dwarf stars have comparatively bluer @xmath41 colors than giants . this is due to opacity differences , mostly from molecular bands of h@xmath61o , which yield significantly higher @xmath62-band and @xmath63-band opacities in m dwarfs @xcite . figure 8 shows the [ ( @xmath41),(@xmath60 ) ] color - color distribution for the stars in our initial sample . stars with parallax measurements are shown in the upper panel , stars without parallax in the lower panel . in both panels , all objects we have identified as probable giants , based on either absolute magnitude ( 2.4.1 ) or reduced proper motion(2.4.2 ) , are plotted in red . the remaining m dwarf candidates are plotted in black . the giants are clearly segregated from the dwarfs along j - h , although there is a small overlap between the two groups . in fact m dwarfs appear to be confined to a relatively compact region around @xmath64 and @xmath65 . a small number of alleged m dwarfs fall outside the cluster ; some have red @xmath41 colors more consistent with the m giants , and indeed they most probably are . other stars have bluer colors which are more typical of f - g - k stars ; we suspect these probably are more massive main sequence stars which made it into our sample probably due to an overestimated photographic magnitude @xmath28 , which would also overestimate the @xmath26 color term , bringing them into the m star subsample . in order to eliminate these leftover misidentified m giants and misclassified f - g - k stars , we further restrict our m dwarf selection to the region in [ ( @xmath41),(@xmath60 ) ] diagram bounded by : @xmath66 among stars with trigonometric parallaxes , we find that 97 of the m dwarf candidates ( black dots in figure 8 upper panel ) fall outside the limits of the region defined above . we conclude that these stars likely have significant errors in their estimated optical or infrared magnitudes , and we exclude them from our census of m dwarfs . among stars with no parallax measurements identified as probable m dwarfs from the reduced proper motion cut ( black dots in figure 8) , we find 487 which fall outside the defined m dwarf boundaries ; these stars are eliminated as well . some of the stars have color consistent with f - g - k main sequence stars ( [email protected] , [email protected] ) and most likely made the initial @xmath582.7 color cut due to an underestimated v magnitude . other excluded sources have @xmath690.7 , and are either giants or distant main sequence stars with large reddening . after these cuts , we are left with a list of 8,889 probable bright m dwarfs , which we retain for building our catalog . among the giants confirmed through parallax ( stars plotted in red in the upper panel of figure 8) a small number fall within the selection box for m dwarfs . specifically , 66 of the 561 giants fall within the region defined by equations 9 - 13 . this means that jhk color cuts alone are susceptible to contamination by background giants . this makes the reduced proper motion selection critical in minimizing giant contamination . note that only 4 of the 561 giants have reduced proper motions within the limit defined for m dwarfs , which increases our confidence that giant contamination in our catalog is negligible . the reduced proper motion cut , on the other hand , may be eliminating bona fide m dwarfs , as mentioned in 2.4.2 above . the lower panel in figure 8 show evidence for this , as 160 of the 626 stars excluded after the reduced proper motion cut ( plotted in red ) do fall within the m dwarf box . overall this suggests that our adopted selection criteria places more of an emphasis on avoiding contamination by giants , at the expense of some incompleteness in the m dwarf targets . from the numbers above , we predict that @xmath70100 m dwarfs might be missing from our catalog because they fail the reduced proper motion cut . in addition , we note that stars identified as giants based on parallax data ( figure 8 top panel , red dots ) and stars excluded from the reduced proper motion cuts ( figure 8 bottom panel , red dots ) show significantly different distributions in @xmath41/@xmath60 color - color space . one average the parallax giants tend to have bluer @xmath42 colors . we suggest that this is due to the fact that most red giants with @xmath140 mas yr@xmath2 are bright enough to have parallaxes from the hipparcos catalog , and that the reduced proper motion cut for stars with no parallax essentially serves to exclude stars with large components of reddening , which are expected to have redder @xmath42 colors than the giants . lllrrrrrrrc pm i00001@xmath716943 & & & & 0.028531 & 69.717117 & 0.136 & 0.136 & -0.002 & & + pm i00003@xmath720802s & & & & 0.087698 & -8.037146 & 0.100 & 0.029 & -0.096 & & + pm i00005@xmath720533 & & & & 0.144895 & -5.551960 & 0.193 & 0.181 & 0.067 & & + pm i00006@xmath711829 & & tyc 1181 - 1683 - 1 & & 0.163528 & 18.488850 & 0.387 & 0.335 & 0.195 & & + pm i00007@xmath711624 & & & & 0.195877 & 16.402805 & 0.125 & 0.013 & -0.125 & & + pm i00007@xmath723510 & & & & 0.195306 & -35.168343 & 0.316 & 0.312 & -0.048 & & + pm i00007@xmath726243 & & & & 0.189068 & -62.728802 & 0.199 & 0.166 & -0.110 & & + pm i00012@xmath711358s & & tyc 600 - 1507 - 2 & & 0.303578 & 13.972055 & 0.147 & 0.025 & 0.144 & [email protected] & va95 + pm i00014@xmath714724 & & & gj 1293 & 0.371546 & 47.414665 & 0.172 & 0.171 & -0.008 & & + pm i00014@xmath721656 & hip 112 & tyc 5838 - 784 - 1 & & 0.357633 & -16.948410 & 0.393 & 0.299 & -0.255 & [email protected] & vl07 + pm i00015@xmath720814 & & & & 0.399365 & -8.244875 & 0.124 & 0.097 & -0.077 & & + pm i00016@xmath723258 & & & & 0.408577 & -32.978249 & 0.238 & 0.238 & 0.005 & & + pm i00016@xmath727613 & & & & 0.408879 & -76.230522 & 0.188 & 0.188 & 0.008 & & + pm i00017@xmath723528 & & & & 0.433436 & -35.476082 & 0.502 & 0.502 & -0.023 & & + pm i00024@xmath710440 & & & & 0.613839 & 4.668418 & 0.122 & 0.058 & 0.105 & & + pm i00024@xmath724601 & hip 191 & tyc 8022 - 637 - 1 & & 0.612922 & -46.028893 & 0.194 & 0.192 & -0.022 & [email protected] & vl07 + pm i00025@xmath726324 & & & & 0.639251 & -63.401974 & 0.822 & -0.488 & -0.661 & & + pm i00026@xmath713821 & & & & 0.667151 & 38.362587 & 0.075 & -0.073 & -0.020 & & + pm i00026@xmath721123 & & & & 0.654399 & -11.393147 & 0.584 & -0.413 & -0.413 & & + pm i00026@xmath723919 & & & & 0.654961 & -39.320053 & 0.200 & 0.199 & 0.023 & & + each of the 8,889 stars in our catalog is identified with the standard proper motion star i d used in the superblink proper motion survey . these superblink names are tabulate in the first column of both table 1 and table 2 . the first three letters ( `` pmi '' ) are the superblink catalog identifier . these are followed by 5 digits which relate to the icrs right ascension of the star in sexagesimal ; the first four digits are the hours and minutes of right ascension , the fifth digit is the seconds of right ascension divided by 6 and rounded down to the nearest integer . these are followed by the declination sign and then four digits which replicate the icrs declination ( hours , then minutes ) of the source . a disambiguation letter ( nsew ) is used when the scheme would leave two stars with the same name , with the letter indicating the relative orientation of the stars in the pair . this scheme follows the recommendation of @xcite and is the convention now adopted in reporting results from the superblink survey , see e.g. @xcite . in table 1 , we additionally provide the names of the stars in both the hipparcos ( column 2 ) and tycho-2 ( column 3 ) catalogs , for all stars which have counterparts in those catalogs . these are provided for convenience because hipparcos names are now commonly used to refer to very bright stars , and because many exoplanet surveys have adopted the hipparcos names in designating the host stars and their exoplanets . the tycho-2 names perhaps do not enjoy the same popularity , but are useful in identifying objects for which the proper motion is known to higher accuracy . our list includes a total of 977 hipparcos and 1,859 tycho-2 stars . some 779 entries have counterparts in both the hipparcos and tycho-2 catalogs , but 192 stars have hipparcos counterparts only , while 1,074 stars are only listed in tycho-2 . in addition , we identify by their catalog names all the stars listed in the third catalog of nearby stars @xcite or cns3 ( column 4 ) . these are also convenient because cns3 names have been widely adopted by the exoplanet community to designate nearby stars . the cns3 counterparts are found after cross - correlation with the catalog of revised positions by @xcite which lists 4,106 objects . a total of 1,529 stars from our catalog of bright m dwarfs are thus matched to their cns3 counterpart . stars with cns3 counterparts include most of the very nearest systems . however , the cns3 does suffer from a significant kinematic bias because it is largely based on the proper motion catalogs of luyten , particularly the lhs catalog which has a relatively high proper motion limit @xmath74450 mas yr@xmath2 . as a result , many stars on our list which we identify as very nearby objects do not have cns3 names ; these are typically stars with relatively small ( @xmath75 mas yr@xmath2 ) proper motions . we note that @xmath76 of the stars in our catalog ( 6,003 entries ) have no counterpart in either the hipparcos , tycho-2 , or cns3 catalog . use of the pm designations for these stars would be convenient , especially given that the pm name indicates the location of the star on the sky . coordinates in the table are tabulated in columns 5 and 6 . right ascension ( @xmath38 ) and declination ( @xmath39 ) are listed in the icrs system , and quoted for the 2000.0 epoch . this means that current positions must be extrapolated using the tabulated proper motions . for stars listed in the hipparcos catalog , we have extrapolated the positions to the 2000.0 epoch , from the values listed in @xcite which were listed for epoch 1991.25 . for all other stars , the 2000.0 coordinates are based on the position of their 2mass counterparts as listed in @xcite and extrapolated to the 2000.0 epoch from the epoch of the 2mass observations . the positions are typically accurate to 0.8@xmath77 , which are the quoted 2mass catalog errors on the absolute astrometry ; proper motion errors have little effect on the accuracy of the extrapolated positions because of the proximity of the 2mass survey epoch to the millennium year ( @xmath78 years ) . the total proper motion @xmath50 of the stars are tabulated in column 7 and listed in seconds of arc per year . the vector components of the proper motion in the directions of r.a . and decl . ( @xmath79 ) are tabulated in columns 8 and 9 . both components are given in seconds of arc per year , following the superblink convention . note that @xmath80 is listed in seconds of arc per year , not seconds of time per year , by which we mean that @xmath81 . these proper motions are from three separate sources . stars with hipparcos counterparts are listed with their proper motion from the hipparcos catalog @xcite . stars not in the hipparcos catalog but with counterparts in the tycho-2 catalog are listed with their proper motion from tycho-2 @xcite . the hipparcos and tycho-2 proper motions are typically accurate to 0.5 - 2.0 mas yr@xmath2 . for stars not listed in either the hipparcos or tycho-2 , the proper motions listed are those measured in the superblink proper motion survey . those proper motions are based on a re - analysis of the digitized sky surveys images , as described in detail in @xcite ; these proper motions have a typical precision of @xmath738 mas yr@xmath2 . one can determine the source of the proper motion by checking whether the star has a hipparcos or tycho-2 name . astrometric parallaxes are recovered from the literature for 1,422 of the m dwarfs in our catalog . see section 2.4.1 for the list of papers and catalogs that were searched to recover the parallaxes . all values are listed in column 10 whenever available , otherwise the field is left blank . a flag is added in column 11 to indicate the source of the parallax , with the key provided in the table footnote . in all cases , measurement errors as reported in the relevant source , are noted in column 10 . lrrrrrrrrrrrrrrc pm i00001@xmath716943 & 0.0461 & & & & & 14.3 & 12.6 & 10.8 & 9.70 & 9.12 & 8.84 & 13.52 & 3.82 & [email protected] & m3 + pm i00003@xmath720802s & & & & & & 13.9 & 11.8 & & 9.12 & 8.47 & 8.27 & 12.93 & 3.81 & [email protected] & m3 + pm i00005@xmath720533 & & & & & & 13.2 & 11.2 & 10.2 & 9.00 & 8.39 & 8.17 & 12.28 & 3.28 & [email protected] & m1 + pm i00006@xmath711829 & & & & 12.91 & 11.28 & 11.9 & 10.3 & 9.5 & 8.44 & 7.79 & 7.64 & 11.28 & 2.84 & [email protected] & k7 + pm i00007@xmath711624 & & & 21.02 & & & 14.0 & 11.9 & & 9.32 & 8.68 & 8.46 & 13.03 & 3.71 & [email protected] & m0 + pm i00007@xmath723510 & & & & & & 13.3 & 11.1 & 10.2 & 9.12 & 8.48 & 8.28 & 12.29 & 3.17 & [email protected] & m1 + pm i00007@xmath726243 & & & & & & 13.7 & 11.7 & 11.0 & 9.89 & 9.23 & 9.07 & 12.78 & 2.89 & [email protected] & m2 + pm i00012@xmath711358s & & & 19.85 & 13.67 & 11.12 & & & & 8.36 & 7.71 & 7.53 & 11.12 & 2.76 & [email protected] & k7 + pm i00014@xmath714724 & & & & & & 13.8 & 11.4 & 10.6 & 9.67 & 9.02 & 8.83 & 12.70 & 3.03 & [email protected] & k7 + pm i00014@xmath721656 & & & & 12.14 & 10.76 & 11.5 & 9.8 & 8.9 & 8.02 & 7.41 & 7.22 & 10.76 & 2.74 & [email protected] & m0 + pm i00015@xmath720814 & & & & & & 14.4 & 11.7 & & 9.79 & 9.12 & 8.91 & 13.16 & 3.37 & [email protected] & m1 + pm i00016@xmath723258 & & & & & & 13.4 & 11.6 & 10.7 & 9.77 & 9.14 & 8.94 & 12.57 & 2.80 & [email protected] & k7 + pm i00016@xmath727613 & & & & & & 14.1 & 12.6 & 10.6 & 9.48 & 8.89 & 8.60 & 13.41 & 3.93 & [email protected] & m3 + pm i00017@xmath723528 & & & & 14.99 & 13.36 & 14.5 & 12.1 & 11.0 & 9.82 & 9.19 & 8.93 & 13.36 & 3.54 & [email protected] & m2 + pm i00024@xmath710440 & & & 21.96 & & & 13.6 & 11.2 & 9.9 & 9.19 & 8.61 & 8.45 & 12.50 & 3.31 & [email protected] & m1 + pm i00024@xmath724601 & & & & 13.92 & 12.43 & & & & 9.18 & 8.52 & 8.34 & 12.43 & 3.25 & [email protected] & m1 + pm i00025@xmath726324 & & & & & & 14.2 & 12.1 & 10.7 & 9.32 & 8.71 & 8.53 & 13.23 & 3.91 & [email protected] & m3 + pm i00026@xmath713821 & & 21.97 & 20.15 & & & 14.8 & 13.1 & 10.6 & 9.71 & 9.20 & 8.91 & 14.02 & 4.31 & [email protected] & m1 + pm i00026@xmath721123 & & & & 14.74 & 13.09 & & & & 9.86 & 9.24 & 9.03 & 13.09 & 3.23 & [email protected] & m0 + pm i00026@xmath723919 & & & & & & 14.0 & 11.4 & 10.9 & 9.84 & 9.19 & 9.00 & 12.80 & 2.96 & [email protected] & m4 + table 2 lists photometric data for the stars , along with photometric parallaxes and estimated spectral types , both based on the photometric measurements . superblink stars names are repeated in column 1 of table 2 , to facilitate the cross - identification with table 1 , although the number of lines are the same . we have cross - correlated our list of bright m dwarfs with both the _ rosat _ all - sky bright source catalog @xcite and the _ rosat _ all - sky survey faint source catalog @xcite . we used a search radius of 15 , which is on the order of the astrometric precision of the rosat catalog . our search identified 1,065 stars with x - ray counterparts . flux values in counts s@xmath2 are listed in column 2 . x - ray fluxes can be useful in identifying young stars in the catalog . m dwarfs can be x - ray emitters if they have significant chromospheric activity . high activity levels are typically linked to rapid rotation , which in m dwarfs is indicative of either spin - orbit coupling in a close binary @xcite or of young age @xcite . it is likely that the youngest m dwarfs in our catalog are to be found among the subsample of objects with significant x - ray emission . follow - up observations of x - ray bright m dwarfs has been successful in identifying scores of low - mass members to young moving groups @xcite . the catalog was cross - correlated against the fifth data release ( dr5 ) of the galex mission . using a @xmath82 search radius , we found galex counterparts for 3,905 of the 8,889 stars on our list . the @xmath83 and @xmath84 magnitudes are listed in columns 3 and 4 . some 762 objects have counterparts in both the @xmath84 and @xmath83 , while 3,115 have counterparts only in the @xmath83 , leaving 28 stars with a counterpart in the @xmath84 only . these ultra - violet ( uv ) magnitudes can also be used to identify young stars . a high uv luminosity is associated with activity in m dwarfs , and can be used as an age diagnostic @xcite . we include optical magnitudes @xmath85 and @xmath86 as they are listed in the hipparcos and tycho-2 catalogs . these are included in columns 5 and 6 respectively . these are available only for stars with counterparts in those catalogs . the @xmath85 and @xmath86 magnitudes are useful because they are generally more accurate ( 0.1 mag or better ) than the photographic magnitudes which we list for all the stars . optical magnitudes are obtained from the usno - b1.0 catalog of @xcite , which is based on scans of historical photographic surveys . the blue magnitudes are extracted mostly from scans of @xmath87 plates from the palomar sky surveys ( poss - i , poss - ii ) and the southern eso schmidt ( serc ) survey . red magnitudes are extracted from scans of @xmath88 plates from poss - i and poss - ii and also from the anglo - australian observatory red survey ( aao - red ) . photographic infrared magnitudes are extracted from @xmath89 plates from poss - ii and serc . cross - correlation with the usno - b1.0 catalog is performed as part of the superblink quality control procedure , and all ambiguous cases are verified by visual examination using overlays of the usno - b1.0 sources on the dss scans . all the recovered photographic magnitudes are listed in columns 7 ( blue ) , 8 ( red ) and 9 ( infrared ) . since the optical @xmath90 , @xmath91 , @xmath92 , magnitudes are based on scans of photographic plates , they are generally not as reliable as magnitudes measured on electronic detectors , such as those from 2mass and galex . photographic magnitudes are typically accurate to only @xmath93 mag . as part of the quality control process , all the stars in the superblink proper motion survey have their counterparts identified in the 2mass catalog of @xcite . as described in 3.2.1 above , the 2mass counterparts provide the absolute astrometry for superblink objects . because of the color restriction of equation 1 , every object in our catalog is required to have a counterpart in 2mass . the corresponding infrared @xmath27 , @xmath62 , and @xmath94 magnitudes are listed in columns 10 , 11 , and 12 , respectively . the magnitudes are generally accurate to @xmath730.08 mag , though one should refer to the 2mass documentation for a detailed discussion of magnitude errors and uncertainties . very bright stars ( @xmath95 ) , in particular , are saturated in 2mass and have larger photometric uncertainties . for convenience , we calculate an effective visual magnitude @xmath28 , which for bright stars is the visual magnitude @xmath86 quoted in the hipparcos or tycho-2 catalogs . for faint stars , it is a combination of the photographic @xmath90 , @xmath91 , and @xmath92 magnitudes , following the algorithm described in @xcite where this effective visual magnitude is labeled as @xmath96 . in this paper , we use the shorter form @xmath28 for convenience , but it should be understood that all the mentions of @xmath28 actually refer to this effective visual magnitude @xmath96 . all values are listed in column 13 . photometric distances are estimated for all the stars based on the color magnitude relationship for m dwarfs calibrated in @xcite . for stars with @xmath31 the relationship is : @xmath97.\ ] ] photometric parallaxes @xmath98 are then calculated based on the estimated @xmath45 and the apparent visual magnitude @xmath28 . values are listed in column 14 . as discussed in @xcite , the relationship has a scatter of 0.7 mag about the mean , which typically results in photometric distances with errors of @xmath99 . in figure 9 , we compare the photometric distances to triangulated distances , for stars with available parallaxes . of the 1,422 m dwarfs with parallaxes , some 1,206 stars have photometric distance estimates within @xmath99 of the astrometric distance , but only 772 have photometric distances that are accurate to @xmath100 . there is also a significant number of outliers which could be unrecognized giants , young field stars , or unresolved doubles , all of which would have distances underestimated by photometry . metal - poor stars ( m subdwarfs ) on the other hand would have their distances overestimated by photometry . spectroscopic follow - up would be required to identify all such objects . thus , while photometric distances are conveniently provided here for all the stars , one should bear in mind that these are mainly for guidance purposes , as these distances carry large uncertainties in many cases . astrometric parallaxes ( listed in table 1 ) should be preferred for stars for which they are available , and should be measured for the remaining objects in our catalog . for all the stars , we provide an estimated spectral subtype which is based on the relationship between @xmath26 color and the spectral subtype ( see fig . we use this simple two - component linear relationship to estimate the subtype @xmath101 : @xmath102,\ ] ] after which the values are rounded up to the nearest integer . we use lowercase letters ( k , m ) instead of the usual uppercase ( k , m ) to indicate that the spectral type is only an estimate ; a value of @xmath103 yields an estimated subtype of k7 . a histogram of the estimated subtype distribution is shown in figure 10 , where it becomes clear that our magnitude - limited catalog of bright m dwarfs is biased toward stars of earlier subtypes . the initial mass function is expected to peak at around 0.3 m@xmath6 or spectral subtype m3-m4 @xcite , so our incompleteness starts at around m0 and becomes severe by m5 . these subtypes are tabulated in column 15 and may be used as a guide in selecting targets of interest . some of the stars on the list , in particular stars listed in the cns3 , already have published literature about them , and the interested user would be advised to verify the existence of formal spectral classification(s ) . we are currently conducting a spectroscopic follow - up survey to obtain formal spectral types for the brightest stars in our catalog , and intend to provide formal spectral classifications for these in the near future . distribution on the sky of all the stars in our catalog is displayed in figure 11 in a galls - peter equal - area projection . to emphasize differences between the brighter and fainter stars in our catalog , we display the distribution in two panels , with one showing only stars with visual magnitudes @xmath104 ( upper panel ) and the other showing stars with visual magnitudes @xmath105 . the separation roughly corresponds to the magnitude limit of the tycho-2 catalog . of the 1,993 stars in the brighter sample , 1,614 are listed in either the tycho-2 or hipparcos catalog . the fainter sample , on the other hand , contains only 437 hipparcos and tycho-2 stars out of a total of 6,896 objects . the distribution of the brightest ( @xmath25 ) m dwarfs appears relatively uniform over the sky , except for an overdensity at low galactic latitude in the general direction of the galactic center ( @xmath106 ) . this overdensity is likely due to contamination by background giants in dense fields . since our catalog is dependent on the hipparcos and tycho-2 catalogs at the bright end , errors in proper motion measurements due to field crowding also possibly result in stars having overestimated proper motions , which would in turn fall within the reduced proper motion range of m dwarfs . close examination of a number of stars at low galactic latitudes indeed raises doubt about the accuracy of the proper motion quoted in hipparcos and tycho-2 in at least a few cases . the distribution of fainter ( @xmath107 ) m dwarfs , on the other hand , shows a net deficit of objects in the southern sky , especially for @xmath108 . as discussed in 2.3 this is due to the limited coverage of superblink in the south , where the survey is only complete for proper motions @xmath18 mas yr@xmath2 , compared to @xmath16 mas yr@xmath2 in the north . the distribution , however , does not show any overdensity / deficit of stars at low galactic latitudes , which suggests both a high completeness and low contamination from background giants . this demonstrates the high efficiency and quality controls of the superblink survey in high density fields . we illustrate in figure 12 the effects of kinematic bias and magnitude limit in our catalog . the upper panels show the estimated absolute magnitudes as a function of distance , while the bottom panels show the transverse motions @xmath109 ( i.e. projected motions in the plane of the sky ) as a function of distance , for all the stars in our catalog . the astrometric distances are used when available , otherwise we use the photometric distances as calculated in 3.4 . one can not speak of incompleteness due to the magnitude limit , since the @xmath34 magnitude limit was _ imposed _ as a primary constraint to the catalog . figure 12 however illustrates the effects of this magnitude limit in restricting the distance range of the stars in the catalog . all the m dwarfs with @xmath34 populate the nearest 75 parsecs . the limiting distance is also a function of absolute magnitude thus spectral subtype . late - type m dwarfs , which have lower absolute magnitudes @xmath45 , have a much shorter distance limit in our catalog , which explains the strong deficit of stars with subtypes later than m4 ( see figure 10 ) . the main source of incompleteness in our magnitude - limited catalog is kinematic bias . because all of our stars are selected among sources with proper motions @xmath16 mas yr@xmath2 , nearby stars with low components of their transverse motion @xmath109 ( i.e. projected motion in the plane of the sky ) may be missed . nearby stars can have low values of @xmath109 either because of a low intrinsic motion relative to the sun , or because their motion vector is nearly perpendicular to the plane of the sky . for example , a star at 50 parsecs from the sun with @xmath110 mas yr@xmath2 has a transverse velocity @xmath1119.48 km s@xmath2 , which will occur with a probability of about 10% for stars in the solar neighborhood ( see 2.2 and fig.1 ) . the superblink catalog currently has a higher proper motion limit ( @xmath112 mas yr@xmath2 ) in large swaths of the southern sky , which makes our census much more prone to a kinematic bias . this explains the deficit of sources in the southern sky ( fig.11 ) . this is only mitigated at the bright end because of our use of the hipparcos and tycho-2 catalogs . to estimate the overall completeness of our catalog , we proceed in two steps . first we use monte - caro simulations to estimate the completeness of the northern sky census , which we assume to suffer only from kinematic bias due to the @xmath110 mas yr@xmath2 proper motion limit . then we assume that the local distribution of field stars is isotropic in the vicinity of the sun , and thus infer how many stars are missing from the southern sky census , i.e. assuming that both the north and south hemisphere should have the same number of bright m dwarfs . we first assume that the transverse velocity @xmath109 is independent of the distance @xmath113 for stars in the solar neighborhood , and assume a statistical distribution of the form @xmath114 . we extract from our data the statistical distributions @xmath115 and @xmath116 from appropriate cuts in @xmath113 and @xmath109 . for @xmath117 , we consider only stars in the northern sky with transverse velocities @xmath118 km s@xmath2 ; this boundary is illustrated in figure 12 ( lower - left panel ) , which shows that the stars that satisfy this constraint do not suffer from the kinematic bias , because all the stars in our catalog which have @xmath118 km s@xmath2 also have proper motions @xmath16 mas yr@xmath2 . the normalized distribution function @xmath117 is shown in figure 13 ( top panel ) . for @xmath119 , we consider only stars with distances @xmath120 parsecs ; again the boundary is illustrated in figure 12 , and shows that these very nearby stars suffer from minimal kinematic bias , because stars with proper motions @xmath110 mas yr@xmath2 ( and missing from the census ) can only have @xmath121 km s@xmath2 , which excludes very few stars . the normalized distribution function @xmath119 is shown in figure 13 ( bottom panel ) where the limit of completeness ( @xmath122 km s@xmath2 ) is shown for reference . given the sharp drop in @xmath119 for transverse motions @xmath123 km s@xmath2 , it is safe to assume that the sample of @xmath120 parsecs stars is essentially complete . we perform monte - carlo simulations based on the statistical distribution @xmath117 and @xmath119 to estimate the probability that an m dwarf with @xmath34 will be detected in our survey . we thus estimate that our census of the northern sky has a completeness of @xmath124% . this means that the 5,361 stars in the northern sky sample are most likely drawn from a hypothetical sample of 5,840@xmath7318 stars . the @xmath125 stars missing due to the kinematic bias are expected to have proper motions @xmath110 mas yr@xmath2 . the above analysis assumes that the superblink survey is complete within the proper motion range and magnitude limit . if the survey itself is incomplete , then the number of missing objects would be larger . the completeness of the superblink survey generally depends magnitude and field density . however in the magnitude range under consideration ( 8@xmath12615 ) a completeness test based on the fraction of tycho-2 stars recovered by superblink _ and _ the fraction of superblink stars listed in tycho-2 , suggests that the combined tycho-2/superblink catalog has a completeness @xmath127 @xcite . if we factor in the possibility that @xmath52% stars in that magnitude range have escaped detection by both tycho-2 and superblink , this would bring the total number of hypothetical bright m dwarfs in the northern sky to @xmath55,950 stars . our southern sky census is significantly less complete than the northern census . while our catalog has 5,361 stars north of the celestial equator , it only has 3,528 in the south . the main reason for the incompleteness is the limited coverage of the superblink proper motion survey in the south . the survey is complete to @xmath16 mas yr@xmath2 for the entire northern sky , but reaches that limit only over a fraction of the south , mainly in the declination range @xmath128 ( see fig.11 ) south of @xmath129 , the survey is only complete to @xmath18 mas yr@xmath2 , but reaches to @xmath130 mas yr@xmath2 in some areas . the incompleteness of the superblink survey is mitigated by the inclusion of the tycho-2 catalog , which is mostly limited by magnitude and actually includes stars with proper motions smaller than the superblink limit of @xmath13140 mas yr@xmath2 . however the magnitude limit of the tycho-2 catalog , @xmath13212.0 , is too bright to include all but a fraction of the m dwarfs in the magnitude range of interest ( @xmath133 ) . under the assumption that the northern sky census is fully representative of the entire volume around the sun , i.e. assuming the m dwarf distribution to be _ isotropic _ in the solar neighborhood , we use the results from the monte - carlo simulation above and estimate that the southern - sky census of m dwarfs brighter than @xmath33 should also include 5,950 stars , for a total of 11,900 m dwarfs over the entire sky . given that our southern sample includes only 3,528 stars , we thus conclude that we are currently missing @xmath52,420 stars in the southern sky . in other words our southern sky census has a completeness of only @xmath559.3% . the need to assemble a comprehensive list of _ bright _ m dwarfs over the entire sky is primarily motivated by the search for extra - solar planets . based on current techniques , m dwarfs are the most promising stars around which to look for earth - mass planet in a habitable zone . radial velocity measurements are most sensitive to the reflex motion from planets in close orbits around their parent stars , or to planets orbiting low - mass stars ; for planets in habitable zones , this favors m dwarfs , whose zone is closer in , and whose masses are relatively low . transit surveys have a higher chance of detecting planets in close orbits , and are more sensitive to planets orbiting smaller stars ; this likewise favors m dwarfs as the most promising targets . the main caveat is that both radial velocity measurements and transit monitoring are most efficient for relatively bright stars . with current observational resources , radial - velocity monitoring can realistically be performed on stars with absolute magnitudes @xmath134 , and near - term developments only promise to increase the depth of surveys by one or two magnitudes , most notably by observing at infrared wavelengths where m dwarfs are noticeably brighter . m dwarfs with apparent infrared magnitudes @xmath34 will soon be within range of radial velocity surveys . the motivation of the superblink proper motion survey is to identify the largest possible number of hydrogen - burning objects up to at least 100 parsecs of the sun . with a low proper motion limit of @xmath16 mas yr@xmath2 and astrometric accuracy @xmath135 mas yr@xmath2 , the catalog suffers from minimal kinematic bias to a distance of 50 - 60 parsecs , which is the range over which one would expect to find m dwarfs with @xmath34 . the current superblink catalog is complete over the entire northern sky , but the southern sky survey is still in progress . at this time , the catalog has a proper motion limit @xmath16 mas yr@xmath2 in the declination range @xmath136 ; at southern declinations the proper motion limit is now @xmath18 mas yr@xmath2 , except for relatively bright stars ( @xmath25 ) for which the catalog extends to @xmath16 mas yr@xmath2 because of the inclusion of data from the tycho-2 and hipparcos catalogs . our search of the current version of the superblink catalog has identified 8,889 bright m dwarfs , of which 5,361 are in the northern sky and 3,528 in the south . the lower number of stars catalogued in the south reflects the limited extent of our proper motion survey in the south . our estimate of the kinematic bias suggests that our northern census is @xmath124% complete . from this we infer that there should be @xmath137 m dwarfs with magnitudes @xmath34 over the entire sky . our catalog currently lists @xmath138% of the entire set . we recommend that these stars be observed by radial - velocity monitoring programs with the highest priority . in addition to exoplanet surveys , completing the census of the solar neighborhood is important to constrain the stellar density in the galaxy , and to determine the distribution of stellar properties . our catalog includes many stars which are identified here for the first time as members of the solar neighborhood . in particular , most of the sources are not listed in the catalog of nearby stars . the key to completing the census of bright m dwarfs in the south is to obtain a proper motion catalog with a lower proper motion limit , currently limited to @xmath18 mas yr@xmath2 south of @xmath129 . we are currently working on an extension of the superblink proper motion survey , which will extend the limits to @xmath16 mas yr@xmath2 to @xmath139 , and to @xmath130 mas yr@xmath2 in the @xmath140 range . we predict that extended surveys will identify an additional @xmath141 bright m dwarfs , if a proper motion limit @xmath16 mas yr@xmath2 can be reached over the entire sky . a revised catalog will be produced as the survey reaches completion . at this time we are conducting a systematic spectroscopic follow - up survey of the brightest stars in this catalog , and searching the literature for previous determinations of spectral subtypes . results from these investigations will be provided in upcoming papers by our team . perryman , m. a. c. & esa 1997 , the hipparcos and tycho catalogues . astrometric and photometric star catalogues derived from the esa hipparcos space astrometry mission , publisher : noordwijk , netherlands : esa publications division

experiment e949 at brookhaven national laboratory studied the rare decay and other processes with an exposure of s . the data were analyzed using a blind analysis technique yielding one candidate event with an estimated background of events . combining this result with the observation of two candidate events by the predecessor experiment e787 gave the branching ratio , consistent with the standard model prediction of . this is a more detailed report of results previously published in physical review letters .

0709.1000

although the standard model ( sm ) has successfully accounted for all low energy cp - violating phenomena thus far observed , it is insufficient as the source of cp - violation needed to explain the cosmological baryon asymmetry in our universe @xcite . according to sakharov @xcite , one of the conditions necessary to generate such an asymmetry is that the elementary interaction violates charge conjugation symmetry ( c ) and the combined cp symmetry ( where p is the parity symmetry ) . however , the size of the asymmetry needed for this can not be derived in model calculations based on the sm @xcite and new sources of cp violation have been sought for many years in particle physics experiments . prominent among these are the rare decays @xmath4 which are sensitive to new physics involving both cp - violating and cp - conserving interactions . in this paper , we present a detailed description of the previously reported measurement of the reaction @xmath5 performed by brookhaven national laboratory ( bnl ) experiment bnl - e949 @xcite . this paper is arranged as follows . we first briefly review cp violation and rare kaon decays , with an emphasis on @xmath6 decays . we then describe previous results on this reaction and discuss the sources of potential background and the methods for suppressing backgrounds . we also discuss the design of the @xmath1 beamline , the detector and the selection criteria used in data analysis and describe the methods used for estimating background levels and for evaluating the acceptance . after examining the signal region , we present the method used for extracting the branching ratio , making full use of our knowledge of the background in the signal region . in the last section , we show how the measurement of @xmath7 impacts the search for new physics beyond the sm . standard model cp violation arises from a complex phase in the three - generation quark mixing matrix @xcite . in the wolfenstein parameterization @xcite of the cabibbo - kobayashi - maskawa ( ckm ) matrix , the parameters can be written in powers of @xmath8 : @xmath9 where @xmath10 , @xmath11 , @xmath12 and @xmath13 are real numbers . cp invariance of the lagrangian for weak interactions is violated when the ckm matrix is complex . the parameter @xmath13 quantifies cp violation in the sm . the unitarity of the ckm matrix implies six unitarity conditions , which can be represented graphically in the form of triangles , all of which must have the same area . the area of these triangles is equal to one half of the jarlskog invariant , @xmath14 @xcite . applying the unitarity property @xmath15 to the ckm matrix in ( [ eq : wolfenstein ] ) yields @xmath16 where the approximations @xmath17 and @xmath18 have been made . this equation can be represented graphically , as shown in fig . [ fig : triangle ] , 0.55 where we have divided all of the sides by @xmath19 . the apex of the triangle is given by two wolfenstein parameters , @xmath20 and @xmath21 , where @xmath22 and @xmath23 @xcite . @xmath24 s and @xmath25 s are so far the only two mesons showing evidence of cp violation in their decay processes . whether or not the observed cp violation can be completely explained by the ckm phase within the sm can be probed by the independent determination of @xmath12 and @xmath13 , from @xmath24 and @xmath25 decays as shown in fig . [ fig : triangle_b_k ] . 0.6 two sensitive methods for making the comparison are : * a comparison of angle @xmath26 from the ratio ( @xmath27)/ ( ) with that from the cp violating asymmetry ( @xmath28 ) in the decay @xmath29 ; and , * a comparison of the magnitude @xmath30 from with that from the mixing frequencies of @xmath31 and @xmath32 mesons , expressed in terms of the ratio of the mass differences , @xmath33 . although the decay is a flavor changing neutral current ( fcnc ) process prohibited at tree level in the sm , it is allowed at the one - loop level . in leading order , it is described by a `` box '' diagram and two `` @xmath34-penguin '' diagrams , as shown in fig . [ fig : pnn-2nd ] . 0.7 + 0.7 0.7 the weak amplitude for this process is represented as @xmath35 where @xmath36 s are the ckm matrix elements , @xmath37 s are the dirac matrices , @xmath38 is the momentum transfer , and @xmath39 s are quark masses . @xmath40 vanishes if all of the quark masses , @xmath39 , are equal , because of the unitarity of the ckm matrix . however , the breaking of flavor symmetry , which results in the variation of quark masses , allows this decay to proceed at a very small rate . the top quark provides the dominant contribution to the branching ratio due to its very large mass in spite of the small coupling of top to down quarks ( @xmath41 ) in the ckm matrix . following ref . @xcite , the branching ratio for is calculated as follows . the effective hamiltonian can be written in the sm as @xmath42 in next - to - leading order ( nlo ) , where @xmath43 and @xmath44 @xmath45 and @xmath46 in ( [ eq : inami ] ) are functions of @xmath47 , and were derived for the first time by inami and lim in 1981 @xcite . the coefficient @xmath48 and the function @xmath49 are the charm and top quark contributions , including qcd corrections at nnlo @xcite . with the top quark mass in the minimal subtraction scheme @xmath50 gev @xcite , @xmath51 is obtained . the perturbative charm contribution gives the largest theoretical uncertainty and can be described in terms of the parameter @xmath52=0.34\pm0.04 , \label{eq : charm_cont}\ ] ] where the error is obtained by varying the charm mass , @xmath53 , the scale factor , @xmath54 and the coupling constant , @xmath55 , by reasonable amounts . one obtains @xmath56 , \label{eq : pnn_br}\ ] ] where @xmath57 ^ 8 , \label{eq : kappa}\ ] ] @xmath58 comprises the long - distance contribution calculated in ref . @xcite , and the @xmath59 s ( @xmath60 ) are from the ckm matrix elements . the @xmath61 represents isospin breaking corrections in relating to the well - measured decay @xcite . in obtaining the numerical value in ( [ eq : kappa ] ) , we used @xcite @xmath62 expression ( [ eq : pnn_br ] ) describes in the @xmath63 plane an ellipse with a small eccentricity , namely @xmath64 where @xmath65 using ( [ eq : pnn_br ] ) and varying @xmath66 , @xmath67 , @xmath68 and @xmath30 , which is constrained by @xmath69 and @xmath70 mixing in the @xmath63 plane , the branching ratio of is predicted to be @xmath71 within the sm . it should be noted that , of the uncertainty of 27% in ( [ eq : br_sm ] ) , the theoretical uncertainty is @xmath726% at present , mainly due to the uncertainty in the charm quark mass . theoretically a precise measurement of ( ) is one of the cleanest ways to extract @xmath30 . this is due to the following factors : * the long - distance contributions to the branching ratio are small @xcite and under control , the most recent calculation gives a contribution of ( + 6@xmath733)% to the branching ratio @xcite ; * the uncertainty from the hadronic matrix element has been reduced to @xmath741% by recent theoretical and experimental developments @xcite , and * the recent nnlo calculation @xcite has reduced the total theoretical uncertainties to @xmath726% , i.e. relatively small and reliably calculated as compared with the uncertainties present in other @xmath25 and @xmath24 decays . if a precise measurement of the neutral analog @xmath75 could also be made , the intrinsic theoretical error on @xmath30 could be reduced to @xmath721% @xcite . as determinations of @xmath24-system parameters become increasingly precise , the uncertainty on the sm prediction for will approach the current theoretical accuracy of @xmath726% . a correspondingly precise measurement of the branching ratio therefore provides a stringent test of the sm and probes for new physics . there have been numerous predictions for in the models beyond the sm and applications of the measured branching ratio to constrain new models . these include the minimal supersymmetric standard model with @xcite and without @xcite new sources of flavor- or cp - violation , generic susy with minimal particle content @xcite , susy with non - universal a terms @xcite , susy with broken r - parity @xcite , topcolor @xcite , topcolor - assisted technicolor models @xcite , multiscale walking technicolor @xcite , four generation models @xcite , leptoquarks @xcite , left - right model with right - handed @xmath76 @xcite , extension of the sm to a gauge theory with @xmath77 mesons @xcite , a multi - higgs multiplet model @xcite , light sgoldstinos @xcite , universal extra dimensions @xcite , 5-dimensional split fermions @xcite , a randell - sundrum scenario @xcite , a littlest higgs model @xcite , non - standard neutrino interactions @xcite , and a minimal 3 - 3 - 1 model @xcite . searches for this process which began over 35 years ago have used stopped-@xmath1 beams . it was believed at the time of the first of these that the branching ratio might be as high as a few @xmath79 @xcite . it was recognized that even at this level , a poor - signature process such as @xmath5 would need effective particle identification , precise kinematic measurement and the ability to veto extra charged and neutral tracks to discriminate it from common decay modes such as @xmath80 and @xmath81 ( referred to as @xmath82 and @xmath83 , respectively ) . the earliest published result was from a heavy liquid bubble chamber experiment @xcite at the argonne zero gradient synchrotron , in which a 90% cl upper limit @xmath84 was obtained . in that paper it was recognized that @xmath83 decay in flight and hadronic @xmath85 interaction in the detector were dangerous sources of potential background . the final analysis of the argonne experiment improved the limit to @xmath86 @xcite , but before it appeared in print , a subsequent counter / spark - chamber experiment at the berkeley bevatron improved the limit to @xmath87 @xcite . however this experiment was sensitive to only the most energetic of @xmath85 , whereas the bubble chamber experiment covered a wide kinematic range . in addition to the background from common @xmath1 decay modes , this experiment considered possible background from @xmath1 charge exchange in the stopping target followed by @xmath88 , and from beam @xmath85 which scattered into the detector . the chicago - berkeley group continued their program with a setup sensitive to @xmath85 in the kinetic energy range 60105 mev , i.e. below that of the potential background process @xmath83 rather than above it . this required reconfiguring their photon veto system so that it became nearly hermetic . combining results from the two configurations , the branching ratio upper limit was improved slightly to @xmath89 @xcite . about a decade later , an experiment at the kek proton synchrotron improved the limit to @xmath90 @xcite . the technique of waveform digitization to record the @xmath91 decay chain was introduced for the first time . this experiment was sensitive only to the @xmath85 with momenta greater than that from @xmath83 ( referred to as the `` @xmath92 '' region ) and its setup resembled that of ref . @xcite . the bnl series of experiments was initiated in the early 1980 s . they were based on a large - acceptance solenoidal spectrometer with a hermetic photon veto situated at the end of a highly pure , very intense stopped-@xmath1 beam @xcite from the bnl alternating gradient synchrotron ( ags ) . the experimental signature of the decay was a single @xmath85 track with @xmath85 momentum less than 227 mev/@xmath93 plus no other particle from a @xmath1 decay . [ fig : mom_spec ] shows momentum spectra of major decay modes of @xmath1 . 0.9 the first phase of e787 in 1988 - 91 achieved a 90% cl upper limit on the branching ratio of @xmath94 @xcite , using data from the @xmath92 region . a separate limit of @xmath95 at 90% cl @xcite was extracted from the kinematic region in which the @xmath85 is softer than that of the @xmath85 from @xmath83 ( referred to as the `` @xmath96 '' region ) . this program completed the identification of backgrounds needed to reach the @xmath97 level of sensitivity and developed methods to reliably measure them . a major upgrade of both the beam line and the detector was undertaken between 1992 and 1994 . the search for @xmath5 resumed in 1995 and continued through 1998 . the limit on the branching ratio from the @xmath96 region was improved by an order of magnitude to @xmath98 at 90% cl @xcite , but the major output of this series of runs was the observation of two clean @xmath5 events @xcite in the @xmath92 region and a measurement of the branching ratio @xmath99 . the bnl - e787 detector was upgraded again over the period from 19992001 . the e949 experiment was proposed to use this detector to run for 60 weeks . after the first 12 weeks of running in 2002 no further funds were provided to complete the experiment . based on the collected bnl - e949 data , the first result was already published in 2004 @xcite . this paper provides an extended and detailed description of the detector and data analysis techniques used to produce the e949 result . e949 ( bnl - e949 ) which succeeded bnl - e787 had a sensitivity goal of detecting ten sm signal events @xcite . e949 employed a low momentum beam of @xmath1 s which were degraded and stopped in the detector . measurement of the decay involved observation of the daughter @xmath85 in the absence of other coincident activity . the @xmath85 was identified by its kinematic features obtained from energy , momentum and range measurements , and by the observation of a decay sequence . since the signal was expected at the @xmath97 level , the detector was designed to have powerful @xmath85 identification for rejection of @xmath82 and @xmath100 decays ( @xmath101 ) , 4-@xmath102 solid angle photon detection coverage for vetoing decays , and efficient @xmath1 identification system for eliminating beam - related backgrounds . the entire e949 spectrometer was surrounded by a 1 tesla solenoidal magnetic field along the beam direction . the coordinate of detector used a cartesian coordinate system in which the origin was at the center of the target ; the @xmath103 axis was along the incident beam direction and the @xmath104 axis in the vertical up direction as shown in fig . [ fig : gamma_detectors ] . under this coordinate system , the azimuthal angle of a track was defined as the arctangent of @xmath105 and , the polar angle @xmath106 was defined as the angle with the @xmath103 axis . many detector components have been discussed elsewhere @xcite . [ fig : gamma_detectors ] shows the e949 detector after upgrades ( 19992001 @xcite ) with the following improved components : photon veto detection efficiency , tracking and trigger efficiency , and data acquisition ( daq ) live time . e949 was designed to run at the same instantaneous rate as e787 , and to achieve a factor of five improvement in sensitivity , through the use of a higher duty factor and reduced @xmath1 momentum for a higher stopping fraction . the higher duty factor was not achieved in the engineering run in 2001 or the first physics run in 2002 due to a broken motor generator set that supplied power to the ags . the regular supply was removed from operation on august 3 , 2001 and the backup was used during the rest of 2001 and 2002 . e949 ran at about twice the beam rate of e787 . 1 . 0.9 the @xmath1 beam was produced by a high - intensity proton beam from the ags at bnl : the entire ags beam of 65@xmath107 protons ( tp / spill ) at a momentum of 21.5 gev/@xmath93 was delivered to the e949 @xmath1 production target . prior to 2001 the ags typically ran at 24 gev/@xmath93 , but at this momentum the longest spill achievable was 0.5 sec . combined with the longer cycle time ( 3.2 s between spills , as compared to 2.3 s ) , the duty factor at 24 gev/@xmath93 was unacceptably low . by lowering the proton momentum to 21.5 gev/@xmath93 , the spill length was increased to a maximum of 2.2 s , resulting in a duty factor of 2.2 s/5.4 s. at this lower proton momentum the production of 710 mev/@xmath93 @xmath1 s was reduced by 10% . the @xmath1 production target was made of 2/3 of an interaction length of platinum ( 6 cm along the beam direction ) , and was located on a water - cooled copper base . at the typical ags running condition 65 tp on the production target per 2.2 s spill , the maximum target temperature was measured to be @xmath72700@xmath108c . the low energy separated beam @xcite ( lesb iii ) collected and transported @xmath1 s emitted at 0@xmath108 ( along with 500 @xmath85 s and 500 protons per @xmath1 ) , and momentum - selected by the first dipole magnet . two electro - magneto - static separators swept @xmath85 s and protons out of the @xmath1 beam axis . the resulting beam was further selected by a second dipole magnet . lesb iii had a total length of 19.6 m from the production target to the e949 target with an angular acceptance of 12 msr and a momentum acceptance of 4.5% fwhm at a mean momentum of 710 mev/@xmath93 . during most of the 2002 running period the first separator voltage was lowered from the standard voltage of 600 kv to @xmath72250 kv due to high voltage discharges . under these conditions a @xmath109 ratio in the beam of 3:1 was achieved with a 40% loss in @xmath1 flux ( typically e787 ran with 4:1 ) . proton contamination was suppressed to a negligible level by the separators . at the same typical ags running condition described previously , @xmath110 @xmath1 s were transported through the beam line . the typical conditions during the 2002 run had 3.5@xmath111 @xmath1 s entering the e949 target every spill . this corresponded to a rate of 1.6@xmath111 @xmath1s / sec during the 2.2 sec spill . the effective spill length was actually 8% shorter ( 2.0 sec ) due to some residual modulation of the beam intensity out of the ags . the typical instantaneous rate of beam particles at the @xmath112erenkov counter was 6.3 mhz of @xmath1 s and 1.52.5 mhz of @xmath85 s . the incoming @xmath1 beam traversed a scintillation counter ( b0 ) , a @xmath112erenkov counter , two beam wire proportional chambers ( bwpc s ) , a passive beo degrader , an active degrader ( ad ) and a beam hodoscope ( b4 ) as depicted in fig . [ fig : gamma_detectors ] . the bwpc s and b0 counter are not shown in fig . [ fig : gamma_detectors ] . the b0 counter , which was a 30.5-cm long , 0.6-cm thick and 7.6-cm wide bicron bc408 plastic scintillator , was located just downstream of the last quadrupole magnet and counted all charged particles in the beam . it was read out by an analog - to - digital converter ( adc ) , a time - to - digital converter ( tdc ) and a 500 mhz transient digitizers based on gallium - arsenide charge - coupled device ( ccd ) @xcite . the @xmath112erenkov counter @xcite located just downstream of the b0 counter identified particles as @xmath1 s or @xmath85 s . the @xmath112erenkov light from the @xmath1 ( @xmath85 ) was transmitted ( internally reflected ) at the downstream surface of the @xmath113erenkov radiator and read out with 14 `` @xmath25 @xmath112erenkov '' ( @xmath114 ) and 14 `` @xmath102 @xmath112erenkov '' ( @xmath115 ) emi9954 kb pmt s . the pmt signals were split , with 90% sent to tdc s via fast lrs3412 discriminators and 10% to a @xmath11610 amplifier . the amplifier output was sent to ccd s . the pulse - height information in every 2 ns interval was recorded to reproduce the time development of the pulses and to detect two particles close in time to each other . the multiplicity output of @xmath117 pmt s was discriminated ( typical threshold : @xmath118 ) to identify @xmath1 s and @xmath85 s in the trigger ( @xmath119 and @xmath120 defined in section [ sec : trigger ] ) . the two bwpc s were located downstream of the @xmath112erenkov counter to monitor the beam profile and identify multiple incoming particles . the first chamber ( bwpc1 ) was located 168.5 cm upstream of the target entrance and contained three planes of sense wires : vertical ( @xmath121-plane ) and @xmath122 to the vertical ( @xmath123- and @xmath124-planes ) . the sense wires were 12-@xmath125m - diameter gold - plated tungsten . the @xmath121- , @xmath123- and @xmath124-planes had 144 , 60 and 60 readout channels , respectively , with a 1.27 mm wire spacing . in the @xmath123- and @xmath124-planes , pair of wires were multiplexed in one readout channel . the active area was 17.8 cm ( horizontal ) by 5.08 cm ( vertical ) . the cathode foils were 25-@xmath125 m thick aluminized mylar coated with carbon . the anode - cathode distance was 3.18 mm , and the total thickness of bwpc1 was approximately 56 mm . the second chamber ( bwpc2 ) was located 1.0 m downstream of bwpc1 and also contained three planes ( @xmath121 , @xmath123 and @xmath124 ) . the direction of the sense wires was vertical ( @xmath121-plane ) and @xmath126 to the vertical ( @xmath123- and @xmath124-planes ) . each plane had 120 active sense wires with a 0.8-mm wire spacing . among the 120 wires , the central 72 ones were multiplexed by 3 and the remaining were multiplexed by 6 in the readout channels , yielding a total of 32 readout channels for each plane . the cathode foils were 8-@xmath125 m single - sided aluminized mylar coated with carbon . the anode - cathode distance was 1.6 mm . both chambers were filled with a recirculated mixture of cf4 ( 80% ) and isobutane ( 20% ) . downstream of the bwpc s a degrader slowed the @xmath1 s so that they stopped in the center of the scintillator fiber target . the upstream section of the degrader was inactive , consisting of 11.11 cm of beryllium oxide ( beo ) and 4.76 mm of lucite . the high density ( 3.0 g/@xmath127 ) and low atomic number of beo was used to minimize multiple scattering . the ad consisting of 40 layers of 2 mm thick disks of bicron bc404 scintillator ( 13.9 cm diameter ) alternating with 2.2-mm thick copper disks ( 13.6 cm diameter ) was divided into 12 azimuthal segments with readout to a single hamamatsu r1924 pmt through fourteen 1-mm - diameter bicron bcf99 - 29-aa - mc wave length shifting ( wls ) fibers . the pmt outputs were provided to tdc s , ccd s and a 4-fold analog sum that was provided to an adc . these measurements enabled the ad to identify the beam particles and to detect activity coincident with @xmath1 decays . downstream of the degrader the b4 hodoscope detected the entrance position of the incoming particle in the target and identified the particle type by measuring its energy loss . the b4 hodoscope consisted of two planes , @xmath123 and @xmath124 , with about a 11.8-cm diameter oriented at a @xmath128 angle with respect to the horizontal axis . each plane had 16 bicron bc404 scintillator fingers with a 7.2-mm pitch . the cross section of each finger had a ` z - shape ' with a 6.4-mm thick middle part and 3.2-mm thick edges . this shape reduced inactive regions and improved the spatial resolution . three bicron bcf99 - 29-aa - mc wls fibers were embedded in each finger and connected to a single hamamatsu h3165 - 10 pmt that was read out by tdc s , adc s and ccd s . at the same position as the b4 hodoscope , but at larger radius was an annular scintillator counter , the ring veto ( rv ) . the rv was designed to veto particles that passed through perimeter of the b4 hodoscope . the rv was composed of two 180@xmath108 arcs of 3.3 mm thick bicron bc404 scintillator with an inner diameter varying from 11.9 cm to 12.0 cm and an outer diameter of 14.6 cm . the two rv elements were readout by hamamatsu h3165 - 10 pmt s and the signals were split three ways to adc s , tdc s and ccd s . the target consisted of 413 bicron bcf10 scintillating fibers of 5-mm square cross section and 3.1-m length that were bundled to form a 12-cm - diameter cylinder . a number of 1-mm , 2-mm and 3.5-mm square scintillating fibers ( called edge fibers " ) filled the gaps near the outer edge of the target . each of the 5-mm fibers was connected to a hamamatsu r1635 - 02 pmt , whereas the adjacent edge fibers were grouped onto 16 pmt s , providing signal read out by adc s , tdc s and ccd s . the fiducial region of the target was defined by two layers of six plastic - scintillating counters surrounding the target . the inner scintillators , called i - counters ( ic s ) , helped to define the fiducial volume and the logical or of the six ic s ( @xmath129 for trigger condition ) used by the trigger for this purpose . the ic s were 6.4-mm thick ( with an inner radius of 6.0 cm ) and extended 24 cm from the upstream face of the target . the outer scintillators , called v - counters ( vc s ) , overlapped the downstream edge of the ic s by 6 mm , and served to detect particles that decayed downstream of the fiducial region of the target . the vc s consisted of six 5-mm thick and 1.96-m long scintillators , and were staggered azimuthally with respect to the ic s . each ic and vc element was instrumented with an emi 9954 kb pmt which was read out by an adc , tdc and a 500 mhz transient digitizer ( td ) based on a flash adc @xcite . approximately 27% of the incident @xmath1 s ( typically @xmath130 @xmath1s / spill ) penetrated far enough into the target to satisfy the online target criteria for @xmath119 defined in section [ sec : pnn_trigger ] . the remaining @xmath1 s either underwent decay - in - flight , nuclear interaction in the degrader or scattered in the material of the beam instrumentation and did not reach the target . it should be noted that some of the @xmath1 s ( @xmath74 25% ) that satisfied the online @xmath119 requirement did nt stop in the target , and a factor for the stopping fraction of @xmath1 s was introduced as described in section [ sec : fs ] . the @xmath1 s deposited an average energy of 100 mev in the scintillating fiber target when coming to rest in the center of the target fiducial volume . the low velocity @xmath1 s typically lost 540 mev in each fiber , while the nearly minimum ionizing ( mip ) @xmath85 s from @xmath1 decays deposited about 1 mev per fiber , as they passed transversely through the fibers . the drift chamber , called the ultra thin chamber " ( utc ) @xcite , was located just outside of the ic . the primary functions of the utc were to measure the momenta of charged particles and to provide tracking between the target and the range stack ( rs ) . the utc had inner and outer radii of 7.85 cm and 43.31 cm , respectively . twelve layers of 58 mm drift cells were grouped into three super - layers ( each consisting of four layers ) with active lengths of 38.8 cm ( inner ) , 44.8 cm ( middle ) and 50.8 cm ( outer ) . the super - layers were filled with a 49.8%:49.8%:0.4% mixture of argon , ethane and ethanol . each anode wire was instrumented with an adc and a tdc . the drift time to the anode wires was used to determine the ( @xmath131 ) positions for the charged track . at the inner and outer radii of each super - layer were cathode foils , with 7 mm helical cathode strips at a @xmath132 pitch angle . each cathode strip was instrumented with an adc and a tdc . the combination of the charge centroid on the cathode strips and anode hits provided the @xmath133 hit position . there were two inactive regions filled with nitrogen gas between the three super - layers . differential pressures of @xmath722 mbar in the five gas volumes supported the cathode foils . the total mass of the utc ( excluding the inner and outer support tubes with the attached foils ) amounted to 2@xmath134 radiation lengths . the utc position resolutions were approximately 175 @xmath125 m for @xmath121 and @xmath135 and 1 mm for @xmath133 . located just outside the utc at an inner radius of 45 cm and an outer radius of 84 cm , the rs consisted of both scintillation counters and embedded straw chambers , providing energy and range measurement of the charged particles , information on the decay sequence and measurement of photon activity . the rs consisted of 19 layers of bicron bc408 plastic scintillator , azimuthally segmented into 24 sectors as shown in fig . [ fig : gamma_detectors ] . layers 218 were 1.9 cm thick and 1.82 m long . layer 19 was 1.0 cm thick and was mainly used to veto longer range muons . each scintillator in layers 219 was coupled through lucite light guides to emi 9954 kb pmt s at both upstream and downstream ends . the innermost counters were 6.4 mm thick and 52 cm long trigger - counters ( t - counters ) , defining the fiducial volume for charged @xmath1 decay products . the t - counters were thinner than layers 219 to suppress rate from photon conversions . seventeen 1-mm - diameter wls fibers ( bicron multi - clad bcf-92 ) with a pitch of 6.9 mm were embedded in each scintillator and coupled to a single hamamatsu r1398 pmt at each end . the signal from each pmt of the rs scintillators was passively split 1:2:2 for adc s , discriminators and fan - in modules , respectively . the discriminator output was sent to a tdc and the trigger . each pmt was read by an adc and a tdc . the tdc s ( lecroy 3377 ) recorded up to 16 hits in a 10.5 @xmath125s time window , and thus allowing for efficient detection of the @xmath136 decay . the analog fan - in summed the signals from 4 pmt s on each end in the same hextant ( four adjacent sectors ) and layer . this analog sum was read out by a single td @xcite and was provided to mean - timers @xcite for good timing on the photon veto in the trigger and a @xmath133-measurement for each layer of the track . the mean - timer output from each layer in a hextant was ored to provide input to the hextant photon veto algorithm in the trigger ( vetoing more than one non - adjacent hextant ) . the td s recorded the charge in 2 ns intervals ( 500 mhz sampling ) in a 2.5 @xmath125s time window with a resolution of 8 bits . the 500 mhz sampling provided sufficient pulse shape information to separate pulses from different events as close as 5 ns apart , and enabled the detection of the @xmath137 decay as described in section [ sec : pimue_id ] . the time window of the td s was narrower than that of the tdc s in order to reduce the data size . two range stack straw chambers ( rssc s ) were located outside rs layer 10 and 14 @xcite . the inner ( outer ) rssc consisted of two staggered layers of 24 ( 28 ) straws per sector with a length 97.8 ( 113.0 ) cm . the average density of an rssc was 0.054 g/@xmath138 . each straw tube was 3.4 mm in radius with a 50-@xmath125m - diameter gold - coated tungsten anode wire at the center . a schematic drawing is given in fig . [ fig : rssc ] . the straw chambers were operated with 67% argon and 33% isobutane mixture with a trace of water in a self - quenching streamer mode at 3,450 v. the local @xmath121-axis of a chamber was defined to be along the width of the chamber . there were a total of 48 chambers , with 2,496 straws installed in the e949 experiment . because of access restriction , a pair of straws from the right and left halves in the same layer were connected at the upstream end to allow downstream - only readout . the position of the hit straws provided @xmath139 position information , while the end - to - end time differences provided the @xmath133 measurement . the @xmath133 resolution of the e787 rssc s was degraded due to a pulse - height dependent time - walk effect . for e949 an amplifier and two discriminators were installed on each channel to allow for low threshold timing discriminator and a high threshold logic discriminator ( above the noise level ) . the @xmath133 resolution was improved from 3.0 cm ( rms ) observed in e787 to 1.5 cm ( rms ) . 0.9 the detection of activity coincident with the charged track was crucial for suppressing background processes that can mimic . photons from and other radiative decays were detected by the hermetic photon system as shown in fig . [ fig : gamma_detectors ] . the photon detectors surrounding the @xmath1 decay vertex with a 4@xmath102 solid angle coverage were located in the barrel , upstream and downstream end caps , and near the beam line . the photon system consisted of essentially every scintillator detector in experiment : the barrel veto ( bv ) , the barrel veto liner ( bvl ) , the rs , the upstream and downstream end caps ( ec s ) , the upstream photon veto ( upv ) , the upstream and downstream collar detectors ( co ) , the downstream microcollar detector ( @xmath125co ) , the downstream photon veto ( dpv ) , the rv , the ic , the vc , the ad and the target . the regions of the target , ic and rs traversed by the charged track were excluded from the photon veto . the ad and dpv were part of the e949 detector upgrade but were only used in the @xmath96 analysis , where photon veto near the beam axis was more important . the 1.9-meter - long , 14.3 radiation length thick bv covering 2/3 of the 4@xmath102 sr solid angle was located in the outermost barrel region with an inner radius of 94.5 cm and an outer radius of 145 cm @xcite . the bv was divided into 48 azimuthal sectors . each sector consisted of four radial layers , in which there were 16 ( innermost ) , 18 , 20 , 21 ( outermost ) layers of 1-mm thick lead and 5-mm thick bicron bc408 plastic scintillator . the light collected in the scintillators accounted for 30% of the total energy deposit in the bv . the azimuthal boundaries of each sector were tilted so that there were no projective cracks for photons from the decay vertex . both ends of every module were read out by an emi 9821 kb pmt and the signals were recorded by an adc and a tdc . the time resolution of individual bv counter was measured to be 1.2 ns . in order to improve the photon veto capability , the bvl , located between the rs and the bv , replaced the outermost layers 20 and 21 of the rs in e787 . each bvl counter was 10 cm wide and 2.2 m long . there were 48 azimuthal sectors , each with 12 layers of 1-mm thick lead and 5-mm thick bicron bc408 plastic scintillator , for a total thickness of 2.29 radiation lengths . both ends of the bvl modules were read out by emi 9821 kb pmt s and the signals were recorded by adc s and tdc s . the eight adjacent sectors ( hextant ) in each end were grouped and read out by td s . the timing resolution of an individual bvl counter was 0.7 ns . a comparison of radiation length coverage with and without the bvl is shown in fig.[fig : bvl ] . a factor of two improvement in the photon veto rejection of k@xmath140 decays was expected from the bvl . 0.8 the ec s had roughly one - third of the 4@xmath102 sr photon coverage @xcite . the upstream ec detector consisted of seventy - five 25 cm long ( 13.5 radiation lengths ) undoped cesium iodide ( csi ) crystals segmented into four rings , and the downstream ec detector consisted of 68 crystals in four rings . to maximize light collection the pmt s were directly coupled to the crystals through a sylgard cookie formed over the pmt and a ultraviolet transmitting optical filter that selectively passed the fast component of the csi scintillation light ( with a decay time of a few tens of nanoseconds at a wavelength of 305 nm ) and blocked the slow component . since the pmt s were situated in the magnetic field , high - field pmt s were used @xcite ( hamamatsu r5543 3 pmt s for the outer three rings and hamamatsu r5545 2 pmt s for the smaller inner ring crystals ) ; the signals were split to adc s , constant fraction discriminators ( cfd ) , and ccd s . the cfd s outputs were ored to provide an online photon veto signal and were also sent to tdc s . the pulses recorded in the ccd s were analyzed offline by a pulse - finding algorithm to provide the timing of the ec signals and to separate two pulses close in time ( to reduce the possibility of accidental hits reducing the efficiency of photon detection ) . since the ec was exposed to a high counting - rate environment near the beam line , good timing was required to reduce acceptance losses and the masking of photons from early accidental hits . the upv mounted to the downstream face of the @xmath112erenkov counter was 3.1 radiation lengths thick , with an outer dimension of 28.4 cm @xmath116 28.4 cm and an inner hole for the beam of 17.5 cm ( horizontal ) @xmath116 4 cm ( vertical ) . it consisted of twelve layers of 2 mm thick bicron bc404 plastic scintillator and 12 mm thick lead plate . the scintillation light was read out by @xmath141 wls fibers coupled to two hamamatsu r1924 pmt s . the upv signals were sent to adc s , tdc s and ccd s . the upstream ( downstream ) co detector was located just upstream ( downstream ) of the ec s @xcite . both of the co s consisted of twenty - four 2-mm thick lead sheets alternating with 25 layers of 5-mm thick bicron bc404 scintillator sheets , providing about 9 radiation lengths along the beam direction . each scintillator layer was segmented into 12 wedges forming 12 identical azimuthal sectors . light from the wedges was readout by wls multi - clad fibers ( bicron bcf99 - 29aa ) . each wedge had 16 fibers glued in grooves in the bc404 scintillator layer . one end of the fiber was polished and aluminized to provide reflective mirror surface . all fibers for each sector ( 16@xmath11625=400 ) were terminated in a connector which was coupled to a 1.17-m long lucite rod via a silicone cookie . the rods transmitted the light to emi 9954 kb pmt s in a low field outside the magnet . signals were sent to adc s and tdc s . the co s detected photons emitted with a small polar angle ( @xmath142cos@xmath143 ) in both the upstream and downstream regions . the @xmath125co was installed between the inner face of the magnet end - plate and the target downstream of the downstream co to give more photon detection coverage . the @xmath125co had an inner ( outer ) diameter of 15.6 cm ( 20.0 cm ) and a length of 53 cm . the eight azimuthal sectors contained eight layers of 2 mm scintillating fibers ( from the original e787 target @xcite ) separated by seven layers of 60 @xmath125 m pb . the 536 fibers from two adjacent sectors were read out by one emi9954 pmt into adc s and tdc s . the e949 trigger selected events from the large number of @xmath1 decays and scattered beam particles with requirements on the range of the @xmath85 track , the presence of a decay in the rs , the absence of other activity at the time of the @xmath85 and the presence of a @xmath1 at an appropriately earlier time . the trigger was composed of two stages : a fast level-0 trigger with a decision time of @xmath72100 ns and a slower level-1 trigger with a decision time of 10100 @xmath125s . the level-0 trigger was based on signals from the beam , target , rs and photon veto systems , processed with a combination of commercial and custom - built boards . the level-1.n trigger was composed of two parts running in parallel , the level-1.1 and level-1.2 triggers , that involved the partial processing of td and adc data , and operated only on events that passed the level-0 trigger . the original e787 trigger has been described previously @xcite . the trigger was upgraded for e949 with the addition of a new programmable level-0 trigger board and mean - timers for the photon veto signals @xcite . e949 collected data with triggers for the signal and for other physics studies and calibrations . in this paper , the triggers for calibration and other physics are referred to as the `` monitor '' samples . for physics and most calibration triggers , level-0 required a signal from the beam system and a charged track in the rs . a beam @xmath1 or @xmath85 was identified in nim logic by a coincidence of hits from the @xmath112erenkov counter , b4 ( @xmath144 ) , target ( @xmath145 ) and ags spill gate ( @xmath146 ) . this @xmath119 signal served as the beam strobe ( @xmath147 ) for the trigger . a charged track @xmath148 was identified by a coincidence of the @xmath129 and t - counter and second layer ( mean - time of the two ends ) in a single rs sector . additional requirements were placed on the crude estimate of the track range , which was the track length derived from the deepest layer of the rs hit in coincidence with the @xmath148 from that sector or the two proceeding sectors ( for a positively charged track ) . another requirements were placed on the photon veto ( ec , rs , bv , bvl ) and the delayed - coincidence ( @xmath149 ) . the @xmath149 was an or over the six coincidences of the individual ic s and the delayed @xmath119 signal such that a track in the ic must typically have occurred at least 1.5 ns later than a prompt signal . the @xmath148 signal , an or of the 24 individual sector @xmath148 s , typically served as the detector strobe ( @xmath150 ) that gated all of the adc s not associated with the beam system or target and provided a common stop for many of the tdc s ; the @xmath148 signal introduced 40 ns of dead time . when an event passed the level-0 trigger , the dead time was extended to 100 ns to allow further processing . additional triggers that did not contain both a beam signal and a @xmath148 were produced on a separate trigger board that fed directly into the trigger bus ( with external coordination of the dead time ) . the level-1.1 trigger ( @xmath151 ) was based on information from the td system for the hextant containing the rs counter in which the charged track was determined to have stopped . stopped @xmath85 s were preferentially selected ( over @xmath152 s ) by looking for the @xmath153 decay by comparing the pulse height to the pulse area ( for early decays ) or by looking for a second detached pulse . this was done by a custom - built asic which had access to the td memories . the asic could also reduce the td data size for readout by discarding the waveform samples outside the `` prompt '' time window , keeping instead a calculation of the pulse area and leading - edge time . the `` prompt '' window typically extended from 0.5 ( 2 ) @xmath125s before ( after ) the @xmath85 track . level-1.1 typically provided a decision in about 1020 @xmath125s . the level-1.2 ( @xmath154 ) used data digitized by the adc s and had two components . one of them , the `` level-1.1 afterburner '' rejected events with an accidental hit near the stopping counter . such hits might defeat the level-1.1 with an apparent double pulse . the other component , the `` @xmath155 afterburner '' , was used for the photon veto , rejecting events with hits in both of the two adjacent hextants when the @xmath148 counter and the stopping counter were found in the same hextant . this usually indicated that a single hextant fully contained the trajectory , thus those observed in the other hextant came from another particle . the level-1.2 trigger introduced a dead time of up to 100 @xmath125s per level-1.1 trigger . the triggers as described below were valid for the bulk of the 2002 running period . the trigger conditions for were designed differently according to the @xmath85 momentum . for high @xmath85 momentum , the @xmath92 trigger condition was defined as : @xmath156 where @xmath157 the @xmath158 required that the @xmath85 reached the @xmath159 or @xmath160 layer of the rs and suppressed the copious 3-body @xmath161 and @xmath162 backgrounds . the @xmath163 signal required that the charged track should not reach the @xmath164 layer in order to suppress @xmath82 background . the `` @xmath165 '' designated the rs sectors that were associated with a @xmath148 ( @xmath148 sector plus the next two clockwise sectors : this was the direction that a positive particle moved in the magnetic field ) . the @xmath166 condition was a fiducial cut on the @xmath133-position of the charged track in each layer , vetoing tracks that exited the fiducial volume . the @xmath167 was a refined calculation of the charged track range , which included the number of target fibers hit and a measurement of the @xmath133-position of the track from flash tdc s on layers 3 and 1113 as well as the deepest layer of penetration of the track ; this rejected events with long range such as the @xmath152 from @xmath82 decay . the photon veto @xmath168 and @xmath155 were from the bv , bvl , ec and rs , respectively , which removed events with photons such as @xmath83 , @xmath169 and @xmath101 . the number of @xmath1 s which met the @xmath119 trigger requirement when the detector was live ( @xmath170 ) , was recorded at the end of each ags spill . the total exposure from the @xmath92 trigger data stream was @xmath171 e949 also provided a @xmath96 trigger for lower momentum @xmath85 s . since the study involves different background mechanisms and a different kinematic region , the result will be presented in a separate paper . in addition to the @xmath78 triggers , there were monitor triggers for calibration and normalization , as well as triggers for other physics modes . all triggers were prescaled to reduce the impact on the total dead time except for the triggers and the @xmath172 " trigger which required the presence of photons in the barrel region for studying @xmath173 and @xmath174 decays @xcite . to monitor detector performance several processes were employed , including @xmath83 and @xmath82 decays , beam particles scattered into the detector fiducial volume , beam @xmath1 , charge exchange events and cosmic rays . these monitor samples were used for calibration and acceptance studies . they were taken simultaneously with the @xmath175 and @xmath176 triggers to reflect any condition changed into the acceptance and background calculations . the @xmath82 decay has the largest branching ratio . since the final state does not contain photons or additional tracks and the daughter @xmath152 does not interact strongly , it was a convenient sample for a variety of acceptance measurements as well as several calibrations . this sample was also used for the normalization of the experiment ; the measurement of the @xmath82 branching ratio effectively normalized the counting of @xmath1 stops to the well - known @xmath82 branching ratio . the @xmath177 trigger condition for @xmath82 was defined as follows @xmath178 the final state of the @xmath83 decay mode contains one charged @xmath85 and two photons from @xmath179 decay . since the @xmath85 momentum is monochromatic , the @xmath83 sample was used to check the measurement of the charged track momentum , range and energy . also the @xmath85 s were used to study particle identification , while the photons were used to study the photon veto . two @xmath83 triggers were defined , a loose one , @xmath180 : @xmath181 and a tighter one , @xmath182 : @xmath183 the @xmath182 monitor trigger was used only for calibration in this analysis . among the incoming beam particles there were many @xmath85 s , including some that scattered into the fiducial volume of the rs . these @xmath184 events were identified as @xmath85 s by the @xmath115 and had an in - time track in the rs . the kinematic features of this @xmath85 sample were almost the same as the signal except that the target pattern was different . this sample was suitable for calibrating the ionization energy loss ( @xmath185 ) of @xmath85 s and for studying the acceptance . the trigger condition for scattered beam particles was defined as : @xmath186 where @xmath187 the charge exchange process ( cex ) , @xmath188 followed by @xmath189 can mimic events when the charged lepton @xmath190 from the @xmath191 decay had a low momentum and was undetected . the largest uncertainty in this background was the determination of the reaction rate . since @xmath192 has an equal fraction of @xmath191 and @xmath193 , the @xmath194 process with a @xmath193 decay to @xmath195 can be used to measure the reaction rate for the determination of the background from @xmath196 . since the @xmath197 could be cleanly identified , a @xmath198 trigger with two charged tracks was defined as : @xmath199 in addition , we also defined triggers for beam @xmath1 and cosmic ray , which were used in trigger efficiency measurement and detector geometrical alignment . analog- and discriminated- signals from the detector were digitized by commercial adc and tdc , and custom - built waveform digitizer ( td and ccd ) systems . when an event was accepted by the trigger system , the digitized data for the event were transferred to a buffer module or a local crate controller . at the end of each spill , the data for the spill were transferred to a host computer . a summary of the digitizing electronics is shown in table [ tab : digitizers ] . .digitizing electronics for e949 . [ cols="^,^,^,^,^",options="header " , ] to get the acceptance in the extended signal region , the estimated acceptance gains given in section [ sec : extended ] were applied to the acceptances for @xmath200 , @xmath201 and @xmath202 , yielding the final acceptance of @xmath203 this value is 10% higher than that in e787 . it is noted that the acceptance of the standard e787 signal region for the e949 data was 84% of the acceptance for the e787 data due to losses incurred by the higher than expected instantaneous rates ( section [ subsect : overview ] ) . based on the total exposure of @xmath1 s ( @xmath170 ) , the single event sensitivity ( @xmath204 ) of the e949 2002 run was given by @xmath205 after the background analysis and the acceptance measurement were completed and satisfactory , all the selection criteria were then be applied to the data . at the stage of examining the signal region , no cut could be changed . one candidate event was observed inside the signal region . a close check also found that this candidate was located in the @xmath206 extended region as described in section [ sec : extended ] . [ fig : candidate_r_vs_e ] shows the range and kinetic energy of the events that passed all of the selection criteria , except for the phase space cuts on both the range and energy . this candidate together with the events observed in e787 are also shown in fig . [ fig : candidate_r_vs_e ] . as indicated in the figure , the signal box definition in e949 was extended in comparison to that in e787 . [ fig : event_display_candidate ] is an event display for this candidate . this event had a momentum of @xmath207 mev/@xmath93 , a kinetic energy of @xmath208 mev and a range of @xmath209 cm . kinematically , this candidate event agreed with all the requirements as a signal , though it existed near the limit expected for signal . the measured quantities of the observed candidate used in the selection criteria were compared to the expected distributions for signal to evaluate the signal probability distributions for the candidate . the probabilities for the single beam @xmath1 requirements , the decay @xmath85 kinematic requirements and the @xmath206 decay sequence cuts showed a fairly flat distribution , which was consistent with the expected signal distribution . there was no observed photon activity for this candidate . 0.8 in this section we describe the method used to obtain the branching ratio and the impact of the e949 and e787 candidates on the unitarity triangle . we also describe the implication of the results on the search for the hypothetical decay where @xmath210 is a stable , massless , non - interacting particle @xcite . we defined a number of cells in the extended signal region of differing signal / background , and calculated the expected signal / background ( @xmath211 ) using individual @xmath212 and @xmath213 functions for each rejection or background type . in total , we had seven types : td rejection , pv rejection , @xmath83 background , @xmath82 range tail background , @xmath152 band background , single - beam background and double - beam background . the variation of each type for different cut position or cell could be expressed as the change of @xmath213 as a function of the corresponding acceptance @xmath212 , yielding seven functions in total used in this analysis . two of them were the td rejection vs. acceptance function ( fig . [ fig : nn_func ] ) and the pv rejection vs. acceptance function ( fig . [ fig : pv_tune_rva ] ) . three of them were the relative background rate in the normalization branch vs. the relative acceptance functions for the kinematic background backgrounds ( fig . [ fig : kin_n_a ] ) . the rest were the relative background rate vs. the relative acceptance functions for the beam backgrounds ( fig . [ fig : beam_n_a ] ) . both relative background rates and the relative acceptance curves were normalized to one at the cut positions . 1.0 1.0 the branching ratio was determined using likelihood analysis incorporating the predicted background functions in the signal region . the likelihood ratio @xmath214 was defined as @xmath215 where @xmath216 and @xmath217 were the estimated signal and background in the @xmath218 cell , @xmath219 was the number of signal candidates in the @xmath218 cell and the product ran over all @xmath220 cells @xcite . in addition the likelihood estimator @xmath221 was defined as the value of @xmath214 given the observed candidates . cells were defined based on the predicted background functions described in previous section . the predicted background functions showed that there was additional background rejection capability within the signal region that could be exploited by sub - dividing the signal region . the number of cells to be used for subsequent analysis , 3,781 , was established prior to the examination of the signal region . the total background in the cell containing the signal candidate was estimated to be @xmath222 dominated by a contamination of @xmath223 events due to the @xmath82 range tail background . the ratio of the acceptance in this cell to the total acceptance in the standard region ( @xmath224 ) was estimated to be 1.21@xmath225 . the expected number of signal events in this cell was @xmath226 where was the branching ratio . the central value of the branching ratio , defined as the value of that maximized @xmath221 @xcite , was @xmath227 . using only the e949 data , @xmath228 where the quoted 68% confidence level ( cl ) interval was determined from the behavior of @xmath214 as described in ref . @xcite and included only the statistical uncertainty . the estimated probability that the e949 candidate was due to background alone was 0.074 . the results from e787 and e949 were combined to calculate the branching ratio for . in the e787 analysis , two candidate events were observed in the signal region @xcite . the number of cells describing the e949 signal region were augmented by 488 cells that defined the signal region for the e787 analysis to produce a likelihood estimator @xmath221 for the combined data . the confidence intervals for the combined e787 and e949 results took into account the estimated systematic uncertainties in the signal acceptance and the background rates . the systematic uncertainty of each background source was estimated to be about 15% based upon the results of the correlation studies . from the study in section [ chap : acc_and_sen ] , the systematic uncertainty on the acceptance was estimated to be about 8% . the systematic uncertainty of each background component and the acceptance were assumed to be uncorrelated and to follow a normal distribution with the magnitudes given above regarded as one standard deviation . with these assumptions , the branching ratio for the combined e787 and e949 result was @xmath229 where the uncertainty denoted the 68% cl interval . the corresponding 90% and 95% cl intervals were @xmath230 and @xmath231 , respectively . the estimated probability that all the candidates observed in e787 and e949 were due to background was 0.001 . the inclusion of the estimated systematic uncertainties had a negligible effect on the cl intervals due to the relatively poor statistical precision inherent in a sample of three candidate events . the experimental signature of a decay was identical to that of except that the kinematic signature afforded by the two - body decay ( @xmath233 mev/@xmath93 , @xmath234 mev , @xmath235 cm ) permitted the definition of a relatively high - acceptance , low - background signal region . the analysis method was identical to that used for the analysis except that the signal region was defined to be within two standard deviations of the expected momentum , energy and range of the @xmath85 with the upper limits of the tightened to @xmath236 mev/@xmath93 , @xmath237 mev , and @xmath238 cm to suppress @xmath239 background . the expected background level was small ( 0.05 events ) , because the region was far from the peak . the acceptance studies for decay paralleled those for the . the single event sensitivity for the e949 decay analysis was estimated to be @xmath240 . the candidate event observed in the signal region for e949 analysis was also in the signal region . however , no candidates were observed in the signal region of e787 @xcite . the combined e787 and e949 sensitivity was 0.196@xmath241 and , using the one observed candidate event without subtraction of the estimated background , the upper limit on the branching ratio was @xmath242 at 90% cl using the feldman - cousins method @xcite . this limit was larger than the previous 90% cl limit of @xmath243 of e787 @xcite due to the e949 candidate event . as described in section [ chap : intro ] , the branching ratio was directly related to the real and imaginary parts of @xmath244 ( equation ( [ eq : pnn_br ] ) ) . in fig . [ fig : lambdatfore949 ] the regions of the complex @xmath245 plane allowed by the branching ratio determined from the combined e787 and e949 results was compared to the regions allowed by other recent measurements with small theoretical uncertainties @xcite . the region favored by other ckm - sensitive measurements is at the edge of the 68% cl region allowed by the measurement . 0.9 the other ckm - sensitive results @xcite used to produce the confidence level intervals in fig . [ fig : lambdatfore949 ] are dominated by measurements of @xmath24 meson decays . the possible discrepancy between the @xmath245 regions allowed by the @xmath24-decay measurements and by @xmath246 could be an indication of physics beyond the sm . as emphasized in ref . @xcite , the clean theoretical interpretation of @xmath247 remains valid in most extensions of the sm in distinct contrast to the @xmath24-decay measurements currently used to determine the ckm parameters . thus a precise measurement of @xmath246 would provide an unambiguous consistency test of the flavor sector of the sm . the rare decay is a flavor - changing - neutral - current process and proceeds via 1-loop diagrams mediated mainly by the top quark . measuring ( ) is one of the cleanest ways to extract @xmath30 . in this paper we have reported results from the bnl experiment e949 , an upgraded version of the bnl - e787 experiment , designed to improve the sensitivity for measurement of decay . all the @xmath1 decays at rest were analyzed using a blind analysis technique in which the signal region was masked until the selection criteria were determined and the background levels were estimated . the development of the cuts and the estimation of the background levels were performed using a bifurcation method and , a likelihood analysis method was developed for interpreting the quality of candidate events . enlargement of the signal region compared to e787 analysis increased the acceptance by 30% with a total background level in the signal region estimated to be @xmath2 events . an examination of the signal region yielded one event near the upper kinematic limit of the decay . based on the candidate event , the branching ratio was determined to be @xmath248 . e787 and e949 results were combined and the branching ratio was determined to be @xmath249 at the 68% cl level based on three events observed in the momentum region @xmath250 mev/@xmath93 . the estimated probability that all the candidates observed in e787 and e949 were due to background was 0.001 . the measured branching ratio is in agreement with the sm prediction of within the uncertainty . we gratefully acknowledge the support and efforts of the bnl collider - accelerator division for the high quality @xmath1 beam delivered . we wish to thank jose ocariz of the ckmfitter group for producing fig . [ fig : lambdatfore949 ] . this research was supported in part by the u.s . department of energy , the ministry of education , culture , sports , science and technology of japan through the japan - u.s . cooperative research program in high energy physics and under grant - in - aids for scientific research , the natural sciences and engineering research council and the national research council of canada , the russian federation state scientific center institute for high energy physics , and the ministry of science and education of the russian federation . s. chen was also supported by program for new century excellent talents in university from the chinese ministry of education . d. rein and l.m . sehgal , phys . * d39 * , 3325 ( 1989 ) ; j.s . hagelin and l.s . littenberg , prog . * 23 * , 1 ( 1989 ) ; m. lu and m.b . wise , phys . b324 * , 461 ( 1994 ) ; s. fajfer , nuovo cim . * a110 * , 397 , ( 1997 ) ; c.q . geng , i.j . hsu , and y.c . lin , phys . * d54 * , 877 ( 1996 ) . this definition of the central value of the branching ratio has the desirable feature that in the limit of @xmath251 observed candidates with high signal - to - background , the central value is the same as that given by the product of the sensitivity ( section [ chap : acc_and_sen ] ) and @xmath251 which is the fundamental definition of the branching fraction .

consider a-dimensional antiferromagnet with a quantum disordered ground state and a gap to bosonic excitations with non - zero spin . in a finite external magnetic field , this antiferromagnet will undergo a phase transition to a ground state with non - zero magnetization , describable as the condensation of a dilute gas of bosons . the finite temperature properties of the bose gas in the vicinity of this transition are argued to obey a hypothesis of _ zero scale - factor universality _ for , with logarithmic violations in . scaling properties of various experimental observables are computed in an expansion in , and exactly in .

cond-mat9401040

numerous experiments @xcite have by now examined the properties of @xmath5 spin chain antiferromagnets which posses the haldane gap @xcite . more recently , these antiferromagnets have been placed in a strong magnetic field @xcite , and found to display evidence of a zero temperature phase transition to a state with a non - zero ground state magnetization . theoretical studies @xcite have also examined this transition at zero temperature , and developed a picture of it as the bose condensation of magnons with azimuthal spin @xmath6 . although this possibility has not been considered before , it is not difficult to see that the condensation of magnons in a finite field should occur in quantum - disordered antiferromagnets in any dimension @xmath0 ( provided , of course , the magnons continue to behave as bosons ) . in this paper we shall present a general theory , in dimensions @xmath7 , of the finite temperature properties of quantum antiferromagnets in the vicinity of such a field - induced quantum transition . we begin by elucidating the precise conditions under which our results apply . consider an antiferromagnet with a quantum disordered ground state . the hamiltonian must posses at least an axial symmetry _ i.e. _ at least one component ( say @xmath8 ) of the total spin must commute with the hamiltonian . the lowest excitation with non - zero spin must be separated from the ground state by a gap , and behave like a quasiparticle with bosonic statistics . these conditions are clearly satisfied by haldane gap antiferromagnets in which the in - plane anisotropy can be neglected ( the compound nenp does have a rather small in - plane anisotropy @xcite , and the restrictions this places on applying our results to on experiments in nenp will be discussed later ) . in @xmath2 , the @xmath9 kagome heisenberg antiferromagnet @xcite is probably the most accessible candidate upon which our results can be tested - it has been argued that the magnons in this system are bosons @xcite . now place this antiferromagnet in a magnetic field pointing along the direction of axial symmetry . the eigenenergy , @xmath10 of single magnon ( boson ) quasiparticles with momentum @xmath11 in a field @xmath12 then takes the form : @xmath13 here @xmath14 is the magnon gap and @xmath15 is the quasiparticle mass , both determined in the zero field antiferromagnet . for the @xmath5 haldane gap chain , we have the azimuthal spin @xmath6 . in the kagome antiferromagnet , the magnon excitations have been argued to be spinons @xcite : we therefore expect @xmath16 . later , we will discuss how the field induced transition in the kagome antiferromagnet may offer a way of experimentally / numerically determining the value of @xmath17 . \label{coherent}\end{aligned}\ ] ] where @xmath24 is the @xmath0-component spatial co - ordinate , @xmath25 is the matsubara time , the chemical potential @xmath26 and @xmath27 is a repulsive interaction of a short range @xmath28 . this field theory has a phase transition exactly at @xmath18 , @xmath29 which has been studied by fisher _ et . they identified the upper - critical dimension as @xmath2 , above which the interaction @xmath27 is irrelevant . for @xmath30 , @xmath27 is relevant , but the exponents were nevertheless found to have trivial values : the dynamic exponent @xmath31 , the correlation length exponent @xmath32 , and the field anomalous dimension @xmath33 . the triviality of the exponents is partially related to the fact that the parameter tuning the system through the transition , @xmath34 , couples to a conserved quantity - the density of bosons ; any such transition @xcite must have @xmath35 . the structure of the @xmath30 , finite-@xmath27 fixed point is thus very unusual : despite describing a non - trivial , interacting , critical field theory , the exponents associated with all the relevant directions away from this fixed point are trivial . in this paper , we shall show that the interactions are crucial in determining the finite temperature properties of the bose gas near this fixed point . the fixed - point interactions are needed to preserve hyperscaling for @xmath1 and lead to highly non - trivial scaling functions for the finite - temperature correlations . for @xmath30 , the combination of the presence of hyperscaling , and the absence of any anomalous exponents in the leading critical behavior , leads to remarkably universal finite temperature properties . as in ref . @xcite , we may use finite - size scaling to deduce scaling forms away from @xmath29 , @xmath18 . however the absence of any anomalous dimensions ( @xmath31 , @xmath49 , @xmath32 ) means that the usual two scale - factor universality @xcite is now reduced to a _ zero scale - factor universality . _ described more precisely in the following subsection . in simple terms , this universality is just the statement that all response functions are universal functions of the _ bare _ coupling constants @xmath34 and @xmath15 . there are no non - universal amplitudes ; the usual case has two non - universal amplitudes , or scale - factors @xcite . the universality can be stated more precisely in terms of the the boson green s function @xmath50 where @xmath51 is the ordering symbol in imaginary time @xmath25 . after fourier transformation as per @xmath52 this yields @xmath53 at the matsubara frequencies @xmath54 , from which the retarded green s function @xmath55 can be obtained by analytic continuation to real frequencies using @xmath56 the lehman spectral representation of the green s function implies that @xmath57 , but does not constrain @xmath58 to be an odd function of @xmath59 . our central result is the zero scale - factor universality of @xmath60 , which is equivalent to the scaling form @xmath61 where @xmath62 is a highly non - trivial , but completely universal complex - valued function ; naturally , @xmath62 is independent of the nature of the boson - boson repulsion . an important property of @xmath62 is that it is _ analytic at all finite , real , values of all three arguments . _ similar scaling forms hold for other correlators of @xmath63 - a particularly instructive observable is the _ local _ green s function , @xmath64 @xmath65 if we take the imaginary part of this equation , it is expected that the resulting on - shell contributions on the right - hand - side will occur only at small momenta ( determined by the frequency @xmath59 ) , and the momentum integral is ultraviolet convergent . we may therefore deduce from ( [ scalegr ] ) the scaling form @xmath66 where @xmath67 is a fully universal , dimensionless , positive function . it is quite instructive to consider the limiting behavior of @xmath67 for small and large frequencies . we expect that @xmath68 should be analytic at @xmath69 at any finite @xmath70 ; this combined with the positivity condition on the spectral weight noted above , implies @xmath71 for small @xmath59 at finite @xmath70 . therefore , from ( [ deff ] ) the scaling function , @xmath67 , must satisfy @xmath72 , and @xmath73 ) ; the coefficient of the @xmath74 term is quite difficult to determine , and will be obtained in this paper only in a special limit . for large @xmath59 , or short times , @xmath60 should display essentially free particle behavior , as the dilute bosons have not had enough time to interact with each other . using the free boson spectral weight we can deduce that @xmath75 let us conclude this subsection by noting the precise conditions under which the system is in the critical region and ( [ scalegr ] ) and ( [ deff ] ) are valid . we must have @xmath76 further the measurement wavevectors must satisfy @xmath77 in @xmath2 the zero - scale - factor universality is violated by a logarithmic dependence on the microscopic interactions . furthermore , the scaling function @xmath62 will have a singularity associated with the finite temperature kosterlitz - thouless transition . zero scale - factor universality does not hold for @xmath78 . the dynamic information contained in @xmath58 is directly observable in neutron scattering experiments . in appendix [ shankar ] we discuss the relationship between the correlators of @xmath63 and antiferromagnetic correlations measured by the neutrons ; this discussion is limited to the case where the quantum disordered phase has confined spinons _ i.e. _ @xmath6 , as in haldane gap antiferromagnets . the relationship for the case of deconfined spinons ( @xmath16 ) is quite different @xcite and will not be considered in this paper explicitly . in the following , @xmath79 will measure the energy _ lost _ by the neutrons in their interaction with the antiferromagnet . consider first a scattering even in which the antiferromagnet undergoes a @xmath80 transition . then , from the discussion in appendix [ shankar ] and the fluctuation - dissipation theorem we may conclude that ( for the scattering cross - section is given by the dynamic structure factor @xmath81 where @xmath82 where @xmath36 is a _ non - universal _ quasi - particle renormalization factor between the magnon operators and the ones that couple to the neutrons . ( the wavevector @xmath11 on the left - hand - side is measured from the antiferromagnetic ordering wavevector ) the scaling result for @xmath83 thus follows directly from ( [ scalegr ] ) . next , consider scattering with @xmath84 for the antiferromagnet . the associated dynamic structure factor is then @xmath85 which is @xmath86 by not resolving the energy of the scattered neutron , it is also possible to measure equal - time correlations embodied in the static structure factors . however the scaling properties of these observables are subtle , and require more careful interpretation . by performing a weighted frequency integral over the scaling limit of @xmath60 , we are implicitly only sensitive to frequencies much smaller than a high frequency cutoff like @xmath87 . thus in the following , our equal - time structure factors actually refer to scattering experiments in which energy transfers greater than @xmath87 are not integrated over . not blocking out these events , will produce a background structure factor which may ( as in the haldane gap region defined below ) overwhelm the universal part of the static structure factor which is considered here . keeping this caveat in mind , we define the structure factor @xmath88 @xmath89 where the result is not sensitive to the precise locations of the limits . the correlator @xmath90 can be defined analogously . from ( [ chipm ] ) and ( [ scalegr ] ) we can deduce the scaling result @xmath91 where @xmath92 is a universal scaling function determined completely by @xmath62 ( similarly for @xmath93 ) . finally , using the fact that @xmath63 and @xmath94 are canonically conjugate fields , it possible to deduce a frequency sum - rule on @xmath95 which leads to @xmath96 where we will henceforth use @xmath97 . we reiterate that all of the above results in this subsection refer only to the case of antiferromagnets with confined spinons . a second useful set of observables are those associated with magnetic fluctuations around @xmath98 . the uniform magnetization is a conserved quantity and , as a result , the quasi - particle renormalization factor @xmath36 does not appear in their scaling forms . the simplest of these is mean - value of the magnetization density @xmath99 itself , which is of course related to magnon density by @xmath100 this relationship , and the considerations of this subsection , apply to both confined ( @xmath101 ) and deconfined ( @xmath102 ) spinons . the mean - value of the magnetization is therefore @xmath103 and obeys the scaling form @xmath104 where the function @xmath105 is again dependent only on @xmath106 . the result ( [ magt0 ] ) follows from ( [ defb2 ] ) . the above discussion shows that the scaling functions of a large number of experimental observables can be obtained directly from the primary scaling function @xmath62 . the remainder of the paper is therefore devoted primarily to describing @xmath62 and associated scaling functions in different parameter regimes . it is convenient to discuss the properties of @xmath62 separately in three distinct regimes , which are analogous to those found by chakravarty _ et . @xcite in the @xmath2 @xmath107 sigma model . these regimes are illustrated in the phase diagrams in figs [ figd1 ] ( @xmath4 ) and [ figd2 ] ( @xmath2 ) . the crossover boundaries between the regimes are delineated by the value of the dimensionless ratio @xmath108 ( upto logarithmic terms in @xmath2 ) . we consider the three cases separately this is the analog of the quantum - disordered regime of refs @xcite . only a dilute gas of thermally - excited bosons is present , and their mutual interactions are weak . properties of the quantum antiferromagnet can be described by a low - magnon - density expansion about the quantum disordered ground state . in @xmath4 , we identify this as the haldane - gap regime in fig [ figd1 ] . this is the quantum - critical regime , in which @xmath31 critical fluctuations are quenched in a universal way by the temperature . the value @xmath111 is the dominant energy scale and universally determines everything : the boson density , spectrum , and interactions . the only small parameter which may be used to determine scaling functions is @xmath112 . as we will see , this is not particularly effective in @xmath4 where , fortunately , exact methods are available . it is instructive to consider the physics of this region as a function of the measurement frequency @xmath59 ( see fig [ omega ] ) . at large frequencies , @xmath113 , or short times , the effects finite temperature have not yet become manifest , and the system displays the physics of the @xmath114 critical field theory i.e. it is a dilute gas of bosonic quasiparticles with repulsive interactions . as one lowers the frequency through @xmath115 there is a crossover to a novel @xmath31 , quantum - relaxational regime ( fig [ omega ] ) . now each boson interacts strongly with thermally excited partners , leading to strong dissipation and overdamped quasiparticles . the behavior in @xmath4 and @xmath2 is quite different and we will therefore consider the two cases separately . in @xmath4 , the ground state is a luttinger liquid , which is itself a critical phase with @xmath117 ( fig [ figd1 ] ) . again , consider the physics as a function of @xmath59 ( fig [ omega ] ) . for sufficiently large @xmath59 ( @xmath118 ) we have the dilute bose gas physics of the @xmath114 critical point , similar to that discussed above for the quantum - critical region . at smaller @xmath59 , there is a crossover ( near @xmath119 ) to a luttinger liquid - like region where we may as well assume that @xmath18 . however at small enough frequencies , @xmath120 , the effects of a finite temperature finally become apparent . the massless modes of the luttinger liquid are then quenched into a @xmath117 quantum - relaxational regime , rather similar to the @xmath31 quantum relaxational regime discussed above . this last crossover is , strictly speaking , a property of the @xmath117 critical point on the luttinger liquid fixed line determined by the value of @xmath34 , and not a property of the @xmath31 critical end - point at @xmath114 . it is thus described by a reduced luttinger liquid scaling function @xmath121 . we will obtain exact results for @xmath121 using an argument based on conformal invariance . we will also discuss an important compatibility condition between the scaling functions @xmath62 and @xmath121 , and show how the more general @xmath62 collapses into the small @xmath34 limit of @xmath121 . in @xmath2 , ( fig [ figd2 ] ) the ground state is a boson superfluid , which survives at finite temperature . there is a kosterlitz - thouless phase transition at a finite @xmath70 , which has been studied in some detail earlier @xcite . in the superfluid phase therefore , one has a large frequency dilute bose gas behavior crossing over to a small frequency goldstone phase with quasi - long - range order ( fig [ omega ] ) . the outline of the rest of the paper is as follows . we will begin , in section [ rganal ] , by a renormalization group analysis of the partition function @xmath36 . this will allow us to demonstrate the logarithmic corrections to scaling in @xmath2 and obtain the leading terms of the in a @xmath3 expansion . these leading terms are consistent with the zero scale - factor universality . we will then , in section [ onedimension ] , turn to a discussion of the exact properties of @xmath62 and @xmath121 in @xmath4 . a brief summary and a discussion of relevance to experiments appear in section [ conc ] . three appendixes contain discussions of some peripheral points . the momentum shell renormalization group equations for @xmath36 have already been obtained by fisher and hohenberg @xcite . however their analysis of the equations was restricted to @xmath2 , @xmath122 , and temperatures at or below the kosterlitz thouless transition . in this section we will extend this analysis to cover the remaining regions in @xmath2 ( fig [ figd2 ] ) , and to dimensions @xmath123 in an @xmath3 expansion ( fig [ figd1 ] ) . the analysis @xcite proceeds by introducing an upper cutoff , @xmath124 in momentum space , and replacing @xmath27 by a contact interaction @xmath125 . degrees of freedom in a shell between @xmath124 and @xmath126 are integrated out , followed by a rescaling of co - ordinates and field variables @xmath127 note that the there is no anomalous dimension in the rescaling factor for @xmath63 as @xmath49 . further , the scaling dimension of @xmath128 is exactly @xmath0 , as it must be for any conserved charge density @xcite . it is convenient to consider the cases @xmath18 and @xmath129 separately : the renormalization group equations are @xcite @xmath130 where @xmath131 and @xmath132 is the surface area of a unit sphere in @xmath0 dimensions . we integrate these equations to a scale @xmath133 where the system is non - critical _ i.e. _ when @xmath134 where @xmath135 is significantly smaller than unity , but not so small that the system is still critical . for @xmath122 , the magnetization , @xmath99 , and the boson density @xmath136 are then given by @xmath137 for @xmath30 , and @xmath138 small , @xmath139 approaches its fixed point value @xmath140 inserting this fixed - point value and the dependence of @xmath141 on the initial value of @xmath34 into ( [ mrg ] ) we find to lowest order in @xmath142 that @xmath42 indeed has the form ( [ magt0 ] ) with the universal number @xmath44 given by @xmath143 in @xmath2 , @xmath139 approaches 0 logarithmically slowly . for large @xmath144 we have @xmath145 inserting this into ( [ mrg ] ) we find @xcite @xmath146 note the logarithmic violation of the perfect scaling of ( [ magt0 ] ) . we will restrict our analysis to the center of the quantum critical region in @xmath2 ( fig [ figd2 ] ) and @xmath123 ( fig [ figd1 ] ) : the initial value of @xmath34 will therefore be fixed at @xmath148 and the initial value of the temperature @xmath70 will be close to 0 . we will only need the finite @xmath70 renormalization group equations for @xmath34 and @xmath70 which are @xcite @xmath149 to leading order in @xmath142 ( leading - logs ) it is sufficient to assume that for @xmath30 ( @xmath2 ) @xmath139 is given by eqn ( [ ufixed ] ) ( eqn ( [ ud2 ] ) ) . we will now integrate the renormalization group equations until a scale @xmath141 where @xmath150 the correlation length @xmath151 is then given by @xmath152 while the boson density is @xmath153 in the last step we have anticipated that to leading order in @xmath142 it is sufficient to evaluate the integral directly in @xmath2 . let us now examine the results of integrating ( [ rgt ] ) for @xmath154 . to leading order in @xmath142 we find @xmath155 using @xmath156 , it is straightforward to perform the integration and obtain from ( [ deflst ] ) the leading result for @xmath157 : @xmath158 from ( [ defxi ] ) we therefore deduce @xmath159 and from ( [ resn ] ) we obtain for the boson density @xmath160 these last two results are consistent with the zero - scale - factor universality of ( [ defb1 ] ) and ( [ defb2 ] ) and yield properties of the scaling functions @xmath92 and @xmath105 at @xmath29 . finally consider properties in the quantum - critical region in @xmath2 . the analog of eqn ( [ muint ] ) is @xmath161 integrating this to leading - log accuracy and using ( [ deflst ] ) we find @xmath162 we therefore have from ( [ defxi ] ) for the correlation length @xmath163^{1/2}}\ ] ] which violates the universality of @xmath164 at @xmath29 by the double logarithms . from ( [ resn ] ) we get for the boson density @xmath165 ^ 4},\ ] ] again logarithmically violating ( [ defb2 ] ) at @xmath29 . we have so far determined that for small @xmath3 the @xmath114 critical field theory has a contact interaction of strength @xmath166 , and all other two / multi - particle interactions can be neglected . remarkably , following haldane @xcite , it also possible to determine the exact critical field theory for @xmath167 or @xmath4 . the critical field theory then has @xmath168 ( the bosons are thus impenetrable ) . moreover , all other boson interactions continue to be irrelevant . in appendix [ canonical ] we consider a one - dimensional bose gas in the vicinity of this strong - coupling fixed point and demonstrate this explicitly . the methods of appendix [ canonical ] and earlier works @xcite use the well - known equivalence between the @xmath4 impenetrable bose gas and free fermions . the field theory of the critical end - point at @xmath114 is therefore given by the free fermion hamiltonian @xmath169 where @xmath170 are canonical fermion fields . correlators of this theory can only depend upon @xmath15 , @xmath34 , and @xmath70 , and the zero scale - factor universality is therefore manifest . the scaling limit of the correlators of the uniform magnetization can now be obtained almost trivially : the uniform magnetization density just measures the number of particles and its correlators are therefore the same as those of @xmath171 . in particular we have for the scaling function for the uniform magnetization in ( [ defb2 ] ) @xmath172 where @xmath173 . this scaling function has the limiting value @xmath174 as @xmath175 in the haldane gap region , and @xmath176 as @xmath177 in the luttinger liquid region ( fig [ figd1 ] ) . this last result combines with ( [ defb2 ] ) to yield ( [ magt0 ] ) with @xmath46 @xcite . observables associated with correlations of the staggered magnetization , like @xmath58 , are much more difficult to obtain - it is necessary to express the impenetrable bose fields in terms of the fermi fields by a continuum jordan - wigner transformation @xcite and then evaluate the correlator - a naive wick s theorem expansion of this correlator will yield an infinite number of terms . recently , korepin and slavnov @xcite , following earlier work of lenard @xcite , have succeeded in resumming this expansion and showing that all space - time dependent , finite temperature , correlators of the impenetrable bose gas can be expressed in terms of the solution and fredholm determinant of a linear fredholm integral equation . thus determination of the universal scaling function @xmath62 in @xmath4 has been reduced to the problem of solving completely an integral equation , and taking the fourier transform of the result . analytic methods can take us no further , and it is necessary to resort to numerical analysis of the integral equations . we have begun such a numerical program , and have so far obtained essentially exact results for the equal - time correlations - these are described below in section [ eqtime ] . it should be possible to extend our results to obtain local , time - dependent correlations ( and hence the scaling function @xmath67 in ( [ deff ] ) ) but we have not yet done so . a general picture of the form of @xmath67 can be obtained from the asymptotic limits quoted in section [ introzerosc ] ; should it become experimentally useful to obtain more precise numerical results for @xmath67 , we shall be happy to provide them . we also note that , recently , korepin and collaborators @xcite have succeeding in determining exact results for certain asymptotic properties of @xmath58 by applying the quantum - inverse scattering method to the integral equations noted above . for the equal - time @xmath58 they obtained results for the leading and next - to - leading terms as @xmath178 , while for unequal time correlators , both @xmath24 and @xmath25 were sent to @xmath179 . unfortunately , these asymptotic results are not very useful in determining experimental observables which require fourier transformation to functions of momenta and frequency ( the large @xmath24 behavior of a function implies little about the small @xmath11 limit of its fourier transform ) . simply fourier transforming the asymptotic terms leads to results which compare very poorly with the exact results which we obtained by the alternative means described below . we comment on some features of these exact asymptotic results in appendix [ itsetal ] . in section [ lutliq ] we will consider the limit @xmath116 where it is possible to make much greater analytic progress in determining the scaling functions . as we have already noted , the lower frequency properties in this region are described by luttinger liquid criticality , and it possible to use conformal invariance arguments to obtain closed - from results . the most convenient procedure for determining equal - time correlations begins with lenard s @xcite result for the density matrix of the impenetrable bose gas , which is tantamount to a formal solution of the integral equation of ref @xcite . his result can be written as @xcite @xmath180 where the operator @xmath181 acts on the real axis between 0 and @xmath24 , and has the matrix elements @xmath182 _ i.e. _ the fermion green s function . the form ( [ lenard ] ) is amenable to rapid numerical evaluation . we discretize the real line between 0 and @xmath24 into @xmath183 points , whence the operators in ( [ lenard ] ) become @xmath184 toeplitz matrices . a straightforward extension of levinson s algorithm @xcite was then used to compute the determinant and inverse of @xmath185 . the computer time required for this step scales only as @xmath186 , and we were able to easily uses values upto @xmath187 . the results for @xmath188 for large @xmath24 were compared with the exact asymptotic results of ref @xcite , with excellent agreement . finally , we performed a spatial fourier transform , and obtained results for the scaling function @xmath189 of the structure factor , @xmath190 defined below ( [ defb1 ] ) and by ( [ bb1 ] ) . our results for some representative values of @xmath191 are shown in fig [ figb1 ] . a computer program to obtain numerical values of @xmath192 ( @xmath193 ) for arbitrary @xmath194 can be obtained from the authors ; the accuracy is limited only by computer time , but it is possible to obtain 3 significant figure accuracy quite rapidly . we note that the zero temperature limit of @xmath190 was computed by vaidya and tracy @xcite : they also pointed out that the @xmath18 , @xmath190 has non - analyticities @xcite ( which are , however , unobservably weak ) at integer multiples of @xmath195 , where @xmath196 is the fermi wavevector of @xmath197 this is the regime where @xmath116 . in the fermionic description of the problem , this means the fermi sea is much deeper than the temperature . in this case the ground state and its pertinent excitations can be described by a theory in which the spectrum is linearized near the fermi points @xmath198 determined by @xmath199 by the same token , as long as the probe frequency @xmath59 and momentum @xmath11 do not cause excitations that probe the sea deeply ( i.e. stay in the linearized region ) the bose gas will exhibit characteristics of the line of finite-@xmath34 , @xmath117 luttinger - liquid critical - points at @xmath18 - the luttinger - liquid scaling function for the green s function will be denoted by @xmath121 . in terms of fig [ omega ] , it means that the @xmath121 will describe the lower frequency crossover around @xmath120 . the ratio @xmath200 can take arbitrary values as long as both @xmath201 and @xmath111 remain significantly smaller than @xmath34 . the depth of the fermi sea will not enter any of the calculations of @xmath121 and @xmath34 will enter only via the fermi velocity @xmath202 . for example , in the low density limit , @xmath202 is given by @xmath203 we shall see all this happen as we analyze the exact results momentarily . the crossover around @xmath204 in fig [ omega ] is _ not _ part of the luttinger liquid criticality , and is instead associated with the @xmath31 critical end - point , and the scaling function @xmath62 . it should be apparent from this discussion that the limits @xmath205 and @xmath206 of @xmath62 do _ not _ commute . note that luttinger - liquid criticality , and associated scaling forms , hold even when the condition that @xmath34 be small in ( [ condmu ] ) is violated . suppose , however , that @xmath207 and @xmath208 are much smaller than @xmath34 , and that @xmath209 itself is small so that eqn.([condmu ] ) is satisfied . _ then zero scale - factor universality of the @xmath31 critical point at @xmath29 , @xmath18 must be simultaneously satisfied . _ this will lead to a compatibility condition condition between the scaling functions @xmath62 and @xmath121 , which we shall shortly examine . let us begin by writing down the scaling forms of the luttinger liquid at @xmath18 . we know that at equal times @xmath210 the constant @xmath211 , and the exponent @xmath212 , will , in general , have a non - universal dependence upon the microscopic couplings . however , knowledge of @xmath211 , @xmath212 , and a zero - sound velocity , @xmath202 , will universally determine all remaining hydrodynamic properties in the luttinger liquid regime . for example the unequal time correlation function will have exactly the same form as above with @xmath24 replaced by the euclidean distance @xmath213 . as for @xmath212 , it has a value that depends on the luttinger coupling . at the point @xmath214 , the bosons are impenetrable and equivalent to free fermions . for @xmath215 , the deviation from the impenetrability condition can be translated into a residual interaction between fermions by integrating out doubly occupied states ( see appendix [ canonical ] ) . this is the marginal coupling of the luttinger liquid . let us note for future reference that @xmath216 for zero luttinger coupling . some readers may have trouble reconciling this with the fact that fermion - fermion correlation functions fall as @xmath217 in free - field theory . however , we have already noted that there is a rather complicated between the fermi and bose fields and we remind the reader of the chain of transformations relating the two . first the hard - core bosons are described by the pauli matrices @xmath218 . the latter are than converted to a single component fermions @xmath170 by a jordan - wigner transformation . these fermions are filled upto some fermi momentum determined by the chemical potential . when linearized near the fermi points , the spinless fermi field turns into a pair of relativistic fields @xmath219 . the hard - core boson correlation function at equal times is @xmath220 if we now write @xmath221 and drop terms that oscillate at @xmath222 , we obtain @xmath223 clearly this is a complicated object in the fermi theory . to evaluate it one uses bosonization . using the standard dictionary it is possible to show that it is proportional to the two - point function @xmath224 where @xmath225 is the field dual to the usual boson field . ( see for example ref @xcite ) . we now consider the correlations at finite temperatures . in general the passage from zero to nonzero temperatures is nontrivial since in the latter case non only the ground state but excited state correlations enter . however in two euclidean dimensions , in a relativistically invariant theory such as this one , we have the remarkable result from conformal field theory that was first pointed out by cardy @xcite . using the conformal mapping between the infinite - plane and the strip of finite - length @xmath226 along the imaginary - time direction , one can obtain from eqn.([defeta ] ) @xmath227^{-\eta /2 } \label{cardy}\ ] ] ( the operator @xmath228 is a primary field , which allows us use conformal invariance as above . ) the subscript @xmath229 has been placed to emphasize that this formula is valid only near the lower frequency crossover in fig . [ omega ] the ease with we obtain this result should not detract us from noting its importance rarely does one have the thermally averaged correlation functions of an interacting system . we shall therefore spend some time analyzing this result . as a first step let us extract form this the correlation function at equal time , for long distances . it is readily seen that @xmath230 now let us consider the regime where ( [ condmu ] ) , is also satisfied and so zero - scale - factor universality holds . thus @xmath211 , @xmath212 , and @xmath202 can no longer be nonuniversal , but must be universal functions of @xmath34 and @xmath15 . let us now invoke the asymptotic ( long distance ) hard - core boson scaling functions at the @xmath231 critical point which are known by exact solution @xcite : @xmath232 where @xmath233 . the constant @xmath234 is a known universal number ; more details on this formula are relegated to appendix [ itsetal ] . we see that this agrees with the luttinger liquid result eqn.([luttcorr ] ) if @xmath235 notice also how the chemical potential entered only via the fermi velocity as anticipated ( excluding the @xmath34 dependence of the pre - factor @xmath211 ) . we now consider the fourier transform of ( [ cardy ] ) to obtain the corresponding @xmath236 at the matsubara frequencies along the imaginary frequency axis @xmath237 given the scaling form of @xmath238 it follows that the two integrals involved in the transform lead us to the scaling form @xmath239 where @xmath121 is a completely universal scaling function , dependent only upon the value of @xmath212 . we will determine @xmath121 in closed form below . now if we take the limit @xmath240 at _ fixed _ @xmath241 and @xmath242 ( while satisfying ( [ condmu ] ) of course ) , the system is described simultaneously by the luttinger liquid result ( [ scalelut ] ) and the zero scale - factor universality of ( [ scalegr ] ) . comparing these two results and using ( [ dval ] ) we obtain immediately the compatibility condition between the reduced scaling function @xmath121 at @xmath243 and the scaling function @xmath62 @xmath244 we are using here , as before , the notation @xmath245 , @xmath246 and @xmath73 . we have demanded here that @xmath34 enters @xmath121 only through @xmath202 . notice that the @xmath247 limit is taken at fixed @xmath248 , and as we have noted before and shall see explicitly below , the @xmath249 limit of @xmath121 does not agree with the @xmath250 limit of @xmath62 which was implicit in ( [ freeparticles ] ) the remainder of this subsection is devoted to obtaining explicit results for @xmath121 and associated scaling functions in minkowski space . there are two ways to proceed . the first approach is to perform the transform for complex ( matsubara ) frequencies and then make the substitution @xmath251 to obtain @xmath252 . the transforms are tedious to perform but the interpretation of the results is instructive and we shall do so soon . the general principles we learn about analytic continuation into the complex plane are usually illustrated with trivial examples ( i.e. , noninteracting propagators ) and here we have one of the few nontrivial and hence instructive cases . the second approach is peculiar to this problem and relevant because conformal invariance methods always give the correlations in coordinate and not momentum space . thus one can continue the results from imaginary to real time first and then take the transform . that calculation may be found in section 3.3 of ref @xcite and has its own pedagogical value . returning to the first approach , we used the identity @xmath253 to put the @xmath254 and @xmath255 terms in ( [ cardy ] ) up in the exponent . the @xmath24 and @xmath25 integrals were then analytically performed , followed finally by the @xmath256 integration . after using ( [ scalelut ] ) , the final result gave us values of the scaling function @xmath121 at the matsubara frequencies along the imaginary frequency axis . we found @xmath257 this result was obtained earlier by schulz @xcite and shankar @xcite in a different context but not analyzed in any detail . first , the above function @xmath258 specifies our knowledge at positive and negative matsubara points . our goal is to construct the physical real frequency correlation function and its singularity structure from it . as it stands , @xmath121 can only be used for numerical purposes and not for studying analytic structure since @xmath259 is neither analytic nor anti - analytic . in other words we can use @xmath121 to calculate values of the putative function at the matsubara points . the mod symbol tells us that the function we are seeking has the same value at any postive matsubara point and its negated image . now , on the real frequency axis we have a retarded correlation function , which we assume is well defined . since the factor @xmath260 converges in the upper half - plane ( uhp ) the function on the real axis has an analytic extension to the uhp which is free of singularities . its values at the matsubara points @xmath261 , with @xmath42 an integer are given by @xmath121 . such a function is readily found : simply drop the mod symbol on @xmath248 in the formula for @xmath121 . let us call the function ( with the mod symbol dropped ) @xmath262.this @xmath263 is the unique analytic function ( with good behaviour in the uhp ) determined by our knowledge at postive matsubara frequencies . being an analytic function it has a continuation to the lower half - plane ( lhp ) which is however not guaranteed to be free of singularities or to have anything to do with the original problem . in particular the poles that the @xmath264 functions have in the lhp are not germane to the physical response function . in fact this continuation to the lhp of @xmath262 does not even agree with the data we have in eqn.([resal ] ) for negative matsubara points : since @xmath265 it is not invariant under the change of sign of frequency as the given data are . however , there is an analytic function which will duplicate the given data in the lhp : it is obtained by replacing @xmath266 by @xmath267 in eqn.([resal ] ) . such a function , @xmath268 , satisifes @xmath269 this function will agree with @xmath270 of eqn.([resal ] ) at points with negative matsubara frequencies and be free of singularities in the lhp . however its poles in the uhp have no physical significance . thus the function at real frequencies is the limit of two different functions as we approach the real axis from above or below . _ the true singularities of the physical response function are due to the mismatch of these two functions and not due to the poles they have in regions where they no longer represent the physical function or a continuation of it . _ so we must consider the difference between these two functions @xmath263 and @xmath271 on the real frequency axis . it is readily shown that the two functions are conjugates of each other there so that the discontinuity is just twice the imaginary part . the physical response function s real singularity is therefore a cut and not poles . this is reasonable since a theory with gapless excitations on top of the ground state ( and hence a cut at zero temperature ) can not lose its spectral weight in these regions by the inclusion of higher states in the thermal average at finite temperatures . ( the converse is possible : a cut free region on the real axis at @xmath18 can close up at @xmath272 . ) in fig [ luta ] we plot @xmath273 as a function of @xmath248 and at a representative set of values of @xmath274 . notice that for large @xmath274 there is a well defined peak at @xmath275 : this is a signature of the propagating modes in the luttinger liquid ground state and represents the behavior of the intermediate region @xmath276 in fig [ omega ] . at smaller values of @xmath274 notice that the peak in fig [ luta ] remains at @xmath277 . this is the @xmath117 quantum relaxational behavior ( fig [ omega ] ) where the strong interaction between the thermally excited luttinger modes has left only overdamped excitations . we now consider a couple of other experimental observables , related to local and equal - time correlations respectively . the local green s function , @xmath64 was defined in ( [ defloc ] ) . in the luttinger liquid regime , we can deduce that , provided @xmath278 , this observable satisfies the scaling form @xmath279 where @xmath280 is a universal function , specified completely by the function @xmath121 in ( [ defloc ] ) . moreover , as the luttinger - liquid criticality has a particle - hole symmetry , @xmath281 must be an even , positive function of @xmath248 . as already noted , we expect on general grounds that @xmath71 for small @xmath59 at finite @xmath70 : therefore @xmath282 at small @xmath248 . we also note that for @xmath283 the _ real _ part of the local green s function will not satisfy an analogous because the integral in ( [ defloc ] ) is then dominated by large momentum contributions . there is again a compatibility condition between the luttinger liquid scaling function @xmath280 and the scaling function @xmath67 in ( [ deff ] ) quite analogous to that for @xmath121 , @xmath62 in ( [ acompat ] ) ; we have @xmath284 as before , the limits @xmath249 and @xmath285 do not commute , and the @xmath286 limit of the exact @xmath281 computed below will not agree with that of @xmath67 in ( [ freeparticles ] ) let us finally present the exact computation of @xmath281 . we use the result ( [ cardy ] ) at @xmath287 , fourier transform to matsubara frequencies , analytically continue and take the imaginary part to obtain the following result for @xmath281 @xmath288 a plot of this function is shown in fig [ lutf ] . for small @xmath248 we have @xmath289 this is the behavior characteristic of the @xmath117 quantum - relaxational regime of fig [ omega ] . it is expected that the limits @xmath290 and @xmath177 _ do commute _ , so combined with ( [ fcompat ] ) , the above result gives us the small @xmath248 behavior of @xmath291 at large values of @xmath292 . in the opposite limit of large @xmath248 we crossover to the critical correlations of the luttinger liquid ground state in which case @xmath293 this last result can also be obtained by a fourier transform of the relativistic zero temperature correlator . the two structure factors @xmath294 and @xmath90 of the antiferromagnet were defined in eqn ( [ chipm ] ) . the luttinger liquid behavior has particle - hole symmetry so in this regime , the two structure factors are essentially equal and will be denoted by by the common value @xmath295 . the scaling form for @xmath296 follows from ( [ introz ] ) , ( [ chipm ] ) and the scaling of @xmath297 in ( [ scalelut ] ) : @xmath298 where the constant @xmath36 was introduced in ( [ introz ] ) and @xmath299 is a universal function obtained below . there is a compatibility condition between the luttinger liquid scaling function @xmath299 and the scaling function @xmath92 in ( [ defb1 ] ) which is quite analogous to that for @xmath121 , @xmath62 in ( [ acompat ] ) : @xmath300 using the other scaling function @xmath189 on the right - hand side would yield , from ( [ bb1 ] ) , an identical result . again the limits @xmath301 and @xmath302 are not expected to commute . finally , the exact determination of @xmath303 : we simply perform a spatial fourier transform of ( [ cardy ] ) at @xmath304 ; we obtain in this manner @xmath305 this result is illustrated in fig [ lutb ] . for small @xmath274 , @xmath303 reaches a constant whose value is easily obtainable from ( [ resbl ] ) . for large @xmath274 we find from ( [ resbl ] ) @xmath306 again this last result could have also been obtained by direct computation at @xmath18 . this paper has studied the universal , finite temperature properties of a dilute bose gas with repulsive interactions in dimensions less than or equal to 2 . in the vicinity of the @xmath18 onset at zero chemical potential , @xmath34 , it was argued that the leading scaling properties obey , for @xmath30 , a hypothesis of zero scale - factor universality . this means that the entire two - point correlator is a universal function of just @xmath34 , @xmath70 and the bare boson mass @xmath15 . the main motivation behind this study is the mapping onto it of the properties of quantum - disordered antiferromagnets in a finite field . in particular , in @xmath4 , haldane gap antiferromagnets undergo a magnetization onset at a critical field which is expected to be in the universality class of the bose gas transition . applicability of our theory requires that there be no spin anisotropy in the plane perpendicular to the applied field . most materials do have some anisotropy - in this case we would require that the temperature , @xmath70 , be larger than the anisotropy gap , before applying our results . we have described the rather complicated properties of numerous scaling functions , which may rather difficult to disentangle experimentally . a useful starting point for neutron scattering experiments appears to be the following . perform the experiment somewhere in the luttinger liquid region where the absolute value of the scattering cross - section is also the largest . measure the local susceptibility @xmath307 and see if collapses onto the scaling form ( [ deff ] ) . for large @xmath308 , we have a rather complete picture of the scaling function @xmath67 : for @xmath59 smaller than or around @xmath309 , we can deduce @xmath67 from ( [ fcompat ] ) and ( [ resfl ] ) , while for extremely large @xmath59 we can use ( [ freeparticles ] ) . another possible application of our results may be to quantum - disordered antiferromagnets in @xmath2 . by measuring the ground - state magnetization in a field , and comparing the result to ( [ muvalh ] ) , ( [ mvaln ] ) and ( [ popovres ] ) it may be possible to determine the spin @xmath17 of the elementary excitations above the ground state . of course , we would also need an independent determination of the quasi - particle mass @xmath15 . we thank i. affleck , i. gruzberg , b.i . halperin , p. hohenberg , and v. korepin for helpful discussions and correspondence . the research was supported by nsf grants no . dmr-9120525 and dmr-9224290 . it is clear that the predominant coupling of neutrons will be to the antiferromagnetic order parameter @xmath310 , with @xmath311 . in a zero - field spin - fluid phase with confined spinons , this order parameter corresponds to a real , massive , bosonic triplet . in this appendix , we want to explore in some more detail the relationship between @xmath310 and the complex bosonic field , @xmath63 in in eqn.([coherent ] ) . we note that the field theory ( [ coherent ] ) will also describe the magnetization onset transition in antiferromagnets with deconfined spinons ; however , in this case , the relationship between the neutron scattering cross - section and the field @xmath63 will be quite different @xcite , and will not be considered in this paper explicitly . for simplicity , we consider @xmath4 , although the analysis is quite general . first we expand the real triplet , @xmath310 , in terms of magnon creation and destruction operators as usual : @xmath312\ ] ] where @xmath313 @xmath14 is the haldane gap , and @xmath314 , @xmath315 the magnon destruction and creation operators for each of three polarizations . let us focus on the two combinations @xmath316 .\ ] ] observe that @xmath317 are adjoints of each other , but commute with each other . let us now consider an effective theory for energies far below the haldane gap . in this case we can make the replacement @xmath318 and obtain @xmath319 \\ & \equiv & { 1 \over \sqrt{2\delta } } [ \psi_{\pm } ( x ) + \psi^{\dag}_{\pm}(x ) ] \end{aligned}\ ] ] where @xmath320 destroys a spin @xmath321 magnon at @xmath24 while @xmath322 creates a spin @xmath323 magnon at @xmath24 . suppose we next argue that , when the applied uniform field is near its critical value , only the spin up magnon ( very light ) will either be easily created or destroyed , so that we may drop the spin down creation and destruction operator in the above expressions . then we obtain @xmath324 observe that now ( up to a scale factor ) @xmath317 , which were previously commuting , _ are now canonically conjugate fields . _ this is just like in the hall effect wherein @xmath24 and @xmath325 , which are commuting coordinates in the full hilbert space become conjugates in the lowest landau level . it is also clear from the discussion that the field @xmath63 in the coherent space integral in eqn.([coherent ] ) is precisely this complex conjugate pair . consider bosons @xmath326 moving on the sites , @xmath327 of a chain described by the hamiltonian @xmath328 where @xmath329 is the number operator , @xmath330 is the hopping matrix element , and @xmath331 is the on - site repulsion between the bosons . in the limit of large @xmath331 , states with more than one boson on a site will only occur rarely , and it should pay to restrict the hilbert space by projecting out such states . however , the elimination will generate a residual interaction of order @xmath332 between the states on the restricted space . this interaction can be determined by the usual second - order perturbation theory and leads to the effective hamiltonian @xmath333 we reiterate that @xmath334 is non - zero only on states with at most one boson per site . notice now that this reduced hilbert space is identical to that of spinless fermions . the transformation between the @xmath326 and the spinless fermion operators @xmath335 is of course the jordan - wigner mapping @xmath336 we now insert ( [ jw ] ) in ( [ heffec ] ) and take the continuum limit with @xmath337 , @xmath338 ( @xmath40 is the lattice spacing ) and obtain @xmath339\ ] ] it is now clear by power counting that the four - fermi coupling term is clearly an irrelevant perturbation to the @xmath114 fixed point . it is in fact also not difficult to see that all interactions between the fermions are irrelevant , and that this result is not special to the boson hubbard model considered here . the key point is of course that a term like @xmath340 , which is the only interaction term which is relevant by power counting about the free fermion fixed point at @xmath29 , vanishes identically because of the fermion anticommutation relations . the significance of the four - fermi coupling changes when we consider the scaling dimensions of operators about the luttinger liquid fixed points . in this case we decompose the fermion field into left- ( @xmath341 ) and right- ( @xmath342 ) moving excitations with a linear dispersion , and obtain the long - wavelength hamiltonian @xmath343\ ] ] where @xmath344 and the fermi wavevector @xmath196 is given by @xmath345 . performing power - couting on the @xmath117 free field part of @xmath346 we now find that the four - fermi coupling is now marginal . note however that the coefficient of this four - fermi coupling is suppressed by a factor of @xmath347 , which vanishes as one approaches the @xmath31 critical end - point . by the usual logarithmic perturbation theory at @xmath18 we can determine that the four - fermi interaction modifies the exponent @xmath212 of section [ lutliq ] by @xmath348 where @xmath349 is a numerical constant of order unity . notice , as expected , that the correction to @xmath212 vanishes as @xmath350 . it is also apparent that the impenetrable limit @xmath351 is equivalent to the vanishing density ( @xmath352 ) limit . at finite @xmath70 , there will be corrections to the correlators with terms like @xmath353 . these can not be neglected when @xmath354 where @xmath355 is of order unity . zero scale - factor universality is thus violated for arbitrarily small @xmath34 , when @xmath70 is smaller still and satisfies ( [ violation ] ) . notice however that the boundary specified by ( [ violation ] ) lies well below @xmath356 crossover to the luttinger liquid regime ( fig [ figd1 ] ) . @xcite have recently obtained some exact asymptotic results for the equal - time boson green s function of the @xmath4 impenetrable bose gas . recall that this model is precisely the scaling limit describing the @xmath31 quantum phase transition with zero scale - factor universality . in this appendix , we show how the requirement that the scaling functions be analytic in @xmath308 , can lead to a considerable simplification of their results . the asymptotic results of its _ et.al . _ can be written in the form @xmath357~~~~\mbox{as $ x \rightarrow \infty$}. \label{gkorrep}\ ] ] where @xmath358 and @xmath359 are functions to be determined ( as before @xmath360 ) . from the arguments in section [ onedimension ] , it is clear that @xmath358 and @xmath359 are also universal crossover functions of the @xmath29 , @xmath18 , quantum phase transition the repulsive , @xmath4 bose gas with _ arbitrary _ , short - range interactions . obtained two separate , closed - form , integral expressions for @xmath361 and @xmath362 valid respectively for @xmath363 and @xmath364 . the two expressions were quite distinct and there appeared to be no straightforward relationship between them . here , we point out that the absence of any singularity in the finite @xmath70 bose gas , in fact requires that @xmath359 and @xmath358 be analytic for all finite , real values of @xmath292 . in other words , the functions @xmath365 and @xmath366 must be analytic continuations of each other ( similarly for @xmath367 and @xmath368 ) . we have in fact succeeded in proving that the expression of its _ et . al . _ for @xmath369 is in fact the analytic continuation of their result for @xmath370 . we have been unable to establish a similar result for @xmath362 , but have performed numerical tests on their expressions , which leave essentially no doubt that @xmath371 is also analytic . with the help of the above considerations , it is possible to deduce from ref @xcite a simple closed - form result for @xmath359 and @xmath358 which is valid for _ all _ @xmath292 @xmath372 the value of the constant @xmath234 was obtained by matching to the @xmath18 result of vaidya and tracy @xcite @xmath373 @xmath374 being glaisher s constant @xcite . note that the analyticity of @xmath375 and @xmath371 for all real @xmath292 is manifest . we have plotted the functions @xmath359 and @xmath358 in fig [ korepfig ] . they obey the asymptotic limits @xmath376 both asymptotic limits of @xmath359 , and the @xmath377 limit of @xmath358 , can be obtained directly from ( [ afint ] ) . the @xmath378 limit of @xmath358 is more difficult to obtain from ( [ afint ] ) , and we used instead the second expression for @xmath358 in ref @xcite . demanding that these two methods of obtaining the limit be identical in fact provides one with an independent derivation of the value of the constant @xmath234 ! we have already noted that taking a fourier transform of the asymptotic results ( [ gkorrep ] ) to obtain the structure factor , yields results which compare very poorly with numerically exact results of fig [ figb1 ] . finally , we note that the requirement of analyticity as a function of @xmath292 should apply to essentially all of the equal - time and unequal - time results of its _ et . @xcite . in every case , they have obtained separate expressions for @xmath380 and @xmath363 : proving that these are analytic continuations of each other will lead to highly non - trivial checks on the results , and should also produce some fascinating mathematical identities . w.j.l . buyers _ et . lett . * 56 * , 371 ( 1986 ) ; m. steiner _ et . al . _ , j. appl . phys . * 61 * , 3953 ( 1987 ) ; j.p . renard _ et . al . * 3 * , 945 ( 1986 ) ; j.p . renard _ et . al . . phys . * 63 * , 3538 ( 1988 ) ; k. katsumata _ et . lett . * 63 * , 1424 ( 1989 ) ; m. hagiwara _ et . * 65 * , 3181 ( 1990 ) . w. lu _ et . lett . * 67 * , 3716 ( 1991 ) ; a. meyer _ et . * 21 * , 1729 ; l.p . legnault _ et . al . _ physica b * 180 * , 188 ( 1992 ) ; m. chiba _ et . b * 44 * , 2838 ( 1991 ) ; m. date and k. kindo , phys . rev . lett . * 65 * , 1659 ( 1990 ) . c. zeng and v. elser , phys . b * 42 * , 8436 ( 1990 ) ; j.b . marston and c. zeng , j. appl . phys . * 69 * , 5962 ( b ) ( 1991 ) ; a.b . harris , c. kallin and a.j . berlinsky , phys . rev b * 45 * , 2889 ( 1992 ) ; j.t . chalker , p.s . holdsworth and e.f . shender , phys . rev . lett . * 68 * , 855 ( 1992 ) ; p. chandra , p. coleman and i. ritchey , phys . b * 47 * , 15342 ( 1993 ) ; a.v . chubukov , phys . rev . lett . * 69 * , 832 ( 1992 ) ; j. von delft and c.l . henley , phys . * 69 * , 3236 ( 1992 ) ; j.n . reimers , a.j . berlinsky and a .- c shi , phys . b * 43 * , 865 ( 1991 ) ; r.r.p . singh and d. huse , phys . * 68 * , 1706 ( 1992 ) ; d.l . huber and w.y . cheong , phys . b * 47 * , 3220 ( 1993 ) . a.v . chubukov and s. sachdev , phys . 71 * , 169 ( 1993 ) - * 71 * , 2680 ( 1993 ) ( e ) ; s. sachdev and jinwu ye , phys . lett . * 69 * , 2411 ( 1992 ) ; a.v . chubukov , s. sachdev and j. ye , phys . b , submitted ; d. stauffer , m. ferer , and m. wortis , phys . * 29 * , 345 ( 1972 ) ; p.c . hohenberg , a. aharony , b.i . halperin , and e.d . siggia , phys . b * 13 * , 2986 ( 1976 ) ; c. bervillier , _ ibid . _ * 14 * , 4964 ( 1976 ) ; m .- c . et.al._ , phys . b * 44 * , 6883 ( 1991 ) .

we present some new ideas about the possibility of life developing around sub - giant and red giant stars . our study concerns the temporal evolution of the habitable zone . the distance between the star and the habitable zone , as well as its width , increases with time as a consequence of stellar evolution . the habitable zone moves outward after the star leaves the main sequence , sweeping a wider range of distances from the star until the star reaches the tip of the asymptotic giant branch . presently there is no clear evidence as to when life actually formed on the earth , but recent isotopic data suggest life existed at least as early as years after the earth was formed . thus , if life could form and evolve over time intervals from to years , then there could be habitable planets with life around red giant stars . for a 1 m star at the first stages of its post main - sequence evolution , the temporal transit of the habitable zone is estimated to be of several 10 years at 2 au and around 10 years at 9 au . under these circumstances life could develop at distances in the range 2 - 9 au in the environment of sub - giant or giant stars and in the far distant future in the environment of our own solar system . after a star completes its first ascent along the red giant branch and the he flash takes place , there is an additional stable period of quiescent he core burning during which there is another opportunity for life to develop . for a 1 m star there is an additional years with a stable habitable zone in the region from 7 to 22 au . space astronomy missions , such as proposed for the terrestrial planet finder ( tpf ) and darwin , that focus on searches for signatures of life on extrasolar planets , should also consider the environments of sub - giants and red giant stars as potentially interesting sites for understanding the development of life . we performed a preliminary evaluation of the difficulty of interferometric observations of planets around red giant stars compared to a main sequence star environment . we show that pathfinder missions for tpf and darwin , such as eclipse and fksi , have sufficient angular resolution and sensitivity to search for habitable planets around some of the closest evolved stars of the sub - giant and red giant class .

astro-ph0503520

low and intermediate mass stars with masses ranging from @xmath6 to @xmath7 m@xmath3 undergo dramatic changes in temperature and luminosity during their evolution . for example , the luminosity of the sun , which is located on the main sequence of the hertzprung - russell diagram , currently is 30 percent greater than its zero - age ( initial ) luminosity ( gough 1981 ) . other major changes take place after the sun evolves off the main sequence . along the red giant branch ( rgb ) the radius of a 1 m@xmath3 star progressively increases up to @xmath8 10 r@xmath3 , and the luminosity increases up to about @xmath9 l@xmath3 ( maeder & meynet 1988 ) . then , after leaving the rgb a star evolves onto the asymptotic giant branch ( agb ) . at this stage in its evolution , its radius increases up to a few hundred times r@xmath3 and the luminosity up to nearly @xmath10 l@xmath3 . the lifetime of a 1 m@xmath3 star , from the moment it leaves the main sequence until the end of the rgb , is of about 2.5 @xmath11 yr , which can be compared to @xmath8 10@xmath11 yr spent on the main sequence . the evolution of a star is accompanied by mass loss processes that play an important role . during the post - main sequence evolution of a 1 m@xmath3 star nearly 0.4 m@xmath3 will be lost to the interstellar medium before it ends its life as a white dwarf ( weidemann 1987 ) . the `` habitable zone '' is conservatively defined as a shell region around a star where liquid water can be found at the surface of a planet . this zone is typically from about 0.95 au to 1.37 au for g stars ( kasting et al . 1993 , forget 1998 ) . a less conservatively defined `` habitable zone '' extends the outer edge of the limit of the habitable zone , to as large as 2.4 au , depending radiative properties of @xmath12 ice clouds ( forget & pierrehumbert 1997 , mischna et al . 2000 ) . consequently , as the temperature and luminosity of a star changes during the course its evolution , the radius and width of this zone will also evolve with time ( see section 2 ) . the habitable zone gradually moves outward after a star leaves the main sequence . depending on the length of time that is necessary for life to form , we show that it is possible that life may exist or develop on a planet that is located within a range of 2 - 9 au from a 1 m@xmath3 star , during the first stages of its post main - sequence evolution and up to 22 au from the star during the second stage after the core he flash . in section 3 this is discussed further , considering also the effects of natural migration of micro - organisms through meteorites as a possible trigger for the emergence of life around evolved stars ( arrhenius 1903 , okeefe et al . 1996 , zubrin 2001 ) . micro - organisms might also transport life from one planet to another planet during the evolution of a star . for example , they could be transported from a planet where life is ending to a more distant planet where favorable conditions for the re - birth of life are encountered because of the `` sufficiently slow '' evolution of the habitable zone . future space missions , such as tpf ( beichman , woolf , & lindensmith 1999 ) and darwin ( fridlund & gondoin 2003 ) , that focus on searches of signatures of life on extrasolar planets primarily around main - sequence stars , should also consider the environments of sub - giant and red giant stars . pathfinder missions such as epic ( clampin et al . 2001 ) , eclipse ( trauger et al . 2003 ) , and fksi ( danchi et al . 2003a , 2003b ) may also be able to observe planets around red giant stars . based on a list of selected objects we evaluate the difficulty of coronagraphic and interferometric observations of planets around red giant stars compared to a main sequence star environment ( section 4 ) . in section 5 we summarize our results and conclude that research engaged in the field of extrasolar planets and life should not omit searches for planets orbiting evolved stars . terrestrial planets in the environment of evolved stars are sites of interest for understanding the development of life . from a biological point of view and given the limits of our present knowledge , the existence of life is strongly associated with the presence of liquid water ( brack 1993 ) . in this respect the `` habitable zone '' is currently defined as the range of distances from a star within which a planet may contain liquid water at its surface . several evaluations of the inner and outer limits of the habitable zone have been proposed , as discussed in the review of forget ( 1998 ) . a conservative estimate , minimizing the width of the habitable zone ( forget 1998 ) , assumes that around our own sun the habitable zone currently ranges from 0.95 to 1.67 au . the inner limit of this conservative estimate is set by an atmospheric water loss effect occurring in the stratosphere of a planet , provided water vapor is injected into it from the planet s surface ( kasting et al . 1993 , kasting 1998 ) . the outer limit is set by the lowest temperature at which the liquid / solid phase change of water occurs . the estimate of this outer limit assumes the existence of a greenhouse effect involving co@xmath13 and h@xmath13o gas ( kasting et al . a less conservatively defined `` habitable zone '' extends the outer edge of the limit of the habitable zone to as large as 2.4 au , depending radiative properties of @xmath12 ice clouds . an optically thick condensing @xmath12 atmosphere with particles having radii larger than 6 - 8 @xmath14 m has a very efficient greenhouse effect that maintains the surface temperature above the freezing point of water ( forget & pierrehumbert 1997 , mischna et al . 2000 ) . in this paper we investigate the evolution of the transit of the habitable zone as a consequence of stellar evolution . a convenient way of estimating the inner and outer limit of the habitable zone at different stages of stellar evolution is to assume that the planet behaves as a gray - body with albedo , @xmath15 , and with perfect heat conductivity ( implying the temperature is uniform over the planet s surface ) . its radiative equilibrium temperature , @xmath16 , is then determined by : @xmath17^{0.25}\ ] ] where @xmath18 is the luminosity of the star , @xmath19 is the stefan - boltzmann constant , and @xmath20 is the distance from the center of the star to the planet . typically a value of approximately 0.2 is chosen for the albedo to be representative of an earth - like planet . for these calculations we determine the temperature at the limits of the habitable zone , and from this we calculate the inner and outer distances for a given star . if the albedo is wavelength independent , then terms due to the albedo cancel out of the calculation . in the following calculations we use both the conservative limits of the habitable zone and the less conservative limit around our sun ( forget 1998 , forget and pierrehumbert 1997 , mischna et al . 2000 ) . for the conservative limit , we assume an effective equilibrium temperature of 269 k and 203 k , which for the sun gives inner and outer limits of 0.95 and 1.37 au , respectively . for the less conservative limit , the equilibrium temperature is 169 k , so that the outer limit is extended to 2.4 au . the three equilibrium temperature extrema , defined as the inner and the 2 possible outer limits , can be used to derive the approximate location and width of the habitable zone around a star at different stages of its evolution . models describing the stellar evolution are proposed by maeder and meynet ( 1988 ) . for intermediate mass stars having masses from 2 to 5 m@xmath3 the evolution is followed up to the end of the early asymptotic giant branch . for stars with masses less than 1.7 m@xmath3 , the evolution is calculated up the he - flash occuring at the end of the rgb part of the hr diagram ( iben 1967 ) . maeder and meynet s ( 1988 ) models are used in the following figures to determine the effect of stellar evolution upon the habitable zone , i.e. , its outward transit and increase in width . figure 1(a ) displays the time evolution of the inner and outer limits defining the habitable zone assuming the initial mass of the star is 1 m@xmath3 . the solid curve represents the evolution of the inner limit of the habitable zone , while the dashed curve depicts the conservative outer limit , and the dotted curve represents the less conservative outer limit . the distance between the center of the star and the habitable zone increases progressively with time , from 1 au to tens of au ( upper limit represented in the figures ) . note that it is easy to show from eq . ( 1 ) that this distance is proportional to the square root of the stellar luminosity , which greatly increases as the star ascends the red giant and asymptotic giant branches of the hr diagram . the width of the habitable zone also increases with time and is also proportional to the square root of the stellar luminosity . thus the width is directly proportional to the distance between the star and the habitable zone . therefore , during the course of stellar evolution , the habitable zone is a shell sweeping progressively outward over a wide range of distances from the star as can be seen in figure 1(c ) . the dashed curve represents the duration of the habitable zone for the conservative assumptions for the inner and outer limits of the habitable zone . the dotted curve assumes the same inner limit of the habitable zone but the less conservative outer limit is used . the duration of the transit during which the habitable zone passes over a planet located at 1 au from a star is of the order of 10@xmath4 years . figure 1(b ) displays the evolution of the stellar radius ( in units of solar radii ) during the time period shown in figure 1(a ) . immediately after the star leaves the main sequence , the habitable zone progressively moves to 2 au ( figure 1(c ) ) . the duration of the transit at this location is approximately 10@xmath4 years . a bump / plateau is observed in the curve up to 9 au ( for the conservative limits ) and up to 13 au ( for the less conservative limits ) , where the duration of habitable conditions lasts from a few to several 10@xmath5 years . at 10 au the duration is smaller , around 10@xmath5 years . at 15 au from the star , the duration of habitable conditions lasts more than 10@xmath21 years , and at the largest distances the duration gradually decreases . thus the evolution of the habitable zone at distances up to 25 au has been considered . during this time period , a 1 m@xmath3 star has not yet reached the agb . the bump / plateau observed up to 9 - 13 au increases the duration of habitability conditions . this bump results from a local maximum in luminosity followed by a local minimum along the rgb related to the first dredge - up . for a 1.5 m@xmath3 star , the evolution of the habitable zone is much more rapid since the star itself evolves faster than a 1 m@xmath3 star . plots of the inner and outer limits of the habitable zone , the radius of the star , and the duration of the habitable zone for this case is shown in figs . 2(a ) , 2(b ) , and 2(c ) , respectively . when the star is on the main sequence , the transit of habitable zone , at 3 au , lasts more than 10@xmath4 years . at 5 au the transit may last from 10@xmath5 to several 10@xmath5 years ( depending on the limits assumed for the habitable zone ) . interestingly , even though the star evolves rapidly , the duration of the passage of the habitable zone at distances less that 15 au from the star remains greater than 10@xmath21 years . bumps are observed in the curve since the variations ( just after the main sequence and along the rgb ascent in the hr diagram ) of the effective temperature and luminosity of the star is non - monotonic . for a 2 m@xmath3 star ( figs . 3(a)-3(c ) ) the habitable zone is at @xmath8 5 au during the main sequence stage . after the star leaves the main sequence the transit of the habitable zone may last more than 10@xmath5 years for distances less than 10 - 15 au . it is interesting to note that the habitable zone jumps at about 1.7 gyrs over a period lasting less than 25 million years ( see figure 3(a ) ) . in this paper we have calculated the duration of the habitable zone for different stellar masses , but we have not considered the _ actual _ mass distribution of red giants in our own solar neighborhood . clearly the longest lived habitable zones are those for the low mass red giant stars that evolve most slowly off the main sequence , as is evident from the present discussion and figs . 1 - 3 , as well as table 1 , described below . although individual masses of red giant stars are not well known in general , and have not been directly determined except in a few cases , neckel ( 1975 ) has shown theoretically that the peak of the red giant mass distribution occurs at progenitor masses of about 1.1 m@xmath3 . an empirical estimate of red giant masses determined by scalo , dominy , and pumphrey ( 1978 ) is in agreement with the theoretical estimate , with values 0.8 1.2 m@xmath3 . this is fortunate since stars selected in a magnitude - limited or distance - limited survey will therefore mostly likely be low mass stars that have long habitable zone lifetimes similar to those calculated in fig . 1 in this paper . also , it is important to note that the calculations in this paper are focused on the evolution of the habitable zone from the time period after the star evolves off the main sequence up the first ascent along the red giant branch during which the star is burning hydrogen in a shell around a growing he core , i.e. , only to the point of the helium flash . after the helium flash , there is an additional long - lived phase of quiescent core helium burning with nearly constant luminosity . scalo and miller ( 1979 ) developed very simple interpolation formulae for the lifetime of a star at its various stages of evolution : main sequence , subgiant , first red giant , and first and second helium burning phases , as a function of initial stellar mass . for completeness we display the lifetimes computed from their formulae for stellar masses of 1.0 , 1.5 , and 2.0 m@xmath3 , which we display in table 1 . for a 1.0 m@xmath3 star , core helium burning provides a nearly constant luminosity phase , adding an additional @xmath2 years to the duration of the habitable zone for low mass stars , with substantial increases to the lifetime of the habitable zone for higher mass stars as well . the luminosity during the core helium burning phase is approximately @xmath22 l@xmath3 , @xmath23 l@xmath3 , and @xmath22 l@xmath3 , for 1.0 , 1.5 , and 2.0 m@xmath3 stars , respectively ( kozlowski & paczynski , 1975 ) . thus an additional period of habitability exists for the region between 7 and 22 au for a 1 m@xmath3 star during core helium burning . the he core flash that precedes this phase is not an explosive event nor does it affect the star s external luminosity very much . this was discussed in the pioneering papers by schwarzschild & harm ( 1962 , 1964 ) , and subsequently analyzed in more detail by mengel & gross ( 1975 ) , paczynski & tremaine ( 1977 ) , paczynski ( 1979 ) , and by deupree ( 1996 ) . the main consequence to the stellar luminosity after the he core flash is an initial _ decrease _ in luminosity for about @xmath24 to @xmath25 years , followed by an increase to approximately the same value as before the flash , i.e. , near the peak luminosity at the tip of the first ascent along the red giant branch ( despain , 1981 ) . over a period of @xmath26 to @xmath27 years the star gradually becomes less luminous , and it settles down to a state of nearly constant luminosity of about @xmath28 l@xmath3 ( despain , 1981 ; chiosi , bertelli , and bressan , 1992 ) statistically the result is a clump " of stars of this luminosity close to the rgb . the constancy of luminosity for this `` clump '' of stars , which is about a factor of 10 lower than at the tip of the rgb , is a result of the fact that stars of different initial masses converge toward a common value of the he core mass ( chiosi , bertelli , & bressan , 1992 ) . for a 1 m@xmath3 star the duration of the rise in luminosity as a star evolves upwards along the rgb is relatively rapid , lasting @xmath8 @xmath29 years as can be seen in fig . 1 . during this period of time the habitable zone moves rapidly outward , such that the inner radius is @xmath30 20 au from the star . once the star settles into the core he burning phase , the region between 7 and 22 au becomes habitable again . thus there is a fresh start " for life in this phase of stellar evolution . we summarize our analysis of the evolution of the habitable zone by noting that there are two opportunities for life in the extended habitable zones around evolved stars . initially life has an opportunity as the star slowly evolves off the main sequence . after the first ascent along the rgb and the he core flash , there is a second chance during the core he burning phase . this latter phase is particularly interesting because of the nearly constant luminosity of the star and the long duration under these conditions . the earth is known to have formed about 4.55 gyr ago . substantial evidence exists that life on earth occurred at least as far as 3.85 gyr ago rests on the basis of isotopic ratios of carbon residues within grains of apatite ( mojzsis et al . 1997 , holland 1997 ) . the oldest purported fossils of microorganisms have an age of 3.50 gyr and may represent photosynthetic cyanobacteria were reported by schopf ( 1993 , 1994 ) . the findings and interpretation of this work was recently debated ( brasier et al . 2002 , dalton 2002 ) . rocks older than 3.60 gyr are all metamorphosed at a high grade and most are strongly deformed , thus morphological fossils if originally present are unlikely to be preserved ( rosing 1999 ) . based on these considerations , it is possible for life to have existed at least as long ago as 3.85 gyr , but this is uncertain because of the imprecise knowledge of the ages of the rocks , and ambiguous interpretations of the isotopic @xmath31 ratios . thus , for our discussion , we consider three time scales for life to form , a conservative estimate of 1 gyr , and a less conservative one of 0.5 gyr , and an optimistic one of 0.1 gyr . recent analyses of isotopic @xmath32c ratios on micron - scale mineralized tubes in archean pillow lavas by furnes et al . ( 2004 ) suggest that microbial life colonized these oceanic volcanic rocks soon after they were formed approximately 3.5 gyr ago . this provides some further support for the optimistic time scales of a few hundred million years or less for life formation . dating of terrestrial fossils reveals that life had evolved on earth by the end of the period of heavy bombardment . evidence for the bombardment comes from theoretical studies of planetary formation and directly from lunar cratering . heavy bombardment may have helped set the stage for the terrestrial origin of life by delivering key biogenic elements ( h , c , n ) to the earth s surface ( chyba et al . 1994 ) . for its first 800 myr , the earth s surface would have been subjected to frequent impacts , possibly delaying the development and spread of life ( maher and stevenson 1988 ) . sleep et al . ( 2001 ) estimate that the time period for the earth to cool off after heavy bombardment to a temperature at which life could form is surprisingly short , in the range of @xmath33 to @xmath27 years . due to the rapid evolution of a star once it has left the main sequence , the transit time of the habitable zone shortens . for an evolved 1 m@xmath3 star , the transit of the habitable zone crossing a planet at 2 au of distance lasts slightly more than 10@xmath4 years , a time scale clearly larger than the time which has been required for the emergence of life on earth . at about 5 au of distance , the duration of habitable conditions lasts from 10@xmath5 years to several 10@xmath5 years , which is about one order of magnitude shorter than the time estimated for occurrence of life on earth . even if the formation of life takes longer than 10@xmath5 years , it is still possible for pre - existing life ( in the form known on earth ) to adapt itself to a new planet containing liquid water . micro - organisms could be transported from a planet where life is ending to one where favorable conditions for its re - birth are encountered due to the transit of the habitable zone . in this sense the transportation of microorganisms through meteorites can be a trigger for life development around evolved stars . the possible migration of life through meteorites has been discussed by arrhenius ( 1903 ) , okeefe et al . ( 1996 ) and zubrin ( 2001 ) , while the possibility of micro - organisms transporting life in a planetary system either from planet to planet or between different planetary systems has also been studied ( mileikowsky et al . 2000 , mastrapa et al . the transportation of material between planetary systems has recently gained support from the possible discovery of extrasolar micro - meteors ( meisel et al . 2002 ) . of course , the time scales are longer because the distances are larger . meteorites or comets may have been expelled from our solar system up to 3.5 billion years ago ( when life first appeared on earth ) with biomolecules or even primitive microorganisms on them . after a journey of a billion years or more in interstellar space , they may have been captured by nearby planetary systems ( up to a few tens of light years ) and may have seeded other planets with suitable conditions for the development of life . mathematical simulations have shown that the probability for expulsion of a meteorite or a comet from a planetary system is not negligible ( gladman et al 1996 ) . current theoretical debate centers on the probability of capture by an extrasolar planet . however , recent calculations by melosh ( 2003 ) suggest that the probability of interstellar transfer of meteoric material between planets on two different stars in the solar neighborhood ( for single stars ) is extremely small . however in the case of red giant environments , the probability of capture of meteoritic material ejected from a planetary body is reasonably large ( melosh , 2003 ) . using this mechanism , life can be transported from one planet to another during the expansion of the habitable zone . it has been demonstrated both by laboratory and space experiments that biomolecules and even micro - organisms can survive space travel in interplanetary conditions , i.e. , low temperature , vacuum , uv radiation , and the shock experienced while impacting a solid planet ( burchell et al 2001 , roten et al 1998 , okeefe et al 1996 ) . all these considerations are part of the hypothesis called panspermia ( arrhenius 1903 , becquerel 1910a , 1910b ) that there may be an `` interfertilisation '' between planets ( mileykowsky 1997 , portner et al 1961 ) . in the case of our own solar system at 2 au , it is possible that in the future ( assuming the presence of liquid water is encountered during a time scale compatible with the time for occurrence of life on earth ) life may appear on the planet mars if some ( presently hypothetical ) frozen liquid water can be melted . liquid water also could be produced on europa as a consequence of the evolution of the sun . however , since the mass of europa is relatively small , it may be difficult for it to maintain an atmosphere for a long time , and hence could evaporate into space . a more massive body like titan is perhaps better adapted to host life ( transit at 10 au from several 10@xmath21 years to about 10@xmath5 years ) . this speculative scenario assumes that no change in the orbits of the planets will occur during the next 6@xmath3410@xmath4 years . from the point of view of dynamical orbital changes related to stellar evolution , this assumption seems reasonable . effects of the stellar mass loss phenomena and tidal dissipation between the planets and the red giant star are not negligible in the dynamics of the orbits ( rasio et al . a sun - like star will lose about 24@xmath35 of its initial mass during the rgb phase and about 20@xmath35 during the agb phase ( schrder and sedlmayr 2001 ) . the transit of the habitable zone that we have calculated , with a reasonably long duration for the development of life , happens along the rgb , at a stage during which the star has a radius less than @xmath8 100 r@xmath3 . in fig . 2 of rasio et al . ( 1996 ) , when a sun like star reaches a radius of about 100 r@xmath3 , only a little more than 10 @xmath35 of the initial mass of the star has been lost , and the effect of this mass loss among the dynamical changes of the planet orbits ( for planets more distant than 1 au not yet subject to tidal effects ) is to increase by 10 @xmath35 the semi - major axis . the work of rasio et al . ( 1996 ) does consider the reimers relation for representing the mass loss , according to parameters used in sackmann et al . it is interesting to note that the increase of the semi - major axis , if occuring for a planet during a time when it is in the habitable zone ( itself expanding ) , will tend to increase the time duration of habitable conditions for the planet . planets around evolved stars of the class of sub - giant and red giant may offer favorable conditions for hosting life at a few au . in the few next decades , the search for life around different type of stars could be made using proposed space interferometers , such as the terrestrial planet finder ( tpf ) ( beichman , woolf , & lindensmith 1999 ) and darwin ( fridlund & gondoin 2003 ) . recently , a giant planet candidate around a subgiant class star has been discovered by means of radial velocity measurements ( frink et al . the detected planet has m @xmath36 = 8.9 m@xmath37 ( making it a planet candidate ) , and is orbiting at 1.3 au from the center of the star . the parent star , hip 75458 , is a k2iii subgiant class star with log l / l@xmath3 = 1.85 solar luminosity ( mallik 1999 ) and with a mass of 1.05 m@xmath3 . it is located 31.5 pc from the earth . figure 1 displays the habitable zone around hip 75458 . the habitable zone , moving outward , is expanding to an outer limit of 14 au and an inner limit at 7 au . habitable conditions at 7 au from the star have been encountered during about 10@xmath5 years , which is a reasonable , if optimistic , time for life to develop . the presence of a jupiter - like planet encourages one to imagine that other bodies ( perhaps even terrestrial ones ) could be orbiting this star . tables 2 - 4 display preliminary lists of evolved stars , and their characteristics , contained within a sphere of radius 30 pc around our sun . there about 94 sources of luminosity class iv ( table 2 ) , 44 sources of luminosity class iii ( table 3 ) and two sources of luminosity class ii ( table 4 ) . the luminosity class iii and iv ( respectively the giants and sub - giants classes ) result from low and intermediate mass stars having just left the main sequence in their evolution . luminosity class iv and iii stars are both of great interest . in the first case , the habitable zone moves outward relatively slowly and allows , for stars with masses @xmath38 2 m@xmath3 , conditions that do not exclude the development of life . in the second case , after the first ascent along the rgb and the he flash , the stars evolve into a stable phase of core he burning which is also favorable to the formation of life . approximately 30 to 60 percent of the luminosity class iv stars in table 3 are expected to be in this phase ( scalo & miller , 1979 ) . it is important to notice that luminosity class iv alone offers 94 nearby sources compared to the approximatly 1000 nearby main sequence stars ( of all spectral types ) . the ratio of all red giant stars ( tables 2 - 4 ) to the main sequence stars is thus about 14% , which at first seems high considering the local space density of all giant stars is about @xmath39 compared to main sequence stars . this value is based on main sequence and red giant star densities given in cox ( 2000 ) and neckel ( 1975 ) . however , a sample chosen based on a brightness or distance limit will select out a larger fraction than expected on the basis of volume estimates because the red giant stars are much brighter than the main sequence stars , and the increase should be proportional to the ratio of the luminosities , or roughly a factor of ten . thus , within 30 pc , a reasonable sample has been shown to exist . this sample allows us to consider evolved stars for the proposed future projects which may contribute to the understanding of the formation of life . the development of large filled - aperture space telescopes coupled to efficient coronagraphs and the development of space interferometers fed by large cooled telescopes , represent technological challenges that may be overcome in the next two decades . a first step in the development of large space telescopes like darwin and tpf are precursor concepts for coronagraphy ( e.g. eclipse , trauger et al . 2003 ; epic , clampin et al . 2001 ) and precursor concepts for interferometry like the fksi mission concept ( fourier kelvin stellar interferometer mission - danchi et al 2003a , 2003b ) . these instruments in the relatively near term could directly observe some types of extrasolar planets around stars . the difficulty of detecting a planet orbiting a red giant star and characterizing its atmospheric signatures lies in the very high contrast ratio between the planet and the star . for a 1.0 m@xmath3 star , due to the squared dependence between the habitable zone size and the stellar luminosity , the habitable zone size associated with a star of luminosity 4 l@xmath3 is of the order of 2 au , while for a star of luminosity 25 l@xmath3 , the corresponding size is 5 au , and , for 100 l@xmath3 , it is about 10 au . the contrast ratio required to observe a planet around a sub - giant or a red giant star increases by a factor of 4 to 100 ( range of luminosity corresponding to a reasonably long duration of the transit ) compared to that required to observe a planet around main - sequence ( fgk ) dwarf stars ( luminosity class v ) . on the other hand the angular resolution needed to separate the planet from the star is decreased due the increased size of the habitable zone . for hip 75458 , the angular separation of the habitable zone is about 0.3 arcsec from the star . this separation is well within the resolution of filled - aperture space telescopes with diameters in the 5 - 10 meter range operating at visible and near - infrared wavelengths . we present a qualitative discussion below that shows how these factors approximately balance out in terms of the detectability of planets around evolved stars . a more detailed discussion is precluded at this time because the architectures for the tpf and darwin missions are still under development , and a detailed detectability study requires detailed knowledge of instrumental parameters , which are not available at present . we will present a more detailed treatment , based on simple model interferometers and coronagraphs in another paper . in the following discussion , for the purpose of developing an appreciation of the detection issues , we use a main sequence star at a reference distance of 10 pc , and compare that to a red giant star at 30 pc . assuming the use of the fksi concept or of any other future nulling interferometer , three physical parameters representing the astrophysical source are of importance in order to estimate the potential of detection . these physical parameters are : ( 1 ) the angular distance between the planet and the star that has to be compared to the inner working angle ( iwa ) of the instrument ; ( 2 ) the contrast ratio between the planet and the star ; ( 3 ) and the stellar diameter , which affects the stellar leakage from a perfect optimum nulling . for a 1 m@xmath3 star , with the habitable zone varying between 2 and 15 au ( middle in the width ) , the corresponding stellar radius variations will be approximately between 2 and 15 r@xmath3 ( see figure 1(a ) and 1(b ) ; note this is an approximation because the effective temperature of the star varies also ) . we now discuss the relative difficulty of detecting a planet around red giant stars compared to main sequence stars . the reason for this comparison is that the designs for tpf and darwin will be based on the requirement to detect earth - like planets around the habitable zone of nearby main sequence ( f , g , k ) dwarf stars ( luminosity class v ) . thus we show that if tpf and darwin can detect earth - like planets around these stars , then these instruments will also be able to detect such planets around subgiant and red giant stars , i.e. , luminosity classes iv and iii . the inner working angle ( iwa ) of a coronagraph is : @xmath40 where @xmath41 and d are the wavelength of light and the diameter of the primary mirror of the telescope , respectively . the corresponding iwa for a simple nulling interferometer is : @xmath42 where b is the baseline length and @xmath41 was defined previously . for the coronagraph systems , @xmath43 varies from approximately 0.32 arcsec for a 1.8 m primary mirror and a wavelength of 0.7 @xmath14 m as planned for the epic and eclipse missions , and is 0.096 arcsec for the 6 m longest dimension for the planned mirror for the coronagraphic version of tpf , now called tpf - c . despite the longer wavelength of light for the infrared interferometers , @xmath44 is typically much smaller than @xmath43 . for the fksi system , @xmath44 is 0.043 arcsec at a center wavelength of 5 @xmath14 m , while @xmath44 is 0.029 arcsec for a 35 m baseline on a structurally connected interferometer , or 0.017 arcsec for a 60 m baseline on a free flying interferometer , both assuming a center wavelength of 10 @xmath14 m . these iwas can now be compared to the angular size of the habitable zones for the two extreme cases for the red giant stars , namely 2 au and 15 au . at a reference distance of 10 pc the habitable zone is located at 0.20 arcsec from the star and 1.5 arcsec for the 2 au and 15 au cases , respectively . for a star at a reference distance of 30 pc , the angular sizes are proportionally smaller , 0.067 arcsec and 0.50 arcsec , respectively . based on the iwas calculated here , the iwa of the small pathfinding coronagraphs like eclipse and epic will be adequate only for the nearest such stars , while the larger tpf - c has resolution sufficient to resolve the habitable zone for essentially all the red giant stars . the interferometer iwas are all smaller than the angular size of the habitable zone for all the cases discussed here , which means that this instrumental parameter is favorable for the detection of planets around red giant stars . the planet - star intensity contrast ratio in reflected light is approximately : @xmath45 where @xmath46 and @xmath47 are the planet and star intensities , respectively , while @xmath48 is the radius of the planet and @xmath49 is the distance from the star to the planet ( cf . charbonneau et al . 1999 ) . for infrared radiation emitted by the planet in thermal equilibrium with the star , the planet - star intensity ratio is : @xmath50\ ] ] where @xmath51 is the effective temperature of the star , and @xmath52 is the equilibrium temperature of the planet , as given by eqn . ( 1 ) previously . the quantities @xmath48 and @xmath53 are the radii of the planet and star , respectively . in reflected light , the planet - star contrast ratio is reduced in comparison with main sequence stars because of the much larger planet - star separation , which amounts to factors of 4 to 225 for the 2 au and 15 au cases , respectively . in the thermal infrared , the contrast ratios are reduced by factors of approximately 4 to 100 ( compared to those of main sequence stars ) as the radius of the star varies from about 2@xmath54 to 10@xmath54 over the course of the evolution of the star during the red giant phase . an additional consideration for the nulling interferometer is the stellar leakage mentioned previously . for a simple bracewell interferometer with a @xmath55 null , the stellar leakage is given by : @xmath56 thus the stellar leakage is proportional to the square of the angular size of the star divided by the angular resolution of the interferometer , or equivalently , the fringe spacing . this means the stellar leakage for a giant star of radius 2@xmath54 at 20 pc is the same as that of a 1@xmath54 main sequence star at 10 pc . however , the observable red giant stars are on average further away from our solar system than the main sequence stars , and the increase in the stellar leakage signal is expected to be at the low end of the range , or comparable to that of the main sequence stars . we can summarize our considerations of the detectability of these stars by noting that the problem is only moderately more difficult by a factor of 4 to 10 in terms of the star - planet contrast ratio , which is in the range of the variation expected for the most optimistic to the worst cases for the main sequence stars , given the variations in spectral type and planet size being considered for tpf / darwin . the inner working angles of the coronagraphs and interferometers are adequate for most of the stars in the tables in this paper . the stellar leakage is approximately equal or only modestly worse than that for main sequence stars since the giant stars are on average at larger distances from our solar system . more detailed studies of detectability issues are planned as more becomes known about the architectures for tpf and darwin . we have estimated the time transit of the habitable zone around evolved stars of initial mass 1.0 , 1.5 and 2.0 m@xmath3 from the time period after which a star moves off the main sequence until it reaches the point of the he flash . after the he flash , there is an additional potential phase for life during the period of quiescent core he burning prior to the star s ascent along the asymptotic giant branch . the case of a 0.85 m@xmath3 star was not analyzed because the time for such a star to leave the main sequence is longer than the age of the universe . a simple hypothesis was used to estimate the inner and outer limits of the habitable zone , i.e. , a blackbody assumption and extrapolation of results presented in kasting et al . ( 1993 ) defining the habitable zone limits around a solar type star . in our approximation the fact that the spectrum of the evolved star differs from the solar spectrum used in kasting et al . ( 1993 ) is neglected . only the luminosity of the stars is considered , while the planet is assumed to be a greybody with perfect absorption at any wavelength . we conclude that life may exist around evolved stars of the sub - giant and red giant classes at several au from the stars . although the temporal evolution of a star is rapid after leaving the main sequence , the temporal transit of the habitable zone does not appear incompatible with the possible duration for the development of life . moreover , micro - organisms could be transported from a planet where life is ending to a planet where favorable illumination conditions for the rebirth of life are encountered . this type of mechanism may be unique , or may be encountered exclusively in the environment of evolved stars . we speculate also that around our sun , life may develop in the future at distances in the range 2 - 9 au . this is deduced from the the temporal transit of the habitable zone , which is estimated to be of the order of several 10@xmath4 years for a planet at 2 au and around 10@xmath5 years for a planet at 9 au . furthermore , after the sun ascends the red giant branch for the first time and undergoes the he flash , there is an additional long stable period of core he burning @xmath8 @xmath2 years during which life can evolve in the habitable zone , which now is from 7 to 22 au from the star . nearly 150 evolved stars ( sub - giants and red giants ) are within 30 pc of our solar system ( see the details in tables 2 - 4 ) . at the present time we do not know if life is common in our universe , or extremely rare , or even unique on earth . if future observing instruments allow the detection of signatures of life on various planets , then the observation of planets , possibly hosting life , around evolved stars may help to set some limit on the time required for life to develop . for example , if life is found on a planet inside a habitable zone transiting in 10@xmath21 years , this proves that life could develop within this period , setting a much smaller upper limit to the formation of life than presently known . in addition , if life appears to be common , another interesting case would be to find a planet containing liquid water within an habitable zone of rather fast transition ( like 10@xmath21 years ) with no signature of life . this case ( fast conditions of habitability ) could contribute to set new lower limits for the emergence of life . the environments of evolved stars may represent some interesting test cases . these test cases can consider the fundamental hypothesis that early life may have been connected to volcanic environments such as deep sea hydrothermal vents . thermophilic life may populate the ocean of superficially frozen planets . the expansion of the habitable zone around the star in evolution will change the habitable conditions of the surface of the planet by increasing its temperature , by increasing the rate of water evaporation and thus by triggering the hydrological cycles ( evaporation , clouds formation , rain process ) . if life was already present underneath the oceans , it could spread on the ground layer very fast . an interesting issue to tackle by observing a freshly habitable planet just entering a rapidly expanding habitable zone is the coupling between the development of life in the ocean and its development on the ground layer . life is known to have developed first in the earth s oceans . on other planets , if life as we know it is found , one may wonder : is the chlorophyl signature ( mostly representing the ground layer activity ) always encountered when bioatmospheric signatures ( related to the global activity : oceans + ground layer ) will be found ? a rapidly expanding habitable zone could help improve our understanding of the time delay coupling between these two different bio - environments : ocean and surface layer . we conclude that future observing programs focused on the search for planets with possible signatures of life should not omit the class of sub - giants and red giant stars . the study of objects of this class will further contribute to the understanding of the formation of life in the universe . filled - aperture space telescopes with diameters in the range of 5 to 10 meters operating at visible and near - infrared wavelengths and space - based mid - infrared interferometers with baselines in the range from 35 to 60 m , provide adequate resolution and sensitivity to detect earth - like planets around red giant stars . we thank p. baudoz , f. selsis , t. guillot and f. thvenin for useful discussions and k .- schrder for constructive remarks on an earlier version of this manuscript . we thank dr . j. rajagopal for a careful reading of the manuscript . we also appreciate constructive comments on the manuscript from our referee , dr . john scalo . arrhenius s.a . , 1903 . die verbreitung des lebens i m weltenraum . die umschau 7 : 481 - 485 . the propagation of life in space , translated by goldsmith d. in the quest for extraterrestrial life . goldsmith d. ( ed . ) . university science books , mill valley , california , 32 - 33 , 308 p. danchi , w.c . , et al . 2003b , in the proceedings `` towards other earths : darwin / tpf and the search for extrasolar terrestrial planets , '' heidelberg , germany , 22 - 25 april 2003 , esa publication sp-539 , p.83 . mileikowsky c. , 1997 , in astronomical and biochemical origins and the search for life in the universe ; proceedings of the 5th international conference on bioastronomy ; iau colloquium no . 161 ; capri , july 1 - 5 , 1996 . cosmovici c.b . , bowyer s. and werthimer d. ( eds ) . editrice compositori , bologna , 545 - 552 , 814 p. llrrrrrcr hd 142 & & & 00 06 19.18 & -49 04 30.7 & 6.22 & 5.70 & g1 & 39.0 + @xmath57 and & hd 358 & sb & 00 08 23.26 & + 29 05 25.6 & 1.95 & 2.06 & b8 & 33.6 + @xmath58 cas & hd 432 & ds & 00 09 10.69 & + 59 08 59.2 & 2.61 & 2.27 & f2 & 59.9 + hd 2071 & & & 00 24 42.55 & -53 59 02.4 & 8.0 & 7.4 & g8 & 36.2 + @xmath58 hyi & hd2151 & v & 00 25 45.07 & -77 15 15.3 & 3.42 & 2.80 & g2 & 133.9 + hd 3141 & & & 00 35 00.87 & + 42 41 41.2 & 8.87 & 8.01 & k0 & 34.5 + @xmath59 tuc & hd 7788 & v & 01 15 46.16 & -68 52 33.3 & 4.73 & 4.25 & f6 & 48.9 + @xmath60 and & hd 8799 & & 01 27 39.38 & + 45 24 24.1 & 5.25 & 4.83 & f5 & 35.3 + hd 9562 & & & 01 33 42.84 & -07 01 31.2 & 6.37 & 5.76 & g2 & 33.7 + hd 10086 & & & 01 39 36.02 & + 45 52 40.0 & 7.30 & 6.61 & g5 & 46.7 + @xmath61 scl & hd 10830 & ds & 01 45 38.76 & -25 03 09.4 & 5.70 & 5.31 & f2 & 36.5 + @xmath57 tri & hd 11443 & sb & 01 53 04.91 & + 29 34 43.8 & 3.90 & 3.41 & f6 & 50.9 + hd 13825 & & & 02 15 24.41 & + 24 16 16.7 & 7.50 & 6.81 & g8 & 37.9 + ag+01 242 & hd 14214 & sb & 02 18 01.44 & + 01 45 28.1 & 6.18 & 5.58 & g0.5 & 40.0 + hd 16417 & & & 02 36 58.61 & -34 34 40.7 & 6.45 & 5.78 & g5 & 39.2 + nsv 893 & hd 16765 & v & 02 41 14.00 & -00 41 44.4 & 6.23 & 5.71 & f7 & 46.2 + 87 cet & hd 17094 & ds & 02 44 56.54 & + 10 06 50.9 & 4.58 & 4.20 & f0 & 38.7 + hd 17190 & & & 02 46 15.21 & + 25 38 59.6 & 8.66 & 7.81 & k1 & 39.0 + @xmath57 for & hd 20010 & & 03 12 04.53 & -28 59 15.4 & 4.36 & 3.85 & f8 & 70.9 + @xmath62 eri & hd 23249 & rs & 03 43 14.90 & -09 45 48.2 & 4.43 & 3.51 & k0 & 110.6 + v891 tau & hd 26913 & by & 04 15 25.79 & + 06 11 58.7 & 7.60 & 6.96 & g5 & 47.9 + v774 tau & hd 26923 & v & 04 15 28.80 & + 06 11 12.7 & 6.87 & 6.33 & g0 & 47.2 + 13 ori & hd 33021 & & 05 07 38.31 & + 09 28 18.4 & 6.79 & 6.17 & g1 & 35.3 + hd 37986 & & & 05 41 53.69 & -15 37 50.0 & 8.16 & 7.37 & g8/k0 & 36.1 + hd 39881 & & & 05 56 03.43 & + 13 55 29.7 & 7.25 & 6.60 & g5 & 35.7 + @xmath58 aur & & al & 05 59 31.72 & + 44 56 50.8 & 1.97 & 1.89 & a2 & 39.7 + hd 45184 & & & 06 24 43.88 & -28 46 48.4 & 7.01 & 6.37 & g2 & 45.4 + 31 gem & hd 43737 & v & 06 45 17.36 & + 12 53 44.1 & 3.79 & 3.40 & f5 & 57.0 + hd 50806 & & & 06 53 33.94 & -28 32 23.3 & 6.75 & 6.04 & g5 & 38.8 + qw pup & hd 55892 & a2 & 07 12 33.63 & -46 45 33.5 & 4.80 & 4.50 & f0 & 47.2 + 55 gem a & hd 56986 & sb & 07 20 07.38 & + 21 58 56.4 & 3.87 & 3.53 & f0 & 55.5 + hd 62644 & & & 07 42 57.10 & -45 10 23.2 & 5.84 & 5.04 & g6 & 41.4 + hd 63433 & & & 07 49 55.06 & + 27 21 47.5 & 7.57 & 6.93 & g5 & 45.8 + 10 cnc & hd 67228 & & 08 07 45.86 & + 21 34 54.5 & 5.91 & 5.30 & g1 & 42.9 + 13 uma & hd 78154 & ds & 09 10 23.55 & + 67 08 02.5 & 5.29 & 4.80 & f6 & 48.9 + @xmath63 vel & hd 82434 & & 09 30 42.00 & -40 28 00.4 & 3.91 & 3.59 & f3 & 53.9 + 23 uma & hd 81937 & v & 09 31 31.71 & + 63 03 42.7 & 3.97 & 3.65 & f0 & 43.2 + 25 uma & hd 82328 & sb & 09 32 51.43 & + 51 40 38.3 & 3.63 & 3.20 & f6 & 74.2 + 40 leo & hd 89449 & ds & 10 19 44.17 & + 19 28 15.3 & 5.24 & 4.80 & f6 & 47.2 + i car & hd 90589 & v & 10 24 23.71 & -74 01 53.8 & 4.32 & 3.99 & f2 & 61.7 + p vel & hd 92139 & sb & 10 37 18.14 & -48 13 32.2 & 4.14 & 3.84 & f4 & 37.7 + 78 leo & hd 99028 & sb & 11 23 55.45 & + 10 31 46.2 & 4.35 & 4.00 & f4 & 41.3 + 83 leo & hd 99491 & & 11 26 45.32 & + 03 00 47.2 & 7.29 & 6.49 & k0 & 56.6 + hd 104304 & & & 12 00 44.45 & -10 26 45.6 & 6.32 & 5.54 & g9 & 77.5 + hd 114174 & & & 13 08 51.02 & + 05 12 26.1 & 7.47 & 6.80 & g5 & 38.1 + hd 114837 & & & 13 14 15.14 & -59 06 11.6 & 5.40 & 4.92 & f7 & 55.5 + hd 118096 & & & 13 34 03.24 & + 33 13 42.2 & 10.35 & 9.23 & k5 & 43.7 + 1 cen & hd 224481 & sb & 13 45 41.25 & -33 02 37.4 & 4.61 & 4.23 & f3 & 6.7 + @xmath64 boo & hd 120136 & v & 13 47 15.74 & + 17 27 24.9 & 4.98 & 4.50 & f6 & 64.1 + @xmath65 boo & hd 121370 & sb & 13 54 41.08 & + 18 23 51.8 & 3.26 & 2.68 & g0 & 88.2 + hd 122862 & & & 14 08 27.17 & -74 51 01.0 & 6.60 & 6.02 & g2.5 & 34.9 + @xmath66 vir & hd 124850 & v & 14 16 00.87 & -06 00 02.0 & 4.60 & 4.10 & f7 & 46.7 + 18 boo & hd 125451 & ds & 14 19 16.28 & + 13 00 15.5 & 5.764 & 5.40 & f5 & 38.3 + 9 lib & hd 130841 & v & 14 50 52.71 & -16 02 30.4 & 2.912 & 2.75 & a3 & 42.3 + hd 136064 & & & 15 14 38.34 & + 67 20 48.2 & 5.66 & 5.10 & f9 & 39.5 + hd 134606 & & & 15 15 15.04 & -70 31 10.6 & 7.60 & 6.86 & g5 & 37.2 + 37 lib & hd 138716 & & 15 34 10.70 & -10 03 52.3 & 5.63 & 4.61 & k1 & 34.5 + nsv 7350 & hd 142860 & v & 15 56 27.18 & + 15 39 41.8 & 4.33 & 3.85 & f6 & 89.9 + 13 dra & hd 144284 & & 16 01 53.35 & + 58 33 54.9 & 4.53 & 4.01 & f8 & 47.8 + @xmath67 sco a & hd 144070 & ds & 16 04 22.3 & -11 22 20 & 5.24 & 4.77 & f5 & 36 + nsv 7915 & hd 150680 & sb & 16 41 17.16 & + 31 36 09.8 & 3.54 & 2.89 & g0 & 92.6 + hd 152311 & & & 16 53 25.22 & -20 24 56.0 & 6.53 & 5.87 & g5 & 35.7 + @xmath62 her & hd 156164 & ds & 17 15 01.91 & + 24 50 21.1 & 3.22 & 3.12 & a3 & 41.6 + hd 157347 & & & 17 22 51.29 & -02 23 17.4 & 6.94 & 6.29 & g5 & 51.4 + hd 158332 & & & 17 27 34.58 & + 26 47 41.9 & 8.54 & 7.73 & k1 & 33.4 + @xmath41 ara & hd 160032 & & 17 40 23.83 & -49 24 56.1 & 5.17 & 4.77 & f3 & 45.7 + 86 her & hd 161797 & & 17 46 27.53 & + 27 43 14.4 & 4.17 & 3.41 & g5 & 119.1 + @xmath68 ser & hd 150680 & & 18 00 29.01 & -03 41 25.0 & 4.99 & 4.63 & f2 & 92.6 + 72 oph & hd 165777 & ds & 18 07 20.98 & + 09 33 49.8 & 3.86 & 3.72 & a4s & 39.4 + sao 85892 & hd 168874 & ds & 18 20 49.26 & + 27 31 49.2 & 7.59 & 7.00 & g2 & 34.0 + hd 177565 & & & 19 06 52.46 & -37 48 38.4 & 6.88 & 6.16 & g5 & 58.2 + 31 aql & hd 182572 & v & 19 24 58.20 & + 11 56 39.9 & 5.93 & 5.16 & g8 & 66.0 + @xmath62 aql & hd 182640 & v & 19 25 29.90 & + 03 06 53.2 & 3.68 & 3.40 & f0 & 65.1 + hd 183877 & & & 19 32 40.33 & -28 01 11.3 & 7.83 & 7.15 & g5 & 38.4 + @xmath58 aql & hd 188512 & v & 19 55 18.79 & + 06 24 24.3 & 4.57 & 3.71 & g8 & 73.0 + hd 190360 & & & 20 03 37.41 & + 29 53 48.5 & 6.44 & 5.71 & g6 & 62.9 + hd 190771 & & & 20 05 09.78 & + 38 28 42.4 & 6.82 & 6.17 & g5 & 54.0 + 27 cyg & hd 191026 & rs & 20 06 21.77 & + 35 58 20.9 & 6.21 & 5.36 & k0 & 41.3 + @xmath62 pav & hd 190248 & v & 20 08 43.61 & -66 10 55.4 & 4.32 & 3.56 & g7 & 163.7 + hd 195564 & & & 20 32 23.70 & -09 51 12.2 & 6.34 & 5.65 & g2.5 & 41.3 + 6 del & hd 196524 & sb & 20 37 32.94 & + 14 35 42.3 & 4.04 & 3.63 & f5 & 33.5 + @xmath65 cep & hd 198149 & & 20 45 17.38 & + 61 50 19.6 & 4.35 & 3.41 & k0 & 69.7 + @xmath64 cyg & hd 202444 & v & 21 14 47.49 & + 38 02 43.1 & 4.11 & 3.72 & f0 & 47.8 + @xmath57 cep & hd 203280 & v & 21 18 34.77 & + 62 35 08.1 & 2.66 & 2.44 & a7 & 66.8 + hd 203985 & & & 21 27 01.33 & -44 48 30.8 & 8.38 & 7.48 & k1 & 42.5 + 51 cap & hd 207958 & & 21 53 17.77 & -13 33 06.4 & 5.45 & 5.08 & f3 & 36.2 + @xmath61 cep & hd 211336 & ds & 22 15 02.19 & + 57 02 36.9 & 4.47 & 4.19 & f0 & 38.9 + hd 212330 & & & 22 24 56.39 & -57 47 50.7 & 5.99 & 5.31 & g3 & 48.8 + 49 peg & hd 216385 & & 22 52 24.08 & + 09 50 08.4 & 5.64 & 5.16 & f7 & 37.3 + @xmath69 ind & hd 216437 & & 22 54 39.48 & -70 04 25.4 & 6.69 & 6.06 & g2.5 & 37.7 + hd 217107 & & & 22 58 15.54 & -02 23 43.4 & 6.90 & 6.18 & g8 & 50.7 + 6 and & hd 218804 & & 23 10 27.20 & + 43 32 39.2 & 6.38 & 6.00 & f5 & 35.4 + 35 cep & hd 222404 & v & 23 39 20.85 & + 77 37 56.2 & 4.26 & 3.22 & k1 & 72.5 + hd 224022 & & p & 23 54 38.62 & -40 18 00.2 & 6.60 & 6.03 & f8 & 35.9 + @xmath57 phe & hd 2261 & sb & 00 26 17.05 & -42 18 21.5 & 3.46 & 2.37 & k0 & 42.1 + @xmath58 cet & hd 4128 & v & 00 43 35.37 & -17 59 11.8 & 3.06 & 2.04 & k0 & 34.0 + @xmath70 cet & hd 11171 & ds & 01 49 35.10 & -10 41 11.1 & 4.981 & 4.66 & f3 & 42.4 + @xmath57 ari & hd 12929 & v & 02 07 10.41 & + 23 27 44.7 & 3.15 & 2.00 & k2 & 49.5 + @xmath66 hyi & hd 21024 & & 03 15 57.66 & -77 23 18.4 & 5.914 & 5.51 & f4 & 34.0 + @xmath71 dor & hd 27290 & pu & 04 16 01.59 & -51 29 11.9 & 4.55 & 4.20 & f4 & 49.3 + @xmath57 tau & hd 29139 & v & 04 35 55.24 & + 16 30 33.5 & 2.39 & 0.85 & k5 & 50.1 + hd 29883 & & & 04 43 35.44 & + 27 41 14.6 & 8.92 & 8.01 & k5 & 44.7 + 67 eri & hd 33111 & v & 05 07 50.99 & -05 05 11.2 & 2.92 & 2.79 & a3 & 36.7 + hd 290054 & & & 05 14 48.13 & + 00 39 43.1 & 11.1 & 9.9 & k2 & 38.1 + @xmath57 aur & hd 34029 & rs & 05 16 41.36 & + 45 59 52.8 & 0.88 & 0.08 & g5 & 77.3 + cal 42 & hd 36705 & ro & 05 28 44.83 & -65 26 54.9 & 7.73 & 6.93 & k1 & 66.9 + @xmath58 col & hd 39425 & & 05 50 57.59 & -35 46 05.9 & 4.28 & 3.12 & k2 & 37.9 + sv * zi 543 & hd 47205 & v & 06 36 41.04 & -19 15 21.2 & 5.01 & 3.96 & k1 & 50.4 + @xmath57 cha & hd 71243 & & 08 18 31.55 & -76 55 11.0 & 4.436 & 4.06 & f5 & 51.4 + q pup & hd 63744 & & 08 18 33.31 & -36 39 33.4 & 4.657 & 4.44 & a7 & 14.4 + sv lmi & hd 82885 & rs & 09 35 39.50 & + 35 48 36.5 & 6.18 & 5.41 & g8 & 89.5 + 46 lmi & hd 94264 & v & 10 53 18.71 & + 34 12 53.5 & 4.87 & 3.83 & k0 & 33.4 + u car & hd 94510 & v & 10 53 29.66 & -58 51 11.4 & 4.713 & 3.79 & k1 & 33.7 + @xmath65 cru & hd 105211 & ds & 12 06 52.90 & -64 36 49.4 & 4.462 & 4.14 & f2 & 50.8 + @xmath71 cru & hd 108903 & v & 12 31 09.96 & -57 06 47.6 & 3.22 & 1.63 & m3.5 & 37.1 + @xmath62 mus & hd 112985 & sb & 13 02 16.26 & -71 32 55.9 & 4.80 & 3.62 & k2 & 35.9 + gj 9427 & sao 63355 & & 13 07 35.06 & + 34 24 06.1 & 10.53 & 9.30 & k3 & 41.9 + hd 118036 & & & 13 34 16.26 & -00 18 49.6 & 8.27 & 7.38 & k4 & 41.0 + @xmath57 boo & hd 124897 & v & 14 15 39.67 & + 19 10 56.7 & 1.19 & -0.04 & k1.5 & 88.9 + @xmath71 boo & hd 127762 & ds & 14 32 04.67 & + 38 18 29.7 & 3.23 & 3.00 & a7 & 38.3 + 107 vir & hd 129502 & & 14 43 03.62 & -05 39 29.5 & 4.26 & 3.90 & f2 & 53.5 + hd 130322 & & & 14 47 32.73 & -00 16 53.3 & 8.80 & 8.05 & k0 & 33.6 + @xmath58 tra & hd 141891 & & 15 55 08.56 & -63 25 50.6 & 3.14 & 2.85 & f2 & 81.2 + 26 sco & hd 151680 & v & 16 50 09.81 & -34 17 35.6 & 3.44 & 2.29 & k2.5 & 49.9 + @xmath59 oph & hd 153210 & v & 16 57 40.10 & + 09 22 30.1 & 4.35 & 3.20 & k2 & 38.0 + @xmath57 oph & hd 159561 & v & 17 34 56.07 & + 12 33 36.1 & 2.23 & 2.10 & a5 & 69.8 + 60 oph & hd 161096 & v & 17 43 28.35 & + 04 34 02.3 & 3.967 & 2.77 & k2 & 39.8 + 10 sgr & hd 165135 & v & 18 05 48.49 & -30 25 26.7 & 3.99 & 2.99 & k0 & 33.9 + 111 her & hd 173880 & ds & 18 47 01.27 & + 18 10 53.5 & 4.477 & 4.34 & a5 & 35.2 + 38 sgr & hd 176687 & ds & 19 02 36.71 & -29 52 48.4 & 2.704 & 2.60 & a2 & 36.6 + hd 187237 & & & 19 48 00.88 & + 27 52 10.3 & 7.51 & 6.88 & g2 & 38.4 + hd 195627 & & & 20 35 34.85 & -60 34 54.3 & 5.031 & 4.75 & f1 & 36.3 + @xmath72 cyg & hd 197989 & & 20 46 12.68 & + 33 58 12.9 & 3.49 & 2.50 & k0 & 45.3 + @xmath73 oct & hd 205478 & sb & 21 41 28.65 & -77 23 24.2 & 4.76 & 3.76 & k0 & 47.2 + @xmath62 cap & hd 207098 & al & 21 47 02.45 & -16 07 38.2 & 3.16 & 2.87 & a7 & 84.6 + hd 216448 & & & 22 52 00.55 & + 57 43 00.9 & 9.05 & 8.02 & k5 & 41.2 + @xmath71 tuc & hd 219571 & & 23 17 25.77 & -58 14 08.6 & 4.369 & 4.00 & f1 & 45.4 + @xmath41 and & hd 222107 & rs & 23 37 33.84 & + 46 27 29.3 & 4.90 & 3.82 & g8 & 38.7 +

by solving the time - dependent radiation transfer problem of stellar radiation through evolving superbubbles within a smoothly varying h i distribution , we estimate the fraction of ionizing photons emitted by ob associations that escapes the h i disk of our galaxy into the halo and intergalactic medium ( igm ) . we consider both coeval star - formation and a gaussian star - formation history with a time spread myr . we consider both a uniform h i distribution and a two - phase ( cloud / intercloud ) model , with a negligible filling factor of hot gas . we find that the shells of the expanding superbubbles quickly trap or attenuate the ionizing flux , so that most of the escaping radiation escapes shortly after the formation of the superbubble . superbubbles of large associations can blow out of the h i disk and form dynamic chimneys , which allow the ionizing radiation to escape the h i disk directly . however , blowout occurs when the ionizing photon luminosity has dropped well below the association s maximum luminosity . for the coeval star - formation history , the total fraction of lyman continuum photons that escape both sides of the disk in the solar vicinity is . for the gaussian star formation history , , a value roughly a factor of two lower than the results of , where superbubbles were not considered . for a local production rate of ionizing photons s , the flux escaping the disk is s for coeval and gaussian star formation , comparable to the flux required to sustain the reynolds layer . rayleigh - taylor instabilities exist early in the ob association s evolutionary stages , possibly causing the shell to fragment and increasing . however , if a significant fraction of h i is distributed in cold clouds with , can be reduced by a factor of if the cloud properties are similar to `` standard clouds '' with a disk geometry . casa , department of astrophysical and planetary sciences , university of colorado , campus box 389 , boulder , co 80309 also at department of physics , metropolitan state college of denver , campus box 69 , denver , co 80217 also at jila , campus box 440 , boulder , co 80309 osservatorio astrofisico di arcetri , i-50125 firenze , italy jila visiting fellow

astro-ph9903331

the diffuse , ionized medium ( dim ) , a.k.a the `` reynolds layer , '' has been established to occupy a significant fraction of the interstellar medium ( ism ) and to have a vertical scale height of @xmath16 1 kpc ( @xcite ) . it is widely believed that ob associations are the sources of radiation responsible for ionizing the reynolds layer ( @xcite ) . it is still unclear how this ionizing radiation is able to travel so far from the ob associations immediate vicinity . in a previous paper ( @xcite , hereafter referred to as ds94 ) , we calculated the geometry of diffuse h ii regions due to ob associations in the galactic plane by assuming a three - component model ( @xcite ) for the vertical distribution of hydrogen . for associations producing an ionizing luminosity @xmath17 s@xmath5 , we found that the h ii regions are `` density bounded '' in the vertical direction , allowing photons to escape into the halo of our galaxy . by integrating over the luminosity function of associations , @xmath18 ( @xcite ) , we estimated that roughly 7% of ionizing photons , corresponding to a number flux @xmath19 @xmath4 s@xmath5 , escape each side of the h i disk layer in the solar vicinity and penetrate the dim layer . if this radiation is absorbed in the dim , assumed to lie entirely above the hi layer , the corresponding photoionization rate would be comparable to the recombination rate implied by the h@xmath20 emission measurements ( @xcite ) . if roughly one - third of this radiation escapes the galactic halo , this flux is also consistent with the estimated flux required to photoionize the magellanic stream and several high velocity clouds in the solar vicinity ( @xcite ) . rather than assuming a purely diffuse ism , @xcite considered the effects of a statistical distribution of small opaque clouds , also decreasing with increasing galactic height and embedded within the diffuse gas . these authors found that the resulting h ii regions are consistent with the observations of dispersion measures and h@xmath20 emission measurements . although these results are encouraging , no models to date consider the role of dynamic chimneys and superbubbles produced by the ob associations . these considerations are the focus of this paper . we recently became aware of a preprint on a similar topic ( @xcite ) . for an ob association , the stars that produce most of the ionizing radiation have strong stellar winds and form supernovae after their relatively short lifetime . therefore , shortly after the formation of an ob association , a superbubble will form , with a thin shell behind the shock front excavating a large cavity of hot gas ( @xcite ) . for a sufficiently large association , the superbubble can break out of the h i disk , forming a dynamic chimney , where the hot cavity penetrates into the halo ( @xcite ) . as we discuss below , the shells of the expanding superbubbles quickly trap the ionizing radiation , so that no radiation can escape the disk until blowout . after blowout , a large fraction ( @xmath16 10% per side ) of ionizing radiation can directly escape the disk of the galaxy by propagating through the dynamic chimney . however , this is a viable mechanism for transporting the ionizing radiation only if an appreciable amount of ionizing radiation is still being produced within the cavity after the dynamic chimney has formed . we find that , for both a coeval star formation history or a non - coeval model having a star formation rate given by a gaussian with a full - width half - maximum @xmath21 myr ( @xcite ) , a small fraction of the ionizing radiation produced by an association over its lifetime is emitted after the time of blowout . therefore , most of the radiation that escapes the disk does so early in the early stages of the superbubble . integrating over the luminosity function , we find that the fraction of ionizing photons _ currently _ being emitted by ob associations that escape each side of the h i disk is @xmath22% for the coeval star - formation history and @xmath23% for the gaussian star - formation history . the aim of this paper is to predict the fraction of ionizing radiation , emitted by ob associations , that escapes the h i disk of the galaxy . as discussed above , ds94 assumed that the reynolds layer lies above the h i layer . therefore , the question of whether the reynolds layer is sustained via hot star photoionization is equivalent to whether enough radiation can penetrate through the h i layer . in this paper , we regard this assumption as too simplistic . the reynolds layer has a scale height of roughly 1 kpc , whereas the h i disk has a scale height of @xmath24 pc ( @xcite ) but extends to at least 1 kpc ( @xcite ) . thus , a significant fraction ( e.g. , @xmath25% ) of the of the ionized gas may lie within the h i layer , reducing the flux of ionizing radiation required to escape the h i disk . a self - consistent model , modeling the distribution of the photoionized gas , is not feasible until the density distribution of the hydrogen gas , both within the disk and in the halo , is determined more accurately . in 2 , we discuss our treatment of the evolution of the superbubble and the radiation transfer of ionizing photons emitted by an association through the expanding superbubble and the diffuse ism of the disk . we also discuss our treatment of the blowout event of the superbubble , where the cavity evolves into a dynamic chimney . in 3 , we determine the time dependence of the fraction of ionizing photons , emitted by a single ob association , that escape the h i disk . integrating over the luminosity function of ob associations , we determine the fraction of ionizing radiation , currently being emitted by ob associations , that escapes the h i disk . in 4 , we discuss key issues not included in our standard model , including shell instabilities , a two - phase ism , `` poisoning '' by photoablated gas , and triggered star formation . we then discuss how these effects may alter our results . in 5 , we discuss the implications of our results and give our conclusions . consider a large association having @xmath26 stars in the @xmath27 mass range . due to the stellar winds from the massive stars and type ii supernovae , the association produces a mechanical luminosity , @xmath28@xmath29 , that varies with time . this energy input drives an expanding superbubble . assuming an initially uniform density distribution of the diffuse ism , and that all of the swept - up mass resides in a thin shell ( the kompaneets approximation , @xcite ) , the radius , @xmath30 , of the cavity is given by ( @xcite ) @xmath31 = \frac{6 { \ensuremath{r_{\mbox{\scriptsize\rm sh}}}}^2}{\pi\rho}{\ensuremath{l_{\mbox{\scriptsize\rm mech}}}}(t),\ ] ] where the mass density @xmath32 , @xmath33 is the number density of hydrogen , and @xmath34 is the proton mass . for a given time - dependent mechanical luminosity prescription , this equation is integrated numerically using a fourth - order runge - kutta algorithm . we note that this formalism is technically valid only for a uniform density distribution , as the shock wave does not evolve with spherical symmetry for non - uniform density distributions . however , as we show in 3 , for most association sizes the expanding shell becomes optically thick to the ionizing radiation before the shock has propagated a distance of a disk scale height , within which using a constant density is a good approximation . therefore , the amount of escaping ionizing radiation is insensitive to the errors produced by neglecting the non - spherical propagation that occurs at large galactic heights . the time dependence of the mechanical luminosity depends on the initial mass function ( imf ) of stars within an ob association and the star formation history of the association . for this paper , we use the ob association evolution models of @xcite , which determine both the mechanical luminosity and the luminosity of ionizing photons as a function of time for individual stars from the latest stellar evolutionary tracks and stellar atmospheres . these authors used an imf given by @xmath35 , where @xmath36 for the stellar mass range @xmath37 ( @xcite ) , for which all stars eventually become supernovae . the massive star - formation history of the association is given by @xmath38 , with the constraint that @xmath39 , where the limits of integration span the entire lifetime of the association . in this paper , we consider two star - formation histories ( sfhs ) : ( 1 ) coeval [ @xmath40 , and ( 2 ) noncoeval with a gaussian distribution [ @xmath41 $ ] . for the gaussian sfh , the imf is assumed to be constant in time . for each sfh , @xmath42 and @xmath43 were obtained from @xcite . here , we used a monte - carlo simulation to determine the mass distribution of individual associations having a total mass of @xmath44 ( corresponding to an average association size of @xmath45 stars , or roughly @xmath46 massive stars ) . the average luminosity curves are then normalized , and , for an arbitrary association size , the luminosity curves are given by the product of the normalized curves and @xmath47 . note that this formalism ignores all possible deviations about the average luminosities due to statistical fluctuations of the stellar mass distribution of small associations . @xcite found that these fluctuations become significant for @xmath48 . however , as shown below , most of the escaping radiation in our galaxy emanates from associations with @xmath49 , so the importance of the statistical fluctuations of small associations is minor for our purposes . even if the small associations were important , when calculating the total fraction of escaping photons , we integrate over many associations , so the use of the average luminosities should yield a good approximation . in figure [ fig : lvst ] , we show the mechanical luminosity and the mach number of the shell as a function of time for an ob association having @xmath50 , for coeval star - formation history and for a gaussian star - formation history with @xmath21 myr . the mach number of the shock front is @xmath51 , where the effective sound speed of the medium is @xmath52 here , @xmath53 is the boltzmann constant , @xmath54 is the temperature of the gas ( assumed equal to @xmath55 k and the shock front is assumed to be isothermal ) , @xmath56 , and @xmath57 is the ratio of the magnetic pressure plus turbulent pressure to the thermal pressure . as discussed in [ sec : blowout ] , the superbubble is modeled as instantaneously blowing out of the disk and becoming a dynamic chimney when the bubble radius exceeds the transition height , @xmath58 . we assume that @xmath58 is twice the scale height of the h i gas distribution , @xmath59 . after blowout , the cylindrical radius continues to expand due to conservation of momentum . once the mach number reaches unity , the expansion of the cavity is halted . the fluctuations in the mechanical luminosity arise from supernovae occurring separately in time due to the different lifetimes of individual stars . although @xmath42 is given by the average of several monte - carlo simulations , the fluctuations have not been smoothed out . we note that , for the gaussian sfh , the mach number is near zero until 1.3 myr , when it then rises quickly . this behavior is due to our initial treatment of the bubble growth for low mechanical luminosities . since we start the simulation at a time @xmath60 before the peak star - formation rate , the initial luminosity is small for small associations . in such cases , the mach number rapidly drops below unity , then subsequently increases as the energy input increases . in reality , the shock front associated with these small associations would stagnate when the ambient pressure of the ism is equal to the internal pressure of the bubble , causing the kompaneets approximation to break down . as the proper hydrodynamic treatment of the growth of a stagnated bubble is outside the scope of this paper , we employed the following procedure for approximating the bubble evolution . for each association size , we determine the time at which the mechanical luminosity exceeded @xmath61 erg s@xmath5 . we then evaluate the mach number at this time using the solution of equation ( [ eq : kompaneets ] ) for a constant luminosity , where the time - averaged luminosity is used . if the mach number is below unity , we assume that the shock has stagnated , set the shell velocity to zero , and continue integrating equation ( [ eq : kompaneets ] ) . although the use of kompaneet s equation is not valid for mach numbers below unity , we argue that the errors introduced are negligible since the time interval for the shell to accelerate to mach numbers above unity is small relative to the sound - crossing time of the bubble . therefore , the shell does not diffuse much into the cavity during the stagnation period . furthermore , during this early interval of the cavity , the photon luminosity of the association is also low , and no radiation escapes the disk of the galaxy regardless of the bubble geometry . finally , by comparing @xmath62 using this method with results when equation ( [ eq : kompaneets ] ) is integrated without intervention , we found that the two were identical for @xmath63 myr . the only ramification of our stagnation treatment is that smaller associations take longer to blow out of the disk . as discussed in [ sec : instabilities ] , the acceleration of stagnated bubbles due to the mechanical luminosity increasing in time can cause rayleigh - taylor instabilities to begin at a much earlier stage , compared to the case of constant luminosity , where acceleration occurs at the time of blowout ( @xcite ) . this instability may dramatically increase the amount of ionizing radiation that escapes the h i disk . for large ob associations that have a sufficiently high mechanical luminosity , superbubbles can blow out of a vertically stratified h i disk , producing dynamic chimneys ( @xcite ) . blowout occurs if the shell of the superbubble is still supersonic when it reaches a height , @xmath64 , where it begins to accelerate due to decreasing density distribution ( @xcite ) . this height depends on the density distribution of the ism and @xmath65 ; for gaussian or exponential distributions it is @xmath66 ( @xcite ) . for most association sizes and star formation histories , the timescale for reaching @xmath67 is considerably longer than the timescale for propagating from this height into the galactic halo ( @xmath68 kpc ) . therefore , rather than calculating the exact evolution of the superbubble during blowout , we model the process as an instantaneous transition from a bubble into a dynamic chimney once the shell reaches a transition height @xmath69 . we note that , due to the existence of the h i gas at high @xmath70 , blowout may first occur at heights considerably larger than @xmath71 or may not occur at all . in addition , magnetic fields could inhibit blowout ( @xcite ) . as discussed below , our results are insensitive to the exact value of @xmath72 , since most ionizing radiation that escapes the disk does so shortly after the formation of the ob association . therefore , the uncertainties regarding the process of blowout are unimportant for our estimation of the flux of escaping ionizing radiation . the geometry of the dynamic chimney is approximated as a cylinder , centered on the ob association , with radius @xmath73 and height @xmath74 . since we are only interested in how much ionizing radiation escapes the h i disk of the galaxy , we do not attempt to model the structure of the superbubble / chimney for heights more than @xmath75 disk scaleheights ( @xmath76 kpc ) . numerical simulations of superbubbles blowing out of stratified disks ( @xcite ) show that a cylinder is a decent approximation for the geometry of the cavity within the h i disk . above the disk , the shell takes the appearance of a mushroom cloud , with the shell clumping and fragmenting due to rayleigh - taylor instabilities . for this paper , we assume that any radiation that reaches a height @xmath74 and is within the dynamic chimney escapes the disk . the question of whether this radiation is absorbed by the `` mushroom cap '' of the shell or escapes due to the shell having a small covering fraction is outside the scope of this paper . even if the covering fraction of the `` mushroom cap '' is not negligible , it is possible that this gas , photoionized by the ob association , is part of the reynolds layer . therefore , it is fair to include this radiation when assessing whether the amount of escaping radiation can ionize the dim . throughout this paper , we assume that ionization equilibrium is satisfied instantaneously . as discussed in ds94 , the ionization front in a gaussian disk layer will reach a galactic height @xmath70 in a time @xmath77\nonumber\\ & & \times \left(\frac{{\ensuremath{\sigma_{\mbox{\scriptsize\rm h}}}}}{0.25\ { \rm kpc}}\right)^3\left(\frac{n_0}{0.3\ { \rm cm}^{-3}}\right ) s_{\rm 49}^{-1}\ { \rm myr } \nonumber,\end{aligned}\ ] ] where @xmath78 is the photon luminosity is units of @xmath79 s@xmath5 , @xmath59 is the scaleheight of gaussian disk , and @xmath80 is the number density of hydrogen in the midplane of the disk . ( this equation should replace equation ( 3 ) of ds94 , whose equation ( 3 ) contains an error . ) for @xmath81 , @xmath82 kpc , and @xmath83 @xmath9 , the i - front reaches a height @xmath84 in a time @xmath85 myr and reaches a height @xmath86 in a time @xmath87 myr , both relatively small compared to the lifetime of the superbubble . as shown in ds94 , ionization equilibrium is reached within a decade or so after the passing of the i - front . [ for gas with an ionization fraction @xmath88 , ionization equilibrium is reached in a time @xmath89 . thus , even though the density distribution of the superbubble is evolving in time ( as discussed below ) , the h ii region is determined in a quasi - static fashion . we assume that the stars of individual ob associations are compactly distributed so that the radiation can be treated as emanating from a point source . we define @xmath90 to be the dimming function of the ob association , which relates the time - dependent photon luminosity of lyman - continuum ( lyc ) photons and the maximum luminosity via @xmath91 for a point source of ionizing radiation , the h ii / h i boundary is given implicitly by @xmath92 where @xmath93 is the angle between the normal of the galactic disk and the radial unit vector , @xmath94 is the case - b recombination rate for hydrogen , and @xmath33 is the number density of hydrogen . azimuthal symmetry has been assumed . for the vertical distribution of h i , ds94 considered both the three - component dickey - lockman model ( @xcite ) as well as several single gaussian distributions ( with different amounts of dark matter ) which were fitted to numerical calculations assuming hydrostatic equilibrium ( @xcite ) . we found that all models predicted the same fraction of escaping photons within 10% . this gaussian distribution which most closely resembled the dickey - lockman model is given by @xmath95 where @xmath70 is the vertical distance above the mid - plane of the disk , @xmath96 @xmath9 and @xmath97 kpc . in this paper , we only consider this single - gaussian distribution . as the superbubble expands into the diffuse ism , all of the swept - up mass is assumed to reside in the thin shell . the thickness of the shell , @xmath98 , is given by equating column densities , @xmath99 which yields @xmath100 where @xmath101 is the number density within the shell , assumed to be a constant with respect to radius , and @xmath102 . defining the transition radius @xmath103 , the geometry of the h ii region for an ob association with luminosity @xmath104 , situated on the mid - plane of the galaxy , is given implicitly by @xmath105\right . \nonumber \\ & & + \left.\frac{\sigma_h^2}{2\mu^2}\left\{{\ensuremath{r_{\mbox{\scriptsize\rm tr}}}}\exp\left[-\left(\frac{\mu { \ensuremath{r_{\mbox{\scriptsize\rm tr}}}}}{\sigma_h}\right)^2\right ] - \right.\right.\nonumber \\ & & \left.\left . { \ensuremath{r_{\mbox{\scriptsize\rm sh}}}}\exp\left[-\left(\frac{\mu{\ensuremath{r_{\mbox{\scriptsize\rm sh}}}}}{\sigma_h}\right)^2\right]\right\ } \right)\nonumber\end{aligned}\ ] ] the density of this shell is given by the isothermal shock condition for compression in the cool post - shock layer , @xmath106 in figures [ fig : hiigeomvst.coeval.n50 ] and [ fig : hiigeomvst.coeval.n300 ] , we show the time evolution of the geometry of the h ii region shortly after the formation of an ob association , for @xmath107 and @xmath108 , respectively , assuming coeval star formation . for both association sizes , the h ii region is initially density bounded in the vertical direction ( see below for discussion of how much radiation escapes the h i disk ) . as the superbubble expands , the column density of the shell increases . since the shell has a higher density than the diffuse gas , and therefore a higher recombination rate , the volume of the h ii region decreases with time . eventually , the h ii region is radiation bounded even in the vertical direction . for @xmath108 , a portion of the h i / h ii interface lies within the shell for @xmath109 myr , and by @xmath110 myr the entire interface is within the shell . the h ii region goes from being density limited to radiation limited in a short time interval . for @xmath107 , the h ii region becomes radiation limited at only a slightly earlier time compared to the @xmath111 association ( @xmath112 myr compared to @xmath113 myr ) even though its photon luminosity is six times lower . this is due to the smaller association having lower shell velocities , causing both a slower accumulation rate of the column density as well as lower gas densities within the compressed shell . if the h ii region is density bounded , then some radiation escapes the h i disk . the fraction of ionizing photons that escape along the differential solid angle @xmath114 is @xmath115 & = & 1 - \frac{4\pi\alpha^{(2)}_h}{s({\ensuremath{n_{\mbox{\scriptsize\rm t}}}},t)}\int\limits_{0}^{\infty } n_h^2(r,\theta , t ) r^2\ { \ensuremath{{d}}}r\nonumber\\ & = & 1 - \frac{4\pi n_0 ^ 2\alpha^{(2)}_h}{s({\ensuremath{n_{\mbox{\scriptsize\rm t}}}},t ) } \left\{\frac{1}{3}\left(\frac{{\ensuremath{n_{\mbox{\scriptsize\rm sh}}}}}{n_0}\right)^2({\ensuremath{r_{\mbox{\scriptsize\rm tr}}}}^3 - { \ensuremath{r_{\mbox{\scriptsize\rm sh}}}}^3 ) + \right . \nonumber\\ & & \left . \left(\frac{\sigma_h}{\mu}\right)^3\left(\frac{\sqrt{\pi}}{4}\left[1-{\ensuremath{{\rm erf}}}\left(\frac{\mu{\ensuremath{r_{\mbox{\scriptsize\rm tr}}}}}{\sigma_h}\right)\right ] + \right.\right.\nonumber \\ & & \left.\left.\frac{{\ensuremath{r_{\mbox{\scriptsize\rm tr}}}}\mu}{2\sigma_h}\exp\left[-\left(\frac{\mu{\ensuremath{r_{\mbox{\scriptsize\rm tr}}}}}{\sigma_h}\right)^2 \right]\right)\right\}.\end{aligned}\ ] ] the critical angle , @xmath116 , is the angle at which the h ii region is radiation bounded for @xmath117 and density bounded for @xmath118 . the critical angle is found by solving @xmath119 . in contrast to ds94 , @xmath120 is time - dependent since both the density distribution and the ionizing luminosity vary with time . finally , we define the fraction of photons , emitted by a single ob association having size @xmath47 , age @xmath121 , and corresponding luminosity @xmath122 , that escape _ each _ side of the h i disk to be @xmath123 for a dynamic chimney with height @xmath74 and cylindrical radius @xmath73 , the angle for which @xmath124 for @xmath125 is given by @xmath126 . the procedure for determining the geometry of the h ii region and the fraction of escaping photons is similar to that of the superbubbles . the only difference is the treatment of the shell . as with the superbubble , we assume that the density of the shell is @xmath127@xmath12 @xmath128 and that the column density of the shell is equal to the column density of swept - up gas within the cavity . however , for dynamic chimneys , the trajectories of the shell deviate from radial lines since the superbubble evolves from a sphere into a cylindrical cavity . technically , the column density of the shell for a vertical height @xmath70 would be given by the line integral of the density of the diffuse ism along the trajectory that leads to the chimney wall at a height @xmath70 . the trajectories can be determined by requiring them to be perpendicular to the surface of the expanding shock wave . for an isothermal atmosphere , @xcite found an analytic solution for the trajectories as a function of time , and , for heights less than a scale height or so , the particle trajectories are found to be roughly radial . for larger heights , the trajectories tend to bend away from the vertical axis and become horizontal near the shell boundary . therefore , we make the approximation that the streamlines are purely radial for @xmath129 and purely horizontal for @xmath130 , where @xmath131 , a free parameter of the model . as discussed below , the results of this paper are insensitive to the value of @xmath72 . with this treatment , for @xmath132 , the shell thickness is given by @xmath133 where @xmath134 and @xmath135 . defining @xmath136 and @xmath137 , the geometry of the h ii region and the fraction of escaping radiation are again given by equations ( [ eq : hiigeom ] ) and ( [ eq : etaesc ] ) , respectively . in figure [ fig : etaesct - coeval ] we show the photon luminosity , emitted by a single ob association having @xmath26 stars , and the luminosity of photons escaping each side of the h i disk as a function of time . here , the stars are assumed to form coevally . notice that the fraction of escaping photons , @xmath138 , rapidly decreases for @xmath139 myr . in fact , for @xmath140 , roughly 90% of the time - integrated flux of escaping radiation occurs within the first 2.7 myr , while only 72% of the time - integrated flux of ionizing radiation has been emitted by this time . the reason for this rapid decrease is that , as the column density of the shell accumulates , the ionizing radiation is very quickly trapped within the shell , causing the fraction of escaping photons to drop to zero . once the ionization front is trapped within the shell , the fraction of escaping photons remains zero unless the superbubble blows out of the h i disk . ( we have not considered shell fragmentation ) . if blowout occurs , photons can directly escape through the chimney ( a small fraction can also escape by penetrating through the upper h ii disk ) , causing @xmath141 to jump up to @xmath142% . however , by the time of blowout , which occurs at @xmath143 myr for @xmath144 and @xmath145 myr for @xmath146 , the photon luminosity is considerably lower than its peak luminosity ( which occurs at @xmath147 myr ) . therefore , the luminosity of escaping radiation is significantly lower compared to its value for @xmath148 myr . in figure [ fig : etaesct - gauss ] we plot the emitted and escaping photon luminosity luminosity for case of the noncoeval , gaussian sfh with @xmath0 myr . for @xmath149 , the photon luminosity @xmath43 is very low , and therefore the fraction of escaping photons is small or even zero , in accordance with the results of our ds94 . as the luminosity increases , @xmath141 increases until the shell of the superbubble becomes thick enough that the shell - integrated recombination rate approaches the photoionization rate . since the shell expands more slowly compared to the coeval sfh ( having the same value of @xmath47 ) , the column density of the shell does not increase as quickly . in addition , due to the lower mach number of the shock front , the shell density is lower than that of the coeval sfh . for these two reasons , @xmath150 does not fall off in time as quickly as in the coeval case after it reaches its peak value . nevertheless , the shell does attenuate significantly the flux of escaping radiation . in figure [ fig : etaesct2 ] , for @xmath50 , we compare the flux of escaping radiation as a function of time with that for the case where no dynamics are considered ( e.g. , as done by ds94 but using the time - dependent photon luminosity ) . for @xmath151 myr , the shell becomes becomes efficient in reducing the escaping flux . therefore , as with the coeval sfh , most of the escaping radiation escapes early , within a short time interval relative to the lifetime of the ob association . the time - integrated flux of photons that escape the disk after blowout is considerably small relative to the lifetime - integrated flux . for example , for @xmath140 , fewer than 3% of the photons that escape the disk over the lifetime of the association escape after blowout ( while fewer than 4% of the photons emitted by the association are produced after blowout ) . for @xmath50 , approximately 1% escape after blowout . these values are not much different than for the coeval sfh . the reason for this insensitivity is that as @xmath152 is increased ( @xmath153 for coeval star - formation ) , the blowout timescale is also increased . thus , even though there are more stars alive at longer times for noncoeval sfhs , there are still relatively few alive at the time of blowout . the time - integrated fraction of escaping photons for each side of the h i disk by an association of size @xmath26 is given by @xmath154 in figure [ fig : time - int - eta ] we plot @xmath155 for both the coeval sfh and the noncoeval gaussian star - formation model . for reference , the maximum photon luminosity of an association over its lifetime is @xmath156 ( @xmath79 s@xmath5 ) for coeval star - formation and @xmath157 ( @xmath79 s@xmath5 ) for the gaussian sfh . note that @xmath158 is smaller for all values of @xmath47 for the gaussian sfh . there are several reasons for this difference . one reason is that , for a given value of @xmath47 , the gaussian sfh has a peak photon luminosity that is roughly a factor of two smaller than that of the coeval model . therefore , even if the shell structures of the superbubble were identical during the interval of maximum luminosity , the maximum value of @xmath150 for the gaussian model would be smaller since @xmath141 increases with increasing photon luminosity ( ds94 ) . in addition , the luminosity profile @xmath43 for the coeval model is much more peaked , with @xmath159% of its lifetime - integrated luminosity emitted within the first @xmath160 myr , compared to the relatively broad distribution of @xmath43 for the gaussian sfh . thus , since @xmath158 is weighted by @xmath43 , and @xmath161 is near its peak value while @xmath43 is near its peak value , the coeval sfhs yield higher values . finally , for coeval sfhs , the peak of @xmath43 occurs shortly after the formation of the association , before the formation of a significant superbubble shell , allowing these photons to escape more easily . we now discuss the fraction of photons , currently being emitted by ob associations of all ages and sizes , that escape the h i disk of the galaxy . we assume a constant formation rate , @xmath162 , of associations per unit area in our galaxy . the probability of a new association having a peak luminosity @xmath163 [ recall that @xmath164 is given by the implicit association luminosity function , @xmath165 . therefore , during a small time interval @xmath166 , the number of associations that will eventually have a peak luminosity @xmath163 is @xmath167 . the number density ( kpc@xmath168 s@xmath5 ) of associations observed today as a function of luminosity , @xmath169 , is related to the implicit luminosity function via @xmath170\ { \ensuremath{{d}}}t.\ ] ] here , @xmath171 is given by @xmath172 , where @xmath173 is the inverse function of @xmath90 , and @xmath174 is given by @xmath175 if the dimming function @xmath90 is multi - valued , then @xmath176 and @xmath177 ( for @xmath178 ) are determined from its maximum values . the observed luminosity function , from h@xmath20 measurements of all known h ii regions in 30 nearby spiral and irregular galaxies , is given by @xmath179 for @xmath180 , where @xmath181 ( @xcite , and references therein ) . @xcite finds a similar result from studying galactic h ii regions . see ds94 for a discussion regarding the uncertainty of the upper and lower limits of the luminosity function . assuming @xmath182 with @xmath183 , the corresponding @xmath169 is close to a pure power law with the same power - law index , with deviations first occurring at @xmath184 . therefore , for the remainder of this paper , unless otherwise stated , we assume that the intrinsic luminosity function is a power law with @xmath185 . a better understanding of the relationship between the implicit luminosity function and the observed luminosity function can be obtained by considering an example with a simple analytical dimming function , @xmath90 . assume @xmath186 and the implicit luminosity function @xmath187 . then , for @xmath188 , @xmath189.\end{aligned}\ ] ] here , @xmath190 , so finally we have @xmath191.\ ] ] therefore for @xmath188 and @xmath192 , @xmath193 . the larger the value of @xmath194 , the closer @xmath104 must be to @xmath195 for appreciable differences to occur between the intrinsic and observed luminosity functions . for @xmath196 , a 10% difference between the two functions first occurs when @xmath197 . since the range of association luminosities is several orders of magnitude , and the value of @xmath195 is very uncertain , we regard this discrepancy as negligible . note that the reason for any differences between the two luminosity functions is due to our prescription that the implicit luminosity function is truncated at @xmath195 . physically , as @xmath104 approaches @xmath195 , only the largest associations recently formed can contribute to the luminosity function , causing the observed luminosity function to drop to zero as @xmath104 reaches @xmath195 . also note that the difference between the two luminosity functions is independent of the functional form of the dimming function . the current production rate of ionizing photons ( # per unit area per unit time ) is given by @xmath198 substituting the expression for the observed luminosity function in terms of the implicit luminosity function [ equation ( [ eq : dnassds ] ) ] , we find , @xmath199 the flux of photons escaping from each side of the h i disk is given by @xmath200\ { \ensuremath{{d}}}t\ { \ensuremath{{d}}}s.\ ] ] here , @xmath201 is the fraction of photons that escape each side of the h i disk emitted by an association of size @xmath26 , where the peak luminosity of this association is @xmath202 = { \ensuremath{s_{\mbox{\scriptsize\rm 0}}}}({\ensuremath{n_{\mbox{\scriptsize\rm t}}}})$ ] . by computing a two - dimensional grid of @xmath138 using the methods outlined in [ sec : etaesc ] , we integrate equation ( [ eq : psiesc ] ) for the two star - formation scenarios . for given values of @xmath104 and @xmath121 , an interpolation routine was used to find the corresponding value of @xmath26 , such that @xmath203 $ ] . finally , the total fraction of photons , currently being emitted , that escape both sides of the h i disk is @xmath204 we find that @xmath205% for the coeval sfh and @xmath23% for the gaussian sfh . in figure [ fig : fesc - vs - parms ] , we show how @xmath206depends on the model parameters . for reasons given in [ sec : etaescavg ] , @xmath206 for the gaussian sfh is considerably lower than that for the coeval sfh . by far , @xmath206 is most sensitive to the power - law index of the luminosity function , @xmath194 . this sensitivity arises because most of the escaping radiation comes from the relatively few large associations with @xmath49 ; small changes in @xmath194 cause rather large changes in the relative number of these large associations . for the coeval sfh , @xmath206 is insensitive to the magnetic field ( the parameter @xmath57 ) . as the superbubble expands , the column density of the shell increases quickly , causing @xmath201 to drop to zero very quickly even for strong magnetic fields . for the gaussian sfh , the superbubble expands more slowly , and different values of @xmath57 cause the shell to trap the ionization front at different times . since this trapping of the i - front occurs near the peak of @xmath43 , @xmath206 is more sensitive to @xmath57 . lastly , we do not plot how @xmath206 depends on the parameters involved with the superbubble blowout ( e.g. , @xmath58 and @xmath74 ) , as we found that varying these parameter over a range between @xmath59 and @xmath207 corresponded to a relative change of @xmath206 less than 1% . this insensitivity occurs because most emitted and escaping photons occur early after the formation of the association . we now discuss the key issues not included in our standard model . we then discuss how these effects may alter our results . as discussed above , the presence of a thick neutral shell of cool , shocked ism produced by multiple supernova explosions is quite effective in absorbing the radiation coming from the ob association . this process significantly reduces the fraction of photons able to escape into the galactic halo until blowout takes place . at that stage , the previously decelerating shell is suddenly accelerated and the interface between the dense shell and the hot cavity gas becomes rayleigh - taylor ( r - t ) unstable . the nonlinear development of such instability induces fragmentation of the shell into a number of clumps , through which the photons can percolate and almost freely escape into the halo . this process generally occurs relatively late ( or not at all ) , when the radiative flux has already decreased by a considerable amount due to the aging of the parent ob association . it is thus worthwhile to investigate if the fragmentation process driven by r - t or other instabilities can take place before blowout , consequently increasing @xmath206 . there are at least three types of instabilities that could affect the shell at the beginning or during the pressure - driven radiative ( pdr ) phase : 1 ) dynamical , 2 ) rayleigh - taylor and 3 ) gravitational . to assess the importance of these processes , it is necessary to discuss in more detail the development of the shell structure for sn bubbles . we restrict our discussion to uniform ambient gas density , which is a reasonable approximation to study the pre - blowout phase . the structure of a wind - blown bubble ( the energy injection by multiple sne can be well approximated as a continuous process ) is constituted by a reverse ( wind " ) shock propagating back into the wind , which is under most conditions adiabatic , and an ambient " shock propagating into the surrounding medium ; the two shocks are separated by a contact discontinuity . initially , the ambient shock is also adiabatic and its radius increases with time as @xmath208 . however , when the cooling time of the shocked ambient gas becomes of the order of the age of the bubble , the ambient shock becomes radiative and the shocked ambient gas collapses into a thin shell . the shell - formation phase is characterized by violent dynamical instabilities which tend to amplify the perturbation due to classical thermal instability in the catastrophically cooling shocked ambient gas . are these instabilities able to disrupt the shell soon after its formation ? at least three different dynamical instabilities may arise in the thin , cool shell : nonradial oscillations of the radiative shock ( @xcite ) , the nonlinear thin - shell ( ntsi ) instability arising when the shell is bounded by two radiative shocks ( @xcite ) , and the pressure - driven thin - shell overstability ( pdtso ) found in a shell bounded by a high thermal pressure on one side and by a radiative shock on the other ( @xcite ) . these dynamical instabilities may produce considerable distortions in the shell , often showing nonradial oscillations . the main difference between the latter two instabilities is that while the ntsi grows exponentially , the pdtso only grows with a power - law rate . recent numerical simulations ( @xcite ) have shown that the pdtso is probably the most dangerous instability , especially for superbubbles where the wind shock is adiabatic rather than radiative for a large fraction of its evolution . however , the pdtso works only for the time during which the mach number of the shell is larger than @xmath209 . as the pdtso grows with time as @xmath210 with @xmath211 , whereas the shock velocity decreases as @xmath212 ( see equation [ [ radius ] ] below ) in the radiative phase , amplifications @xmath213 are typically possible , i.e. the overstability saturates . the above numerical studies concluded on this basis that fragmentation of the shell due to these instabilities is unlikely , although @xcite emphasize that in a denser ism the shock enters the radiative phase at a higher mach number , thus allowing for a longer growth time of the instability . even if the shell can survive dynamical instabilities , it can be fragmented by r - t instabilities generated _ before _ blowout . as the radiative bubble expands , its pressure drops and eventually becomes equal to the that of the ism . in this case , the wind shock stalls , but the shocked wind is still adiabatic and forces the shell to continue its expansion . however , the expansion rate is now so slow that the acceleration of the shell is dominated by the galactic gravitational field . thus , the polar region of the shell will be r - t unstable even in the absence of a blowout . we calculate the time at which the pressure - confined phase occurs ( and hence the r - t instability starts to grow ) as follows . for constant luminosity , the evolution of the shell radius in the radiative phase is given by ( @xcite ) @xmath214 where @xmath215 is the wind mechanical luminosity , @xmath80 is the ambient gas number density , and @xmath216 . from the pressure balance condition @xmath217 , where @xmath218 is the ism effective sound speed , this transition occurs at @xmath219 where @xmath220 is the ism effective sound speed . the growth time of the r - t instability on spatial scales @xmath221 is @xmath222^{-1/2 } \simeq \left({2\pi g \over \lambda_s } \right)^{-1/2 } \sim 1{\rm ~myr}\ ] ] if the density contrast between the shell and the hot cavity gas @xmath223 and @xmath224 cm s@xmath168 . as @xmath225 is much shorter than @xmath226 , the shell is fragmented soon after it enters the pressure - confined phase . for the case of a gaussian sfh , the mechanical luminosity is initially very low , causing the shock to stagnate much earlier than the constant - luminosity case . in addition , the shell can be re - accelerated early in its evolutionary stage shortly after the first supernovae of the association due to the sudden increase in the mechanical luminosity . we have numerically evaluated @xmath227 due to this process assuming the kompaneets approximation is valid , and find that the product of @xmath227 and the duration of the re - acceleration ( typically @xmath228 myr ) is @xmath229 . thus , the growth of the instability is moderately non - linear . a more detailed study of this process , including the role of poisoning , will be given in @xcite . therefore , in contrast to the case of a constant mechanical luminosity , the shell can be r - t unstable before blowout . in this case , a fragmenting shell would allow a higher amount of ionizing radiation to escape the h i disk since fragmentation occurs during the time interval in which the photon luminosity of the association is near its peak value . the shell might also be fragmented by a gravitational instability . several studies have investigated this possibility , both under interstellar ( @xcite ) and intergalactic conditions ( @xcite ) . here , we simply state the the growth time of the most unstable mode is @xmath230 again , we see that this mechanism is able to fragment the shell on time scales of interest and similar to the ones found for the r - t instability . we conclude that , even in the absence of blowout , fragmentation of the shell is likely about @xmath231 yr after the beginning of the energy injection . even if the shells of superbubbles do not blow out , do they fragment while the ob association is still producing ionizing photons ? simple conditions on the occurrence of blowout have been derived in different contexts by @xcite and @xcite . neglecting the role of magnetic fields , @xcite studied in detail the case for the milky way and concluded that the blowout requires @xmath232 . for a marginally confined bubble , we find from equation ( [ tpres ] ) and equation ( [ ggrow ] ) that @xmath233 yr and @xmath234 yr for the r - t and the gravitational instability , respectively . note that , as @xmath235 is decreased , fragmentation occurs earlier in time . thus , even bubbles that do not blow out of the disk can fragment on time scales short enough that at least some ionizing photons escape . determining the covering fraction of the fragmented clumps is outside the scope of the present estimates and will require numerical simulations of the fragmentation process . as a final caveat , we stress that a dynamically important magnetic field could qualitatively modify the above conclusions by suppressing some of the instabilities ( @xcite ) and by introducing new ones . although our models assume that , outside the superbubbles , the neutral gas is distributed diffusely , there is observational evidence that a significant fraction of the gas is in cold h i clouds , with densities much larger than the space - averaged density . due to the higher recombination rates , these clouds modify our radiation - transfer results and therefore merit some consideration . since most radiation that escapes the h i disk comes from young ob associations , with shells having small column densities , we considered the ramifications of a clumpy medium by ignoring the effects of evolving superbubbles . a more proper treatment would consider the time - dependent evolution of the clouds due to passage of the shell of the superbubble as well as the expansion and evaporation of the clouds after being photoionized . as an approximate model , we let a mass - fraction of the gas , @xmath236 , be contained in cold , dense clouds . the remaining fraction of the gas , @xmath237 , is distributed diffusely using the dickey - lockman density distribution ( dickey & lockman 1990 ) . we consider two cloud geometries : ( 1 ) spheres of radius @xmath238 ; or ( 2 ) cylindrical disks with radius @xmath238 and an aspect ratio @xmath239 , where @xmath240 is the height of the cylinder . each of the cylindrical disks is assumed to be `` face - on '' with respect to the line of sight from the ob association . for either geometry , all clouds are assumed to have identical column densities , sizes , and densities . the number of clouds per unit area of the disk is given by @xmath241 where @xmath242 is the volume of each cloud , @xmath243 is the density of hydrogen within each cloud , and @xmath244 @xmath4 is the observed total column density of hydrogen through the entire disk at the solar circle . we assume that the clouds are distributed randomly within the disk of the galaxy using a probability distribution , @xmath245 , that is constant in the plane and proportional to the dickey - lockman density distribution in the vertical direction . the _ average _ number of clouds intersecting a line of sight having an inclination angle @xmath93 ( @xmath102 ) , emanating from an ob association situated on the midplane , is given by @xmath246 as in 2.3.2 , the fraction of lyc photons from a single ob association that escape each side of the disk is given by @xmath247 where @xmath248 here , @xmath249 is given by the diffuse density distribution for radii outside any cloud and is equal to @xmath243 for radii inside any cloud . for a given value of @xmath93 and @xmath104 , we evaluate @xmath250 by averaging over 100 monte - carlo simulations . for each simulation , the number of clouds along the line of sight is drawn from a poisson distribution having an average number of clouds given by equation ( [ eq : avgclouds ] ) . the distances from the origin of each of these clouds to the ob association are picked randomly using the probability distribution @xmath245 . for spherical clouds , the average chord length between the line of sight and the cloud is @xmath251 . for the cylindrical clouds , we assume that each cloud is orientated face - on , such that @xmath252 . the observed column density of each cloud in the line of sight is @xmath253 . using the average values of @xmath250 , we determine @xmath254 via equation ( [ eq : etaesc2 ] ) . finally , as in ds94 , we determine the total fraction of ionizing radiation that escapes the disk is determined by integrating over the luminosity function , @xmath255 lcllcl disk & @xmath256 & @xmath257 & @xmath258 & @xmath259 & @xmath260 disk & @xmath256 & @xmath257 & @xmath258 & @xmath261 & @xmath262 disk & @xmath256 & @xmath257 & @xmath258 & @xmath263 & @xmath258 disk & @xmath256 & @xmath257 & @xmath264 & @xmath259 & @xmath262 disk & @xmath256 & @xmath257 & @xmath264 & @xmath261 & @xmath258 disk & @xmath256 & @xmath257 & @xmath264 & @xmath263 & @xmath265 disk & @xmath256 & @xmath266 & @xmath264 & @xmath259 & @xmath267 disk & @xmath256 & @xmath266 & @xmath264 & @xmath261 & @xmath268 disk & @xmath256 & @xmath266 & @xmath264 & @xmath261 & @xmath260 disk & @xmath269 & @xmath266 & @xmath258 & @xmath259 & @xmath270 disk & @xmath269 & @xmath266 & @xmath258 & @xmath261 & @xmath271 disk & @xmath269 & @xmath266 & @xmath258 & @xmath263 & @xmath272 disk & @xmath269 & @xmath266 & @xmath264 & @xmath259 & @xmath271 disk & @xmath269 & @xmath266 & @xmath264 & @xmath261 & @xmath264 disk & @xmath269 & @xmath266 & @xmath264 & @xmath263 & @xmath273 disk & @xmath269 & @xmath274 & @xmath264 & @xmath259 & @xmath258 disk & @xmath269 & @xmath274 & @xmath264 & @xmath261 & @xmath271 disk & @xmath269 & @xmath274 & @xmath264 & @xmath263 & @xmath272 sphere & @xmath256 & @xmath266 & @xmath258 & @xmath259 & @xmath275 sphere & @xmath256 & @xmath266 & @xmath258 & @xmath261 & @xmath276 sphere & @xmath256 & @xmath266 & @xmath258 & @xmath263 & @xmath277 sphere & @xmath269 & @xmath266 & @xmath258 & @xmath259 & @xmath278 sphere & @xmath269 & @xmath266 & @xmath258 & @xmath261 & @xmath277 sphere & @xmath269 & @xmath266 & @xmath258 & @xmath263 & @xmath279 sphere & @xmath269 & @xmath274 & @xmath257 & @xmath259 & @xmath277 sphere & @xmath269 & @xmath274 & @xmath257 & @xmath261 & @xmath280 sphere & @xmath269 & @xmath274 & @xmath257 & @xmath263 & @xmath281 in table [ tab : fescclump ] , we give @xmath206 for various values of @xmath282 , and @xmath283 . it is apparent that @xmath206 is sensitive to these parameters , which are poorly constrained observationally . for a given value of @xmath284 , @xmath206 decreases as the average number of clouds within a line of sight increases . we find that the i - front is trapped within most clouds . only clouds very close to the ob association are not able to trap the i - front and become completely ionized , a consequence of the high recombination rate within the cloud . essentially all escaping radiation emanates from lines of sight that are free of clouds . for comparison , for the case where @xmath285 , @xmath286 for @xmath287 , and @xmath263 , respectively ( these results are identical to those of ds94 ) . we conclude that , if an appreciable amount of the h i gas is in cold clouds having a disk geometry and column densities and hydrogen densities similar to the h i clouds studied by @xcite , then @xmath206 can be a factor of @xmath10 smaller than previously estimated . for a fixed amount of gas locked in cold clouds , spherical clouds occupy a smaller solid angle as compared to thin cylindrical clouds , allowing more radiation to escape . note that , since we did not consider superbubbles in these calculations , these numbers are only indicative of the _ relative _ decrease of @xmath7 when cold clouds are included in the model . although the survey used for calibrating the production rate of ionizing photons is heavily biased towards associations far away from the parent molecular cloud , most if not all associations were born in a molecular cloud . therefore , early in the lifetime of the ob association , it is possible that a considerable amount of matter is photoablated from the molecular cloud . this material increases the interior density of the bubble , leading to radiative cooling , potentially at a rate equal to the the injected mechanical luminosity , a process independently dubbed `` poisoning '' by @xcite and @xcite . this poisoning process can alter the evolution of a young superbubble , as well as the propagation of the i - front through the cavity . to date , the photoablation rate has not been calculated for a time - varying luminosity of an ob cluster with a proper radiation spectrum . in addition , for a given mass - injection rate , solving the evolution of the injected gas within the cavity is a complex problem . these complexities lead to uncertainties in the radiative cooling rate , both in magnitude and in its time evolution . therefore , in order to estimate the degree to which poisoning alters the evolution of the superbubble and @xmath206 , we introduced a factor , @xmath288 , representing the fraction of the mechanical luminosity injected into the cavity that is drained due to radiative cooling from both photoablated gas and evaporated gas from the dense shell . thus , @xmath289 is the portion of energy available to do @xmath290 work . in what we consider an extreme case , we let @xmath291 for the duration in which @xmath292 is less than @xmath293 pc [ see @xcite for a discussion of when cooling due to poisoning becomes negligible ] . in this case , for our non - coeval model , we find that @xmath206 _ increases _ by roughly 35% . this increase occurs because the superbubble shell expands more slowly , and the density of the shell is smaller ( it is proportional to the mach number squared of the shell ) , and because the column density of the shell increases more slowly . on the other hand , it is possible that enough material is photoablated into the hot cavity that , after recombining , it contributes to the absorption of ionizing radiation emitted by the ob association , partially offsetting the increase in @xmath206 due to the slower shell expansion . however , to model this process requires a knowledge of the relative configurations of the molecular cloud , the ob association , and the disk of the galaxy . also , the evolution of the injected gas , especially the expansion of the gas as it expands from the molecular cloud or shell boundaries and into the the line - of - sight between the ob association and the halo , is a numerical problem beyond the scope of this paper . the fraction of ionizing radiation escaping the h i disk would be increased if an appreciable number of ob associations were formed within pre - existing dynamic chimneys . the two - dimensional filling factor of dynamic chimneys is estimated by @xmath294 where @xmath295 is the fraction of ob associations that are large enough to produce a dynamic chimney , @xmath296 is the lifetime of the cavity , and @xmath297 is the radius of the cavity . as discussed in [ sec : instabilities ] , associations having an average mechanical luminosity @xmath298 erg s@xmath5 are thought to be capable of blowing out of the galactic disk ( @xcite ) . this corresponds to an ob association size of @xmath299 , or a peak luminosity @xmath300 s@xmath5 . the fraction of associations that produce a dynamic chimney is then @xmath301 where we assumed @xmath302 . for @xmath185 and @xmath303 s@xmath5 , @xmath304 . the lifetime of the dynamic chimney is estimated to be the sound crossing time , @xmath305 therefore , the probability that an ob association is born within a dynamic chimney is given by @xmath306 and it is unlikely that @xmath7 is enhanced significantly by this process if ob associations are born randomly in space . however , as suggested by @xcite , the birth of ob associations could be triggered by the passage of the shock wave associated with another superbubble . indeed , recent far ultraviolet , h@xmath20 , and h i observations of several dwarf galaxies show evidence of secondary star - formation sites on the dense h i rims of superbubbles ( @xcite ) . we find that @xmath307% for the coeval star - formation history and @xmath308% for the gaussian star - formation history . for @xmath309 s@xmath5 , @xmath310 , and @xmath196 , we obtain @xmath311% for the coeval sfh and @xmath312% for the gaussian sfh . adopting a production rate , @xmath313 @xmath4 s@xmath5 , the corresponding photon fluxes are @xmath314 @xmath4 s@xmath5 for the coeval sfh and @xmath315 @xmath4 s@xmath5 for the gaussian sfh . for comparison , @xmath316% using the formalism of ds94 , where the photon luminosity was assumed constant in time and dynamically evolving superbubbles were not considered . in order to determine the degree to which the differences between our new results and our old results are due to the treatment of expanding superbubbles and dynamic chimneys , we also calculated @xmath7 for the case where we neglect entirely the superbubble dynamics but include the time - varying photon luminosities . for the same parameters as given above , we find @xmath317% for the gaussian sfh and @xmath318% for the coeval sfh , values similar to the old model . these small differences are due solely to the new treatment of the luminosity function of associations with time - varying photon luminosities , as discussed in 3.3.1 . therefore , the inclusion of the evolution of superbubbles _ reduces _ the fraction of escaping radiation , as the superbubble shells trap more radiation . for coeval sfhs , the inclusion of the dynamics only changes @xmath206 from @xmath319% to @xmath320% , since most radiation is emitted and escapes before a substantial shell has formed . on the other hand , for the gaussian sfh , @xmath206 drops from @xmath321% to @xmath322% . this reduction of @xmath7 occurs because the photon luminosity peaks after a shell has formed , causing fewer photons to escape compared to the case when superbubbles are neglected . in ds94 , we argued that @xmath206 must be @xmath323% in order for the reynolds layer to be sustained by radiation from ob associations . however , using costar model atmospheres , @xcite found that the yield of ionizing photons from ob associations is 50% higher than previously estimated , and the production rate of ionizing photons by ob associations in the solar vicinity is @xmath324 @xmath4 s@xmath5 . thus , the required value of @xmath206 to sustain the reynolds layer is reduced to @xmath325% . the flux of escaping radiation for the gaussian sfh is about 25% smaller than this required value . however , as discussed in 1 , since a portion of the reynolds layer may reside within the h i disk , it need not be sustained solely by the radiation escaping the disk . throughout this paper , we have assumed that the formation rate of ob associations is constant in time . for starburst galaxies , the observed star formation rate is much higher than the time - averaged value . therefore , since @xmath326 is at a maximum shortly after the formation of an ob association , starburst galaxies having a morphology similar to our galaxy should have higher values of @xmath206 than those predicted for our galaxy . in addition , starburst galaxies may have lower h i column densities and lower gravitational fields , also contributing to a higher escape fraction of ionizing radiation . these considerations will be the focus of our subsequent work . if ob associations are indeed the sources of radiation responsible for ionizing the reynolds layer , it is possible that an appreciable amount of radiation can escape the galaxy entirely and contribute to the intergalactic radiation field ( @xcite ) . if the average escape fraction from low - redshift , star - forming galaxies exceeds @xmath229% , then the contribution of ionizing radiation to the intergalactic medium by galaxies can exceed the contribution for qsos ( @xcite ) . we note , however , that our models predict only the flux of radiation penetrating the galactic halo , before a portion of it is absorbed by the portion of the reynolds layer situated high above the galactic plane . in order to predict the fraction of radiation that escapes the galaxy entirely , we need a self - consistent model of the density distribution of hydrogen gas within the halo , including the contribution due to high velocity clouds , whose covering fraction is uncertain . @xcite found that approximately 37% of 21-cm sightlines near quasars showed hvcs with n@xmath327 @xmath4 . however , observations with the parkes multibeam survey ( @xcite ) show that the spatial distribution of h i in the hvcs is filamentary , suggesting that lyman continuum could leak through . furthermore , the actual hvc coverage fraction will vary considerably around the galaxy , depending on proximity to spiral arms and upwelling superbubbles . for specificity , if we assume @xmath328 , then the upward ionizing flux is @xmath329 @xmath4 s@xmath5 above the h i disk ( @xmath330 kpc ) . if this flux is incident on the hvcs , and if we use the relationship between h@xmath20 emission and ionizing flux , for optically thick clouds and photoionization equilibrium ( @xcite ) , then the predicted h@xmath20 emission intensity is @xmath331 this intensity is roughly a factor of two larger than the measured intensities from the magellanic stream ( @xcite ) and the smith hvc ( @xcite ) . in summary , we have shown that the escape fraction of ionizing radiation from star - forming regions of disk galaxies is strongly affected by the presence of superbubbles and a cloudy ism . the shells of the expanding superbubbles quickly trap or attenuate the ionizing flux , so that most of the escaping radiation escapes shortly after the formation of the superbubble . in our models , the escape fraction above 1 kpc range from @xmath332% . for a gaussian sfh , @xmath206 is roughly a factor of two lower than the results of @xcite , where superbubbles were not considered . if @xmath333% , then the upward flux is sufficient to sustain the reynolds layer . we have also examined the radiative transfer in a two - phase ( cloud / intercloud ) medium . we neglect the filling factor of hot , low - density gas , consistent with recent o vi studies . if a significant fraction of h i is distributed in cold clouds with @xmath8 @xmath9 and column densities @xmath334 @xmath4 , @xmath7 can be reduced by a factor of @xmath10 if the clouds have a disk geometry . thus , the amount of ionizing photons that escape the disk into the igm is likely to be highly variable , and will depend on the luminosity distribution of ob associations , the cloud structure , and the massive star formation history . if the escaping lyc radiation is ever observed directly , it will confirm these theoretical expectations but also verify the likelihood that stellar ionizing radiation plays an important role in reionization of the high - redshift igm . we thank ralph sutherland for the ob association luminosity curves . we thank mark giroux , phil maloney , john bally , and brad gibson for informative discussions . we also thank the referee for useful suggestions that led to major improvements to the paper . this work was supported by the astrophysical theory program at the university of colorado , through grants nag5 - 4063 from nasa and ast96 - 17073 from nsf .

the dilute model is a solvable irf ( interaction round a face ) model with three local states and adjacency conditions encoded by the dynkin diagram of the lie algebra . it can be regarded as a solvable version of an ising model at the critical temperature in a magnetic field . one therefore expects the scaling limit to be governed by zamolodchikov s integrable perturbation of the conformal field theory . indeed , a recent thermodynamic bethe ansatz approach succeeded to unveil the corresponding structure under certain assumptions on the nature of the bethe ansatz solutions . in order to check these conjectures , we perform a detailed numerical investigation of the solutions of the bethe ansatz equations for the critical and off - critical model . scaling functions for the ground - state corrections and for the lowest spectral gaps are obtained , which give very precise numerical results for the lowest mass ratios in the massive scaling limit . while these agree perfectly with the mass ratios , we observe one state which seems to violate the assumptions underlying the thermodynamic bethe ansatz calculation . we also analyze the critical spectrum of the dilute model , which exhibits massive excitations on top of the massless states of the ising conformal field theory .

hep-th9610003

the ising model @xcite is , without doubt , one of the most frequently studied and best understood lattice models of classical statistical mechanics . although onsager s solution @xcite of the two - dimensional ising model without external field dates back half a century already , no analytic solution of the 2d ising model in a magnetic field has been found . however , the situation is somewhat different if considered from the viewpoint of field theory . the critical 2d ising model corresponds to the @xmath2 conformal field theory ( cft ) @xcite of a massless majorana fermion . it was zamolodchikov @xcite who noticed that a ( symmetry - breaking ) perturbation of this cft with the relevant spin density operator ( which has conformal dimensions @xmath3 ) preserves infinitely many conservation laws and therefore leads to an integrable quantum field theory . the corresponding minimal theory contains eight massive particles with factorized ( purely elastic ) scattering @xcite ; the particle masses and the @xmath4-matrix elements are related to the exceptional lie algebra @xmath5 . this integrable field theory describes an appropriate scaling limit of the 2d ising model in a magnetic field . numerically , the predictions for the lowest mass ratios have been verified by several authors @xcite . as these results rely on relatively small size transfer matrix calculations or on monte carlo simulations , they provide rather crude checks for the lowest mass ratios only . furthermore , the larger masses can not be obtained directly as these lie above the two - particle threshold of the lightest particles and hence are buried in the continuum of scattering states . however , one can extract indirect information about the higher masses and the scattering amplitudes from the finite - size behavior of the lower masses @xcite . another , rather different , approach employs a truncated conformal hilbert space @xcite , which however does not make the connection to the lattice model . similar theoretical and numerical investigations have also been performed for a variety of lattice models such as the quantum ising chain @xcite , the lee - yang model @xcite , the three - state potts model @xcite , the tricritical ising respectively blume - capel model @xcite , the ashkin - teller model @xcite , the @xmath6 models @xcite , and the integrable rsos models @xcite . recently , this has been extended to the study of form factors and correlation functions @xcite and non - integrable perturbations @xcite . the discovery of a new class of lattice models @xcite which are solvable in the presence of a symmetry - breaking field @xcite has changed the situation considerably . although we still can not solve the 2d ising model in a magnetic field , we now know a solvable model ( the so - called dilute @xmath0 model ) which belongs to the same universality class : it has a critical point of ising type @xcite , and it is solvable in a symmetry - breaking field that corresponds to the spin density perturbation of the cft @xcite . therefore , one expects this model to show the same properties in the scaling limit as the non - integrable ising model in a magnetic field . note , however , that the two models are rather different lattice models : the dilute @xmath0 model contains three local states , whereas the ising spins take two values only . the dilute @xmath7 models are solvable by bethe ansatz ( ba ) @xcite . by a thermodynamic bethe ansatz ( tba ) approach @xcite , the mass ratios and @xmath4-matrix of the @xmath5 field theory have been obtained . however , this approach relies on assumptions on the nature of the ba solutions , which could not be substantiated by numerical solution of the ba equations ( bae ) at criticality . connections between the @xmath0 model and the exceptional lie algebra @xmath5 are also revealed by the existence of corresponding rogers - ramanujan identities @xcite . for our numerical investigation of the off - critical bae , our main motivation was to check the assumptions on the conjectured string structure of the ba solutions . this could be achieved for the states with lowest masses , yielding at the same time very accurate numerical results not alone for the mass ratios , but also for the complete scaling functions . however , we also found _ one discrepancy _ with the predictions of ref . @xcite . within the range of scaling parameter values considered , the characteristic string type for a particle of mass @xmath8 does not match the proposed structure . as a by - product of our investigation , we found that the critical spectrum also contains _ massive _ excitations besides the well - known conformal spectrum . this is one property that distinguishes the dilute @xmath0 model from the ising model which does not have such excitations . however , this does not contradict universality as that only concerns the universal properties of the systems in the vicinity of the critical point , which are the same for both models . the ba solutions corresponding to the massive excitations show the same type of strings as those conjectured for the massive scaling limit . already for relatively small systems , these solutions lie very close to singularities of the bae , which makes the numerical treatment difficult . estimates of the corresponding mass ratios , which show no resemblance of the @xmath5 structure , are presented . this paper is organized as follows . in section [ s : a3 ] , the dilute @xmath0 models is introduced and the bae are presented . the critical spectrum predicted by cft and the behavior in the massive scaling limit are described . furthermore , the assumptions on the form of the ba solutions formulated in ref . @xcite are summarized . section [ s : nr ] contains the results of our numerical investigations . after a brief description of our numerical approach , we commence with the critical case , where not only the scaling dimensions predicted by cft are considered , but also additional massive excitations in the spectrum are identified . then , we show how the ba solutions behave as one goes away from criticality . finally , we present our numerical scaling functions for the lowest gaps and the numerical results for the mass ratios in the massive scaling limit . because an immediate specialization to the @xmath0 case does not considerably simplify the equations , we keep the first part of this section on a more general level and consider the dilute @xmath7 models . this also serves us to avoid an unnecessary repetition in a forthcoming publication @xcite . the dilute @xmath7 models are irf ( interaction round a face ) models @xcite of rsos ( restricted solid - on - solid ) type @xcite on the square lattice with adjacency conditions encoded by the dynkin diagram of the lie algebra @xmath7 . in contrast to the usual ( non - dilute ) rsos models built on this adjacency graph ( these are the abf models @xcite ) , the effective adjacency graph of the dilute @xmath7 models contains loops which connect each node to itself , see fig . [ f : ead ] . in other words , one considers configurations of @xmath9 local states or `` heights '' ( labeled @xmath10 ) on the vertices of the square lattice subject to the additional requirement that states on neighboring ( adjacent ) lattice sites may differ _ at most _ by one . one then defines a statistical model by assigning boltzmann weights to the elementary plaquettes ( faces ) of the lattice . the face weights of the dilute @xmath7 models are @xcite @xmath11 @xmath12 here , @xmath13 labels the heights , and the possible values of the variable @xmath14 are determined by @xmath9 @xcite . the crossing factors @xmath15 are given by @xmath16 and @xmath17 , @xmath18 are standard theta functions of nome @xmath19 with @xmath20 @xcite . these face weights satisfy the yang - baxter equation and therefore lead to an integrable lattice model . the corresponding row transfer matrices @xcite for fixed value of @xmath19 form a one - parameter commuting family in the spectral parameter @xmath21 . in general , the dilute @xmath7 models have four different branches for any value of @xmath9 @xcite , distinguished by two possible values of @xmath14 with two regimes each for the spectral parameter @xmath21 . here , we are only interested in the dilute @xmath0 model ( @xmath22 ) in a particular branch @xcite where the model behaves like the ising model in a magnetic field . for this case , @xmath23 and the spectral parameter @xmath21 lies in the interval @xmath24 . it is the parameter @xmath19 that acts like the magnetic field , and the ising critical point corresponds to @xmath25 . the eigenvalues @xmath26 of the row transfer matrix for a system of size @xmath27 with periodic boundary conditions have the following form @xcite @xmath28 where @xmath29 , and where the @xmath30 , @xmath31 , form a solution of the set of @xmath27 coupled bae @xmath32 here , @xmath33 labels a sector related to the braid limit eigenvalues of the row transfer matrix at criticality ( @xmath25 ) . whereas it is believed to be true that all eigenvalues of the transfer matrix are of the form ( [ e : eval ] ) , the converse is certainly wrong ( at least off - criticality ) the bae ( [ e : bae ] ) in general allow for many additional solutions which do not correspond to proper eigenvalues of the transfer matrix . for any numerically found solution of eq . ( [ e : bae ] ) one thus has to check whether it gives a proper eigenvalue of the transfer matrix ( see section [ s : nr ] below ) . the theta function @xmath17 shows the ( quasi-)periodicity properties @xcite @xmath34 where @xmath35 ( @xmath36 ) with @xmath37 . consider a solution @xmath38 of the bae ( [ e : bae ] ) in sector @xmath39 . clearly , nothing is changed if a multiple of @xmath40 is added to any of the roots @xmath30 . more interesting is the result of adding @xmath41 ( @xmath42 ) to a root @xmath30 . in this case , one obtains a solution of ( [ e : bae ] ) with @xmath43 replaced by @xmath44 . moreover , from eq . ( [ e : eval ] ) it is evident that these two solutions correspond to the same eigenvalue . this means that the `` sectors '' @xmath33 lose their significance in the off - critical case ( @xmath45 ) since we can always adjust @xmath39 by adding or subtracting suitable multiples of @xmath46 to some of the roots . of course , this also yields many possibilities in which the phase factors cancel , and hence eq . ( [ e : bae ] ) is recovered without any alteration , which means that a single solution of the bae can be presented in many ways differing by addition and subtraction of suitable multiples of @xmath46 to some of the roots . for the case of interest , it turns out that the largest eigenvalue @xmath47 of the transfer matrix is given by a purely imaginary solution of the bae ( [ e : bae ] ) in the sector @xmath48 . therefore , in place of the roots @xmath30 , we prefer to use @xmath49 in what follows ; thus the largest eigenvalue corresponds to a set of real ba roots @xmath50 . general solutions to the bae will however involve complex roots , which in the large @xmath27 limit typically arrange into so - called _ strings _ ( subsets of roots with approximately the same real part , in terms of the @xmath50 ) . this is shown schematically in fig . [ f : string ] , where the horizontal line represents the real axis . note that the `` string content '' of a particular solution might well depend on the variable @xmath19 , an example of this behavior will be given in section [ ss : bar ] below . the dilute @xmath0 model has a critical point of ising type at @xmath25 . thus its critical limit ( given by @xmath25 , @xmath51 ) is described by the @xmath2 cft with scaling dimensions @xmath52 . this has drastic consequences for the spectrum of the transfer matrix in the critical limit . consider the scaled spectral gaps @xmath53 where @xmath54 are the eigenvalues of the transfer matrix with periodic boundary conditions at the isotropic point @xmath55 , @xmath56 being the largest eigenvalue . in the critical limit , the spectrum of scaling dimensions @xmath57 consists of `` conformal towers '' of states labeled by pairs @xmath58 with scaled energy @xmath59 , conformal spin @xmath60 and momentum @xmath61 , where @xmath62 . they form representations of two commuting virasoro algebras with central charge @xmath2 . the degeneracies of the descendant states can be read off from the character functions of their irreducible representations with highest weight @xmath63 . these can be written in the form @xcite @xmath64 where @xmath65 is the generating function of the number of partitions @xmath66 moreover , the central charge @xmath2 manifests itself in the finite - size corrections of the largest eigenvalue @xcite @xmath67 where @xmath68 denotes the bulk free energy for @xmath55 . taking into consideration the parameter @xmath19 , one can approach the critical point by simultaneously performing the two limits @xmath69 and @xmath51 keeping the scaling variable @xmath70 constant . here , @xmath71 corresponds to the critical limit discussed above , @xmath72 to the massive limit where zamolodchikov s @xmath5 field theory results apply . in the scaling limit , the appropriately scaled spectral gaps @xmath73 become functions of the scaling variable @xmath74 alone . for the largest eigenvalue , we can also define a scaling function @xmath75 \label{e : f0}\ ] ] for the finite - size corrections of the largest eigenvalue , where @xmath76 $ ] is the bulk free energy again . note that the face weights ( [ e : fw ] ) for nome @xmath19 and @xmath77 are related by symmetry , thus it suffices to consider positive values of the nome . in the massive limit ( @xmath72 ) , the ratios @xmath78 approach the particle mass ratios of the corresponding massive field theory . the masses of the eight stable particles are proportional to the entries of the perron - frobenius eigenvector of the cartan matrix of the lie algebra @xmath5 and their ratios ( ordered by magnitude ) are given by @xmath79 where @xmath80 defines the mass scale . the dynkin diagram of @xmath5 is shown in fig . [ f : dd ] , where our labeling of nodes follows the ordering of the masses . in the tba calculation of ref . @xcite , each of the corresponding eight massive particles is associated to a particular string in the ba solution . in addition , it is assumed that , apart from these eight strings and the one - strings forming the vacuum solutions , no other string types occur in thermodynamically relevant quantities . with these assumptions , the tba equations imply that the hole - type excitations in the one - strings have vanishing density . the density of one - strings can be eliminated from the calculation , which finally leads to the @xmath81 scattering matrix of the @xmath5 factorized scattering theory @xcite . in table [ t : st ] , the nine string types are given , labeled by @xmath82 in the order used in ref . @xcite , which corresponds to the usual labeling of the @xmath5 dynkin diagram rather than that used in fig . [ f : dd ] . from the data of table [ t : st ] , the string of type @xmath83 consists of @xmath84 roots of the form @xmath85 where @xmath86 denotes the center of the string on the real line . for the numerical treatment of the bae ( [ e : bae ] ) we used a modified newton method @xcite . the calculations were performed in extended precision fortran ( with 16 byte real numbers ) on an ibm workstation . as a first step , we solved the bae at criticality ( @xmath25 ) for small system size and compared the corresponding eigenvalues ( [ e : eval ] ) to those obtained by direct diagonalization of the transfer matrix using high precision arithmetics . this also served as a test of the performance of our programs . in particular , the _ complete _ set of solutions corresponding to zero - momentum eigenstates of the transfer matrix for systems of size @xmath87 was obtained . in this way , the structure of the solutions for the largest eigenvalues could be identified . these solutions were followed as a function of the elliptic nome @xmath19 . the results of this analysis were then generalized to larger system size by looking for solutions of the same type which differ by additional real roots only . since we are mainly interested in the mass ratios in the scaling limit , we exclusively considered zero momentum states . for solutions of the bae , the momentum is obtained from the eigenvalue @xmath88 ( [ e : eval ] ) at spectral parameter @xmath89 , where the transfer matrix reduces to a shift operator . because we want to follow solutions for varying @xmath19 , which means that we have to repeat the calculation for a large number of values of @xmath19 , we limited ourselves to systems of size @xmath90 . however , in certain cases the numerical accuracy of the calculations becomes the major problem . this happens for instance for ba solutions which contain roots that are very close to singularities of the bae , or for solutions where differences between roots ( respectively their real parts ) become extremely small . these situations typically show up if one considers large values of the elliptic nome @xmath19 ( where the meaning of `` large '' depends on the system size ) , but also for the massive excitations of the critical dilute @xmath0 model discussed in section [ ss : me ] below . as far as we can see , there are only two ways to go beyond the limitations imposed by these numerical problems ; either by using higher precision arithmetics , or , to a lesser extent , by adjusting the program according to each specific type of ba solution ( which we did to some extent by treating the ubiquitous complex conjugate pairs of ba roots with imaginary parts close to @xmath91 in a special way ) . in the critical limit ( @xmath25 , @xmath51 ) , the largest eigenvalues of the transfer matrix organize according to the characters of representations of the virasoro algebra with central charge @xmath2 as described in section [ ss : cs ] above . schematically , the resulting values for the scaling dimensions @xmath92 in the zero - momentum sector and their distribution into the three sectors labeled by @xmath93 are shown in fig . [ f : conf ] . in table [ t : c ] , numerical values for the central charge and the lowest scaling dimensions obtained from the corresponding solutions of the bae for systems of different sizes are presented . here , eq . ( [ e : cc ] ) with the exact value of the bulk free energy @xcite of the dilute @xmath0 model was used . all zero - momentum states with a critical scaling dimension of @xmath94 have been found , compare fig . [ f : conf ] , except one of the four excitations with conformal dimensions @xmath95 for which , despite some effort , we have not been able to find the corresponding solution of the bae . in particular , this includes all cases without degeneracy , i.e. , where only a single zero - momentum state of a certain scaling dimension occurs in the conformal tower . to keep the three remaining ba solutions with scaling dimension @xmath96 apart , we distinguish them by subscripts @xmath97 , @xmath98 , and @xmath99 . actually , it turns out that the two solutions denoted by @xmath98 and @xmath99 yield the same eigenvalue even for finite systems , which remains true also off - criticality . thus , we do not need to consider them separately unless we are interested in the ba solutions , which of course are different , see section [ ss : bar ] below . while the numerical data are in perfect agreement with the exact values , it should be noted that the string structure of the critical ba solutions have no apparent similarity to those proposed in ref . we shall come back to this point later . analyzing the spectrum of largest eigenvalues of the transfer matrix , one observes that , in addition to the conformal spectrum discussed above , the spectrum contains _ massive _ excitations . more precisely , these are states with eigenvalues @xmath100 for which the quantities @xmath101 converge to a non - zero limit ( mass ) as @xmath51 , compare eq . ( [ e : xj ] ) for the conformal states where this limit gives zero . apparently , a number of different masses is involved , which are again characterized by particular strings in the ba solutions . strikingly , the strings which we observed are among those listed in table [ t : st ] . we have found explicit solutions containing strings of type @xmath102 , and @xmath103 . in all cases , the roots forming these strings are even for small systems located extremely close to the singularity of the bae which they approach for @xmath51 . this asks for a careful numerical treatment , and severely limits the system sizes we can treat . with the numerical accuracy of our program , we were able to get reliable results for systems of at most 2030 sites , depending on the particular solution . an example is given in table [ t : me ] . here , the solutions obtained by adding a number of `` massive '' two - strings ( @xmath104 ) to the ground - state solution of a system of size @xmath105 are presented . from the eigenvalues for larger systems , one can see that each additional `` massive '' two - string just adds the same mass , thus substantiating our interpretation in terms of massive particles . apparently , it is possible to add arbitrary numbers of these `` massive '' strings not only to the ground - state , but to _ each _ solution corresponding to a conformal state . this means that the spectrum ( at the critical point @xmath25 ) contains infinitely many copies of the complete conformal spectrum , shifted with respect to the ground - state by a mass determined by the collection of `` massive '' strings . as to the observed `` massive '' strings , they fall into two classes : for @xmath106 , and @xmath103 , we have a single solution of the bae , whereas for @xmath107 and @xmath108 the solutions belong to eigenvalues which are doubly degenerate ( for finite systems ) . labeling the masses associated to the `` massive '' string type @xmath83 by @xmath109 , we obtain the numerical estimates @xmath110 for the ratios of the masses with respect to the smallest mass which is associated to the two - string ( @xmath104 ) . note that these numbers are calculated from the eigenvalues of the transfer matrix with spectral parameter @xmath55 . this situation is similar to what is observed in the hubbard model , where one has two types of excitations : so - called `` holons '' ( carrying charge but no spin ) and `` spinons '' ( carrying spin but no charge ) @xcite . generically , both excitations are massless , and the corresponding field theory consists of two coupled @xmath111 cft . however , if the filling fraction ( number of electrons per lattice site ) is chosen to be precisely one , the holons become massive while the spinons stay massless . at low energies , this theory is then described by a @xmath111 cft with additional massive excitations @xcite . as mentioned above , the string structure of the ba roots at criticality does not seem to agree with the predictions of ref . however , if one follows a particular solution as a function of the nome @xmath19 , one finds that in many cases one encounters singularities of the bae where the string structure changes . the simplest example is given by the second largest eigenvalue , which corresponds to the conformal spin density field with dimensions @xmath112 in the critical limit and to the lightest massive particle in the massive scaling limit . at criticality , the corresponding solution of the bae differs from that of the largest eigenvalue ( which contains real roots @xmath50 only ) by a single root with imaginary part @xmath113 . however , as can been seen from fig . [ f : bas ] , as @xmath19 is increased from @xmath25 , the complex root and one real root approach each other , until their real parts agree . at that point , the real parts of the two roots stay the same , but they move in the imaginary direction until they form a complex conjugate pair of roots ( hence a two - string ) , which then persists for large values of @xmath19 . comparing systems of different size , this mechanism stays always the same , the additional real roots just play the rle of spectators . moreover , the transition from one string type to another takes place at approximately the same value of the scaling parameter @xmath74 ( [ e : mu ] ) as the system size is increased . more precisely , the scaling parameter values approach a non - zero limit as the system size tends to infinity . this implies that although @xmath69 in the scaling limit , for large values of @xmath74 ( and in particular for the massive scaling limit @xmath72 ) , the two - string is the relevant type of solution . the same scenario applies to the other excitations we considered . in table [ t : ss ] , we compiled numerical values of the non - real ba roots for systems of size @xmath114 , both at criticality ( @xmath25 , i.e. @xmath71 ) and for a rather large value @xmath115 of the scaling parameter ( corresponding to @xmath116 for @xmath114 ) . note that for @xmath117 the real part of the roots @xmath50 can be shifted by multiples of @xmath118 ( @xmath36 ) as discussed in section [ ss : bae ] above . using this property , we arranged the solutions such that the values of the non - real roots have a real part in the interval @xmath119 . actually , as table [ t : ss ] suggests , all non - real roots are located in the vicinity of @xmath120 for sufficiently large scaling parameters @xmath74 . this also distinguishes these solutions from the `` massive '' strings at criticality considered above , whose centers are located close to the origin . furthermore , the values of the nome @xmath19 where changes in the string patterns for @xmath114 occur are also indicated in table [ t : ss ] . all five single - particle states show at least one such point , whereas some of the two - particle states do not change as @xmath19 is varied . although the relation between the scaling exponents on one end and the content of massive particles on the other is determined by the field theory , we are not aware that it has been calculated . our results given in table [ t : ss ] give an unambiguous connection for the lowest states , but they do not suggest an obvious pattern that can be generalized to the higher excitations . comparing with the proposed string types of ref . @xcite given in table [ t : st ] and by eq . ( [ e : strings ] ) , one finds that the solutions for large scaling parameter agree with the predictions apart from the two states which contain a particle of mass @xmath8 . here , in place of a five - string with imaginary parts @xmath121 , @xmath122 and @xmath113 , the solution found numerically on first view looks more like a seven - string with imaginary parts @xmath123 , @xmath124 , @xmath125 and @xmath113 . however , in comparison to the other single - particle states , the real parts of the roots forming this `` string '' are quite far apart , which might be an indication for another change of pattern occurring at a larger value of @xmath19 , which of course we can not exclude on the basis of our numerical data . let us have a closer look at the single particle state with mass @xmath8 . figures [ f : bam4_10 ] and [ f : bam4_20 ] show the ba roots of the corresponding solutions for systems with @xmath126 and @xmath127 sites , respectively . for convenience , we plotted the real and imaginary parts of the roots , normalized by @xmath128 and by @xmath40 as in fig . [ f : bas ] , against @xmath129 . there are two obvious points where non - analyticities can be seen in the imaginary parts , which both are situated at approximately the same value of the scaling parameter for the two different systems . however , fig . [ f : bam4_10 ] may suggest that another change will occur at a larger value of @xmath74 than numerically accessible by our routines , as the real parts of one real root and one of the two - strings approach each other . but , comparing with the same region of fig . [ f : bam4_20 ] , no indication of this behavior remains , and the same holds true for the larger systems ( up to @xmath114 ) which we examined . from this , we conclude that either there is no further change in the string pattern , or it has to occur for a very large value of the scaling parameter @xmath74 , which however seems rather unlikely to us . though this result is somewhat inconclusive , let us examine the string pattern of this solution in more detail . in particular , we are interested in the question whether there are holes in the one - strings in this case . to see this , we consider the `` phase function '' @xmath130 \label{e : ph}\end{aligned}\ ] ] which is basically the logarithm of the bae ( [ e : bae ] ) , but now @xmath131 denote the roots of our particular ba solution , and we consider @xmath132 as a function of the complex variable @xmath133 . by definition , @xmath134 for all the roots @xmath50 , @xmath31 . restricting to real values of @xmath133 , @xmath135 therefore takes integer values at all real solution @xmath136 , hence for all one - strings . fig . [ f : m4ph ] shows the @xmath135 for the solution under consideration . here , the size of the system is @xmath127 , and @xmath137 . the horizontal lines are drawn at integer values ( the precise numbers are not relevant as they depend on the choice of branch in eq . ( [ e : ph ] ) ) , and the crosses located on intersections of these lines with the graph of @xmath135 denote the position of the real roots . the vertical lines are placed at the real parts of the remaining seven non - real roots . as observed above , the single root with imaginary part @xmath113 lies somewhat separated from the three complex conjugate pairs which are very close together ( such that the three vertical lines appear as one thicker line in the figure ) . moreover , the location of this single root coincides _ precisely _ with a hole in the one - strings , which may seem to disagree with the findings of ref . however , the behavior of the roots as a function of parameters like @xmath19 suggests that the roots with imaginary part @xmath113 are simply an alternative locus of the one - string . on the basis of the numerical data , it thus seems that the string structure associated to the mass @xmath8 consists of the rather complicated pattern of a seven - string involving three complex conjugate pairs of roots ( with imaginary parts @xmath123 , @xmath124 and @xmath125 ) and a neighboring root at @xmath138 which comes together with a hole in the one - string distribution . of course , in the infinite size limit the distance between the real parts of these roots becomes infinitesimal . it is not obvious what happens in the tba calculation of ref . @xcite if their string type @xmath139 ( see table [ t : st ] ) is replaced by the observed pattern , especially as we did not see the strings associated to the three largest masses . nevertheless , we believe that the main ideas underlying the tba treatment in ref . @xcite are correct , and that only details of the calculation would be affected . before we move on to the discussion of the scaling function , two short remarks regarding the ba solutions are in order . it should be noted that the `` massive '' strings mentioned previously in section [ ss : me ] , which yield the massive excitations _ at criticality _ , do not undergo similar changes . these strings are always located very close to singularities of the bae , and move even closer as the nome @xmath19 is increased . moreover , their centers lie at the origin , see e.g. table [ t : me ] , whereas for the states belonging to the @xmath5 integrable field theory the strings cluster around the value @xmath120 . finally , let us mention that a similar behavior of ba roots has recently been observed in the solution of the xxz heisenberg chain , where the boundary twist acts as the varying parameter @xcite . let us now turn to the results for the scaling functions @xmath140 ( [ e : f0 ] ) and @xmath141 ( [ e : fj ] ) . for the latter , all zero - momentum states with a critical scaling dimension of @xmath94 are considered , except the one missing excitation with conformal dimensions @xmath95 . in all cases , we use data from systems of size @xmath142 , @xmath143 and @xmath114 . for these sizes and the values of @xmath19 we considered , it turns out that the corrections to scaling are so small that one can hardly recognize them in our figures . in fig . [ f : gsf ] , the ( natural ) logarithm of @xmath140 ( which as defined in eq . ( [ e : f0 ] ) is positive ) is shown as a function of @xmath129 ( [ e : mu ] ) . apart from the behavior for small @xmath19 , the plot is nicely linear , showing the exponential decrease of @xmath144 as a function of @xmath145 down to about @xmath146 where the difference between the eigenvalues for the finite sizes and the bulk limit value becomes smaller than our numerical precision , which also proves the performance of our numerical routines . small deviations from scaling can be seen for larger values of @xmath19 . the scaling functions for the excitations @xmath141 ( [ e : fj ] ) are displayed in fig . [ f : sf ] . the curves shown are piecewise linear plots connecting data points obtained from systems with @xmath114 . for convenience , we again used @xmath129 ( [ e : mu ] ) on the horizontal axis . qualitatively , the scaling functions agree with the results of previous numerical calculations for the ising model in a magnetic field @xcite and with the results of the truncated fermionic space approach , compare the figures given in ref . the one - particle states can be recognized by their characteristic minima @xcite . this becomes clearer when we consider their ratios @xmath147 ( [ e : rj ] ) which are presented in fig . [ f : ratios ] . here , we again show individual data points which were obtained from systems of size @xmath142 , @xmath143 , and @xmath114 , which obviously were large enough to keep corrections to scaling small . the ratios @xmath147 , given here against the scaling variable @xmath74 , are labeled by the corresponding conformal dimensions at criticality . we also indicated the single - particle mass ratios ( [ e : masses ] ) of the @xmath5 field theory . clearly , the agreement is convincing , more detail on the approach of the massive scaling limit is contained in table [ t : rat ] . the spectrum of the dilute @xmath0 model has been studied by numerical solution of the bethe ansatz equations , both at and off - criticality . this gives the correspondence between the lowest states of the conformal spectrum of the critical ising model and those of the massive @xmath5 field theory ( note that the connection given in ref . @xcite contains an obvious mistake ) . at criticality , the spectrum consists for one part of massless excitations described by @xmath2 minimal cft as in the critical ising model , but in addition to these it contains massive excitations . obviously , the masses of these excitations are linked to the appearance of particular strings in the bethe ansatz solutions , which can be interpreted as massive particles . in our numerical analysis , we have seen at least five such `` massive '' strings . it is intriguing that all these strings ( showing up in solutions _ at criticality _ ) are among the conjectured list of `` thermodynamically significant string types '' in the tba analysis @xcite of the massive scaling limit . this is strongly suggestive of a connection between the lie algebra @xmath5 and the masses of these excitations . however , these are not the states which yield zamolodchikov s @xmath5 field theory of the ising model in a magnetic field . those correspond to massless excitations at the critical point , which develop a mass due to the existence of the symmetry - breaking field . at criticality , the string structure of these states does not agree with the predictions of @xcite to solve this puzzle had been the original motivation for this work . our numerical results suggest a scenario that may solve this apparent contradiction : the string type of the relevant excitations undergoes a number of changes as the field is switched on , and for large systems these reorganizations take place at particular values of the scaling parameter @xmath74 . therefore , the string structure entering the massive scaling limit @xmath72 is that observed for large values of the elliptic nome @xmath19 . numerically , we have been able to identify the single particle states up to the fifth mass , and a number of two - particle states . clearly , our results for the scaling functions are in complete agreement with both the analytic values of the mass ratios and with earlier numerical work on the ising model in a magnetic field . as to the bethe ansatz solutions , apart from one exception , the string structure for large field is that conjectured in @xcite . the one exception concerns the string type associated to the particle of mass @xmath8 . we found two states which contain this particle one being the single particle state with conformal dimensions @xmath148 , the other a two - particle state which also contains the lightest particle and has scaling dimensions @xmath149 at criticality . in both cases , the observed string structures agree , but they differ from the one proposed in @xcite , compare table [ t : ss ] . of course , it can not be ruled out completely that another change of string type appears at a larger value of the scaling parameter , but we found no trace of such a behavior for scaling parameters up to @xmath150 . this clearly demands further clarification , and maybe the investigation of the dilute @xmath151 model can lead the way . in the scaling limit , this model is described by an @xmath152 theory of factorized scattering , but has so far eluded a tba approach analogous to that of @xcite . in this case , we have been able to identify _ all _ seven single - particle states and observed interesting string solutions , details will be published soon @xcite . a number of interesting questions arise in connection with the massive excitations of the critical dilute @xmath0 model . this is one property of the dilute @xmath0 model which distinguishes it from the proper ising model in a magnetic field , since the critical ising model does not show such excitations . moreover , we suppose that this phenomenon is not particular to the specific model and will show up in the other dilute models as well . it would be interesting to understand the physical nature of these excitations , and to obtain analytic predictions for the observed mass ratios . also , their dispersion relations have not been studied because we concentrated on momentum zero states throughout this work . u.g . gratefully acknowledges financial support of the samenwerkingsverband fom / smc mathematische fysica during his stay in amsterdam , where most of this work was done . the authors thank m. baake , b. m. , p. a. pearce , k. schoutens and s. o. warnaar for interesting discussions and helpful comments . p. ginsparg , _ applied conformal field theory _ , in : fields , strings and critical phenomena , les houches , session xlix , 1988 , edited by e. brzin and j. zinn - justin , ( elsevier , amsterdam , 1990 ) , pp . j. l. cardy , _ conformal invariance and statistical mechanics _ , in : fields , strings and critical phenomena , les houches , session xlix , 1988 , edited by e. brzin and j. zinn - justin , ( elsevier , amsterdam , 1990 ) , pp . 169245 . a. b. zamolodchikov , _ integrable field theory from conformal field theory _ , in : integrable systems in quantum field theory and statistical mechanics , advanced studies in pure mathematics 19 , edited by m. jimbo , t. miwa and a. tsuchiya , ( kinokuniya , tokyo , 1989 ) , pp . 641647 . a. b. zamolodchikov and al . b. zamolodchikov , _ factorized @xmath4-matrices in two dimensions as the exact solutions of certain relativistic quantum field theory models _ , ( n.y . ) * 120 * , 253291 ( 1979 ) . e. ogievetsky , n. reshetikhin and p. wiegmann , _ the principal chiral field in two dimensions on classical lie algebras : the bethe - ansatz solution and factorized theory of scattering _ , nucl . phys . * b280 * , 4596 ( 1987 ) . + @xmath159 & 1 & 2 & 1 & ( -1,1 ) + @xmath160 & 7 & 4 & 0 & ( -3,-1,1,3 ) + @xmath161 & 2 & 4 & 0 & ( -4,-2,2,4 ) + @xmath8 & 8 & 5 & 1 & ( -12,-8,0,8,12 ) + @xmath162 & 3 & 6 & 1 & ( -7,-5,-1,1,5,7 ) + @xmath163 & 6 & 7 & 1 & ( -14,-6,-2,0,2,6,14 ) + @xmath164 & 4 & 8 & 0 & ( -10,-8,-4,-2,2,4,8,10 ) + @xmath165 & 5 & 10 & 1 & ( -13,-11,-7,-5,-1,1,5,7,11,13 ) 10&0.499@xmath166681&0.125@xmath166080&1.014@xmath166498&2.196&3.148&4.272&4.417 & 5.449&6.608&6.771&6.804&7.942 + 20&0.499@xmath166920&0.125@xmath166020&1.003@xmath166543&2.142&3.034&4.062&4.191 & 5.098&6.142&6.274&6.274&7.196 + 50&0.499@xmath166987&0.125@xmath166003&1.000@xmath166563&2.128&3.005&4.010&4.135 & 5.015&6.022&6.148&6.148&7.030 + 75&0.499@xmath166994&0.125@xmath166001&1.000@xmath166250&2.126&3.002&4.004&4.130 & 5.007&6.010&6.135&6.135&7.013 + 100&0.499@xmath166997&0.125@xmath166001&1.000@xmath166141&2.126&3.001&4.002&4.128 & 5.004&6.005&6.131 & 6.131&7.007 @xmath181&@xmath188 & @xmath181&@xmath189 & @xmath181&@xmath190 & @xmath181&@xmath191 & @xmath181&@xmath192 & @xmath181&@xmath193 + & @xmath181&@xmath194 & @xmath174&@xmath195 & @xmath174&@xmath196 & @xmath174&@xmath197 & @xmath174&@xmath198 + & @xmath199&@xmath200 & @xmath199&@xmath201 & @xmath199&@xmath202 & @xmath199&@xmath203 & @xmath199&@xmath204 + & & @xmath181&@xmath205 & @xmath181&@xmath206 & @xmath174&@xmath207 & @xmath174&@xmath208 + & & @xmath199&@xmath209 & @xmath199&@xmath210 & @xmath199&@xmath211 & @xmath199&@xmath212 + & & & @xmath181&@xmath213 & @xmath181&@xmath214 & @xmath174&@xmath215 + & & & @xmath199&@xmath216 & @xmath199&@xmath217 & @xmath199&@xmath218 + & & & & @xmath181&@xmath219 & @xmath181&@xmath220 + & & & & @xmath199&@xmath221 & @xmath199&@xmath222 + & & & & & @xmath181&@xmath223 + & & & & & @xmath199&@xmath224 @xmath232&@xmath227&@xmath233 & @xmath159 & @xmath234&@xmath235 & & @xmath231&@xmath236&@xmath237&@xmath238&@xmath239&@xmath236&@xmath240 & @xmath241 & & @xmath242 + & & & & & & & & & & & & & & & & @xmath243&&@xmath244 @xmath245&@xmath227&@xmath246 & @xmath160 & @xmath247&@xmath238&@xmath103&@xmath236&@xmath248 & @xmath231&@xmath236&@xmath249&@xmath238&@xmath108&@xmath236&@xmath250 & @xmath251 & & @xmath252 + & & & & @xmath253&@xmath235 & & @xmath231&@xmath236&@xmath249&@xmath238&@xmath254&@xmath236&@xmath255 & & & @xmath256&@xmath227&@xmath257 & @xmath161 & @xmath258&@xmath235 & & @xmath231&@xmath236&@xmath237&@xmath238&@xmath259&@xmath236&@xmath260 & @xmath261&&@xmath262 + & & & & @xmath263&@xmath238&@xmath239&@xmath236&@xmath264 & @xmath231&@xmath236&@xmath237&@xmath238&@xmath265&@xmath236&@xmath266 & @xmath267&&@xmath252 @xmath268&@xmath227&@xmath269 & @xmath8 & @xmath270&@xmath238&@xmath271&@xmath236&@xmath272 & @xmath231&@xmath236&@xmath273&@xmath238&@xmath231&@xmath236&@xmath274 & @xmath275&&@xmath276 + & & & & @xmath277&@xmath238&@xmath278&@xmath236&@xmath279 & @xmath231&@xmath236&@xmath280&@xmath238&@xmath103&@xmath236&@xmath281 & @xmath282&&@xmath283 + & & & & @xmath284&@xmath235 & & @xmath231&@xmath236&@xmath280&@xmath238&@xmath259&@xmath236&@xmath285 & & & + & & & & & & & & & @xmath231&@xmath236&@xmath237&@xmath235 & & & & @xmath286&@xmath227&@xmath287 & @xmath288 & @xmath289&@xmath238&@xmath199&@xmath236&@xmath290 & @xmath231&@xmath236&@xmath291&@xmath238&@xmath239&@xmath236&@xmath240 & & & + & & & & @xmath292&@xmath238&@xmath199&@xmath236&@xmath293 & @xmath231&@xmath236&@xmath294&@xmath238&@xmath239&@xmath236&@xmath240 & & & @xmath295&@xmath227&@xmath296 & @xmath297 & @xmath298&@xmath238&@xmath199&@xmath236&@xmath299 & @xmath231&@xmath236&@xmath237&@xmath238&@xmath239&@xmath236&@xmath240 & @xmath300&&@xmath301 + & & & & @xmath302&@xmath238&@xmath199&@xmath236&@xmath303 & @xmath231&@xmath236&@xmath304&@xmath238&@xmath254&@xmath236&@xmath305 & @xmath306&&@xmath307 + & & & & @xmath308&@xmath235 & & @xmath231&@xmath236&@xmath304&@xmath238&@xmath108&@xmath236&@xmath309 & & & @xmath310&@xmath227&@xmath311 & @xmath162 & @xmath312&@xmath238&@xmath259&@xmath236&@xmath313 & @xmath231&@xmath236&@xmath249&@xmath238&@xmath314&@xmath236&@xmath315 & @xmath316&&@xmath317 + & & & & @xmath318&@xmath238&@xmath314&@xmath236&@xmath319 & @xmath231&@xmath236&@xmath249&@xmath238&@xmath239&@xmath236&@xmath240 & @xmath320&&@xmath321 + & & & & @xmath322&@xmath235 & & @xmath231&@xmath236&@xmath249&@xmath238&@xmath323&@xmath236&@xmath324 & & & @xmath325&@xmath227&@xmath326 & @xmath288 & @xmath327&@xmath238&@xmath239&@xmath236&@xmath328 & @xmath231&@xmath236&@xmath329&@xmath238&@xmath239&@xmath236&@xmath330 & & & + & & & & @xmath331&@xmath238&@xmath239&@xmath236&@xmath332 & @xmath231&@xmath236&@xmath333&@xmath238&@xmath239&@xmath236&@xmath330 & & & @xmath334&@xmath227&@xmath335 & @xmath336 & @xmath337&@xmath238&@xmath239&@xmath236&@xmath338 & @xmath231&@xmath236&@xmath249&@xmath238&@xmath239&@xmath236&@xmath330 & @xmath339&&@xmath261 + & & & & @xmath340&@xmath238&@xmath239&@xmath236&@xmath341 & @xmath231&@xmath236&@xmath237&@xmath238&@xmath239&@xmath236&@xmath342 & @xmath343&&@xmath344 + & & & & @xmath345&@xmath235 & & @xmath231&@xmath236&@xmath237&@xmath238&@xmath199&@xmath236&@xmath346 & @xmath347&&@xmath348 @xmath334&@xmath227&@xmath349 & @xmath297 & @xmath350&@xmath238&@xmath314&@xmath236&@xmath351 & @xmath231&@xmath236&@xmath352&@xmath238&@xmath239&@xmath236&@xmath330 & @xmath353&&@xmath354 + & & & & @xmath355&@xmath235 & & @xmath231&@xmath236&@xmath356&@xmath238&@xmath278&@xmath236&@xmath357 & & & + & & & & @xmath340&@xmath238&@xmath239&@xmath236&@xmath341 & @xmath231&@xmath236&@xmath356&@xmath238&@xmath271&@xmath236&@xmath358 & & & + & & & & @xmath345&@xmath235 & & & & & & & & & & & @xmath334&@xmath227&@xmath359 & @xmath297 & @xmath337&@xmath238&@xmath239&@xmath236&@xmath338 & @xmath231&@xmath236&@xmath360&@xmath238&@xmath278&@xmath236&@xmath357 & & & + & & & & @xmath361&@xmath238&@xmath278&@xmath236&@xmath362 & @xmath231&@xmath236&@xmath360&@xmath238&@xmath271&@xmath236&@xmath358 & & & + & & & & @xmath363&@xmath238&@xmath271&@xmath236&@xmath364 & @xmath231&@xmath236&@xmath365&@xmath238&@xmath239&@xmath236&@xmath330 & & & @xmath366&@xmath227&@xmath367 & @xmath368 & @xmath369&@xmath238&@xmath370&@xmath236&@xmath371 & @xmath231&@xmath236&@xmath372&@xmath238&@xmath231&@xmath236&@xmath373 & @xmath374&&@xmath375 + & & & & @xmath376&@xmath235 & & @xmath231&@xmath236&@xmath280&@xmath238&@xmath103&@xmath236&@xmath377 & & & + & & & & @xmath378&@xmath238&@xmath239&@xmath236&@xmath379 & @xmath231&@xmath236&@xmath280&@xmath238&@xmath259&@xmath236&@xmath380 & & & + & & & & @xmath381&@xmath238&@xmath239&@xmath236&@xmath382 & @xmath231&@xmath236&@xmath237&@xmath238&@xmath239&@xmath236&@xmath330 & & & + & & & & & & & & & @xmath231&@xmath236&@xmath237&@xmath235 & & & & & 10 & 1.618@xmath166749 & 1.983@xmath166943 & 2.403@xmath166558 & 2.083@xmath166461 & 2.616@xmath166910 & 2.910@xmath166107 & 2.317@xmath166908 & 3.123@xmath166561 & 2.811@xmath166391 & 3.360@xmath166590 + & 20 & 1.618@xmath166037 & 1.983@xmath166923 & 2.403@xmath166649 & 2.081@xmath166017 & 2.617@xmath166335 & 2.930@xmath166942 & 2.320@xmath166478 & 3.086@xmath166893 & 2.781@xmath166193 & 3.369@xmath166241 + 10 & 50 & 1.618@xmath166024 & 1.983@xmath166977 & 2.403@xmath166797 & 2.079@xmath166804 & 2.617@xmath166468 & 2.934@xmath166096 & 2.314@xmath166580 & 3.078@xmath166762 & 2.773@xmath166762 & 3.375@xmath166544 + & 75 & 1.618@xmath166024 & 1.983@xmath166984 & 2.403@xmath166815 & 2.079@xmath166550 & 2.617@xmath166482 & 2.934@xmath166583 & 2.313@xmath166631 & 3.077@xmath166950 & 2.772@xmath166773 & 3.376@xmath166215 + & 100 & 1.618@xmath166025 & 1.983@xmath166990 & 2.403@xmath166834 & 2.079@xmath166551 & 2.617@xmath166497 & 2.934@xmath166617 & 2.313@xmath166523 & 3.077@xmath166537 & 2.772@xmath166605 & 3.376@xmath166509 + & 10 & 1.623@xmath166692 & 1.989@xmath166446 & 2.412@xmath166441 & 2.017@xmath166536 & 2.623@xmath166691 & 2.958@xmath166322 & 2.085@xmath166027 & 3.006@xmath166016 & 2.671@xmath166041 & 3.411@xmath166712 + & 20 & 1.618@xmath166144 & 1.987@xmath166688 & 2.405@xmath166009 & 2.025@xmath166316 & 2.618@xmath166144 & 2.954@xmath166218 & 2.129@xmath166306 & 3.006@xmath166016 & 2.669@xmath166638 & 3.404@xmath166917 + 20 & 50 & 1.618@xmath166034 & 1.987@xmath166597 & 2.404@xmath166866 & 2.026@xmath166212 & 2.618@xmath166034 & 2.954@xmath166289 & 2.132@xmath166554 & 3.003@xmath166342 & 2.669@xmath166034 & 3.404@xmath166805 + & 75 & 1.618@xmath166034 & 1.987@xmath166591 & 2.404@xmath166865 & 2.026@xmath166286 & 2.618@xmath166034 & 2.954@xmath166306 & 2.132@xmath166780 & 3.003@xmath166242 & 2.668@xmath166960 & 3.404@xmath166808 + & 100 & 1.618@xmath166034 & 1.987@xmath166589 & 2.404@xmath166865 & 2.026@xmath166315 & 2.618@xmath166034 & 2.954@xmath166311 & 2.132@xmath166871 & 3.003@xmath166212 & 2.668@xmath166941 & 3.404@xmath166808 + & 10 & 1.634@xmath166823 & 1.992@xmath166979 & 2.428@xmath166402 & 2.003@xmath166678 & 2.634@xmath166823 & 2.970@xmath166263 & 2.021@xmath166663 & 2.994@xmath166370 & 2.649@xmath166653 & 3.428@xmath166399 + & 20 & 1.618@xmath166424 & 1.988@xmath166671 & 2.405@xmath166376 & 2.012@xmath166146 & 2.618@xmath166424 & 2.956@xmath166329 & 2.071@xmath166838 & 2.992@xmath166915 & 2.644@xmath166399 & 3.405@xmath166376 + 30 & 50 & 1.618@xmath166036 & 1.988@xmath166497 & 2.404@xmath166869 & 2.013@xmath166483 & 2.618@xmath166036 & 2.956@xmath166009 & 2.078@xmath166248 & 2.992@xmath166952 & 2.645@xmath166130 & 3.404@xmath166869 + & 75 & 1.618@xmath166034 & 1.988@xmath166489 & 2.404@xmath166867 & 2.013@xmath166604 & 2.618@xmath166034 & 2.956@xmath166007 & 2.078@xmath166742 & 2.992@xmath166957 & 2.645@xmath166220 & 3.404@xmath166867 + & 100 & 1.618@xmath166034 & 1.988@xmath166486 & 2.404@xmath166867 & 2.013@xmath166644 & 2.618@xmath166034 & 2.956@xmath166007 & 2.078@xmath166950 & 2.992@xmath166958 & 2.645@xmath166248 & 3.404@xmath166867 + @xmath226{0ex}{4ex}\infty$ ] & @xmath383 & 1.618@xmath166034 & 1.989@xmath166044 & 2.404@xmath166867 & 2.000@xmath166000 & 2.618@xmath166034 & 2.956@xmath166295 & 2.000@xmath166000 & 2.989@xmath166044 & 2.618@xmath166034 & 3.404@xmath166867 +

in this paper , we propose a systematic design of space - time block codes ( stbc ) which can achieve high rate and full diversity when the partial interference cancellation ( pic ) group decoding is used at receivers . the proposed codes can be applied to any number of transmit antennas and admit a low decoding complexity while achieving full diversity . for transmit antennas , in each codeword real and imaginary parts of complex information symbols are parsed into diagonal layers and then encoded , respectively . with pic group decoding , it is shown that the decoding complexity can be reduced to a joint decoding of real symbols . in particular , for transmit antennas , the code has real symbol pairwise ( i.e. , single complex symbol ) decoding that achieves full diversity and the code rate is . simulation results demonstrate that the full diversity is offered by the newly proposed stbc with the pic group decoding . 4.6ex mimo systems , space - time block codes , partial interference cancellation , decoding complexity

1004.2773

full diversity and low decoding complexity have been considered as two fundamental properties which a good space - time block code ( stbc ) should possess for multiple - input multiple - output ( mimo ) wireless communications . the first orthogonal stbc ( ostbc ) was proposed by alamouti which can achieve full transmit diversity for two transmit antennas@xcite . inspired by the alamouti scheme , seminal studies focused on the designs of ostbc for its unique orthogonal code structure which ensures a single symbol maximum likelihood ( ml ) decoding @xcite@xcite@xcite . however , ostbc suffers from the reduced symbol rate with an increase of the number of transmit antennas , especially when complex constellations are used @xcite . in spite of full diversity advantage , ostbc fails to achieve full channel capacity in mimo channels @xcite . to address the problem of low symbol rate and capacity loss in ostbc , linear dispersion code ( ldc ) was proposed as a full - diversity scheme that is constructed linearly in space and time @xcite @xcite . the ldc design can be viewed as a linear combination of a fixed set of dispersion matrices with the transmitted symbols ( or equivalently , combining coefficients ) . diagonal algebraic space - time ( dast ) block codes in @xcite and threaded algebraic space - time ( tast ) codes in @xcite were also proposed as two typical algebraic designs which can obtain both full diversity and full rate with moderate ml decoding complexity . however , it is noted that the aforementioned high - rate codes rely on ml decoding to collect full diversity which has high decoding complexity . efficient designs of stbc with low decoding complexity were proposed , such as coordinate interleaved orthogonal design ( ciod ) with single - symbol ml decoding in @xcite and quasi - orthogonal stbc ( qostbc ) in @xcite@xcite , but their rates are restricted by the rates of ostbc . recently , full diversity achieving stbc based on linear receivers , such as the minimum mean square error ( mmse ) receiver and zero - forcing ( zf ) receiver , were studied and proposed @xcite , @xcite . however , it was shown in @xcite that the rates of these stbc based on linear receivers are not more than one . to address the complexity and rate tradeoff , a general decoding scheme with code design criterion , referred to as partial interference cancellation ( pic ) group decoding algorithm , was proposed in @xcite . in the pic group decoding , the symbols to be decoded are divided into several groups after a linear pic operation and then each group is decoded separately with ml decoding . therefore , the pic group decoding can be viewed as an intermediate decoding between ml decoding and zf decoding . apparently , pic group decoding complexity depends on the number of symbols to be decoded in each group . moreover , a successive interference cancellation ( sic)-aided pic group decoding was proposed in @xcite . based on the design criterion of stbc with pic group decoding derived in @xcite , a systematic design of stbc achieving full diversity under pic group decoding was developed in @xcite . in subsequent work , a new design of stbc having an alamouti - toeplitz structure was proposed in @xcite which provides a lower pic decoding complexity compared with the design in @xcite . however , the decoding complexity of the stbc in @xcite is equivalent to a joint decoding of @xmath3 complex symbols . in this paper , we propose a design of stbc with pic group decoding that can achieve both full diversity and low decoding complexity . the decoding complexity is equal to a joint decoding of @xmath3 _ real _ symbols for @xmath0 transmit antennas , i.e. , only half decoding complexity of the stbc in @xcite . for the proposed stbc , real and imaginary parts of @xmath1 complex information symbols are parsed into @xmath2 diagonal layers and encoded by linear transform matrices , respectively . the full diversity can be achieved by the proposed stbc with @xmath6 under pic group decoding and with any @xmath2 under pic - sic group decoding , respectively . the code rate is equal to @xmath7 . in particular , for @xmath4 transmit antennas the code has real symbol pairwise ( i.e. single complex symbol ) decoding . furthermore , the code rate is @xmath5 . it should be noted that the existing stbc with single complex symbol ( or real symbol pairwise ) decoding , such as qostbc @xcite , @xcite , and ciod etc . in @xcite have symbol rates not larger than one . also the codes with linear receivers have single complex symbol decoding but their rates can not be above one either @xcite . simulation results show that the proposed code outperforms the ciod in @xcite and the qostbc with the optimal rotation in @xcite for @xmath4 transmit antennas at the same bandwidth efficiency . moreover , our code guarantees full diversity without performance loss compared with other pic group decoding based stbc in @xcite , @xcite and @xcite , but a half decoding complexity is reduced . it should be mentioned that the major difference between the code in @xcite and the one proposed in this paper is that a complex - valued linear transform matrix is used for input complex signal vector to construct the code in @xcite , whereas in this paper two real - valued linear transform matrices are used for real and imaginary parts of the signals , respectively . by doing so , half decoding complexity can be reduced . the rest of this paper is organized as follows . the system model is outlined in section ii . in section iii , a systematic design of stbc is proposed and a few code design examples are also given . the full diversity is proved under pic group decoding in section iv . in section v , simulation results are presented . finally , we conclude the paper in section vi . the following notations are used throughout this paper . column vectors ( matrices ) are denoted by boldface lower ( upper ) case letters . superscripts @xmath8 , @xmath9 and @xmath10 stand for conjugate , transpose , and conjugate transpose , respectively . @xmath11 denotes the field of complex numbers and @xmath12 denotes the real field . @xmath13 denotes the @xmath14 identity matrix , and @xmath15 denotes the @xmath16 matrix whose elements are all @xmath17 . additionally , @xmath18 and @xmath19 represent the real part and the imaginary part of variables , respectively . consider a mimo system with @xmath0 transmit and @xmath20 receive antennas . data symbols are first encoded into a space - time block code @xmath21 of size @xmath22 where @xmath23 is block length of the codeword . in this paper , @xmath21 can be represented in a general dispersion form @xcite as follows : @xmath24 where the data symbols @xmath25 are selected from a normalized complex constellation @xmath26 such as qam , and @xmath27 are constant matrices called dispersion matrices . then , the received signal from @xmath20 receive antennas can be arranged in a matrix @xmath28 as follows @xmath29 where @xmath30 is the channel matrix of size @xmath31 with the entries being independent and identically distributed ( i.i.d ) @xmath32 . the channels are assumed to experience the quasi - static fading . @xmath33 is the noise matrix whose elements are also i.i.d distributed @xmath32 . @xmath34 denotes the average signal - to - noise ratio ( snr ) per receive antenna , and the transmitted power is normalized by the factor @xmath35 such that the average energy of the coded symbols transmitting from all antennas during one symbol period is one . we suppose that channel state information is available at receiver only . to decode the transmitted sequence @xmath36 , we need to extract @xmath36 from @xmath37 . through some operations , we can get an equivalent signal model from ( [ ry ] ) as @xcite@xcite @xmath38 where @xmath39 is a received signal vector , @xmath40 is a noise vector , and @xmath41 is an equivalent channel matrix . denote @xmath42 , @xmath43 , and @xmath44 . then , we can rewrite ( [ eqyc ] ) as a real matrix form given by @xmath45 = \sqrt{\frac{\rho}{\mu}}\mathcal{h } \left[\begin{array}{c } \mathbf{s}_r\\ \mathbf{s}_i \end{array}\right ] + \left[\begin{array}{c } \mathbf{w}_r\\ \mathbf{w}_i \end{array}\right ] , \end{aligned}\ ] ] where @xmath46 has @xmath47 real column vectors @xmath48 for @xmath49 . in @xcite , a new decoding scheme was proposed , referred to as pic group decoding which aims to address the rate and complexity tradeoff of the code while achieving full diversity . in the pic group decoding , the equivalent channel matrix @xmath41 is divided into a number of column groups @xmath50 with @xmath51 columns for group @xmath52 , @xmath53 , and @xmath54 . then , for group @xmath52 a group zf is applied to cancel the interferences coming from all the other groups , i.e. , @xmath55 , followed by a joint decoding of symbols corresponding to the group @xmath52 . note that the interference cancellation ( i.e. , the group zf ) mainly involves with linear matrix computations , whose computational complexity is small compared to the joint decoding with an exhaustive search of all candidate symbols in one group . to evaluate the decoding complexity of the pic group decoding , we mainly focus on the computational complexity of the joint decoding of each group under the pic group decoding algorithm . the joint decoding complexity can be characterized by the number of frobenius norms calculated in the decoding process . in the pic group decoding algorithm , the complexity is then @xmath56 . it can be seen that the pic group decoding provides a flexible decoding complexity which can vary from the zf decoding complexity @xmath57 to the ml decoding complexity @xmath58 . an sic - aided pic group decoding algorithm , namely pic - sic group decoding was also proposed in @xcite . similar to the blast detection algorithm @xcite , the pic - sic group decoding is performed after removing the already - decoded symbol set from the received signals to reduce the interference . if each group has only one symbol , then the pic - sic group decoding will be equivalent to the blast detection . in @xcite , full - diversity stbc design criteria were derived when the pic group decoding and the pic - sic group decoding are used at the receiver . in the following , we cite the main results of the stbc design criteria proposed in @xcite . [ prop1 ] ( * ? ? ? * theorem 1 ) [ _ full - diversity criterion under pic group decoding _ ] for an stbc @xmath59 with the pic group decoding , the full diversity is achieved when 1 . the code @xmath59 satisfies the full rank criterion , i.e. , it achieves full diversity when the ml receiver is used ; _ and _ 2 . for any @xmath60 , @xmath61 , any non - zero linear combination over @xmath62 of the vectors in the @xmath60th group @xmath52 does not belong to the space linearly spanned by all the vectors in the remaining vector groups , as long as @xmath63 , i.e. , @xmath64 where @xmath65 is the index set corresponding to the vector group @xmath66 and @xmath67 . [ prop2 ] @xcite [ _ full - diversity criterion under pic - sic group decoding _ ] for an stbc @xmath59 with the pic - sic group decoding , the full diversity is achieved when 1 . the code @xmath59 satisfies the full rank criterion , i.e. , it achieves full diversity when the ml receiver is used ; _ and _ 2 . at each decoding stage , for @xmath68 , which corresponds to the current to - be decoded symbol group @xmath69 , any non - zero linear combination over @xmath62 of the vectors in @xmath68 does not belong to the space linearly spanned by all the vectors in the remaining groups @xmath70 corresponding to yet uncoded symbol groups , as long as @xmath63 . in this section , a systematic design of linear dispersion stbc is presented and two design examples are given for four and six transmit antennas , respectively . suppose that @xmath0 is even . our proposed stbc @xmath71 is of size @xmath22 , and given by @xmath72 where the codeword matrices for @xmath73 and @xmath74 are given by @xmath75 , ~\mathbf{b}_{m , t , p}\label{b } = \left[\begin{array}{cc } \mathbf{c}^1_i & \mathbf{c}^2_i \\ \mathbf{c}^2_i & -\mathbf{c}^1_i \end{array}\right].\end{aligned}\ ] ] note that @xmath73 and @xmath74 are both real matrices of size @xmath22 ( @xmath76 ) . @xmath77 and @xmath78 are real and imaginary parts of @xmath79 @xmath80 which is given by @xmath81,\end{aligned}\ ] ] with @xmath82 and the @xmath60 th diagonal layer from left to right written as the @xmath83 vector @xmath84 , given by @xmath85^t.\end{aligned}\ ] ] moreover , the real and imaginary parts of @xmath86 are given by respectively @xmath87 where @xmath88 can be different linear transform matrices chosen from @xcite and @xcite , and the @xmath83 vector @xmath89 are given by @xmath90^t.\end{aligned}\ ] ] the rate of the code @xmath91 is @xmath92 which is the same as that of stbcs with pic group decoding proposed in @xcite and @xcite . it should be mentioned that the code structure ( [ new ] ) is similar to the one in @xcite . the main difference is that in @xcite a linear transform matrix is used for input complex symbol vectors in the code construction and the matrix does not have to be real - valued , whereas in the design of @xmath91 in ( [ new ] ) , two real linear transform matrices @xmath93 and @xmath94 are used . later , we will see that the proposed code in ( [ new ] ) with pic group decoding of real - valued signals yields lower decoding complexity than the one in @xcite . \1 ) for four transmit antennas @xmath95 consider the case with @xmath95 transmit antennas . according to the design in ( [ new ] ) , we have @xmath96 where @xmath97 , \end{aligned}\ ] ] with @xmath98^t = \mathbf{\theta}_{a } \left[\begin{array}{cc } s_{\{2(i-1)+1\},r } & s_{\{2(i-1)+2\},r } \end{array}\right]^t$ ] for @xmath99 , and @xmath100 \end{aligned}\ ] ] with @xmath101^t = \mathbf{\theta}_{b } \left[\begin{array}{cc } s_{\{2(i-1)+1\},i } & s_{\{2(i-1)+2\},i } \end{array}\right]^t$ ] for @xmath99 . for simplicity , the same linear transform matrix @xmath102 is used for @xmath103 and @xmath104 as @xmath105,\end{aligned}\ ] ] with @xmath106 @xcite . then , the codeword matrix of @xmath107 is written as @xmath108 , \end{aligned}\ ] ] the rate of the code @xmath107 is @xmath5 and equal to that of @xmath109 in ( * ? ? ? * eq . ( 29 ) ) and @xmath110 in ( * ? ? ? * eq . ( 37 ) ) . \2 ) for six transmit antennas @xmath111 for given @xmath112 , the code @xmath113 with six transmit antennas is designed as follows @xmath114 where @xmath115 \end{aligned}\ ] ] with @xmath116 = \mathbf{\theta}_a \left[\begin{array}{c } s_{\{3(i-1)+1\},r}\\ s_{\{3(i-1)+2\},r}\\ s_{\{3(i-1)+3\},r } \end{array}\right],\end{aligned}\ ] ] and @xmath117 \end{aligned}\ ] ] with @xmath118 = \mathbf{\theta}_b \left[\begin{array}{c } s_{\{3(i-1)+1\},i}\\ s_{\{3(i-1)+2\},i}\\ s_{\{3(i-1)+3\},i } \end{array}\right],\end{aligned}\ ] ] for @xmath99 . the same linear transform matrix @xmath119 is used for @xmath103 and @xmath104 as @xcite @xmath120\end{aligned}\ ] ] then , the codeword @xmath113 can be written as @xmath121 . \end{split}\ ] ] the code rate for @xmath113 is @xmath122 . in this section , we prove that our proposed stbc can obtain full diversity under pic group decoding and have a lower decoding complexity compared with @xcite and @xcite . define @xmath123 as the difference between symbols @xmath124 and @xmath125 . following the proof of ( * ? ? ? * theorem 1 ) , three cases should be considered separately in terms of @xmath126 and @xmath127 as follows 1 . both @xmath128 and @xmath129 [ case1 ] + consider @xmath128 . after some row / column permutations , a different codeword matrix @xmath130 can be written as follows @xmath131\ ] ] where @xmath132 is a @xmath133 matrix and @xmath134,i=1,2 . \end{aligned}\ ] ] + from ( [ roxir ] ) and ( [ roxii ] ) , we deduce that there exists at least one vector @xmath135 such that @xmath136 , @xmath137 , because the signal space diversity is obtained from the linear transform matrix @xmath103 . then , we have that @xmath130 is full rank , which can be proved with a similar proof given in ( * ? ? ? * theorem 1 ) . hence , @xmath73 can guarantee full diversity with ml decoding . likewise , it is obvious that @xmath74 can also achieve full diversity since @xmath129 . + therefore , the code @xmath91 can achieve full diversity under ml decoding . @xmath128 only [ case2 ] + as we mentioned in case [ case1 ] ) , @xmath73 can achieve full diversity under ml decoding if @xmath128 . considering @xmath73 forms the real part in ( [ new ] ) , the code @xmath91 can achieve full diversity under ml decoding . @xmath138 only + similar to case [ case2 ] ) , @xmath74 being the imaginary part of our proposed code can achieve full diversity under ml decoding , which is sufficient to prove that @xmath91 has a property of full diversity . by observing all three cases , we conclude that the proposed code in ( [ new ] ) can achieve full diversity under ml decoding . compared with the pic grouping schemes derived in @xcite and @xcite , the separated linear transform of real and imaginary parts of the information symbols in the proposed code contributes to the real symbol decoding . in the following , we show the main result of the proposed stbc when a pic group decoding with a particular grouping scheme is used at the receiver , as follows . [ tpic ] consider a mimo system with @xmath0 transmit antennas and @xmath20 receive antennas over block fading channels . the stbc as describe in ( [ new ] ) with two diagonal layers in each submatrix is used at the transmitter . the real equivalent channel matrix is @xmath139 . if the received signal is decoded using the pic group decoding with the grouping scheme @xmath140 , where @xmath141 for @xmath142 , i.e. , the size of each real group is equal to @xmath3 , then the code @xmath143 achieves the full diversity . [ m4co ] for the proposed code with @xmath95 transmit antennas in ( [ m4 ] ) , real symbol pairwise ml decoding is achieved in each group , which is equivalent to single complex symbol ml decoding . table i shows the comparison of pic group decoding complexity between the new code in ( [ m4 ] ) and the codes in @xcite and @xcite . according to this table , it is obvious that the proposed code for @xmath95 transmit antennas further reduce the decoding complexity to real symbol pairwise ( i.e. , single complex symbol ) decoding in each pic group . in order to prove _ theorem [ tpic ] _ , let us first introduce the following definition and lemma . [ orde ] let @xmath144 be @xmath145 groups of vectors . vector groups @xmath144 are said to be orthogonal if for @xmath146 , @xmath147 is orthogonal to the remaining vector groups @xmath148 . [ leq ] consider the system described in _ theorem [ tpic ] _ with @xmath149 as follows @xmath150= \sqrt{\frac{\rho}{\mu}}\mathcal{h } \left[\begin{array}{c } \mathbf{s}^1_{1,r}\\ \mathbf{s}^1_{2,r}\\ \mathbf{s}^2_{1,r } \\ \mathbf{s}^2_{2,r}\\ \mathbf{s}^1_{1,i}\\ \mathbf{s}^1_{2,i}\\ \mathbf{s}^2_{1,i}\\ \mathbf{s}^2_{2,i } \end{array } \right ] + \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2\\ \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right],\end{aligned}\ ] ] where the @xmath83 vector @xmath151 are given by ( [ eqn : sip ] ) for @xmath152 and @xmath153 . the equivalent channel matrix @xmath154 is expressed as @xmath155 \\ & = & \left[\begin{array}{cccccccc}\label{picl } \mathbf{g}_1 & \mathbf{g}_2 & \ldots & \mathbf{g}_8 \end{array } \right],\end{aligned}\ ] ] where @xmath156 , { \rm{and}~ } \mathcal{h}_{p , i}^i=\left[\begin{array}{c } \mathbf{0}_{(p-1)\times ( m/2)}\\ \mathrm{diag}(\mathbf{h}_i^{i})\\ \mathbf{0}_{(2-p)\times ( m/2 ) } \end{array } \right],\end{aligned}\ ] ] for @xmath153 and @xmath152 . the channel coefficient vector @xmath157 is evenly divided into two groups with @xmath158^t$ ] and @xmath159^t$ ] . @xmath160 is the channel gain from the @xmath161th transmit antenna to the single receive antenna for @xmath162 . a proof of _ lemma 1 _ is given in appendix . note that ( * ? ? ? * corollary 1 ) proves that the full diversity conditions only need to be proved for one receive antenna case . thus , we only consider the miso system model ( i.e. @xmath149 ) . first , after some column / row permutations , ( [ lemmaeq ] ) can be rewritten as @xmath163 & [ \mathbf{g}_5^ { ' } & \mathbf{g}_7^ { ' } ] & [ \mathbf{g}_2^ { ' } & \mathbf{g}_4^ { ' } ] & [ \mathbf{g}_6^ { ' } & \mathbf{g}_8^ { ' } ] \end{array } \right],\nonumber\\ & = & \left[\begin{array}{cccc } \mathcal{f}^{r}_1 & -\mathcal{f}^{i}_1 & \mathbf{0}_{2\times m } & \mathbf{0}_{2\times m } \\ \mathcal{f}^{i}_1 & \mathcal{f}^{r}_1 & \mathbf{0}_{2\times m } & \mathbf{0}_{2\times m } \\ \mathcal{f}^{r}_2 & -\mathcal{f}^{i}_2 & \mathcal{f}^{r}_1 & -\mathcal{f}^{i}_1 \\ \mathcal{f}^{i}_2 & \mathcal{f}^{r}_2 & \mathcal{f}^{i}_1 & \mathcal{f}^{r}_1\\ \vdots & \vdots & \mathcal{f}^{r}_2 & -\mathcal{f}^{i}_2\\ \vdots & \vdots & \mathcal{f}^{i}_2 & \mathcal{f}^{r}_2 \\ \mathcal{f}^{r}_{\frac{m}{2 } } & -\mathcal{f}^{i}_{\frac{m}{2 } } & \vdots & \vdots \\ \mathcal{f}^{i}_{\frac{m}{2 } } & \mathcal{f}^{r}_{\frac{m}{2 } } & \vdots & \vdots \\ \mathbf{0}_{2\times m } & \mathbf{0}_{2\times m } & \mathcal{f}^{r}_{\frac{m}{2 } } & -\mathcal{f}^{i}_{\frac{m}{2}}\\ \mathbf{0}_{2\times m } & \mathbf{0}_{2\times m } & \mathcal{f}^{i}_{\frac{m}{2 } } & \mathcal{f}^{r}_{\frac{m}{2 } } \end{array } \right],\end{aligned}\ ] ] where both @xmath164 and @xmath165 are @xmath166 real matrix given by @xmath167,\\ \mathcal{f}^i_j&=&\left[\begin{array}{cc}\label{fi } \mathbf{f}_{j , j_i } & \mathbf{f}_{j,{\{j+\frac{m}{2}\}}_i } \\ -\mathbf{f}_{j,{\{j+\frac{m}{2}\}}_i } & \mathbf{f}_{j , j_i } \end{array } \right],\end{aligned}\ ] ] for @xmath168 . @xmath169 is a @xmath170 real vector with @xmath171 being the @xmath172th row of the linear transform matrix @xmath173 for @xmath174 and with @xmath175 and @xmath176 being the real and imaginary part of @xmath160 for @xmath177 , respectively . it is worthwhile to mention that from ( [ lemmaeq ] ) , @xmath178 in @xmath179 $ ] and @xmath180 $ ] are related to @xmath93 , while @xmath181 in @xmath182 $ ] and @xmath183 $ ] are associated with @xmath94 . in appendix , it is shown that the orthogonality between each groups is irrelevant to the linear transform matrices @xmath93 and @xmath94 . therefore , for simplicity @xmath173 is used for both @xmath93 and @xmath94 . next , we prove that any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath185 . for any @xmath186 , i.e. , @xmath187 where @xmath188 is a column vector . for any nonzero @xmath189 , we have following three cases . a ) : : if @xmath190 and @xmath191 , then it must exist a minimum index @xmath161 ( @xmath192 ) such that @xmath178 is nonzero and a minimum index @xmath193 ( @xmath194 ) such that @xmath195 is nonzero . therefore , @xmath196 must be all zeros and @xmath197 must be all zeros , too . b ) : : if @xmath190 and @xmath198 , then it must exist a minimum index @xmath161 ( @xmath192 ) such that @xmath178 is nonzero . therefore , @xmath196 must be all zeros and @xmath199 must be all zeros , too . c ) : : @xmath191 and @xmath200 , then it must exist a minimum index @xmath193 ( @xmath194 ) such that @xmath195 is nonzero . therefore , @xmath197 must be all zeros and @xmath201 must be all zeros , too . next , we first focus on the case of a ) . the proof is presented in terms of @xmath161 and @xmath193 . a1 ) : : @xmath202 + in this case , ( [ eqhp1 ] ) can be expressed as @xmath203.\end{aligned}\ ] ] where @xmath204 . by observing the @xmath205th row to the @xmath206th row in ( [ eqhp2 ] ) , the vector groups @xmath207 are all zeros , and @xmath184 is orthogonal to the vector groups @xmath208 . thus , it is obvious that in these four rows , any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath207 . + furthermore , according to _ definition [ orde ] _ , the vector groups @xmath209 are orthogonal in these four rows . consequently , in these four rows , any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath185 . considering all rows in ( [ eqhp2 ] ) , any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath185 . a2 ) : : @xmath210 + in this case , ( [ eqhp1 ] ) can be expressed as @xmath211.\end{aligned}\ ] ] + it is seen that from the @xmath212th row to the @xmath213th row in ( [ eqhp3 ] ) the groups @xmath207 are all zeros . similarly , we have that in these four rows , any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath207 . additionally , the vector groups @xmath209 are orthogonal . similar to case a1 ) , we have that any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath214 . a3 ) : : @xmath215 + in this case , ( [ eqhp1 ] ) can be expressed as + @xmath216.\end{aligned}\ ] ] + as for this case , the vector groups @xmath207 are all zeros from the @xmath205th row to the @xmath206th row , and the vector groups @xmath209 are orthogonal in ( [ eqhp4 ] ) . similar to the proof for case a1 ) , we have that any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the remaining vectors in @xmath214 . to summarize all the cases a1)-a3 ) , we then conclude that for @xmath190 and @xmath191 any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath185 . if the case b ) occurs , i.e. , @xmath190 and @xmath198 , then ( [ eqhp1 ] ) can be written as a similar form to ( [ eqhp4 ] ) by replacing @xmath217 by @xmath132 for all @xmath193 . the proof is the same as that of case a3 ) . if the case c ) occurs , i.e. , @xmath191 and @xmath200 , then ( [ eqhp1 ] ) can be written as a similar form to ( [ eqhp3 ] ) by replacing @xmath218 by @xmath132 for all @xmath161 . the proof is the same as that of case a2 ) . therefore , we have proved that for any @xmath219 any non - zero linear combination of the vectors in @xmath184 over @xmath62 does not belong to the space linearly spanned by all the vectors in the vector groups @xmath185 . similarly , we can prove that any non - zero linear combination of the vectors in @xmath220 over @xmath62 does not belong to the space linearly spanned by all the vectors in the remaining vector groups , for @xmath221 . note that @xmath220 is a row permutation of @xmath52 for @xmath142 , respectively . we prove that any non - zero linear combination of the vectors in @xmath52 over @xmath62 does not belong to the space linearly spanned by all the vectors in the remaining vector groups , for @xmath142 . according to _ proposition 1 _ , the proof of _ theorem [ tpic ] _ is completed . in the preceding discussion , the new code @xmath91 in ( [ new ] ) is proved to achieve the full diversity under pic group decoding when @xmath6 only . in the following , we will further show that @xmath91 with any value @xmath2 can obtain full diversity under pic - sic group decoding @xcite . consider a mimo system with @xmath0 transmit antennas and @xmath20 receive antennas over block fading channels . the stbc as described in ( [ new ] ) with @xmath2 diagonal layers is used at the transmitter . the equivalent channel matrix is @xmath222 . if the received signal is decoded using the pic - sic group decoding with the grouping scheme @xmath223 and with the sequential order , where @xmath224 for @xmath225 , i.e. , the size of each real group is equal to @xmath3 , then the code @xmath226 achieves the full diversity . the code rate of the full - diversity stbc can be up to @xmath3 symbols per channel use . the proof is similar to that of _ theorem 1_. note that @xmath227 for the code @xmath91 in _ lemma 1 _ can be written as an alternative form similar to the one in ( [ eqhp1 ] ) except the expansion of column dimensions . with aid of _ proposition 2 _ , it is simple to follow the proof for the case of @xmath6 in section iv - b to prove _ theorem 2_. the detailed proof is omitted . in this section , we present some simulation results for four transmit antennas and four receive antennas . in all simulations , the channel model follows that described in section [ model ] . in fig . [ fig:1 ] , four kinds of stbcs are compared : guo - xia s code proposed in ( * ? ? ? * eq . ( 40 ) ) , @xmath109 in ( * ? ? ? * eq . ( 29 ) ) , @xmath110 in ( * ? ? ? * eq . ( 37 ) ) and the new code @xmath107 given in ( [ m4 ] ) . note that all the codes presented in fig . [ fig:1 ] have the same rate of @xmath5 , and 64-qam constellation is used so that we keep the same bandwidth efficiency of 8 bps / hz for each code . [ fig:1 ] shows the bit error rate ( ber ) for four codes based on pic group decoding . firstly , as expected guo - xia s code , @xmath109 , @xmath110 and the new code @xmath107 can achieve full diversity at high snr . then , one can observe that @xmath107 has a very similar performance to @xmath110 and guo - xia s code since we use the same real linear transform matrix for the case @xmath95 . however , compared with @xmath110 and guo - xia s code , the code @xmath107 further increases the number of pic groups and allows two real symbols ( i.e. single complex symbol ) to be decoded in each pic group without performance loss . in fig . [ fig:2 ] , ciod of rate 1 in ( * ? ? ? * eq . ( 85 ) ) and qostbc of rate 1 in ( * ? ? ? * eq . ( 39 ) ) with ml decoding are compared with the code @xmath107 with pic group decoding . in order to make a fair performance comparison , the symbols are chosen from a 256qam signal set for ciod and qostbc , and 64qam for the code @xmath107 . thus , the code @xmath107 has the same bandwidth efficiency with ciod and qostbc at @xmath228 bps / hz . note that qostbc with optimal transformation has a very similar performance to ciod . moreover , one observe that the code @xmath107 outperforms both ciod and qostbc by @xmath4 db . as for this case , the decoding complexity of new code ( real symbols pair - wise ) is equivalent to that of qostbc ( real symbols pairwise ml decoding ) and ciod ( single complex symbol ml decoding ) . [ fig:3 ] presents the performance comparison between the code @xmath229 in @xcite and the proposed code @xmath230 with pic and pic - sic group decoding , respectively . here , 64qam is used to keep the same bandwidth efficiency of @xmath231 bps / hz . it can be observed that @xmath229 has a very similar performance to @xmath230 under both pic and pic - sic group decoding . in addition , it is shown that both @xmath229 and @xmath230 can achieve full diversity under pic - sic group decoding , but lose full diversity when pic group decoding is employed which is validated by theorem 1 . in this paper , we proposed a systematic design of stbc that can achieve full diversity with the pic group decoding . by coding the real and imaginary parts of the complex symbols vector independently , the proposed code has a reduced pic group decoding complexity , which is equivalent to a joint decoding of @xmath3 real symbols for @xmath0 transmit antennas . the full diversity of the proposed stbc with @xmath2 diagonal layers was proved for pic group decoding with @xmath6 and pic - sic group decoding with any @xmath2 , respectively . it is worthwhile to mention that for @xmath4 transmit antennas the code admits real symbol pairwise decoding and the code rate is @xmath5 . simulation results show that our proposed code can achieve full diversity with a lower decoding complexity than other existing codes . consider the system described in _ theorem [ tpic ] _ with receive antenna @xmath149 . according to the system model given in ( [ ry ] ) , the matrix form of @xmath232 is represented as @xmath233 with the expansion of ( [ eqn : y ] ) , we rewrite @xmath234 as a matrix form @xmath235 we substitute the codeword matrices ( [ ab ] ) into ( [ eqn : yr1 ] ) and([eqn : yi1 ] ) . then , we can obtain @xmath236\\ = & \sqrt{\frac{\rho}{\mu } } \left(\left[\begin{array}{cc } \mathbf{c}_r^{1 } & \mathbf{c}_r^{2}\\ -\mathbf{c}^{2}_r & \mathbf{c}^{1}_r\\ \end{array}\right ] \left[\begin{array}{c } \mathbf{h}_{r}^1\\ \mathbf{h}_{r}^2 \end{array}\right ] - \left[\begin{array}{cc } \mathbf{c}_i^{1 } & \mathbf{c}_i^{2}\\ \mathbf{c}^{2}_i & -\mathbf{c}^{1}_i\\ \end{array}\right ] \left[\begin{array}{c } \mathbf{h}_{i}^1\\ \mathbf{h}_{i}^2 \end{array}\right]\right ) + \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2 \end{array}\right]\\ = & \sqrt{\frac{\rho}{\mu } } \left[\begin{array}{c } \mathbf{c}_r^{1}\mathbf{h}_{r}^1+\mathbf{c}_r^{2 } \mathbf{h}_{r}^2-\mathbf{c}_i^{1}\mathbf{h}_{i}^1-\mathbf{c}_i^{2 } \mathbf{h}_{i}^2\\ -\mathbf{c}_r^{2}\mathbf{h}_{r}^1+\mathbf{c}_r^{1 } \mathbf{h}_{r}^2-\mathbf{c}_i^{2}\mathbf{h}_{i}^1+\mathbf{c}_i^{1 } \mathbf{h}_{i}^2 \end{array}\right ] + \left[\begin{array}{c } \mathbf{w}_{1,r}\\ \mathbf{w}_{2,r } \end{array}\right ] , \end{split}\ ] ] @xmath237\\ = & \sqrt{\frac{\rho}{\mu } } \left(\left[\begin{array}{cc } \mathbf{c}_r^{1 } & \mathbf{c}_r^{2}\\ -\mathbf{c}^{2}_r & \mathbf{c}^{1}_r \end{array}\right ] \left[\begin{array}{c } \mathbf{h}_{i}^1\\ \mathbf{h}_{i}^2 \end{array}\right ] + \left[\begin{array}{cc } \mathbf{c}_i^{1 } & \mathbf{c}_i^{2}\\ \mathbf{c}^{2}_i & -\mathbf{c}^{1}_i\\ \end{array}\right ] \left[\begin{array}{c } \mathbf{h}_{r}^1\\ \mathbf{h}_{r}^2 \end{array}\right]\right ) + \left[\begin{array}{c } \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right]\\ = & \sqrt{\frac{\rho}{\mu } } \left[\begin{array}{c } \mathbf{c}_r^{1}\mathbf{h}_{i}^1+\mathbf{c}_r^{2 } \mathbf{h}_{i}^2+\mathbf{c}_i^{1}\mathbf{h}_{r}^1+\mathbf{c}_i^{2 } \mathbf{h}_{r}^2\\ -\mathbf{c}_r^{2}\mathbf{h}_{i}^1+\mathbf{c}_r^{1 } \mathbf{h}_{i}^2+\mathbf{c}_i^{2}\mathbf{h}_{r}^1-\mathbf{c}_i^{1 } \mathbf{h}_{r}^2 \end{array}\right ] + \left[\begin{array}{c } \mathbf{w}_{1,i}\\ \mathbf{w}_{2,i } \end{array}\right ] , \end{split}\ ] ] where @xmath238 , @xmath239 and @xmath240 . let @xmath241 $ ] and @xmath242 $ ] . furthermore , according to the code structure in ( [ c ] ) , ( [ eqn : yr ] ) and ( [ eqn : yi ] ) can be rewritten as @xmath243\\ = & \sqrt{\frac{\rho}{\mu}}\left[\begin{array}{c } \sum^2_{p=1}\mathbf{c}_{p , r}^{1}\mathbf{h}_{r}^1+\sum^2_{p=1}\mathbf{c}_{p , r}^{2 } \mathbf{h}_{r}^2-\sum^2_{p=1}\mathbf{c}_{p , i}^{1}\mathbf{h}_{i}^1-\sum^2_{p=1}\mathbf{c}_{p , i}^{2 } \mathbf{h}_{i}^2\\ -\sum^2_{p=1}\mathbf{c}_{p , r}^{2}\mathbf{h}_{r}^1+\sum^2_{p=1}\mathbf{c}_{p , r}^{1 } \mathbf{h}_{r}^2-\sum^2_{p=1}\mathbf{c}_{p , i}^{2}\mathbf{h}_{i}^1+\sum^2_{p=1}\mathbf{c}_{p , i}^{1 } \mathbf{h}_{i}^2 \end{array}\right ] + \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2 \end{array}\right ] , \end{split}\ ] ] @xmath244\\ = & \sqrt{\frac{\rho}{\mu } } \left[\begin{array}{c } \sum^2_{p=1}\mathbf{c}_{p , r}^{1}\mathbf{h}_{i}^1+\sum^2_{p=1}\mathbf{c}_{p , r}^{2 } \mathbf{h}_{i}^2+\sum^2_{p=1}\mathbf{c}_{p , i}^{1}\mathbf{h}_{r}^1+\sum^2_{p=1}\mathbf{c}_{p , i}^{2 } \mathbf{h}_{r}^2\\ -\sum^2_{p=1}\mathbf{c}_{p , r}^{2}\mathbf{h}_{i}^1+\sum^2_{p=1}\mathbf{c}_{p , r}^{1 } \mathbf{h}_{i}^2+\sum^2_{p=1}\mathbf{c}_{p , i}^{2}\mathbf{h}_{r}^1-\sum^2_{p=1}\mathbf{c}_{p , i}^{1 } \mathbf{h}_{r}^2 \end{array}\right ] + \left[\begin{array}{c } \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right ] , \end{split}\ ] ] where @xmath245,\ , p=1,2;\,\ , i=1,2 . \end{aligned}\ ] ] equivalently , we have @xmath246= \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2 \end{array}\right]+\\ & \sqrt{\frac{\rho}{\mu}}\left[\begin{array}{c } \mathcal{h}^1_{1,r}\mathbf{x}_{1,r}^{1}+\mathcal{h}^1_{2,r}\mathbf{x}_{2,r}^{1 } + \mathcal{h}^2_{1,r}\mathbf{x}_{1,r}^{2}+\mathcal{h}^2_{2,r}\mathbf{x}_{2,r}^{2}- \mathcal{h}^1_{1,i}\mathbf{x}_{1,i}^{1}-\mathcal{h}^1_{2,i}\mathbf{x}_{2,i}^{1 } -\mathcal{h}^2_{1,i}\mathbf{x}_{1,i}^{2}-\mathcal{h}^2_{2,i}\mathbf{x}_{2,i}^{2}\\ -\mathcal{h}^1_{1,r}\mathbf{x}_{1,r}^{2}-\mathcal{h}^1_{2,r}\mathbf{x}_{2,r}^{2}+\mathcal{h}^2_{1,r}\mathbf{x}_{1,r}^{1}+\mathcal{h}^2_{2,r}\mathbf{x}_{2,r}^{1}-\mathcal{h}^1_{1,i}\mathbf{x}_{1,i}^{2}-\mathcal{h}^1_{2,i}\mathbf{x}_{2,i}^{2}+\mathcal{h}^2_{1,i}\mathbf{x}_{1,i}^{1}+\mathcal{h}^2_{2,i}\mathbf{x}_{2,i}^{1 } \end{array}\right ] , \end{split}\ ] ] @xmath247=\left[\begin{array}{c } \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right]+\\ & \sqrt{\frac{\rho}{\mu}}\left[\begin{array}{c } \mathcal{h}^1_{1,i}\mathbf{x}_{1,r}^{1}+\mathcal{h}^1_{2,i}\mathbf{x}_{2,r}^{1 } + \mathcal{h}^2_{1,i}\mathbf{x}_{1,r}^{2}+\mathcal{h}^2_{2,i}\mathbf{x}_{2,r}^{2}+ \mathcal{h}^1_{1,r}\mathbf{x}_{1,i}^{1}+\mathcal{h}^1_{2,r}\mathbf{x}_{2,i}^{1 } + \mathcal{h}^2_{1,r}\mathbf{x}_{1,i}^{2}+\mathcal{h}^2_{2,r}\mathbf{x}_{2,i}^{2}\\ -\mathcal{h}^1_{1,i}\mathbf{x}_{1,r}^{2}-\mathcal{h}^1_{2,i}\mathbf{x}_{2,r}^{2}+\mathcal{h}^2_{1,i}\mathbf{x}_{1,r}^{1}+\mathcal{h}^2_{2,i}\mathbf{x}_{2,r}^{1}+\mathcal{h}^1_{1,r}\mathbf{x}_{1,i}^{2}+\mathcal{h}^1_{2,r}\mathbf{x}_{2,i}^{2}-\mathcal{h}^2_{1,r}\mathbf{x}_{1,i}^{1}-\mathcal{h}^2_{2,r}\mathbf{x}_{2,i}^{1 } \end{array}\right ] \end{split}\ ] ] where @xmath156 , { \rm{and}~ } \mathcal{h}_{p , i}^i=\left[\begin{array}{c } \mathbf{0}_{(p-1)\times ( m/2)}\\ \mathrm{diag}(\mathbf{h}_i^{i})\\ \mathbf{0}_{(2-p)\times ( m/2 ) } \end{array } \right],\end{aligned}\ ] ] for @xmath153 and @xmath152 . next , we gather the equations @xmath248 , @xmath249 , @xmath250 and @xmath251 to form a real system as follows @xmath252&= & \sqrt{\frac{\rho}{\mu } } \left[\begin{array}{cccccccc } \mathcal{h}^1_{1,r}&\mathcal{h}^1_{2,r}&\mathcal{h}^2_{1,r}&\mathcal{h}^2_{2,r}&-\mathcal{h}^1_{1,i}&-\mathcal{h}^1_{2,i}&-\mathcal{h}^2_{1,i}&-\mathcal{h}^2_{2,i}\\ \mathcal{h}^2_{1,r}&\mathcal{h}^2_{2,r}&-\mathcal{h}^1_{1,r}&-\mathcal{h}^1_{2,r}&\mathcal{h}^2_{1,i}&\mathcal{h}^2_{2,i}&-\mathcal{h}^1_{1,i}&-\mathcal{h}^1_{2,i}\\ \mathcal{h}^1_{1,i}&\mathcal{h}^1_{2,i}&\mathcal{h}^2_{1,i}&\mathcal{h}^2_{2,i}&\mathcal{h}^1_{1,r}&\mathcal{h}^1_{2,r}&\mathcal{h}^2_{1,r}&\mathcal{h}^2_{2,r}\\ \mathcal{h}^2_{1,i}&\mathcal{h}^2_{2,i}&-\mathcal{h}^1_{1,i}&-\mathcal{h}^1_{2,i}&-\mathcal{h}^2_{1,r}&-\mathcal{h}^2_{2,r}&\mathcal{h}^1_{1,r}&\mathcal{h}^1_{2,r } \end{array } \right ] \left[\begin{array}{c } \mathbf{x}^1_{1,r}\\\mathbf{x}^1_{2,r}\\ \mathbf{x}^2_{1,r } \\\mathbf{x}^2_{2,r } \\ \mathbf{x}^1_{1,i } \\ \mathbf{x}^1_{2,i } \\\mathbf{x}^2_{1,i } \\\mathbf{x}^2_{2,i } \end{array } \right]\\ & + & \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2\\ \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right].\end{aligned}\ ] ] in order to obtain the equivalent signal model in ( [ eqyr ] ) , using ( [ roxir ] ) and ( [ roxii ] ) we can rewrite ( [ eqn : ye1 ] ) as @xmath253= \sqrt{\frac{\rho}{\mu}}\mathcal{h } \left[\begin{array}{c } \mathbf{s}^1_{1,r}\\ \mathbf{s}^1_{2,r}\\ \mathbf{s}^2_{1,r } \\ \mathbf{s}^2_{2,r}\\ \mathbf{s}^1_{1,i}\\ \mathbf{s}^1_{2,i}\\ \mathbf{s}^2_{1,i}\\ \mathbf{s}^2_{2,i } \end{array } \right ] + \left[\begin{array}{c } \mathbf{w}_{r}^1\\ \mathbf{w}_{r}^2\\ \mathbf{w}_{i}^1\\ \mathbf{w}_{i}^2 \end{array}\right],\end{aligned}\ ] ] where the equivalent real channel matrix @xmath254 is given by @xmath255\\ & = & \left[\begin{array}{cccccccc}\label{pic } \mathbf{g}_1 & \mathbf{g}_2 & \ldots & \mathbf{g}_8 \end{array } \right ] . \end{aligned}\ ] ] according to _ definition [ orde ] _ , we obtain that the groups @xmath256 are orthogonal , and the groups @xmath257 are orthogonal as well . the authors would like to thank tianyi xu for his reading and comments on this manuscript . v. tarokh , n. seshadri , and a. calderbank , `` space - time codes for high data rate wireless communications : performance criterion and code construction , '' _ ieee trans . inf . theory _ , 44 , pp . 744765 , mar . v. tarokh , h. jafarkhani , and a. r. calderbank , `` space - time block codes from orthogonal designs , '' _ ieee trans . inf . theory _ , vol . 45 , pp . 14561467 , july 1999 . also , `` corrections to ` space - time block codes from orthogonal designs ' , '' _ ieee trans . inf . theory _ 46 , p. 314 , jan . 2000 . k. lu , s. fu , and x .- xia , `` closed form designs of complex orthogonal space - time block codes of rates @xmath258 for @xmath259 or @xmath260 transmit antennas , '' _ ieee trans . inf . theory _ 43404347 , dec . 2005 . h. wang and x .- g xia , `` upper bounds of rates of complex orthogonal space - time block codes , '' _ ieee trans . inf . theory _ , vol 49 , pp . 27882796 , oct . 2003 . x. guo and x .- g . xia , `` on full diversity space - time block codes with partial interference cancellation group decoding , '' _ ieee trans . inf . theory _ , 55 , pp . 43664385 , oct . 2009 . also , `` corrections to ` on full diversity space - time block codes with partial interference cancellation group decoding ' , '' http://www.ece.udel.edu/~xxia/correction_guo_xia.pdf . w. zhang , t. xu , and x .- xia , `` two designs of space - time block codes achieving full diversity with partial interference cancellation group decoding , '' _ ieee trans . inf . theory _ , submitted . http://arxiv.org/abs/0904.1812v3 w. zhang , l. shi , and x .- xia , `` full diversity space - time block codes with low - complexity partial interference cancellation group decoding , '' _ ieee trans . _ , submitted . http://arxiv.org/abs/1003.3908 [ http://arxiv.org/abs/1003.3908 ] j. boutros and e. viterbo , `` signal space diversity : a power and bandwidth efficient diveristy technique for the rayleigh fading channel , '' _ ieee trans . inf . theory _ , 14531467 , july 1998 .

this paper studies the extreme gaps between eigenvalues of random matrices . we give the joint limiting law of the smallest gaps for haar - distributed unitary matrices and matrices from the gaussian unitary ensemble . in particular , theth smallest gap , normalized by a factor , has a limiting density proportional to . concerning the largest gaps , normalized by , they converge in to a constant for all . these results are compared with the extreme gaps between zeros of the riemann zeta function .

1010.1294

we address here the following question : what is the asymptotic size and the limit laws of the smallest and largest gaps or spacings in the spectra of random matrices ? typical spacings between eigenvalues of random matrices have been well understood for invariant ensembles for quite some time . more recently , the behavior of these typical spacings has even been proved to be universal for much larger classes of random matrices ( @xcite , @xcite , @xcite ) . much less is known for atypically large or small spacings . this question was first considered for the smallest spacings in the unpublished phd thesis of j.vinson ( @xcite ) and raised by persi diaconis in @xcite for the largest ones . it was also discussed in an interesting debate during a conference at the courant institute in 2006 , in honor of percy deift . we solve here completely the question of the smallest spacings for the simplest invariant ensembles , i.e the cue and the gue . we give the scaling and the limit laws for the joint distribution of the smallest spacings . the answer is simple to state : it is given by the trivial poissonian ansatz where the spacings would be treated as i.i.d random variables . the answer we find for the largest spacings is less complete , since we can only obtain at this point a first - order approximation which gives the asymptotic size of the largest spacings and not their limit laws . we believe that the same poissonian ansatz should work as well for the largest gaps , but this question is left open . the question of the universality of the behavior of the extremal spacings is also left open . dyson @xcite showed that the repulsion between eigenvalues of the gaussian unitary ensemble ( the gue ) could be described asymptotically in terms of the _ determinantal _ point process associated with the sine kernel @xmath6 for this limiting determinantal point process , the probability of having no eigenvalue in an interval of size @xmath7 is known to be the fredholm determinant @xmath8 where @xmath9 is the convolution operator acting on @xmath10 with kernel @xmath11 . the density of the spacing between two successive points is then given by ( see @xcite ) @xmath12 this spacing distribution was shown to appear in a number - theoretic context : some statistics produced by odlyzko @xcite presented a correspondence between the histogram of normalized gaps between zeros of the riemann zeta function and @xmath13 . this gave further evidence for the analogy discovered by dyson and montgomery : they realized that the local dependence of the zeros of @xmath14 , previously calculated by montgomery , involved the sine kernel ( see @xcite , @xcite for an historical account and other steps of this fruitful analogy ) . at the same mean or typical gap scale , a precise analysis of the joint distribution of the gaps between eigenvalues was performed by katz - sarnak @xcite and soshnikov @xcite , for the circular unitary ensemble ( the cue ) . less attention was paid to eigenvalues statistics at smaller and larger scales . this paper concerns the extreme gaps . this study was initiated by j. vinson @xcite : he showed that the smallest gap between elements of the @xmath15 , multiplied by @xmath1 , converges in law to a random variable with distribution function @xmath16 , as the size @xmath17 of the unitary matrix increases . in his thesis , similar results for the smallest gap between eigenvalues of a generalization of the @xmath18 were obtained . vinson also gives interesting heuristics suggesting that the largest gap between @xmath15 eigenvalues should be of order @xmath19 , with poissonian fluctuations around this limit . using a different technique , soshnikov @xcite investigated the smallest gaps for determinantal point processes on the real line , with a translation invariant kernel : amongst points included in @xmath20 $ ] , this extreme spacing multiplied by @xmath21 converges weakly to the distribution with distribution function @xmath16 , as @xmath22 . heuristically , the above extreme gaps asymptotics can be obtained using the known asymptotics @xcite of the spacing distribution @xmath23 and treating the gaps as independent random variables . the difficulty to obtain rigorous results lies in showing that this poissonian ansatz is asymptotically correct for the extreme gaps , and in making the above estimates uniform in the dimension @xmath17 . we first consider the _ joint _ law of the smallest gaps ( theorem [ thm : unitary ] , corollary [ cor : smallestunitary ] ) between eigenvalues of unitary matrices . this relies both on soshnikov s method and a convergence of the process of small gaps to a poisson point process . the same reasoning equally applies to the small gaps between eigenvalues from the gaussian unitary ensemble ( theorem [ thm : g ] , corollary [ cor : smallestgue ] ) . the proofs of the small spacings asymptotics are in section 2 . the first order asymptotics of the largest gaps is then proved . concerning unitary random matrices ( theorem [ thm : largestgaps ] ) , this makes use of two important tools . a key ingredient , by deift & al @xcite , is the uniform asymptotics about the probability for a given arc of the circle to be free of eigenvalues . the proof also requires the negative correlation property for the event that two disjoint arcs are free of eigenangles . on account of the @xmath18 ( theorem [ thm : largestgapsgue ] ) , we also make a essential use of the negative correlation , and the large gap probability is evaluated by comparing the @xmath18 fredholm determinant with the unitary one . these large gaps asymptotics are proved in section 3 . the extreme spacings between random eigenvalues are important quantities for statistical physics , computational mathematics and number theory . for this reason , diaconis @xcite mentions the open question of maximal spacings , answered in theorem [ thm : largestgaps ] . after making our results explicit , successively for unitary matrices and the gue ensemble , we give applications of our extremal spacings statistics at the end of this introduction . let @xmath24 , a haar - distributed ( measure @xmath25 ) unitary matrix over @xmath26 . suppose @xmath24 has eigenvalues @xmath27 s , with ordered eigenangles @xmath28 consider the point process on @xmath29 @xmath30 our first result is about the convergence of @xmath31 to a poisson point process , thanks to this normalization by @xmath32 . [ thm : unitary ] suppose @xmath33 . as @xmath34 , the process @xmath31 converges to a poisson point process @xmath35 with intensity @xmath36 for any bounded borel sets @xmath37 and @xmath38 . the intensity is proportional to @xmath39 because of the rotational invariance of the haar measure . the corresponding factor will be less trivial in the case of the @xmath18 ensemble . our method to prove theorem [ thm : unitary ] relies on the _ s - modified _ random point field technique initiated by soshnikov @xcite : one can calculate the correlation functions of the process obtained by keeping only the @xmath40 s for which @xmath41 contains exactly one other eigenvalue . contrary to @xcite , we do not use the notion of cluster functions , because we characterize the convergence to poisson random variables thanks to the convergence of the factorial moments ; this allows also to consider easily non translation invariant kernels , like in the @xmath18 case . moreover , theorem [ thm : unitary ] gives information about the joint distribution of the number of gaps taking values in disjoint intervals ( convergence in terms of point processes ) . in particular we can compute the limiting joint law of the smallest gaps . let @xmath42 be the @xmath0 smallest eigenangles gaps ( i.e. of the form @xmath43 , where the indexes are modulo @xmath17 and gaps are defined as small as possible , i.e. @xmath44 $ ] ) . for the sake of brevity , write @xmath45 the limiting joint law of the @xmath46 s is a corollary of theorem [ thm : unitary ] . [ cor : smallestunitary ] for any @xmath47 , under the haar measure on @xmath48 , @xmath49 in particular , the @xmath50 smallest normalized space @xmath51 converges in law to @xmath46 , with density @xmath52 note that this result , for @xmath53 , is proved in vinson s thesis @xcite by a different method : he characterizes the number of small gaps as a symmetric function of the eigenvalues , and computes its moments . it is not clear how his method can be extended to provide the joint law of the @xmath0 smallest gaps . + we now turn to our next question about extreme gaps , i.e the asymptotic behavior of the largest gaps , which were guessed by vinson , based on the supposed asymptotic independence of distant gaps . we obtain thanks to the precise asymptotics of one gap probability , obtained by the steepest descent method for riemann hilbert problems in @xcite , and the negative association property of determinantal point processes ( see e.g. @xcite ) . note that both results are posterior to vinson s thesis . consider @xmath54 the largest gaps between successive eigenangles of @xmath55 , i.e. of the form @xmath56 , where the indexes are modulo @xmath17 and @xmath57 . then , as @xmath17 goes to infinity , the largest gap converges in @xmath58 to a constant , for any @xmath5 : @xmath59 actually the above limit holds for all the @xmath60 largest gaps if @xmath61 is subpolynomial . [ thm : largestgaps ] let @xmath62 be positive integers . then for any @xmath5 , @xmath63 as @xmath34 , @xmath64 the problem of the exact fluctuations will be addressed in a future work . ] . note that , for independent uniform eigenangles on the unit circle , the largest gap is of order @xmath65 , more than in the above theorem , as expected from the repulsion of the eigenvalues in the determinantal case . similar results hold for the @xmath18 . for this ensemble , the distribution of the eigenvalues has density @xmath66 with respect to the lebesgue product measure , on the simplex @xmath67 . the empirical spectral distribution @xmath68 converges in probability to the semicircle law ( see e.g. @xcite ) @xmath69 like for the unitary group , we first consider the smallest gaps , studying the point process @xmath70 for any arbitrarily small fixed @xmath71 ( this is a technical restriction allowing the use of the plancherel - rotach asymptotics of the hermite polynomials ) . [ thm : g ] as @xmath34 , the process @xmath72 converges to a poisson point @xmath73 process with intensity @xmath74 for any bounded borel sets @xmath75 and @xmath76 . the following corollary about the smallest gaps is an easy consequence of the previous theorem . as for the unitary group , introduce @xmath77 the @xmath0 nearest spacings in @xmath78 , i.e. of the form @xmath79 , @xmath80 , with @xmath81 , @xmath82 $ ] , @xmath83 . let @xmath84 [ cor : smallestgue ] for any @xmath47 , with the above notations and for the @xmath18 ensemble measure ( [ eqn : gue ] ) , @xmath85 in particular , the @xmath50 smallest normalized space @xmath86 converges in law to @xmath46 , with density @xmath52 the above result gives for @xmath53 the probability density of the normalized smallest gap in a given interval @xmath78 : @xmath87 . the following precises the distribution of this smallest spacing location in the bulk of the spectrum . [ cor : smallestlocation ] let @xmath88 be the index of the smallest gap between eigenvalues of the @xmath18 in a compact subset @xmath89 with nonempty interior : @xmath90 as @xmath34 , @xmath91 converges weakly to the probability measure with density proportional to @xmath92 we now turn to the largest gaps for the @xmath18 ensemble . the result is completely different inside the bulk and on the edge . indeed , for eigenvalues strictly inside the support of the limiting measure , the maximal spacings have order @xmath19 ( see the following theorem [ thm : largestgapsgue ] ) , while the eigenvalues on the border have an average distance of higher order , @xmath93 : for any @xmath0 , @xmath94 converges weakly as @xmath34 to a multivariate tracy - widom distribution ( see e.g. @xcite ) . strictly inside the bulk , the result is analogous to the circular case , the only difference being the normalization , due to the average density of eigenvalues . let @xmath95 be the largest gaps of type @xmath96 with @xmath97 , a compact subset of @xmath98 with nonempty interior . [ thm : largestgapsgue ] let @xmath99 be positive integers . then for any @xmath5 @xmath100 when seen in a window of size proportional to the average gap , the spacings between the zeros of dirichlet l - functions are distributed like particles of a determinantal point process with sine kernel , this is the montgomery - odlyzko law . here we want to discuss the accuracy of this analogy when looking at rare events , the extreme gaps between the zeta zeros , relying on theorems [ thm : unitary ] and [ thm : largestgaps ] . due to the availability of many numerical data , we focus on the riemann zeta function @xmath101 ( @xmath102 ) , which admits an analytic continuation to @xmath103 . let @xmath104 be the non trivial zeta zeros , @xmath105 , with @xmath106 then @xmath107 has an average value 1 . the quantity @xmath108 has been widely studied . conditionally to the generalized riemann hypothesis , the best known result is @xmath109 @xcite . from the @xmath18 hypothesis for the zeta zeros , it is expected that @xmath110 . however , to the best of our knowledge , more precise conjectures about the growth speed of large gaps between zeta zeros were not proposed . from theorem [ thm : largestgaps ] , amongst @xmath17 successive gaps with fermionic repulsion , the maximal gap has size about @xmath111 times the average gap , suggesting @xmath112 in particular @xmath113 odlyzko s numerical data @xcite give 3.303 for the maximal value of @xmath114 , @xmath115 , while @xmath116 , giving a difference of @xmath117 with the observed gaps . r0.5 [ fig : maximal ] further tests can be performed at distinct heights along the critical axis , thanks to numerical data of gourdon @xcite : he computed @xmath118 successive zeta zeros at height @xmath119 along the critical axis for each @xmath120 . the extreme normalized gaps are given on the joint graph , where the expectation from our random matrices result is the straight line @xmath121 . for example , amongst the @xmath122 gaps following the height @xmath123 , @xmath124 , i.e. a difference of @xmath125 with the expected value . concerning the smallest gaps , does the poisson intensity @xmath126 from theorem [ thm : unitary ] appear in the context of the zeta zeros ? note @xmath127 the normalized gaps . we know that , as @xmath34 , the set of gaps @xmath128 converges weakly to a poisson point process with intensity @xmath126 . we therefore expect that , as @xmath34 , @xmath129 converges to a poisson point process with the same intensity . r0.5 the joint graph gives the histogram of the @xmath130 smallest gaps , normalized as previously , amongst the @xmath131 first zeta zeros , based again on numerical data from @xcite . more precisely , the histogram gives the 3000 smallest values of @xmath132 . the straight line is the function @xmath133 ( the step of the histogram is 5 ) . this presents a good relevance of the @xmath18 hypothesis for the riemann zeta function , even at the scale of rare events , here the extreme spacings . the most classical method to diagonalize a matrix is the well - known qr algorithm . in the case of hermitian matrices , an alternative approach was proposed by deift & al @xcite , based on the isospectral property of the toda flow . more precisely , given a @xmath134 hermitian matrix @xmath135 , the first step is to reduce it in a tridiagonal form @xmath136 ( which is a robust and fast operation ) , conjugating with successive householder reflections , @xmath137,\ ] ] keeping the same spectrum as @xmath135 . after such a tridiagonalization , the @xmath138 and @xmath139 s are real . the above matrix is important in the analysis of the toda lattice : flaschka , hnon , and manakov proved independently in the 70 s that the following evolution of @xmath17 particles on a line ( @xmath140 , @xmath141 , @xmath142 ) , @xmath143 is an integrable system . more precisely , after the change of variables @xmath144 the differential equation takes the lax pair form @xmath145 where @xmath146.\ ] ] in particular and importantly , the spectrum of @xmath147 does not depend on time . moser proved that @xmath148 , @xmath149 as @xmath150 , with @xmath67 . this implies that @xmath151 converges to @xmath152 , hence @xmath136 converges to a diagonal matrix , whose entries give the eigenvalues of @xmath153 . deift @xcite asked about the speed of convergence of the toda flow till its equilibrium . more precisely , for a given @xmath154 , what is the necessary time @xmath155 such that @xmath156 for all @xmath157 ? as @xmath158 the speed of convergence to the spectrum is governed by the minimal gap between eigenvalues . a good choice for a _ typical _ hermitian matrix is a matrix from the @xmath18 , with independent ( up to symmetry ) complex gaussian entries ( of variance 2 on the diagonal , 1 elsewhere ) . from corollary [ cor : smallestgue ] , the minimal gap for such matrices is of order @xmath159 for a given required precision @xmath160 , the toda flow time necessary to evaluate the eigenvalues is expected to grow as @xmath161 with the dimension as a function of @xmath17 ] . we first review some results about convergence of processes , here with values in @xmath29 and a finite number of atoms . a point process @xmath163 is said to converge in distribution to @xmath164 is for any bounded continuous function @xmath165 @xmath166 to show the convergence of @xmath31 to a poisson point process @xmath35 , we only need to show the convergence in law of @xmath167 to @xmath168 for all bounded intervals @xmath169 and @xmath78 , the independence for disjoint @xmath170 s being an automatic consequence : this is a very practical property of poisson point processes , detailed in proposition [ prop : facmom ] below . moreover , the convergence of @xmath167 will be shown thanks to the convergence of the factorial moments to those of a poisson random variable . this is particularly adapted to our situation because as we will see the correlation functions of point processes are defined through factorial moments , and are explicit in the case of determinantal point processes . note that this is the same technique employed in @xcite , where the following result is given , in the case of point processes with values in @xmath171 . [ prop : facmom ] let @xmath163 be a sequence of point processes on @xmath29 , and @xmath35 a poisson point process on @xmath29 with intensity @xmath172 having no atoms ( and @xmath173-finite ) . assume that for any bounded intervals @xmath169 and @xmath78 and all positive integers @xmath174 @xmath175 then the sequence of point processes @xmath31 converges in distribution @xmath35 . we need to check the three conditions of the following theorem [ thm : kallenberg ] by kallenberg , which is written here in the specific case @xmath176 , @xmath177 and @xmath178 the set of compact rectangles @xmath170 ( @xmath169 and @xmath78 intervals ) , with the notations from @xcite . conditions ( 1 ) , ( 2 ) and ( 3 ) will be verified if @xmath179 indeed , if ( [ eqn : convlaw ] ) holds , for any @xmath155 @xmath180 and as @xmath181 is almost surely finite this goes to @xmath152 as @xmath150 ( is is a consequence of the dominated convergence theorem ) ; this proves part ( 1 ) . equations ( 2 ) and ( 3 ) are also consequences of the above convergence in law . to prove ( [ eqn : convlaw ] ) , as @xmath181 is a poisson random variable , the convergence of all the moments is sufficient . a moment is a finite linear combination of factorial moments , which concludes the proof . [ thm : kallenberg ] let @xmath35 be a point process on @xmath29 and assume @xmath35 is almost surely simple ( i.e. the atoms of the measure @xmath35 all have weight 1 almost surely ) . then @xmath31 converges weakly to @xmath35 if and only if the three following conditions are satisfied for any compact intervals @xmath169 and @xmath78 in @xmath171 : 1 . @xmath182 ; 2 . @xmath183 ; 3 . @xmath184 . references for the main properties of correlation functions of determinantal point processes are @xcite and @xcite . we follow this last survey to present the notions used in the following . if @xmath185 is a simple point process on a complete separate metric space @xmath186 , consider the point process @xmath187 on @xmath188 . one can define this way a measure @xmath189 on @xmath188 by @xmath190 for any borel set @xmath191 in @xmath188 . most of the time , there is a natural measure @xmath192 on @xmath186 , in our cases @xmath193 or @xmath194 and @xmath192 is the lebesgue measure . if @xmath195 is absolutely continuous with respect to @xmath196 , there exists a function @xmath197 on @xmath188 such that for any borel sets @xmath198 in @xmath186 @xmath199 hence one can think about @xmath200 as the asymptotic ( normalized ) probability of having exactly one particle in neighborhoods of the @xmath201 s . more precisely under suitable smoothness assumptions , and for distinct points @xmath202 in @xmath193 , @xmath203 note that @xmath197 is not a probability density . moreover specifically , if @xmath204 are disjoint in @xmath186 and @xmath205 , @xmath206 see @xcite for a proof . in particular , @xmath207 note the analogy with formula ( [ eqn : facmompoisson ] ) we want to prove . for unitary matrices or the @xmath18 ensemble , our method to prove convergence of small spacings counting measures is the same : * for given compact intervals @xmath169 and @xmath78 consider the modified process obtained from @xmath208 by keeping only the points @xmath209 in @xmath78 such that @xmath210 . * show that the correlation function @xmath211 of this new process uniformly converges to @xmath212 . this is possible thanks to the determinantal aspect of @xmath213 and the hadamard - fischer inequality , lemma [ lem : fischer ] . * conclude that the factorial moments converge to those of the expected poisson random variables thanks to ( [ eqn : facmompoisson ] ) and ( [ eqn : facmomcorrelation ] ) for the smallest gaps asymptotics , the following inequality will be repeatedly used . a concise proof can be found in @xcite . [ lem : fischer ] let @xmath135 be a positive - definite @xmath134 ( hermitian ) matrix . for any @xmath214 , let @xmath215 ( resp . @xmath216 ) be the submatrix of @xmath135 using rows and columns numbered in @xmath217 ( resp @xmath218 ) . then @xmath219 we begin with the proof of theorem [ thm : unitary ] . we know that for a unitary matrix @xmath33 , the density of the eigenangles @xmath220 , with respect to the lebesgue measure on the corresponding simplex is @xmath221 moreover , a remarkable fact about the point process @xmath222 is that it is determinantal : all its correlation functions @xmath223 , @xmath142 are determinants based on the same kernel : @xmath224 this classical property relies on gaudin s lemma , see @xcite . in the following for any bounded interval @xmath37 we write @xmath225 . we want to show that for an interval @xmath38 @xmath226 with @xmath227 . we consider the point process @xmath228 and its thinning @xmath229 obtained from @xmath213 by only keeping the eigenangles @xmath40 for which @xmath230 . the following lemma means that @xmath231 is properly estimated by @xmath232 . it is analogous to lemma 3 in @xcite . [ lem : nosucsmall ] for any interval @xmath38 , as @xmath34 , @xmath233 . let @xmath234 be such that @xmath235 , and @xmath236 . if @xmath237 , then @xmath238 . hence @xmath239 where the last inequality comes from the definition ( [ eqn : multdimpcess ] ) , where @xmath191 is the set of points @xmath240 with @xmath241 and @xmath242 . to prove that this positive random variable converges in law to @xmath152 , we consider its expectation @xmath243 and show it goes to 0 . thanks to the multilinearity of the determinant , @xmath244 as @xmath245 and @xmath246 , the first column of this determinant is @xmath247 and the two others are @xmath248 . thus @xmath249 the integration domain is @xmath250 , concluding the proof let @xmath251 , @xmath252 , be the correlation functions of the point process @xmath229 . if , for any @xmath174 , the convergence of the factorial moment @xmath253 can be shown , then theorem [ thm : unitary ] will be proved , thanks to the above lemma [ lem : nosucsmall ] . the way to show ( [ eqn : convfacmom ] ) relies on three steps , to apply a simple dominated convergence argument : * if all @xmath40 s are distinct @xmath251 converges to @xmath254 as @xmath34 ( lemma [ lem : simpleconvergence ] ) ; * in the set @xmath255 @xmath251 is uniformly bounded ( lemma [ lem : uniformbounded ] ) ; * even if @xmath251 is not uniformly bounded in the complement of @xmath256 in @xmath257 ( @xmath258 ) , the contribution to the integral is negligible because the volume of @xmath258 decreases sufficiently fast ( lemma [ lem : neglectset ] ) . [ lem : simpleconvergence ] let @xmath259 be distinct elements in @xmath257 . then @xmath260 first note that , as all the @xmath40 s are distinct , for sufficiently large @xmath17 the point @xmath261 is in @xmath256 ( see ( [ eqn : defomega ] ) ) . this means that if @xmath259 are points of @xmath229 , the point in each of the @xmath262 is not another one of the @xmath263 s . this makes the combinatorics easy : the correlation functions of @xmath229 can be explicitly given in terms of those of @xmath213 , as noted in @xcite , by an inclusion - exclusion argument : for sufficiently large @xmath17 @xmath264 note that there is no convergence issue here as @xmath265 if @xmath266 . we first show that the term corresponding to @xmath267 in the above sum gives the expected asymptotics . the determinantal aspect of the process makes things easy : @xmath268 is a @xmath269 determinant , and only the terms in the @xmath270 diagonal blocks make a significant contribution ( which leads to the idea of asymptotic independence ) . more precisely , we write formally @xmath271 as @xmath272 and @xmath273 , if @xmath274 all terms of the corresponding @xmath270 above matrix are @xmath275 . moreover , the above determinant is unchanged by subtracting an odd column to the following even column , and then by subtracting an odd line to the following even line . in this way , the diagonal @xmath270 matrices becomes @xmath276 where the last equality relies on @xmath245 , @xmath277 and @xmath278 . as a consequence , in the expansion of the determinant over all permutations of @xmath279 , the terms corresponding to entries only in the diagonal @xmath270 block matrices have order at most @xmath280 , while all other terms have a strictly lower order ( at most @xmath281 ) . as the integration domain of @xmath282 is @xmath283 , the only permutations hopefully giving a non - zero limit need to come from the block diagonal @xmath270 matrices . indeed they give a non - trivial limit : their contribution is exactly @xmath284 a simple change of variable @xmath285 allows to conclude , thanks to the easy limit , uniform on compacts , @xmath286 our last task is to show that in the limit ( [ eqn : inclusionexclusion ] ) is equivalent to its @xmath267 term . by iterations of the hadamard - fisher inequality , lemma [ lem : fischer ] , @xmath287 the contribution of all terms with @xmath288 in ( [ eqn : inclusionexclusion ] ) is therefore bounded by @xmath289 where the integration domain @xmath290 has size @xmath291 and @xmath292 . the first term of the above product converges , as previously proved ( it corresponds to @xmath267 ) , so the whole term goes to 0 as @xmath34 , concluding the proof . [ lem : uniformbounded ] there is a constant @xmath234 depending only on @xmath169 such that , for any @xmath293 and @xmath294 ( see ( [ eqn : defomega ] ) ) , @xmath295 as previously mentioned , the formula ( [ eqn : inclusionexclusion ] ) is true whenever , for all distinct @xmath296 and @xmath297 , @xmath298 , i.e. @xmath299 is in @xmath256 . using the hadamard - fisher inequality as in the proof of the previous lemma , @xmath251 is therefore bounded by @xmath300 with @xmath290 . once again , the hadamard - fisher inequality gives @xmath301 this gives the upper bound , uniform in @xmath302 , @xmath303 converging to @xmath304 as previously seen . a better upper bound in the previous proof can be obtained as follows . by a direct ensembles argument , @xmath251 is bounded by @xmath305 this also comes from the fact that the inclusion - exclusion series ( [ eqn : inclusionexclusion ] ) is alternate . we know by fisher - hadamard that this upper bound is lower than @xmath306 , which is interpreted as follows : @xmath307 converges to its limit from below , which is a sign of repulsion before the asymptotic independence . [ lem : neglectset ] let @xmath258 be the complement of @xmath308 in @xmath257 ( see ( [ eqn : defomega ] ) ) . then @xmath309 _ proof . _ let @xmath310 , and note @xmath311 the set of these points . r.5 for notational convenience , one can suppose @xmath312 . a set of points @xmath313 is said to be a cluster of @xmath314 if , for all @xmath315 , @xmath316 . let @xmath317 be the set of points which can not be included in a maximal cluster : @xmath318 this way we get a partition @xmath319 where there are @xmath320 maximal clusters @xmath321 . suppose that @xmath322 where @xmath323 . then the following obvious bound holds : @xmath324 where we used the hadamard - fisher inequality . this last product of @xmath325 elements is bounded , uniformly in @xmath261 , as it is equal to the @xmath325th power of @xmath326 which converges . concerning the first product , we analyze each term in the following way . for any @xmath327 , the correlation @xmath328 is a @xmath329 determinant of @xmath330 . suppose that all arguments of @xmath328 are included in an interval of size @xmath331 . by subtracting the first row from the @xmath332 others one obtains that the first row is @xmath247 and the others are @xmath333 , because @xmath277 . hence @xmath334 . as a consequence , as all points of a cluster @xmath335 are within distance @xmath291 , @xmath336 this leads to the upper bound @xmath337 but the size of the set of points @xmath310 with such a clusters configuration is @xmath338 , because there needs to be @xmath339 small gaps ( i.e. of size @xmath291 ) between successive @xmath40 s . hence the total contribution of such clusters to the integral over @xmath258 is @xmath340 which goes top 0 because @xmath341 ( with equality if all clusters have only one point ) and @xmath342 ( @xmath343 is not possible : this would mean that @xmath344 ) . + as previously explained , the three above lemmas conclude the proof of theorem [ thm : unitary ] . note that the events @xmath345 and @xmath346 are almost surely the same ( @xmath347 ) . the independence property of the limit poisson point process @xmath35 in disjoint subsets therefore yields @xmath348 where we noted that , for any interval @xmath349 , @xmath350 is a poisson random variable with parameter @xmath351 . a straightforward simplification of the above products gives the expected result . concerning the limiting density of @xmath352 , we proceed in two steps . first , from formula ( [ eqn : jointlarge ] ) , the joint density of @xmath353 is , in the limit and on the simplex @xmath354 , proportional to @xmath355 consequently , @xmath356 so the density of @xmath46 is proportional to @xmath2 . the small gaps asymptotics for the @xmath18 are obtained exactly in the same way as for the unitary group . the only difference is that the determinantal kernel is not translation invariant anymore , leading to some complications . more precisely , let @xmath357 be the hermite polynomials , more precisely the successive monic orthogonal polynomials with respect to the gaussian weight @xmath358 . following @xcite , where the following results on the determinantal aspect of the @xmath18 can be found , we introduce the functions @xmath359 then the set of point @xmath360 with law ( [ eqn : gue ] ) is a determinantal point process with kernel ( with respect to the lebesgue measure on @xmath171 ) given by @xmath361 we will now discuss why lemma [ lem : nosucsmall ] till lemma [ lem : neglectset ] still hold in the case of the @xmath18 ensemble , restricted to @xmath362 for a given @xmath71 . note that the hadamard - fischer inequality , an important tool for the proof of all the following lemmas , still holds for the gaussian unitary ensemble because the set of its eigenvalues is a determinantal process . * we now note @xmath363 and its thinning @xmath229 obtained from @xmath213 by only keeping the eigenvalues @xmath364 for which @xmath365 . the absence of successive small gaps , lemma [ lem : nosucsmall ] , only requires in its proof that @xmath366 , @xmath367 . this is proved in lemma [ lem : infinitenorms ] . * the analogue of the simple convergence of the correlation function associated to the thinned point process , lemma [ lem : simpleconvergence ] , is now , for any distinct points @xmath368 in @xmath362 , @xmath369 the proof , in the same way as the unitary case , requires uniform bounds on the partial derivatives of the kernel , given in lemma [ lem : infinitenorms ] , and the asymptotics @xmath370 which is a direct consequence of lemma [ lem : taylorexpansion ] . * the uniform boundness result , lemma [ lem : uniformbounded ] , only requires the hadamard - fischer inequality , lemma [ lem : fischer ] , which applies to any determinantal point process . * finally , we give the analogue of lemma [ lem : neglectset ] in the following way . let @xmath371 and @xmath372 be its complement in @xmath257 , where @xmath78 is any borel set in @xmath362 . @xmath373 the only estimate necessary to this result is that the first order partial derivative of the kernel is uniformly @xmath374 , which is one of the estimates of the following lemma [ lem : infinitenorms ] . [ lem : infinitenorms ] let @xmath71 . uniformly for @xmath375 , @xmath376 is @xmath374 , the first order partial derivatives of @xmath377 are @xmath378 and the second order ones are @xmath379 . moreover , under the additional condition @xmath380 , @xmath376 is uniformly bounded , by a constant depending only on @xmath381 and @xmath382 . from the plancherel - rotach asymptotics for the hermite polynomials ( theorem 8.22.9 in @xcite ) , for any nonnegative integer @xmath0 , @xmath383 is @xmath384 , uniformly in @xmath385 . consequently , if @xmath386 , from the first line of ( [ eqn : kernelhermite ] ) , @xmath377 is uniformly @xmath387 . to prove the first assertion of the lemma , we use the stability property of the functions @xmath388 by derivation ( see @xcite lemma 3.2.7 ) : @xmath389 injecting this expression of @xmath390 in the last two lines of ( [ eqn : kernelhermite ] ) and using @xmath391 yields @xmath392 uniformly for @xmath375 . iterating this procedure , any partial derivative of @xmath377 of order @xmath0 are @xmath393 times a linear combination of factors of type @xmath394 , which shows that any partial derivative of order @xmath0 is @xmath395 . the plancherel - rotach asymptotics for hermite polynomials also yield a precise evaluation of the correlation functions evaluated at close points , in particular the following result . [ lem : taylorexpansion ] let @xmath71 , @xmath396 . then as @xmath34 , uniformly in @xmath385 , @xmath397 , @xmath398 the intuition for this result is as follows . from the well - known convergence to the sine kernel , @xmath399 if this convergence is sufficiently uniform , one expects for small @xmath400 a taylor expansion @xmath401 injecting this expansion in @xmath402 gives the expected result . to justify the accuracy of the sine kernel approximation for @xmath400 close to 0 , we rely on corollary 1 in @xcite ( more general asymptotics in the full complex plane for , in particular , hermite polynomials were obtained in @xcite ) : using the plancherel - rotach asymptotics of hermite polynomials the authors show that . ] @xmath403 uniformly for @xmath404 in @xmath405 where @xmath406 , @xmath407 , @xmath408 , @xmath409 , @xmath410 . as @xmath411 and @xmath412 , a simple expansion yields @xmath413 for @xmath414 , this implies @xmath415 and the expected result for @xmath416 . the above discussion concludes the proof of theorem [ thm : g ] . its corollary [ cor : smallestgue ] follows exactly in the same way as corollary [ cor : smallestunitary ] from theorem [ thm : unitary ] . corollary [ cor : smallestlocation ] is a straightforward consequence of the scission of the limiting lvy measure : if @xmath417 is a poisson point process with measure @xmath418 , conditionally to the @xmath419 s the @xmath420 s are distributed independently of each other , independently of the @xmath419 s , as a poisson point process with intensity proportional to @xmath421 . in particular , in our situation , the abscissa associated to the minimal ordinate is distributed with density proportional to @xmath422 to evaluate the extreme gaps , we first investigate the queuing distribution of one given spacing . a large part of the literature concerns the probability of having no eigenvalues in a given interval ( e.g. @xcite ) . a rigorous derivation of the queuing distribution for the large gaps was only given recently thanks to the steepest descent method for riemann hilbert problems @xcite or by operator theory tools for toeplitz determinants @xcite . all the asymptotics we need in the following are direct consequences of the precise analysis of @xmath428 . such estimates were first given by widom @xcite , for the following equation ( [ eqn : est1 ] ) without the error term , ( [ eqn : est2 ] ) first by krasovsky @xcite and ( [ eqn : est1 ] ) together with ( [ eqn : est2 ] ) by deift & al @xcite : they prove that for some sufficiently large @xmath429 and any @xmath154 , uniformly in @xmath430 , @xmath431 for an explicit constant @xmath432 , which remained conjectural for 20 years . from this , we can give upper bounds on @xmath433 , the expectation of the number of gaps greater than @xmath400 . from lemma [ lem : meh ] , @xmath437 evaluated for @xmath438 the first term is evaluated thanks to ( [ eqn : est2 ] ) : it is @xmath439 uniformly in @xmath440 . concerning the second term , ( [ eqn : est1 ] ) and the inequality @xmath441 on @xmath442 imply that it is @xmath443 , concluding the proof . note that , substituting @xmath444 in the asymptotics ( [ eqn : est1 ] ) and ( [ eqn : est2 ] ) yields @xmath445 where the @xmath446 terms are uniform in @xmath447 $ ] , for any given positive @xmath448 and @xmath449 . hence , if @xmath450 , the expected number of gaps greater than @xmath451 goes to @xmath152 . if , @xmath452 , this goes to @xmath453 . the transition in the number of large gaps at @xmath454 is a strong clue for theorem [ thm : largestgaps ] , but it is not sufficient : concluding the proof requires additional knowledge about the correlations between gaps , lemma [ lem : negcor ] , assumed for the moment . take @xmath154 and @xmath5 . we note @xmath455 where @xmath99 . then @xmath456 and we aim to show that the last three terms converge to 0 ( the value @xmath457 is arbitrary , any angle strictly smaller than @xmath458 would be appropriate for this proof ) . first , decompose the unit circle into 8 fixed angles of size @xmath459 . if @xmath460 , on of these arcs is free of eigenvalues . from ( [ eqn : est1 ] ) , the probability of this event decreases exponentially , so the last term in ( [ eqn:3terms ] ) converges to 0 . moreover , integrating by parts , @xmath461 because @xmath462 needs to be shorter than @xmath463 . the probability that @xmath462 is greater that @xmath400 is obviously shorter than @xmath464 , the expectation of the number of gaps greater than @xmath465 . hence the above quantity goes to @xmath152 thanks to the uniform estimate of lemma [ lem : boundexpectation ] . finally , showing that @xmath466 requires an additional argument , the negative correlation property for empty sets events , lemma [ lem : negcor ] . in particular , this negative correlation implies the following result . [ lem : varianceexpectation ] consider a set of disjoint arcs on the unit circle . let @xmath467 be the number of such intervals free of eigenangles , i.e. those @xmath468 s such that @xmath469 . then @xmath470 . this is a straightforward consequence of lemma [ lem : negcor ] : as @xmath213 is a determinantal point process , for disjoints @xmath78 and @xmath471 , @xmath472 . hence , noting @xmath473 the number of initial arcs , @xmath474 as expected . consider now a number @xmath473 of disjoint intervals @xmath475 of length @xmath476 in @xmath194 . we can find @xmath477 of them . let @xmath467 be the number of such intervals free of eigenangles , i.e. those @xmath468 s such that @xmath469 . if there are less than @xmath60 gaps larger than @xmath476 , either there are less than @xmath60 intervals @xmath468 s free of eigenangles , or there is a gap between successive eigenangles containing two intervals : @xmath478 this last term is bounded by the expectation of the number of gaps greater than @xmath479 ; from ( [ eqn : transition ] ) this goes to 0 if @xmath480 ( true for @xmath160 sufficiently small ) . concerning @xmath481 , first note that by the estimate ( [ eqn : transition ] ) , @xmath482 , so @xmath483 . this allows to use chebyshev s inequality , for sufficiently large @xmath17 : @xmath484 where we used lemma [ lem : varianceexpectation ] in the last inequality . this last term is equivalent to @xmath485 , thus going to @xmath152 , which concludes the proof . for the proof of theorem [ thm : largestgapsgue ] , precise asymptotics like ( [ eqn : est1 ] ) and ( [ eqn : est2 ] ) related to unitary groups are not available in the @xmath18 context . this difficulty can be overcome , our main observation being lemma [ lem : compfredholm ] : the probability that an interval is free of eigenvalues is equivalent in the @xmath486 and @xmath48 cases , up to a normalization , if the interval size is shorter than the expected extreme gap size . the proof relies on a comparison of the fredholm determinants associated to @xmath377 and @xmath330 . the rest of the proof is similar to the one concerning @xmath48 . indeed , consider an @xmath154 , @xmath5 , @xmath487 the @xmath60th largest gap in @xmath78 , and @xmath488 we then decompose @xmath489 this last expectation is , by integration by parts , @xmath490 the probability @xmath491 is lower than the expectation of the number of gaps greater than @xmath492 . lemma [ lem : inthetailgue ] therefore yields @xmath493 as @xmath34 . concerning @xmath494 , we proceed as for the unitary group : for @xmath495 with @xmath496 , consider @xmath497 disjoint intervals of length @xmath498 included in @xmath78 . let @xmath499 be the number of these intervals free of eigenvalues . then @xmath500 from lemma [ lem : inthetailgue ] this last probability goes to @xmath152 as @xmath34 if @xmath480 , @xmath501 . moreover , from lemma [ lem : expectgue ] , @xmath502 for some @xmath503 depending only on @xmath160 and @xmath78 . moreover , from the negative correlation property lemma [ lem : negcor ] and the same reasoning as lemma [ lem : varianceexpectation ] , @xmath504 as @xmath62 , from ( [ eqn : expboundgue ] ) @xmath505 for sufficiently large @xmath17 , which allows to use chebyshev s inequality : @xmath506 the last inequality being ( [ eqn : varianceexpectationgue ] ) . from ( [ eqn : expboundgue ] ) this last term is equivalent to @xmath507 and going to 0 , as @xmath34 , concluding the proof . concerning the gaussian unitary ensemble , the same type of asymptotics hold , being a direct consequence of formula ( [ eqn : gueexpansion ] ) . note that when taking @xmath516 the speed of convergence to the sine kernel is @xmath517 for the unitary group , much better that @xmath518 for the @xmath18 , whose correlation kernel is not translation invariant . by inclusion - exclusion , the probability that a determinantal point process with kernel @xmath11 has no points in a measurable subset @xmath169 is the fredholm determinant ( see e.g. lemma 3.2.4 in @xcite ) @xmath522 to compare empty sets probabilities , we therefore need to compare fredholm determinants . a classical inequality ( see e.g. @xcite chap iv , ( 5.14 ) ) is @xmath523 is the trace norm of a nuclear operator @xmath136 . however , for positive ( like @xmath169 or @xmath524 ) operators @xmath525 the trace norm is @xmath526 , but for non - positive operators , like @xmath527 , the trace norm is difficult to evaluate , even for a compactly supported continuous kernel @xmath11 . however , in such a case , the hilbert - schmidt norm is computable : @xmath528 ) , we will use the modified carleman - fredholm determinant @xmath529 and the inequality @xmath530 ) , after a simple change of variables , the probability that there are no eigenvalues in @xmath531 $ ] is equal to @xmath532 in the same way , the probability that there are no eigenvalues in @xmath533 $ ] is @xmath534 hence the inequality ( [ eqn : ineqhs ] ) will be applied with @xmath169 and @xmath524 integral operators with respective kernel @xmath535 and @xmath536 from lemma [ lem : asymptsine ] , the infinite norm between the two kernels above is @xmath537 , so by ( [ eqn : hsnorm ] ) , integrating on a domain of area @xmath538 , @xmath539 moreover , consider a parameter @xmath540 and decompose @xmath541 from lemma [ lem : asymptsine ] , if @xmath542 then @xmath543 is smaller than @xmath544 , and when @xmath545 it is bounded by @xmath546 . consequently , @xmath547 by choosing @xmath548 . in the same way , hence @xmath550 , here on the diagonal , @xmath551 and the same for @xmath552 . finally , write @xmath553 and @xmath554 show that the first term is bounded by @xmath555 . moreover , @xmath556 is a probability , so bounded by 1 , and the estimates of @xmath557 and @xmath552 , formula ( [ eqn : estimatetrace ] ) , show that @xmath558 is @xmath559 , concluding the proof . the first step allows reasoning on fixed intervals instead of gaps . more precisely , note @xmath82 $ ] with @xmath496 , and consider the intervals of length @xmath565 by successive slips of size @xmath566 : @xmath567,\ 0\leq k\leq p_n=\lfloor ( b - a)/s_n\rfloor.\ ] ] there an injective map associating to any eigenvalues gap of size at least @xmath568 an interval @xmath569 included in this gap , for example the one with lower index . consequently , @xmath570 where we use the abbreviation @xmath571 . the second step consists in obtaining uniform upper bounds for these empty intervals probabilities . for this purpose , the negative correlation property , lemma [ lem : negcor ] , is used by partitioning the interval @xmath569 : @xmath572 where * @xmath573 has length @xmath574 , with @xmath575 to be chosen positive but sufficiently small compared to @xmath160 ; * @xmath576 has length @xmath577 ; * the intervals @xmath578 all have length @xmath579 and their number is @xmath580 ; * @xmath189 is a residual interval . the negative correlation property yields @xmath581 this now can be upper - bounded using lemma [ lem : compfredholm ] because all intervals are shorter than @xmath582 : for example , noting @xmath520 one extremity of @xmath573 , @xmath583=0\right)+n^{{\mathrm{o}}(1)-1}\\ \leq \operatorname{\mathbb{p}}^{\operatorname{u}(n)}\left([0,(1-{\varepsilon}')\alpha_n]=0\right)+n^{{\mathrm{o}}(1)-1}=n^{-(1-{\varepsilon}')^2+{\mathrm{o}}(1)}\end{gathered}\ ] ] with @xmath446 not depending on the index @xmath0 , and where the last estimate relies on ( [ eqn : est1 ] ) . in the same way , @xmath584 gathering all these results , @xmath585 we can chose @xmath575 sufficiently small such that @xmath586 ( e.g. @xmath587 ) . for such a choice , for sufficiently large @xmath17 not depending on @xmath588 the above exponent is smaller than @xmath589 for some @xmath396 . such a function is smaller than @xmath590 on @xmath563 , for some @xmath591 . [ lem : expectgue ] let @xmath82 $ ] with @xmath496 , @xmath560 , @xmath592 and @xmath561 . consider a maximal number @xmath593 of disjoint intervals @xmath475 of length @xmath594 included in @xmath78 let @xmath595 be the number of those intervals containing no eigenvalues . then @xmath596 for some @xmath503 depending only on @xmath160 and @xmath78 we first take a restricted interval to avoid the fluctuations in the spectral measure : there is @xmath597 with @xmath598 such that @xmath599 then , for a given interval @xmath600 $ ] included in @xmath601 , from lemma [ lem : compfredholm ] , @xmath602,1\leq i\leq n\right)+n^{{\mathrm{o}}(1)-1}\ ] ] with @xmath446 uniform in @xmath520 . from ( [ eqn : restrictedi ] ) this is greater than @xmath603,1\leq i\leq n\right)+n^{{\mathrm{o}}(1)-1}=n^{-\left(1-\frac{{\varepsilon}}{2}\right)^2+{\mathrm{o}}(1)}.\ ] ] from ( [ eqn : est1 ] ) . there are @xmath604 intervals @xmath468 s included in @xmath601 , so as @xmath34 @xmath605 with @xmath606 for example . in the context of a point process on a finite set @xmath607 , let @xmath186 and @xmath608 be distinct disjoint subsets of @xmath607 , and write @xmath609 for the event that the elements of @xmath186 are free of particles . shirai and takahashi @xcite showed that for determinantal point processes , the empty sets events are negatively correlated : @xmath610 this negative association property has received a considerable attention in the past few years , in the context of asep for example . still for discrete determinantal point processes , formula @xmath611 was generalized to all increasing events @xcite , and general criteria for the negative association property were given in @xcite . the following continuous analogue of ( [ eqn : negdiscr ] ) holds . it can be shown by a simple discretization , relying on results from @xcite . we give another justification , which relies on a work of georgii and yoo @xcite . [ lem : negcor ] let @xmath213 be the point process associated to the eigenvalues of haar distributed unitary matrix ( resp . an element of the @xmath18 ) . let @xmath612 and @xmath613 be compact disjoint subsets of @xmath614 ( resp . @xmath171 ) . then @xmath615 the general negative correlation result , corollary 3.3 in @xcite , requires a locally trace class operator @xmath11 with a restriction on its spectrum . more precisely , consider the unitary case , the @xmath18 proof , being similar . for an operator @xmath11 acting on @xmath616 , if @xmath617 $ ] , there exist a unique determinantal point process @xmath618 with kernel @xmath11 ( see @xcite ) ; under the additional hypothesis @xmath619 , georgii and yoo proved that for any compact and disjoint borel sets @xmath620 , @xmath621 in the case of haar distributed unitary matrices , @xmath622 is a nuclear operator with kernel @xmath623 , and defines a projection : @xmath624 is in its spectrum , and the general statement does not directly apply . to care of this minor problem , look at the restriction @xmath625 of @xmath330 to a compact subset @xmath186 of @xmath194 ( @xmath626 , @xmath627 being the projection on @xmath186 ) . suppose that the set @xmath628 has a nonempty interior . as a projection of @xmath330 , @xmath625 is still non - negative and trace class . as for any determinantal point process , @xmath629 in the sense of fredholm determinants of a trace class operator . suppose that @xmath630 has an eigenvalue @xmath631 . then by choosing @xmath632 , ( [ eqn : freddet ] ) yields @xmath633 in each case , this is absurd because the joint law of the eigenvalues is absolutely continuous with respect to the lebesgue measure on @xmath634 , and @xmath635 has a nonempty interior : both quantities need to be strictly positive . hence @xmath636 is a trace class operator with spectrum in @xmath637 , and the result from @xcite applies . another way to prove lemma [ lem : negcor ] is as follows : the inequality can be stated for the determinantal point process with kernel @xmath638 with @xmath639 , and then the inequality remains true for @xmath330 by continuity of the application @xmath640 in the set of trace class operators , by ( [ eqn : ineqtrace ] ) . anderson , a. guionnet , o. zeitouni , an introduction to random matrices , cambridge university press , 2009 . g. ben arous , a. guionnet , `` wigner matrices '' , chapter 21 , handbook on random matrix theory , editors : g.akemann , j.baik , p.di francesco , to appear oxford university press , 2010 . g. ben arous , a. kuptsov , rem universality for random hamiltonians , preprint . j. borcea , p. brndn . t. liggett , negative dependence and the geometry of polynomials , j. amer . ( 2009 ) , 521 - 567 . h. bui , large gaps between consecutive zeros of the riemann zeta - function , preprint . d. bump , lie groups , springer , 2004 . p. deift , some open problems in random matrix theory and the theory of integrable systems . contemporary mathematics ( 2008 ) 458:419 - 30 . p. deift , t. kriecherbauer , k.t - r mclaughlin , s. venakides , x. zhou : uniform asymptotics for polynomials orthogonal with respect to varying exponential weights and applications to universality questions in random matrix theory . pure appl . 52 , 1335 - 1425 ( 1999 ) p. deift , a. its , i. krasovsky , x. zhou , the widom dyson constant for the gap probability in random matrix theory , journal of computational and applied mathematics volume 202 , issue 1 , 1 may 2007 , pages 26 - 47 . p. deift , t. nanda , c. tomei , ordinary differential equations and the symmetric eigenvalue problem , siam j. numer . ( 1983 ) , pp . 1 - 22 . b. delyon , j. yao , on the spectral distribution of gaussian random matrices , acta mathematicae sinica , english series vol . 22 , no . 2 ( 2006 ) 297 - 312 . p. diaconis . patterns in eigenvalues : the 70th josiah williard gibbs lecture . bull . 40 , 155178 ( 2003 ) . f. dyson , statistical theory of the energy levels of complex systems , i , ii , iii , j. math . phys . 3 140 ( 1962 ) . t. ehrhardt , dyson s constant in the asymptotics of the fredholm determinant of the sine kernel , comm . 262 , 317 - 341 ( 2006 ) . l. erds , b. schlein , h .- universality of random matrices and local relaxation flow , preprint . georgii , h.j . yoo , conditional intensity and gibbsianness of determinantal point processes , journal of statistical physics , vol . 118 , nos 1/2 , january 2005 . i. goldberg , s. goldberg , n. krupnik , traces and determinants of linear operators , operator theory : advances and applications , vol . 116 , birkhuser , 2000 . x. gourdon , computation of zeros of the zeta function , 2004 , unpublished . r. horn , c. johnson , matrix analysis , cambridge university press , 1985 . k. johansson , random matrices and determinantal processes , lectures given at the summer school on mathematical statistical mechanics in july 05 at ecole de physique , les houches . o. kallenberg , random measures , 3rd , rev . and enl . ed . new york : academic press , 1983 . katz , p. sarnak , random matrices , frobenius eigenvalues and monodromy , american mathematical society colloquium publications , 45 . american mathematical society , providence , rhode island , 1999 . j.p . keating , n.c . snaith , random matrix theory and @xmath641 , comm . 214 , p 57 - 89 , 2000 . i. v. krasovsky , gap probability in the spectrum of random matrices and asymptotics of poly orthogonal on an arc of the unit circle , international mathematics research notices 2004 ( 25 ) : 1249 - 1272 . r. lyons , determinantal probability measures , publ . hautes tudes sci . 98 ( 2003 ) , 167212 . mehta , random matrices , third edition . pure and applied mathematics ( amsterdam ) , elsevier / academic press , amsterdam , 2004 . montgomery , the pair correlation of zeros of the zeta function , analytic number theory ( st . louis , 1972 ) , proc . sympos . pure math . 24 , amer . ( providence ) , 181 - 193 ( 1973 ) . a.m. odlyzko , on the distribution of spacings between zeros of the zeta function , math . , 48 ( 1987 ) , pp . 273 - 308 . t. shirai , y. takahashi , random point fields associated with certain fredholm determinants ii : fermion shifts and their ergodic and gibbs properties , ann . volume 31 , number 3 ( 2003 ) , 1533 - 1564 . a. soshnikov , level spacings distribution for large random matrices : gaussian fluctuations annals of mathematics , vol . 148 , pp.573 - 617 , 1998 . a. soshnikov , determinantal random point fields russian mathematical surveys , vol.55 , no.5,923 - 975 , 2000 a. soshnikov , statistics of extreme spacings in determinantal random point processes moscow math . j. , vol . 5 , no.3 , 705 - 719 , 2005 . g. szego , orthogonal polynomials , 4th ed . , american mathematical society , colloquium publications , vol . 23 , american mathematical society , rhode island , 1975 . t. tao , v. vu , random matrices : universality of local eigenvalue statistics , to appear in acta mathematica . j. vinson , closest spacing of eigenvalues , phd thesis , princeton university , 2001 . h. widom , the strong szeg limit theorem for circular arcs , indiana univ . j. 21 , 277 - 283 ( 1971 ) .

we have investigated the influence of a velocity shear surface on the linear and non - linear development of the cd kink instability of force - free helical magnetic equilibria in 3d . in this study we follow the temporal development within a periodic computational box and concentrate on flows that are sub - alfvnic on the cylindrical jet s axis . displacement of the initial force - free helical magnetic field leads to the growth of cd kink instability . we find that helically distorted density structure propagates along the jet with speed and flow structure dependent on the radius of the velocity shear surface relative to the characteristic radius of the helically twisted force - free magnetic field . at small velocity shear surface radius the plasma flows through the kink with minimal kink propagation speed . the kink propagation speed increases as the velocity shear radius increases and the kink becomes more embedded in the plasma flow . a decreasing magnetic pitch profile and faster flow enhance the influence of velocity shear . simulations show continuous transverse growth in the nonlinear phase of the instability . the growth rate of the cd kink instability and the nonlinear behavior also depend on the velocity shear surface radius and flow speed , and the magnetic pitch radial profile . larger velocity shear radius leads to slower linear growth , makes a later transition to the nonlinear stage , and with larger maximum amplitude than occur for a static plasma column . however , when the velocity shear radius is much greater than the characteristic radius of the helical magnetic field , linear and non - linear development can be similar to the development of a static plasma column .

1104.0549

relativistic jets occur in black hole binary star systems ( microquasars ) ( e.g. , mirabel & rodorguez 1999 ) , occur in active galactic nuclei ( agn ) ( e.g. , urry & padovani 1995 ; ferrari 1998 ; meier et al . 2001 ) , can be associated with neutron stars and pulsar wind nebulae , e.g. , crab nebula jet ( weisskopf et al . 2000 ) , and are thought responsible for the gamma - ray bursts ( grbs ) ( e.g. , zhang & mszros 2004 ; piran 2005 ; mszros 2006 ) . it is thought that these jets are powered and collimated hydromagnetically ( e.g. , blandford 2000 ) . such poynting flux dominated outflows arise on magnetic field lines threading the horizon of a rotating black hole ( blandford & znajek 1977 ; blandford & payne 1982 ) , and form as magnetic flux is pumped by the inflowing gas or generated by dynamo action in a surrounding accretion disk ( e.g. , kudoh & kaburaki 1996 ) . the observed high degree of jet collimation is generally attributed to hoop stress from the toroidal magnetic field acting in concert with an external pressure . jet generation simulations using general relativistic magnetohydrodynamics ( grmhd ) ( e.g. , de villiers et al . 2003 , 2005 ; hawley & krolik 2006 , mckinney & gammie 2004 ; mckinney 2006 ; mckinney & blandford 2009 ; beckwith et al . 2008 ; hardee et al . 2007 ; komissarov & barkov 2009 ; penna et al . 2010 ) codes show development of magneto - rotational instability ( mri ) ( balbus & hawley 1998 ) and angular momentum transfer in the accretion disk , leading to diffusion of matter and magnetic field inwards , and unsteady outflows near a centrifugally supported funnel wall . in general , grmhd simulations with spinning black holes indicate jet production consisting of a poynting - flux high lorentz factor spine with @xmath0 , and a matter dominated sheath with @xmath1 possibly embedded in a lower speed , @xmath2 , disk / coronal wind . circumstantial evidence such as the requirement for large lorentz factors suggested by the tev bl lacs when contrasted with much slower observed motions ( ghisellini et al . 2005 ) suggests such a spine - sheath morphology , although alternative interpretations are also possible ( georganopoulos & kazanas 2003 ; stern & poutanen 2008 ; bromberg & levinson 2009 ; giannios et al . 2009 ) . strongly magnetized relativistic outflows are typically produced from rotating bodies ( neutron stars , black holes and accretion disks ) . a toroidal magnetic field ( @xmath3 ) is wound up in such outflows and in the far zone becomes dominant because the poloidal field ( @xmath4 ) falls off faster with expansion and distance . in configurations with strong toroidal magnetic field , the current - driven ( cd ) kink mode is unstable . this instability excites large - scale helical motions that can strongly distort or even disrupt the system . for static cylindrical force - free equilibria , the well - known kruskal - shafranov criterion indicates that the instability develops if the length of the plasma column , @xmath5 , is long enough for the field lines to go around the cylinder at least once ( e.g. , bateman 1978 ) : @xmath6 . for relativistic force - free static configurations , the linear instability criteria have been studied by several authors ( istomin & pariev 1994 , 1996 ; begelman 1998 ; lyubarskii 1999 ; tomimatsu et al . 2001 ; narayan et al . 2009 ) . in a more realistic case , rotation and shear could significantly affect the instability criterion . twisted structures are observed in many agn jets on sub - parsec , parsec and kiloparsec scales ( e.g. , gmez et al . 2001 ; lobanov & zensus 2001 ) . non - relativistic and relativistic simulations of magnetized jet formation and/or propagation have showed helical structures attributed to cd kink instability ( e.g. , lery et al . 2000 ; ouyed et al . 2003 ; nakamura & meier 2004 ; nakamura et al . 2007 ; moll et al . 2008 ; moll 2009 ; mckinney & blandford 2009 ; carey & sovinec 2009 ) . in the absence of cd kink instability and resistive relaxation , helical structures may be attributed to the kelvin - helmholtz ( kh ) helical instability driven by velocity shear at the boundary between the jet and the surrounding medium ( e.g. , hardee 2004 , 2007 ) or triggered by precession of the jet ejection axis ( begelman et al . it is still not clear whether current driven , velocity shear driven or jet precession is responsible for the observed structures , or whether these different processes are responsible for the observed twisted structures at different spatial scales . in a previous paper we performed 3d relativistic mhd simulations and investigated the development of cd kink instability in a static plasma column with a force - free helical magnetic field ( mizuno et al . we found that the initial configuration was strongly distorted but not disrupted by the kink instability . although static configurations ( or more generally rigidly moving flows considered in the proper reference frame ) are the simplest ones for studying the basic properties of the kink instability , in a realistic case , rotation and shear motions could significantly affect the cd kink instability . in this paper we investigate the influence of a velocity shear surface on the stability and nonlinear behavior of relativistic sub - alfvnic flow . because sub - alfvnic flow is stable to the kelvin - helmholtz ( kh ) instability , we can focus on the development of cd kink instability . in this paper we describe the numerical method and setup used for our simulations in 2 , present our results in 3 , in 4 compare our results to instability expectations and conclude , and in the appendix present multi - dimensional numerical code tests . in order to study time evolution of the cd kink instability in the relativistic mhd ( rmhd ) regime , we use the 3d grmhd code `` raishin '' in cartesian coordinates . raishin is based on a @xmath7 formalism of the general relativistic conservation laws of particle number and energy - momentum , maxwell s equations , and ohm s law with no electrical resistance ( ideal mhd condition ) in a curved spacetime ( mizuno et al . 2006 ) . in the raishin code , a conservative , high - resolution shock - capturing scheme is employed . the numerical fluxes are calculated using the hll approximate riemann solver , and flux - interpolated constrained transport ( flux - ct ) is used to maintain a divergence - free magnetic field . the raishin code performs special relativistic calculations in minkowski spacetime by choosing the appropriate metric . the raishin code has proven to be accurate to second order and has passed a number of numerical tests including highly relativistic cases and highly magnetized cases in both special and general relativity ( mizuno et al . the results of multidimensional numerical test problems are shown in the appendix . the multidimensional numerical test problems show that the mc slope - limiter scheme performs best on most of the tests and we have used this scheme for reconstruction . in our simulations we choose a force - free helical magnetic field for the initial configuration ( mizuno et al . a force - free configuration is a reasonable choice for the strong magnetic field cases that we study here . in general , the force - free equilibrium of a static cylinder is described by the equation @xmath8 in particular we choose the following form for poloidal ( @xmath9 ) and toroidal ( @xmath3 ) components of magnetic field determined in the laboratory frame @xmath10^{\alpha}}~,\ ] ] @xmath11^{\alpha } } \sqrt { { [ 1 + ( r / a)^{2}]^{2 \alpha } -1 - 2 \alpha ( r / a)^2 \over 2 \alpha -1}}~,\ ] ] where @xmath12 is the radial position in cylindrical coordinates normalized by a simulation scale unit @xmath13 , @xmath14 parameterizes the magnetic field amplitude , @xmath15 is the characteristic radius of the column , and @xmath16 is the pitch profile parameter . the pitch parameter , @xmath17 , determines the radial profile of the magnetic pitch , and larger @xmath18 indicates increasing helical pitch of the magnetic field lines . with our choice for the force - free field , the pitch parameter can be written as @xmath19^{2 \alpha } -1 -2 \alpha ( r / a)^{2 } } } ~.\ ] ] if the pitch profile parameter @xmath20 , the magnetic helicity increases with radius . when @xmath21 , the magnetic helicity is constant and if @xmath22 , the magnetic helicity decreases . this definition for the pitch parameter and the force - free helical field has been chosen to be the same as that used in previous non - relativistic work ( appl et al . 2000 ; baty 2005 ) and in our previous relativistic work ( mizuno et al . 2009a ) for purposes of comparison . for our modestly relativistic sub - alfvnic flow speeds , the jet density and magnetic fields within the jet when determined in the jet flow frame are reduced slightly relative to their values in the laboratory frame by the flow lorentz factor ( e.g. , anile 1989 ; komissarov 1999 ) . the simulation grid is periodic along the axial z direction . as a consequence the allowed axial wavelengths are restricted to @xmath23 , with @xmath24 a positive integer and @xmath25 is the grid length . the grid is a cartesian ( @xmath26 ) box of size @xmath27 with grid resolution of @xmath28 . the grid resolution is the same in all directions . in simulations we choose a characteristic radius , @xmath29 . in terms of @xmath15 , the simulation box size is @xmath30 and the allowed axial wavelengths are restricted to @xmath31 . we impose outflow boundary conditions on the transverse boundaries at @xmath32 ( @xmath33 ) . this simulation grid is the same as that used for case b in mizuno et al . ( 2009a ) . we checked the influence of grid resolution for the case of a static plasma column using four different grid resolutions from 20 to 60 computational zones per simulation length unit @xmath34 in mizuno et al . ( 2009a ) . we found that growth does depend on grid resolution , and our choice of @xmath35 is sufficient to capture the growth of cd kink instability . we consider a low gas pressure medium with constant @xmath36 in units of @xmath37 for the equilibrium state , and a non - uniform density profile decreasing with the magnetic field strength as , @xmath38 with @xmath39 . we choose a density decreasing @xmath40 in order to keep the alfvn speed inside the velocity shear surface above the flow speed . the equation of state is that of an ideal gas with @xmath41 , where @xmath42 is the specific internal energy density and the adiabatic index @xmath43 .. this condition is generally fulfilled in our simulations even though the sound speed is greater than the relativistic limit of @xmath44 at large radial position ( see fig . the region where the cd kink instability occurs is near the axis and in this region an adiabatic index @xmath43 is appropriate . we have checked the dependence of our simulation results on different adiabatic indicies ( see appendix a in mizuno et al . we found no significant difference in results for different adiabatic indicies . ] the specific enthalpy is @xmath45 . the magnetic field amplitude is @xmath46 in units of @xmath47 leading to a low plasma-@xmath48 near the axis . the sound speed is @xmath49 and the alfvn speed is given by @xmath50^{1/2}$ ] . we choose two different jet velocities : ( case s ) slow , @xmath51 and ( case f ) fast , @xmath52 . in order to study the effect of the velocity , we perform the simulations with four different velocity shear surface radii : @xmath53 ( @xmath54 ) , @xmath55 ( @xmath15 ) , @xmath56 ( @xmath57 ) , @xmath58 ( @xmath59 ) . results are compared to those for a static plasma column ( no flow ) as the reference . we also investigate the effect of different magnetic pitch profiles with : ( case cp ) constant pitch , @xmath60 , and ( case dp ) decreasing pitch , @xmath61 . the radial profiles of the magnetic field components , the magnetic pitch , and the sound and alfvn speeds determined in the lab frame for the different cases are shown in figure 1 . for all cases the sound and alfvn speeds on the axis are @xmath62 and @xmath63 , respectively , when determined in the lab frame . we have chosen constant and decreasing pitch cases only because previous simulation studies of the cd kink instability of a static plasma column show that increasing magnetic pitch is not too different from the constant pitch case ( mizuno et al . the jet velocity is sub - alfvnic inside the velocity shear radius for almost all cases with the exception of the constant pitch case with the largest shear radius @xmath64 and fastest jet velocity @xmath52 . in this paper we focus on the development of cd kink instability as the sub - alfvnic jet is stable to the velocity shear driven kelvin - helmholtz ( kh ) instability . to break the symmetry the initial mhd equilibrium configuration is perturbed by a radial velocity in all region with profile given by @xmath65 the amplitude of the perturbation is @xmath66 with radial width @xmath67 ( @xmath57 ) , and we choose @xmath68 and @xmath69 . this is identical to imposing @xmath70 , because of the symmetry between @xmath71 and @xmath72 pairs . the various different cases that we have considered are listed in table 1 . lcccc [ table1 ] cpsa/2 & 1.0 & 0.2 & 0.5 & constant + cpsa & 1.0 & 0.2 & 1.0 & constant + cps2a & 1.0 & 0.2 & 2.0 & constant + cps4a & 1.0 & 0.2 & 4.0 & constant + cpfa/2 & 1.0 & 0.3 & 0.5 & constant + cpfa & 1.0 & 0.3 & 1.0 & constant + cpf2a & 1.0 & 0.3 & 2.0 & constant + cpf4a & 1.0 & 0.3 & 4.0 & constant + cp0 & 1.0 & 0.0 & 0.0 & constant + dpsa/2 & 2.0 & 0.2 & 0.5 & decrease + dpsa & 2.0 & 0.2 & 1.0 & decrease + dps2a & 2.0 & 0.2 & 2.0 & decrease + dps4a & 2.0 & 0.2 & 4.0 & decrease + dp0 & 2.0 & 0.0 & 0.0 & decrease + figure 2 shows the time evolution of a density isosurface for constant helical pitch with @xmath51 and @xmath74 ( cps2a ) where the time , @xmath75 , is in units of @xmath76 ( light travel time across the largest velocity shear surface radius , @xmath77 , considered in this study ) . displacement of the initial force - free helical magnetic field by growth of the cd kink instability leads to a helically twisted magnetic filament wound around the density isosurface . in the nonlinear phase , helically distorted density structure shows continuous transverse growth and propagates in the flow direction . this propagation of the helical kink structure does not occur for a static plasma column ( mizuno et al . 2009a ) . in order to investigate the dependence of kink growth and propagation on the location of the velocity shear surface , we consider four velocity shear radii from @xmath78 to @xmath59 ( see table 1 ) . the effect of different radii on the growth of the cd kink for constant magnetic pitch and @xmath51 is shown in figure 3 . in figure 3 we show the time evolution of the volume - averaged kinetic , @xmath79 , and magnetic , @xmath80 , energy transverse to the @xmath81-axis determined within a cylinder of radius @xmath82 as an indicator of the growth of the cd kink instability . the wavelength of the kink is @xmath83 . note that according to the kruskal - shafranov criterion , the instability develops at @xmath84 . the instability growth rate reaches a maximum at @xmath85 , the exact coefficient being dependent on the transverse distribution of the density and magnetic pitch , and also possibly the location and magnitude of the velocity shear . for the case of constant pitch and uniform density appl et al . ( 2000 ) found @xmath86 and a corresponding growth rate of @xmath87 . in general , one can use the estimate @xmath88 in the rest frame of the kink . for a moving kink we might expect the temporal growth rate in the lab frame to be reduced with @xmath89 where @xmath90 is the moving kink lorentz factor , e.g. , narayan et al . change in the evolution of @xmath79 and @xmath80 indicate an initial linear growth phase at @xmath91 with duration depending on the velocity shear radius , and followed by a nonlinear evolution phase . in all cases , the initial growth phase is characterized by an exponential increase in @xmath79 by about 3 orders of magnitude to a maximum amplitude followed by a slow decline in the nonlinear phase . by fitting the linear portion of the slope in @xmath79 between the amplitudes of @xmath92 and @xmath93 we can determine an e - folding time where @xmath94 and @xmath95 is the time interval . the e - folding times can be found in table 2 in 4 . in general , the e - folding time increases as the velocity shear radius increases for @xmath96 . the e - folding time at @xmath97 decreases but is still significantly longer than for the static plasma column . the time evolution trend of @xmath80 is opposite to the time evolution of @xmath79 . @xmath80 gradually decreases in the early linear growth phase , then exhibits an initial rapid decrease into the nonlinear phase to a minimum followed by a slight increase at later times . at the smallest velocity shear radius the behavior of @xmath79 and @xmath80 are very similar to that of a static plasma column . increased difference in behavior from that of a static plasma column appears as the velocity shear radius increases to @xmath98 and @xmath57 . for these cases as the velocity shear radius increases the growth rate of the cd kink slows , and @xmath79 achieves a somewhat higher maximum amplitude with a later transition to the nonlinear stage . as the shear radius increases @xmath80 exhibits a more gradual decline in the transition to the nonlinear stage than for the static plasma column . however , for the largest velocity shear radius , @xmath64 , the difference relative to the static plasma column in the linear and early non - linear phase is reduced , and the growth rate of the cd kink instability is faster and with lower maximum @xmath79 than for the cases with velocity shear radius @xmath98 and @xmath74 . but note that @xmath79 becomes slightly larger than the static case at the longest comparable simulation times . when the velocity shear radius is much larger than the characteristic radius of the helical magnetic field , we would expect the growth of cd kink instability to approach that of a static plasma column moving with respect to the observer . however , in the flow reference frame the alfvn speed for the sub - alfvnic jet is not the same as that for a static plasma column in addition to a small relativistic clock effect . thus , we do not expect the initial growth rate of largest jet radius case to perfectly match that of the static plasma column as determined in the observer ( simulation ) rest frame . the effect of the velocity shear is greatest for the case with velocity shear radius @xmath74 but also significant when @xmath98 and @xmath59 . figure 4 shows a density isosurface at @xmath99 for the constant pitch cases with @xmath51 at different velocity shear radii . in the linear growth phase , the behavior of the growing kink is almost the same for different velocity shear radius . however , in the nonlinear phase , the behavior of the kink is different for different velocity shear radius . for the smallest radius , @xmath78 , the kink does not propagate along the z - axis significantly . transverse amplitude growth dominates this case and is very similar to that of a static plasma column . in general the flow appears to follow the helical twist indicated by the density isosurface . flow velocities in the transverse x - y plane are larger than flow along the z - axis . when @xmath98 flow in the x - y plane is reduced somewhat relative to flow along the z - axis and the kink propagates slowly along the z - axis . as the radius increases the kink propagates along the z - axis more rapidly ( see more detail in figure 5 ) while continuing transverse growth . flow is less twisted helically as the velocity shear radius increases and at @xmath100 we see a helical kink embedded within and moving with the flow . recall that the time evolution of the volume averaged transverse kinetic energy , i.e. , fig . 3 , were the most similar to a static plasma column for smallest and largest shear radius . the principal difference between the smallest and largest velocity shear radius cases lies in the flow morphology relative to the kink morphology indicated by the density isosurface . in figure 5 we follow the time evolution of the @xmath101 and @xmath102position of the maximum density in the @xmath103 plane at @xmath104 . transverse growth of the kink is revealed by displacement of this maximum away from zero in the @xmath103 plane . in all cases significant displacement of the density maximum begins at @xmath105 , when growth is still within the linear regime . an oscillation of the maximum indicates rotation around the z - axis in the @xmath103 plane and is related to the propagation speed of the kink . since the kink wavelength is @xmath106 in all simulations , the kink propagation speed is @xmath107c = ( 3/\delta t)c$ ] where @xmath95 is the length of time for one rotation in simulation time units . at the smallest shear radius ( red solid lines ) , no oscillation is evident and transverse growth is very similar to that of the static plasma column ( black dash double dotted lines ) . the lack of measurable propagation implies that the flow moves through the growing helical twist at nearly the flow speed , as suggested by the velocity vectors in panel ( a ) of figure 4 . when the shear radius is @xmath98 ( orange dashed lines ) , only a partial oscillation occurs by the end of the simulation . our best estimate of @xmath108 comes from panel ( b ) in figure 5 with a maximum and minimum displacement of @xmath109 occurring at @xmath110 and @xmath111 , respectively . this implies a kink propagation speed of @xmath112 . the kink is propagating along jet axis while transverse growth continues but the propagation speed is slow . in this case the flow is still helically twisted but with less than the apparent helical twist of the density isosurface shown in panel ( b ) of figure 4 . more than one complete oscillation is evident for a shear radius @xmath74 ( see the green dotted lines ) and visual inspection suggests that the oscillation period becomes longer with time . at @xmath113 three measurements indicate @xmath114 , where @xmath115 " indicates the range of the measurements . this implies a kink propagation speed @xmath116 . at @xmath117 three measurements indicate @xmath118 . this implies a kink propagation speed @xmath119 . the kink is propagating rapidly along the axis but as transverse growth continues the kink slows . in this case the flow is only modestly helically twisted and with much less than the apparent helical twist of the density isosurface shown in panel ( c ) of figure 4 . multiple oscillations are evident for a shear radius @xmath64 ( see blue dash - dotted lines ) and again the oscillation period becomes longer with time . five measurements at @xmath113 provide @xmath120 , and five measurements at @xmath121 provide @xmath122 . here the initial kink speed @xmath123 is approximately equal to the flow speed . at later times the kink speed @xmath124 has slowed significantly . at early time the smaller transverse amplitude of the kink is embedded within the flow and the flow exhibits little helical twist . the oscillation amplitude , indicative of the transverse amplitude of the kink , increases up to about @xmath125 and then declines slightly with time . this suggests a decrease in transverse amplitude growth as the growing kink encounters the velocity shear surface and slows . we see from these results that kink propagation and flow morphology are strongly dependent on the velocity shear radius relative to the characteristic radius of the force - free plasma column . at one extreme the fluid flows through the kink with helicity comparable to that of the kink and at the other extreme the kink is embedded within a more uniform flow . the kink propagation speeds are listed in table 2 in 4 . in figure 6 we show the time evolution of the volume - averaged kinetic , @xmath79 , and magnetic , @xmath80 , energy transverse to the @xmath81-axis determined within a cylinder of radius @xmath82 , e.g. , figure 3 . here we see an initial linear growth phase , @xmath126 , and subsequent nonlinear evolution phase . the exponential increase in @xmath79 by about 3 orders of magnitude to a maximum amplitude followed by a slow decline in the nonlinear phase and behavior of @xmath80 is similar to what was found for the constant pitch cases . by fitting the linear portion of the slope in @xmath79 between the amplitudes of @xmath92 and @xmath93 we can determine an e - folding time where @xmath94 and @xmath95 is the time interval . the e - folding times can be found in table 2 in 4 . in this case , the e - folding time is increased relative to the static plasma column by about the same amount only for velocity shear radii @xmath127 . however , the decreasing helical pitch results in more rapid growth in the linear phase , and makes a transition to the nonlinear phase in about 60% of the time for the constant pitch cases shown in figure 3 . this more rapid growth is similar to results for a static plasma column ( mizuno et al . thus , the growth rate trends found for the static plasma column are maintained in the presence of sub - alfvnic flow . the maximum amplitude of @xmath79 is comparable to that found for constant helical pitch . however , the amplitude of @xmath79 declines less in the non - linear phase relative to the static case than was found for the constant pitch cases . we note however that the constant pitch cases might have exhibited similar behavior at longer timescales . here the decline in @xmath79 is clearly the least when @xmath74 with decline increasing for @xmath128 , @xmath15 and @xmath54 , respectively . the same trend is evident in figure 3 for constant helical pitch . in this set of simulations we follow the development of the kink to a much longer time relative to the transition time from the linear growth phase to the non - linear phase . again we find that the largest effects occur for velocity shear radius @xmath74 but now confirm relatively large effects for @xmath98 and @xmath100 . however , even for @xmath78 we clearly see the influence of velocity shear when compared with the results for constant pitch . we conclude that non - linear development is more influenced by a velocity shear surface in the case of decreasing helical pitch . figure 7 shows a density isosurface for the decreasing pitch cases with @xmath51 for different velocity shear radius at @xmath99 . the decreasing pitch cases shown here all appear similar to results shown for static plasma columns ( mizuno et al . 2009a ) . in the linear growth phase the properties are almost same for different shear radius but after transition to the nonlinear phase the behavior of the kink is different for different shear radius . in general , the behavior is similar to what was found for constant helical pitch . for the smallest radius , @xmath78 , the kink does not propagate along the z - axis significantly . in general the flow appears to follow the helical twist indicated by the density isosurface . flow velocities in the transverse x - y plane are larger than flow along the z - axis . when @xmath98 flow in the x - y plane is reduced somewhat relative to flow along the z - axis and the kink propagates slowly along the z - axis . as the radius increases the kink propagates along the z - axis more rapidly ( see more detail in figure 8) while continuing transverse growth . flow is less twisted helically as the velocity shear radius increases . however , we note that when @xmath100 we see considerably more indication of flow helicity than for the constant pitch case . in figure 8 we follow the time evolution of the @xmath101 and @xmath102position of the maximum density in the @xmath103 plane at @xmath104 , e.g. , figure 5 for constant pitch cases . in all the decreasing pitch cases significant displacement of the density maximum begins at @xmath129 , when growth is still within the linear regime , and significant displacement occurs earlier than for the constant pitch cases . at the smallest shear radius ( red solid lines ) , no oscillation is evident and transverse growth is very similar to that of the static plasma column ( black dash - double dotted lines ) . the lack of measurable propagation implies that the flow moves through the growing helical twist at nearly the flow speed , as suggested by the velocity vectors in panel ( a ) of figure 7 . when the shear radius is @xmath98 ( orange dashed lines ) , only a partial oscillation occurs by the end of the simulation . our best estimate of @xmath130 comes from panel ( a ) in figure 7 with a minimum and zero displacement of @xmath131 occurring at @xmath132 and @xmath133 , respectively . this implies a kink propagation speed of @xmath134 . comparison to the comparable constant pitch case suggests a reduction in the propagation speed but our level of accuracy does not make this a firm conclusion . the kink is propagating along jet axis while transverse growth continues but the propagation speed is slow . in this case the flow is still helically twisted but with less than the apparent helical twist of the density isosurface shown in panel ( b ) of figure 7 . more than one complete oscillation is evident for a shear radius @xmath74 ( see the green dotted lines ) and visual inspection suggests that the oscillation period becomes longer with time . at @xmath135 two measurements indicate @xmath136 , where @xmath115 " indicates the range of the measurements . this implies a kink propagation speed @xmath137 . at @xmath138 two measurements indicate @xmath139 . this implies a kink propagation speed @xmath140 . the kink is propagating along the axis but as transverse growth continues the kink slows . here the kink speed both at early and late times is significantly less than that found for the comparable constant pitch case . multiple oscillations are evident for a shear radius @xmath64 ( see blue dash - dotted lines ) and again the oscillation period becomes longer with time . four measurements at @xmath135 provide @xmath120 , and three measurements at @xmath138 provide @xmath141 . here the initial kink speed @xmath123 is approximately equal to the flow speed . at later times the kink speed @xmath142 has slowed significantly . at early time the smaller transverse amplitude of the kink is embedded within the flow and the flow exhibits little helical twist . the oscillation amplitude , indicative of the transverse amplitude of the kink , increases up to about @xmath133 . here we see transverse amplitude growth continuing to longer times and achieving larger amplitude than the comparable constant pitch case . the kink speed at later times is significantly less than that found for the comparable constant pitch case . we see from these results that kink propagation and flow morphology are again strongly dependent on the velocity shear radius relative to the characteristic radius of the force - free plasma column . in general , the presence of a velocity shear surface influences the propagation speed more than was found for the constant pitch cases . the slower kink propagation speeds found for the decreasing pitch cases are likely the result of the faster amplitude growth of the kink to a larger transverse amplitude . thus , at the larger velocity shear radii the growing kink approaches the velocity shear surface more rapidly and/or more closely . the kink propagation speeds are listed in table 2 in 4 . figure 9 shows the time evolution of the volume - averaged kinetic energy ( @xmath79 ) and magnetic energy ( @xmath80 ) transverse to the @xmath81-axis within a cylinder of radius @xmath144 for constant pitch with @xmath52 , similar to figures 3 and 6 . change in the evolution of @xmath79 and @xmath80 indicate an initial linear growth phase at @xmath145 with duration of the linear growth phase depending on the velocity shear radius . in all cases , the initial growth phase is characterized by an exponential increase in @xmath79 by about 3 orders of magnitude to a maximum amplitude followed by a slow decline in the nonlinear phase . by fitting the linear portion of the slope in @xmath79 between the amplitudes of @xmath92 and @xmath93 we can determine an e - folding time . the e - folding times are listed in table 2 in 4 . in general , the e - folding time increases as the velocity shear radius increases for @xmath96 . the e - folding time at @xmath97 decreases but is still significantly longer than for the static plasma column . evolution is qualitatively similar but quantitatively different from the constant pitch slower jet cases . in particular , the maximum value for @xmath79 is about a factor of two larger than found for the constant pitch cases with @xmath51 , and is likely the result of the initial flow kinetic energy being about a factor of two higher . as was found for the slower jet cases , the growth rate is slower and transition to the nonlinear phase occurs later as the shear radius increases from @xmath78 ( red lines ) to @xmath146 ( green dotted lines ) . as was found previously , the largest effects of velocity shear appear when @xmath146 with smaller but still significant effects when @xmath147 . the @xmath100 simulation terminated before the maximum amplitude in @xmath79 was achieved , but it is clear that the maximum amplitude in @xmath79 will occur at significantly later time than when @xmath147 . this result is different from the comparable slower flow case . this difference is likely the result of relativistic effects and will be considered further in 4 . the three - dimensional helical structure of these faster jet cases is qualitatively similar to that of slower jet cases with constant pitch . density isosurfaces , magnetic field lines and velocity vectors appear similar to those shown in figures 2 & 4 . in figure 10 we show the time evolution of the @xmath101 and @xmath102position of the maximum density in the @xmath103 plane at @xmath104 , e.g. , figure 5 for the slower flow constant pitch cases . here significant displacement of the density maximum begins at @xmath148 , when growth is still within the linear regime . this is similar to the slower flow constant pitch cases . at the smallest shear radius ( red solid lines ) , there is a suggestion of an oscillation in @xmath109 ( panel b ) with a maximum negative displacement at @xmath149 and @xmath150 , although no oscillation is evident in @xmath131 ( panel a ) . we can use this apparent oscillation to set an upper limit to the propagation speed of @xmath151 . thus , the flow moves through the growing helical twist at nearly the flow speed , as was found for the previous cases . when the shear radius is @xmath98 ( orange dashed lines ) , a full oscillation occurs and visual inspection suggests that the oscillation period increases with time . at @xmath135 two measurements indicate @xmath152 , where @xmath115 " indicates the range . this implies a kink propagation speed @xmath153 . at @xmath138 two measurements indicate @xmath154 . this implies a kink propagation speed @xmath155 . the kink is propagating along the axis but as transverse growth continues the kink slows . here the kink speed at late times is similar to that found for the comparable slower flow speed constant pitch case . multiple oscillations are evident for a shear radius @xmath74 ( see the green dotted lines ) and visual inspection suggests that the oscillation period becomes slightly longer with time . at @xmath113 measurements indicate @xmath156 , where @xmath115 " indicates the range . this implies a kink propagation speed @xmath157 . at @xmath121 measurements indicate @xmath158 . this implies a kink propagation speed @xmath159 . the kink is propagating relatively rapidly along the axis at early times but as transverse growth continues the kink slows . multiple oscillations are evident for a shear radius @xmath64 ( see blue dash - dotted lines ) but there is no evidence that the oscillation period becomes longer with time for @xmath160 when the simulation terminated . the multiple oscillations provide @xmath161 . here the kink speed @xmath162 is approximately equal to the flow speed . the oscillation amplitude , indicative of the transverse amplitude of the kink , increases up to the end of this simulation . this suggests that the kink remains embedded in the flow up to the end of this simulation . these results are similar to those found for the slower flow cases with constant pitch . when the shear radius is larger , the propagation speed of the helical kink becomes faster . quantitative comparison with individual slower jet cases ( fig . 5 ) , on average indicates a more rapid propagation speed for each shear radius . the kink propagation speeds are listed in table 2 in 4 . we have investigated the development of the cd kink instability of a force - free helical magnetic field with a sub - alfvnic velocity shear surface located at various radii relative to the characteristic radius of the magnetic field . we restricted this study to sub - alfvnic shear as this regime is appropriate to the magnetically dominated flows thought to exist in the acceleration and collimation regions of relativistic jets . in this magnetically dominated parameter regime the flow is stable to the velocity shear driven kelvin - helmholtz instability so that we could focus solely on the effect of the shear flow on growth and propagation of the current driven kink and the velocity flow field accompanying the helically twisted kink . the growth of cd kink instability in the initial exponential growth phase is slower than found for a static plasma column . in general , the reduction in the growth rate is larger for constant magnetic pitch than for decreasing magnetic pitch , and is larger if the velocity is larger . in all cases , effects resulting from the presence of a velocity shear surface are largest when the velocity shear surface lies at or not too far outside , @xmath163 , the characteristic radius , @xmath15 , of the force - free magnetic field . for the slower speed , @xmath73 , and constant pitch case it was clear that the initial growth was least affected when the velocity shear surface was inside , @xmath164 , or far outside , @xmath77 , the characteristic radius . for the higher speed , @xmath143 , and constant pitch case initial growth was again least affected when @xmath164 but relativistic effects slowed the initial growth when @xmath77 . transition to the nonlinear stage occurs at a later time but with larger maximum amplitude in the volume averaged transverse kinetic energy , @xmath79 , as the velocity shear radius is increased for @xmath165 , when compared to a static plasma column . however , when the velocity shear radius is far from the characteristic radius , @xmath166 , the maximum amplitude is comparable to that of the static plasma column although that maximum is reached after longer time . in the absence of relativistic effects which slow the observed rate of growth , it is clear that the presence of a velocity shear surface has the strongest influence on both the linear and non - linear behavior of the growing kink when the velocity shear surface is located at or not too far outside the characteristic radius . in general , decreasing the magnetic pitch or increasing the flow velocity enhances the influence of the velocity shear surface . the location of the velocity shear surface has profound consequences for kink propagation and the associated flow field . for the velocity shear surface well inside the characteristic radius , transverse growth is similar to the static plasma column . in this case the plasma flows through the growing helical kink . for the velocity shear surface well outside the characteristic radius , the kink is embedded within and moves with the flow until the kink amplitude becomes large and the kink approaches the velocity shear surface . in this case the initial transverse growth is similar to that of a static plasma column advected with the flow and with growth computed in the proper reference frame . as the growing helically twisted kink approaches the velocity shear surface the kink slows and the flow field becomes slightly helically twisted . for velocity shear radii on the order of the characteristic radius there is a more intimate interaction between the growing kink and the flow field . in general , the kink propagates more slowly than the flow and slows as the amplitude increases . thus , the flow field becomes more helically twisted as the kink amplitude increases . for these cases the flow helicity remains less than kink helicity . the lorentz factors for slower and faster flows are @xmath167 1.02 and 1.05 , respectively . the growth rate of the cd kink instability depends on the alfvn velocity ( e.g. , appl et al . 2000 ) and also depends on relativistic time dilation . in the flow reference frame , the toroidal and axial magnetic field components are reduced by the lorentz factor and the mass density is reduced by the lorentz factor squared . the alfvn velocity is decreased slightly in the flow reference frame because @xmath168 is slightly larger . therefore in faster flow cases , we would expect the growth rate to be reduced primarily by time dilation related to the lorentz factor of the moving kink , @xmath90 . our results for kink e - folding times in the linear growth phase are summarized in table 2 along with the kink propagation speeds for the various different simulations . lcccc [ table1 ] cp0 & 0.0 & 0.0 & 3.75 & 0.0 + cpsa/2 & 0.2 & 0.5 & 3.75 & @xmath169 0 + cpsa & 0.2 & 1.0 & 4.10 & 0.06 + cps2a & 0.2 & 2.0 & 4.35 & 0.15 - 0.10 + cps4a & 0.2 & 4.0 & 4.05 & 0.20 - 0.16 + dp0 & 0.0 & 0.0 & 2.45 & 0.0 + dpsa/2 & 0.2 & 0.5 & 2.40 & @xmath169 0 + dpsa & 0.2 & 1.0 & 2.45 & 0.04 + dps2a & 0.2 & 2.0 & 2.75 & 0.12 - 0.06 + dps4a & 0.2 & 4.0 & 2.70 & 0.20 - 0.10 + cp0 & 0.0 & 0.0 & 3.75 & 0.0 + cpfa/2 & 0.3 & 0.5 & 3.70 & @xmath170 0.05 + cpfa & 0.3 & 1.0 & 4.30 & 0.12 - 0.06 + cpf2a & 0.3 & 2.0 & 5.60 & 0.23 - 0.17 + cpf4a & 0.3 & 4.0 & 5.30 & 0.30 + quantitative comparison between our results and instability predictions is difficult because no sufficiently general stability analysis has been performed for magnetically dominated jets . the stability analysis performed by narayan et al . ( 2009 ) considers the case of a `` ... rigid impenetrable wall at the outer cylindrical radius , @xmath171 . '' , that unfortunately is not appropriate to our simulations . nevertheless , we can make some comparison with previous results for static plasma columns and consider the implications for the spatial development of the instability . in all cases , the initial axisymmetric structure is strongly distorted by the kink instability , even though not disrupted . in general , the addition of a sub - alfvnic velocity shear surface leads to slower temporal development of the instability than for the static case . comparison between the static and moving kink constant pitch cases shows that the e - folding times ( see table 2 ) are significantly longer , up to @xmath172 and @xmath173 longer for slow and fast velocity shear , when the shear surface is located at twice the characteristic radius of the plasma column . when the velocity shear surface is located far outside the characteristic radius of the plasma column the kink is advected at about the flow speed in the linear growth regime and the e - folding times are shortened somewhat to about @xmath174 and @xmath175 longer than for the comparable static case . with the kink moving with the flow frame we would expect the growth rate in the flow frame to be related to the growth rate in the lab frame by the lorentz factor of the flow . however , comparison between our static plasma column kink e - folding times and our moving kink e - folding times is complicated by length contraction in addition to time dilation . time dilation increases the e - folding times from fluid to lab frames by the lorentz factor . length contraction means that our simulation box imposes a wavelength that appears longer in the fluid frame than in the lab frame by the lorentz factor . in either frame the wavelength is longer than the fastest growing wavelength . examination of our previous static case numerical results for the growth of @xmath176 at wavelengths of @xmath83 and @xmath177 ( see figure 2 in mizuno et al . ( 2009a ) for constant pitch cases a and b ) indicates that the e - folding time increases about @xmath178 faster than proportional to the wavelength over this wavelength range . thus , length contraction means that our e - folding time observed for a static kink should convert approximately to our e - folding time observed for a moving kink by @xmath179 . the e - folding times for cases cps4a and cpf4a when compared to the comparable static case , cp0 , are significantly longer than the predicted 4% ( @xmath180 ) and 11% ( @xmath181 ) increase . we can only assume that a much larger velocity shear radius is required to further reduce the e - folding times for the propagating kinks to that predicted . on the other hand , this also means that temporal kink growth is significantly slowed even for a velocity shear surface at four times the characteristic radius . it is interesting to note that the fluid inertia increases by @xmath182 and if the growth time is also increased by the fluid inertia then the e - folding times in the lab frame should be increased by @xmath183 = 1.16 and 1.31 relative to the static case , and this increase comes much closer to the observed increase . we speculate that the large increase in e - folding times measured in the lab frame is partly a result of the increased inertia of the relativistically moving fluid . the characteristic time for the instability to affect strongly the initial structure varies from @xmath184 for the decreasing pitch case to @xmath185 for the constant pitch cases . for the constant pitch cases the characteristic time is roughly @xmath186 , with values for @xmath187 being dependent on the structure of the undisturbed state . in a jet context our perturbations remain static or can propagate with the flow frame depending on the location of the velocity shear surface . in order to check whether the instability would affect a jet flow , one has to compare @xmath188 with a propagation time . if we identify @xmath187 with the fastest growing wavelength , our present results suggest a scaling like @xmath189 with @xmath190 for a moving kink and with velocity shear surface a few times the characteristic radius . here @xmath60 would correspond to time dilation only and @xmath191 to time dilation plus inertial effects from the relativistically moving fluid . in this case the condition for the instability to affect the jet structure might be written as @xmath192 where we set @xmath193 and @xmath194 is a function of @xmath195 and is sensitively dependent on the location of the velocity shear surface provided @xmath196 . this result suggests that the characteristic scale for kink development could be longer or very much shorter than for a kink simply advected with a broad flow for which @xmath197 . in order to find a more general criteria one has to know how the characteristic radius , @xmath15 , and lorentz factor , @xmath90 , increase with distance . if the jet is narrow enough so that @xmath198 , where @xmath199 is the angular velocity at the base of the jet , one can use the scaling ( tchekhovskoy et al . 2008 ; komissarov et al . 2009 ; lyubarsky 2009 ) @xmath200 and assume that @xmath201 . in this case one finds that the criterion for the kink instability can be written as @xmath202^{1/2 } a~.\ ] ] the instability could affect the jet structure only if the jet expands slowly enough and/or the kink moves slowly enough . assuming the parabolic shape for the jet , @xmath203 , where @xmath204 and @xmath205 are dimensionless numbers , one finds that the instability develops only if @xmath206 . in this case the characteristic scale for the development of the instability can be written as @xmath207}[1 - ( \epsilon \xi \chi^k)^{-2}]^{1/2[1 - ( \alpha + 1)k]}~.\ ] ] for @xmath208 , @xmath209 and @xmath210 we recover the case for a kink advected with the flow field , eq.(9 ) in mizuno et al . ( 2009a ) , @xmath211 for a = 100 . here as @xmath212 , @xmath213 and @xmath214 so the characteristic scale for development of instability can be very short . for @xmath22 , i.e. , potential inertial effects , the characteristic time is lengthened and this has the potential for lengthening the characteristic scale for the development of the instability . this shows up in the constraint on @xmath215 , i.e. , only if the jet expands slowly enough . the 3d relativistic jet generation simulation performed by mckinney & blandford ( 2009 ) indicates relatively rapid , less than 100 gravitational radii , but non - disruptive kink development over 500 gravitational radii . our previous and present simulations for static and moving kinks suggest that the rapid but non - disruptive kink development in the jet generation simulation could be a result of a velocity shear surface located less or on order of the characteristic magnetic radius and a density increasing with radius . this combination would result in a slowly moving kink , hence rapid initial spatial development , but with non - linear growth slowed by the density increase and accompanying alfvn speed decline with radius , increasing the alfvn crossing time and slowing spatial development . a non - linearly stabilizing increasing density profile is what might be expected for a poynting flux jet core confined within a senser slowly moving sheath , as appears to be predicted by jet generation simulations . of course , a proper investigation of spatial growth requires stability simulations designed to study spatial kink development using more realistic flow , magnetic , and density profiles . in this paper , we considered sub - alfvnic jet flow and focused on the development of the cd kink instability . if the velocity shear is super - alfvnic , the flow can be kh unstable and cd unstable . baty & keppens ( 2002 ) have investigated the interaction between kh and cd driven instabilities of a magnetized force - free cylindrical configuration via 3d mhd simulations in the non - relativistic regime . they found that the cd unstable modes provided a stabilizing effect on kh instability driven vortical structure . however , they assumed a relatively weak magnetic field in their simulations and studied the super - alfvnic regime where the kh instability grows faster than the cd kink instability ( e.g. , appl et al . 2000 ; baty 2005 ) . if the magnetic field is strong but the velocity shear is weakly super - alfvnic , the growth rate of the kh instability can be less than or comparable to that of the cd instability . even in the super - alfvnic regime the kh instability can be suppressed if a jet is embedded in a slower moving magnetized sheath that reduces the velocity shear to being effectively sub - alfvnic ( hardee 2007 ; mizuno et al . 2007 ) . such a sheath may exist around jets in the acceleration and collimation region . thus , investigation of the weakly super - alfvnic parameter regime will be particularly important to understanding the development of the twisted structures that are observed on relativistic jets . in future work we will investigate the spatial development of cd instability and the coupling between cd instability and kh instability . thanks y. lyubarsky , k. shibata , b. zhang , m. a. aloy , and j. m. stone for helpful discussions . this work is supported by nsf awards ast-0506719 , ast-0506666 , ast-0908010 , and ast-0908040 , and nasa awards nng05gk73 g , nnx07aj88 g , and nnx08ag83 g , and us - israeli bsf award 2006170 . the simulations were performed on the columbia supercomputer at nas division at nasa ames research center , the sgi altix ( cobalt ) at the national center for supercomputing applications in teragrid project which is supported by the nsf and the altix3700 bx2 at yitp in kyoto university . in this section we first summarize previous code results against known solutions and second we check our numerical simulation code results against those obtained by other codes using the same test problems . in rmhd there are not many problems with known solutions because a closed solution for the general rmhd riemann problem has not yet been found . the code has been verified against a number of test and physical problems with known solutions . in mizuno et al . ( 2006 ) , a 1d linear alfvn wave propagation test and a 1d magnetized bondi accretion flow test showed second - order convergence for our simulation code , and relativistic mhd shock - tube tests showed that our simulation code correctly handles the wave structure of both shocks and rarefactions even in an extreme relativistic regime . the code has been used to model several 1d and 2d physical problems with known solutions . the code was successfully used to solve the 1d riemann problem for the deceleration of an arbitrarily magnetized relativistic flow injected into a static unmagnetized medium ( mizuno et al . 2009b ) . in particular , this physical problem involved comparing conditions for existence of the reverse shock against known relativistic shock jump conditions . the code has also been successfully used to solve in 2d the effect of magnetic fields on an hd / mhd boost mechanism proposed by aloy & rezzolla ( 2006 ) . the results to this physical problem involving a relativistic flow that flows parallel to an overpressure generated shock and rarefaction wave combination agreed with and extended the previous hd findings to mhd ( mizuno et al . 2008 ) . we also have used our simulation code for several multidimensional ( 3d ) physical problems involving comparison with known theoretical predictions . the code has been used to study spatial development of the kelvin - helmholtz instability on relativistic super - alfvnic and trans - alfvnic cylindrical jets , and results ( see mizuno et al . 2007 ) were compared to predictions made by a linear stability analysis ( hardee 2007 ) . the code has also been used to study temporal development of the current driven instability of a relativistic static plasma column , i.e. , with magnetic energy density comparable to the plasma energy density including the rest mass energy , and results were successfully compared to linear stability analysis predictions and to previous non - relativistic numerical results ( see mizuno et al . 2009a ) . in both completely different instability regimes , the code delivered results that agreed with stability analysis predictions and with previous numerical results . these successful comparisons verify the code against known solutions to physical multidimensional problems . in general , we have not yet shown results for multidimensional test problems . in what follows we check our simulation code against multidimensional test problems using the mc slope - limiter scheme which we have used for the simulations in this paper . in all test simulations we use the hll approximate riemann solver to calculate numerical fluxes and the flux - ct scheme to maintain a divergence - free magnetic field . we consider the advection of a weak magnetic field loop in an initially uniform velocity field . in thermal pressure dominance the loop is transported as a passive scalar . the 2d test that we perform is a relativistic version of the magnetic loop advection problem . this tests the dissipative properties of the numerical scheme and the correct discretization balance of multidimensional terms ( gardiner & stone 2005 , 2008 ; mignone et al . 2010 ) . a successful test results in the preservation of the initial loop . following non - relativistic mhd tests ( gardiner & stone 2005 , 2008 ; mignone et al . 2010 ) , we employ a periodic computational box defined as @xmath216 and @xmath217 discretized on @xmath218 computational zones . density and gas pressure are initially constant , and set @xmath219 and @xmath220 respectively . the sound speed @xmath221 for an adiabatic index @xmath43 . the magnetic field is defined through its magnetic vector potential as @xmath222 where @xmath223 , @xmath224 , @xmath225 , @xmath226 , @xmath227 , and @xmath228 . this modification of the vector potential in the @xmath229 region , with respect to the original version of the test problem performed by gardiner & stone ( 2005 ) , is done to remove the singularity in the loop s center that can cause spurious oscillations and erroneous evaluations of the magnetic energy ( mignone et al . we perform a case with no advection and an advection case in which the velocity of the flow is set up as @xmath230 , @xmath231 and @xmath232 . the simulations are evolved until @xmath233 when the loop has crossed through the periodic boundaries and returned to the center of the grid in the advection case . in figure 11 the magnetic energy density ( @xmath234 ) and magnetic file lines are shown for the no advection case ( upper ) and advection case ( lower ) with @xmath235 . in the no advection case , the circular shape of the loop is perfectly preserved . there is no visible distortion or dissipation around the center and outer boundaries of the field loop . in the advection case , the circular shape of the loop is relatively well preserved although some flattening of the magnetic field lines is seen at left upper and right lower outer boundaries to the field loop . the magnetic energy density image shows internal structure comparable to that found in the newtonian limit by gardiner & stone ( 2005 ) using first and second order ctu+ct schemes ( fig . 3 and 6 in gardiner & stone 2005 ) . mignone et al . ( 2010 ) performed a similar 2d magnetic loop advection test in the newtonian limit using higher order schemes ( third and fifth order ) . not surprisingly , the higher order schemes do a better job of maintaining uniformity in the magnetic energy density and in the loop structure . higher numerical resolution leads to a less diffusive result with less than @xmath236 magnetic energy loss at the highest numerical resolution . our result using @xmath237 finds almost the same magnetic energy dissipation , about 7% magnetic energy loss , as that found by gardiner & stone ( 2005 ) ( see @xmath233 in fig . 7 of gardiner & stone 2005 ) in the newtonian limit using a comparable second - order scheme . our quantitative results compare favorably with the higher order schemes tested by mignone et al . ( 2010 ) . here our result using @xmath237 ( @xmath238 loss ) is similar to their result using a third - order scheme ( @xmath239 loss ) at the same resolution ( see @xmath233 in fig.a.9 of mignone et al . our result using higher resolution @xmath235 shows less magnetic energy dissipation ( @xmath240 loss ) at time @xmath233 . the three dimensional version of this problem is particularly challenging and see gardiner & stone ( 2008 ) for comparable second - order scheme results in the newtonian limit . correct evolution depends on how accurately the divergence - free condition is preserved and how well multidimensional mhd terms are balanced . the test consists of a computational box defined as @xmath241 , @xmath217 and @xmath242 discretized on @xmath243 computational zones with periodic boundary condition in all directions . following the two - dimensional case , density and gas pressure are initially constant , and we set @xmath219 and @xmath220 . we show a no advection case and an advection case in which the velocity of the flow is set as @xmath244 . as for the two - dimensional case the vector potential @xmath245 is used to initialize the magnetic field in the coordinate system @xmath246 which is related to computational coordinate system @xmath247 via the rotation @xmath248 where @xmath249 . the initial vector potential is given by eq . ( a1 ) , with @xmath250 . figure 13 shows three - dimensional isovolume images of the magnetic energy density for the no advection ( left ) and advection case ( right ) with @xmath235 . in the no advection case , the cylindrical shape of field loops is perfectly preserved . some small dissipation is seen around the center and outer boundaries of the field loops . in the advection case , the cylindrical shape of the field loops is relatively well preserved although there is some distortion and dissipation around the inner and outer boundaries of the field loops . this result is similar to that seen in the 2d field loop advection case ( see fig . 11 ) and appears comparable to the results shown in figure 2 in gardner & stone ( 2008 ) . mignone et al . ( 2010 ) performed a similar 3d magnetic loop advection test in the newtonian limit using higher order schemes ( third and fifth order ) . not surprisingly , the higher order schemes do a better job preserving the loop structure and also display sharper boundaries . a quantitative measure of the magnetic field dissipation rate in 3d field loop advection problem is indicated by the time evolution of the volume averaged magnetic energy density normalized to its initial value as shown in figure 14 . as found for the 2d magnetic advection problem , higher numerical resolution leads to a less diffusive result with @xmath240 magnetic energy loss at the highest numerical resolution of @xmath235 . our result using @xmath237 finds a slightly greater magnetic energy dissipation ( @xmath251 loss ) when compared to the @xmath252 magnetic energy loss found by gardiner & stone ( 2008 ) ( see fig . 1 in gardiner & stone 2008 ) in the newtonian limit using a comparable second - order scheme . our magnetic energy loss using @xmath237 ( @xmath251 loss ) is greater than that found by mignone et al . ( 2010 ) ( see fig.a.9 in mignone et al . 2010 ) using a third - order scheme ( @xmath253 loss ) at the same resolution . however , our result using higher resolution @xmath235 shows less magnetic energy dissipation ( @xmath240 loss ) . in summary , we obtain results very similar to previous non - relativistic results obtained using comparable second - order schemes ( gardiner & stone 2005 ; gardiner & stone 2008 ) . at present there are no published 2d or 3d relativistic loop test results for comparison . we conclude that our simulation code successfully passes both 2d and 3d test problems for the second - order mc slope - limiter and flux - ct schemes . we note that test simulations without the flux - ct scheme fail to pass these test problems . the cylindrical explosion test problem consists of a strong shock propagating into a magnetically dominated medium . results for different cylindrical explosion problems in rmhd have been reported by several authors ( e.g. , dubal 1991 ; van putten 1995 ; komissarov 1999 ; del zanna et al . 2003 ; leismann et al . 2005 ; mignone & bodo 2006 ; antn et al . 2010 ) . here , unlike the 2d loop test we can compare our results to identical relativistic test results obtained by other rmhd second - order codes . this test verifies that the generation of a high lorentz factor flow is handled correctly . a successful test results in a dramatic difference in pressure and density between the ambient gas and the explosion zone . we have chosen a setup following that in leismann et al . ( 2005 ) which is very similar to that in komissarov ( 1999 ) : a cylinder with high pressure ( @xmath254 ) , density ( @xmath255 ) , and radius @xmath256 is located in the center of a square cartesian grid which initially contains a uniform , strong magnetic field . between a radius of @xmath256 to @xmath257 the density and pressure smoothly decrease to those of a homogeneous ambient medium ( @xmath258 and @xmath259 ) . initially , the magnetic field is in the @xmath260direction , @xmath261 , and the velocity is zero everywhere . the simulations are carried out until @xmath262 on grid resolutions of @xmath263 and @xmath264 , spanning a region of @xmath265^{2}$ ] . the outer fast shock has an almost circular shape . the innermost region is also almost circular and bounded by a reverse fast shock . between these two fast shocks there are two density shells bounded on the outside by compressed magnetic field . our test results are qualitatively identical to the results found by other second - order rmhd codes ( see figs . b3 & b4 in leismann et al . 2005 ; see fig . 4 in del zanna et al . 2008 ) . in order to evaluate the quantitative difference between our simulation code and others , one - dimensional gas pressure and magnetic pressure profiles along the @xmath266 axis are shown in figure 16 . the location of discontinuities is identical to the results in leismann et al . ( 2005 ) and del zanna et al . the maximum value of the gas pressure is slightly larger than in del zanna et al . because of different numerical resolution but the same as in leismann et al . at comparable numerical resolution . slight differences in the one - dimensional magnetic pressure profile around the maximum result from differences in numerical diffusivity in the different codes . in summary , we obtain qualitatively identical and quantitatively almost identical results when compared to previous second order rmhd simulation code results . we conclude that our simulation code passes this test problem successfully . del zanna et al . ( 2003 ) adapted the two - dimensional rotor problem from non - relativistic mhd performed by balsara & spicer ( 1999 ) and tth ( 2000 ) to the relativistic case . here we consider a three dimensional version of the relativistic rotor problem ( mignone et al . 2009 ) . in a successful test the complicated pattern of shocks and torsional alfvn waves launched by the rotor is handled correctly . the initial condition consists of a sphere with radius @xmath267 centered at the origin of the computational box taken to be the unit cube @xmath268^{3}$ ] . the sphere is heavier ( @xmath269 ) than the surrounding medium ( @xmath270 ) and rapidly rotates around the @xmath81 axis with an angular velocity @xmath271 , i.e. , with velocity components @xmath272 . the gas pressure , magnetic field and adiabatic index are constant everywhere , @xmath273 , @xmath274 , and @xmath43 . exploiting the point symmetry , we have carried out simulations until @xmath275 at grid resolutions of @xmath276 and @xmath277 . this test problem has resolution - dependent complexity since the maximum lorentz factor in the initial set up depends on the grid resolution . figure 16 shows two - dimensional density images in the @xmath103 plane at @xmath278 ( perpendicular to the rotation axis ) and the @xmath279 plane at @xmath280 ( including the rotation axis ) for the @xmath277 grid resolution case , at @xmath275 . when the sphere starts rotating , torsional alfvn waves propagate outwards . the initial spherical structure collapses into a disk - like structure in the equatorial plane ( @xmath278 ) which generates shock waves propagating the vertical direction that can be seen in figure 16 . in the @xmath103 plane , surrounding matter is pushed into a thin elliptical shell enclosed by a tangential discontinuity . in mignone et al . ( 2009 ) the thin shell shows a distinct octagonal - like shape when they use a five wave hlld riemann solver scheme to calculate numerical fluxes that makes a less diffusive transition at rotational ( alfvn ) discontinuities . the whole structure is embedded in a radially expanding spherical fast rarefaction wave front . in the @xmath279 plane our spindle shaped structure appears nearly identical to the results shown in mignone et al . ( 2009 ) . our simulation test results are very similar to the results using an hll approximate riemann solver scheme performed by mignone et al . ( 2009 ) ( see the lower panels of fig . 12 in mignone et al . 2009 ) except for their now smoother , when compared to the hlld result , octagonal thin shell structure . we note that other previous relativistic and non - relativistic results for the mhd 2d rotor test problem using the hll scheme ( del zanna et al . 2003 ; balsara & spicer 1999 ) found a thin elliptical shell structure similar to our result in the xy plane ( see the top - left panel in fig . 5 of del zanna et al . therefore the thin elliptical shell structure appears commonly developed in the rotor test problem if the hll scheme is employed . here we must speculate that the difference between the thin elliptical shell structure found by us and others using the hll scheme , and the smoothed octagonal shell structure found by mignone et al . ( 2009 ) using the hll scheme is associated with a difference in the numerical diffusivity resulting from details of the piecewise linear reconstruction and constrained transport schemes used by mignone et al . ( 2009 ) . in profiles along the @xmath266 axis , the density in the central region is larger at lower resolution . lower resolution gives a lower shell density and underestimates the propagation speed . our simulation test results are very similar to those found using the hll approximate riemann solver scheme in mignone et al . ( 2009 ) , see the plus and star symbols of fig . 14 in mignone et al . ( 2009 ) , except for the position of the thin shell . this difference is caused by the difference between the octagonal shape of the thin shell found by mignone et al . and the elliptical shape of the thin shell found here . a protrusion associated with the octagonal - like thin shell along the @xmath266 axis extends beyond the smoother elliptical shell . in summary , we obtain very similar results to previous studies of the 3d and 2d rotor test problem using the hll approximate riemann solver scheme in rmhd . we conclude that our simulation code passes this test problem as successfully as other second - order codes .

we search for stars with proper motions in a set of twenty deep subaru images , covering about 0.28 square degrees to a depth of , taken over a span of six years . in this paper , we describe in detail our reduction and techniques to identify moving objects . we present a first sample of 99 stars with motions of high significance , and discuss briefly the populations from which they are likely drawn . based on photometry and motions alone , we expect that 9 of the candidates may be white dwarfs . we also find a group of stars which may be extremely metal - poor subdwarfs in the halo .

0809.2692

the basic structure of our milky way galaxy seems clear : a thin disk of young stars , gas and dust circles the center quietly , immersed within a thicker disk of older stars . both disks sit inside a nearly spherical halo of very old , metal - poor stars which do not share the overall rotation of the disks . surrounding everything is an extended distribution of dark matter . our knowledge of the details within this big picture , on the other hand , is not so clear . recent large - scale projects , such as the sloan digital sky survey , have measured the properties of high - luminosity stars throughout the halo ( see , for example , @xcite , and @xcite ) , while the rapid development of infrared detectors has allowed projects such as the two micron all sky survey ( @xcite ) and spitzer space telescope ( @xcite ) to pierce the dusty disks and measure the properties of their stars . nonetheless , some portions of the milky way remain largely unexplored . in particular , we know little of the low - mass , low - luminosity stars of the halo . the white dwarfs and metal - poor subdwarfs of the halo glow so faintly , from so great a distance , that they are rarely seen and more rarely recognized . as a recent review ( @xcite ) points out , however , these shy and elusive stars may dominate the microlensing events observed towards the galactic bulge and the magellanic clouds . there are two ways to search for these stars : cover a very large area on the sky to a shallow depth , or use a `` pencil - beam '' survey to examine a tiny region much more deeply . the first approach ( see , for example , @xcite and @xcite ) will find objects in many directions , but only out to a small distance from the sun ; the second approach ( see , for example , @xcite , @xcite , and @xcite ) probes farther into the halo , but only in a specific direction . one way to characterize surveys is to combine their area with the distance out to which they would detect some specific star to generate an `` effective volume '' for that type of star . in table [ table : survey_volume ] , we compare the projects mentioned above by this metric , using a star of absolute magnitude @xmath1 , appropriate for a cool white dwarf . we assumed a color @xmath2 to convert limiting magnitudes for @xmath3-based surveys to @xmath4-band . l l l r survey & area ( sq.deg . ) & limiting mag @xmath4 & volume ( pc@xmath5 ) + oppenheimer et al . & 4165 & 19.8 & 80000 + carollo et al . 1150 & 20 & 15000 + nelson et al . & 0.021 & 26.5 & 2100 + mendez & 0.0013 & 26.0 & 64 + kalirai et al . & 0.0031 & 29 & 9800 + this work & 0.28 & 26 & 14000 + our project could be described as a `` pencil - beam '' survey , but it uses a very thick pencil . we examine images in the area of the subaru deep field ( sdf ) ( @xcite ) acquired over a period of six years to search for moving objects . these images were acquired primarily to study high - redshift galaxies ( @xcite , @xcite ) , but have also been used to find high - redshift supernovae ( @xcite ) . the images are nearly as deep as those in some hst - based surveys , but cover a significantly wider area , yielding a large effective volume . we can refer to the sdf catalogs compiled by @xcite and @xcite for information on our candidates in multiple passbands ( @xmath6 , @xmath4 , @xmath7 , @xmath8 and @xmath9 ) . unlike most other pencil - beam surveys , we have measurements at many epochs : our dataset contains images taken on 20 nights . we can therefore measure the proper motion of our candidates very well , and place strong constraints on the uncertainties in our measurements . this paper is the first in a series on proper motions in several small fields studied with subaru . we will concentrate on techniques , leaving detailed analysis of the results for later papers . in section 2 , we describe the observations , their reductions , and the combination of individual frames into a single combined image for each night . in section 3 , we walk through our procedure for finding moving objects , and discuss our criteria for separating good candidates from bogus ones ; we end up with a first sample of stars which have very well measured proper motions . in section 4 , we compute the reduced proper motions for objects in this sample , and compare their properties to those of objects drawn from a simulated survey of our field . finally , in section 5 , we list our plans for future work on this dataset , and in other fields with multiple epochs of deep subaru imaging . astronomers at the observatoire de besanon have created a model of the stellar populations in the milky way ( @xcite ) with a very convenient web - based interface . the model consists of four populations of stars thin disk , thick disk , spheroid , outer bulge ( the innermost portions of the milky way are poorly constrained ) plus white dwarfs added to each component separately . the parameters of each component are adjusted to produce the best fit to the observed stellar populations and their dynamics . recent work by @xcite finds that the besanon model does a very good job of reproducing observed star counts in two of three deep fields down to @xmath10 . those authors criticized the besanon populations for being too sharply defined , forcing them to smooth the model colors by small amounts ( @xmath11 mag ) in order to match observed color - magnitude diagrams . however , since our main concern is to classify objects very broadly using a mixture of kinematics , magnitudes and colors , we will adopt the besanon model and use it as a reference throughout this paper to help us interpret properties of our sample . the sdf is a region at high galactic latitude ( @xmath12 ) roughly half a degree on a side . @xcite describe very deep optical images taken with the subaru 8.2-meter telescope and suprime - cam camera ( @xcite ) . we investigated this region using a set of @xmath8-band images with shorter exposure times . table [ table : journal ] lists the date for all nights used in our analysis . note that our images taken on 2003 april 30 had a shorter exposure time than the rest ; since these images also had some of the worst seeing , we gave those measurements very little weight in the final proper motion calculations . we split the images taken on the night of 2006 may 3 into two sets and treated each independently , as if taken on different nights . since the data taken on 2007 feb 15 had the best seeing and the largest number of detected objects , we adopted it as the fiducial set for matching ( see section 3 ) . l c r ut date & julian date - 2,450,000 & exptime ( seconds ) + 2001 april 25 & 2024.34 & 3600 + 2001 may 20 & 2049.44 & 3240 + 2002 april 12 & 2376.44 & 9780 + 2002 may 7 & 2401.52 & 5670 + 2003 april 1 & 2730.43 & 12330 + 2003 april 3 & 2732.29 & 2730 + 2003 april 25 & 2754.42 & 4800 + 2003 april 26 & 2755.36 & 4680 + 2003 april 30 & 2759.60 & 964 + 2003 may 1 & 2760.39 & 4583 + 2005 march 5 & 3434.54 & 5100 + 2005 march 6 & 3435.54 & 5400 + 2006 may 3(a ) & 3858.41 & 3000 + 2006 may 3(b ) & 3858.52 & 2400 + 2007 february 13 & 4144.59 & 2700 + 2007 february 14 & 4145.59 & 4200 + 2007 february 15 & 4146.59 & 4500 + 2007 february 16 & 4147.60 & 3600 + 2007 may 16 & 4236.32 & 4180 + 2007 may 17 & 4237.27 & 3780 + during each night of observing , we took a series of short ( typically 180-second to 360-second ) exposures , shifting the telescope position slightly to fill in small gaps between the ten ccds on the focal plane . using the _ sdfred _ package ( @xcite ) and neko software ( @xcite ) , we followed the procedures described in section 4 of @xcite to turn all the raw frames taken during the night into a single , large mosaic . briefly , we cleaned the raw images by subtracting a bias deduced from the overscan regions and dividing by a normalized flatfield frame made from a median of many night - time target images . using the parameters derived in @xcite , we corrected for optical distortions in the focal plane . images from all chips were convolved to form a uniform point - spread function ( psf ) across the entire array . we determined a sky background by calculating the local sky at a series of grid points spaced at intervals of roughly @xmath13 arcseconds and using bi - linear interpolation between the grid points ; we then subtracted this sky background from each image . we used stars shared by adjacent ccds to determine the weights to use when combining data from individual images to make the final mosaic . the result for each night is one ( or , in the case of 2006 may 3 , two ) large image covering the entire sdf . the quality of final combined images varied from night to night . the full width at half maximum ( fwhm ) ranged from @xmath14 to @xmath15 , but , since the plate scale was @xmath16 per pixel , no data was undersampled . the limiting magnitude also varied with the conditions , but was usually @xmath17 . selecting objects with proper motions from a set of images requires several steps : finding and measuring the properties of stars in individual images , matching stars found at different epochs , computing the change in position of each star over time , and deciding which changes are due to real movement . we will now describe these steps in detail . in order to find star - like objects in each image , we used the `` stars '' program within the _ xvista _ package ( @xcite ) . the position of each object was calculated by fitting a gaussian to the background - subtracted , intensity - weighted marginal sums in each direction ( see @xcite for details ) . it is not crucial to separate stars from galaxies at this early stage , since we will later discard any objects which do not move significantly ; therefore , we accepted any object with a sharp core , @xmath18 , as a `` star . '' the number of `` stars '' found each night ranged from about 20,000 to about 100,000 , depending on the exposure time and seeing . the suprime - cam field is wide enough that even small uncorrected distortions near the edge of the field might move the apparent position of a star enough from one epoch to the next to hide real , but small , proper motions . in order to reduce any residual distortions , we broke the field into smaller units we will call `` sectors . '' each sector is a square @xmath19 pixels , or @xmath20 arcseconds , on a side . we allowed a small overlap of @xmath21 arcseconds between adjacent sectors so that stars near the edges would not be missed . we designated one epoch , 2007 february 15 , as `` fiducial , '' to serve as the basis of our matching procedure . for every other image , we used the _ match _ package ( @xcite ) to match the objects in each sector to objects in the corresponding sector of the fiducial image . in order to count as an initial match to the fiducial image , a star had to lie within @xmath22 of the position of an object in the fiducial frame ; we imposed this limit in order to avoid spurious matches between unrelated objects . given the six - year span of our survey , this places an upper limit of about @xmath23 per year on our proper motion candidates . we may increase this limit to look for fast - moving objects in the future . we will demonstrate later ( see figure [ fig : pm_histogram ] ) that this requirement does not have a strong effect on the results . there were typically three hundred to eight hundred pairs of matching items found within each sector . we transformed the ( pixel ) coordinates of each star to the ( pixel ) coordinates of the fiducial image in the following iterative manner . first , we used all the matched pairs in the sector to find the coefficients of a linear transformation @xmath24 via a least - squares technique . next , we computed the residuals between the positions of the members of each pair in the fiducial coordinate system . we discarded pairs with large residuals ; specifically , any pair with a residual more than 10 times the 35th percentile . we then went back to compute new coefficients of the linear transformation with the surviving pairs . we repeated this procedure three times in each sector . ignoring a few sectors with very few objects , the mean residual difference in position for surviving items matched to the fiducial frame was @xmath25 . however , most of the objects contributing to this residual , like most of the objects in each image , are faint , and some of the matches are spurious . the uncertainties in the positions of bright objects are considerably smaller , as we will show below . we performed trials using a cubic transformation between the two coordinate systems , but found that the residuals were not significantly smaller than those based on a linear transformation . the final steps in our matching procedure were to discard duplicate entries for objects in the overlapping areas between sectors , and to discard any objects which appeared in fewer than five epochs . the result was a set of positions in the fiducial coordinate system for objects appearing in at least five epochs . in order to estimate the uncertainty in the calculated positions , we computed the mean position of each objects in each coordinate ( row and col ) and its standard deviation ; we then discarded measurements more than two standard deviations from the mean and recalculated mean and standard deviation . as shown in the first two rows of table [ table : pos_uncert ] , the typical clipped standard deviation rose from @xmath26 for bright , unsaturated objects to @xmath27 for faint objects . l c r r r r r r r sample , method & direction & @xmath28 & @xmath29 & @xmath30 & @xmath31 & @xmath32 & @xmath33 & @xmath34 + all objects , clipped & row & 0.010 & 0.006 & 0.006 & 0.007 & 0.012 & 0.028 & 0.047 + stdev from mean pos & col & 0.011 & 0.007 & 0.007 & 0.008 & 0.013 & 0.028 & 0.047 + & & & & & & & & + moving candidates , & row & 0.007 & 0.007 & 0.007 & 0.009 & 0.022 & 0.036 & 0.052 + scatter from fit & col & 0.008 & 0.008 & 0.009 & 0.011 & 0.023 & 0.036 & 0.052 in order to create a sample of objects for which proper motions could be measured accurately , we selected all objects which appeared in the fiducial image and at least four others . a total of @xmath35 objects satisfied this requirement . faint objects were less likely to be selected , since they might not be detected on nights with poor seeing . in order to check the completeness of this sample as a function of magnitude , we inserted a set of @xmath36 artificial stars with magnitudes ranging from @xmath37 into the images . we then re - analyzed the entire set of images as before . figure [ fig : completeness ] shows the fraction of artificial stars which were detected and placed into the sample for further study . since the fraction falls to 50% at @xmath38 , we estimate that our search may be considered complete to that magnitude . ( 80mm,80mm)completeness_a.eps we subjected this sample to a round of tests . for each coordinate , row and column , we made a linear fit to position as a function of julian date . our fitting routine , following @xcite , provides values for the the slope @xmath39 of this line , the 95 percent confidence interval @xmath40 in the value of the slope , and the scatter @xmath41 around the line . the scatter is another estimate of the one - dimensional uncertainty in the position of a single measurement ; we show its values in the lower rows of table [ table : pos_uncert ] . in order to verify that our fitting method yields both the correct motion and an appropriate uncertainty , we ran a monte carlo simulation . for each integer magnitude between @xmath42 , and for each value of 1-d annual proper motion @xmath43 , we created 100 artificial stars . for each star , we drew 15 epochs randomly from our list of observations ( see table [ table : journal ] ) , and computed a set of positions , using the true proper motion plus some random error in each direction drawn from a gaussian distribution consistent with our measurements of @xmath41 for the given magnitude . we then submitted this list of simulated positions to our fitting routines , and compared their results to the true proper motions . we found that over this entire range of magnitudes and motions , our estimates for the proper motion and its uncertainty were accurate . next , we computed a significance of the slope for each coordinate : @xmath44 choosing objects based on the significance of their motion in one direction alone would discriminate against objects moving diagonally across the ccd , which was aligned with the equatorial coordinate system . therefore , we combined the significance values to create an unbiased measure of motion , @xmath45 . we expect that real proper motions should show an asymmetry , due to the relative motions around the galactic center of the sun and the stars in the sdf , while spurious motions due to random errors in position measurements should be the same in all directions . in the upper panel of figure [ fig : significance ] , we plot the observed motions of objects with motions of low significance ; they are distributed around zero with circular symmetry . on the other hand , objects with highly significant motions ( shown in the lower panel of figure [ fig : significance ] ) are biased towards the south - east . the observed asymmetry matches that of the stars in a simulation made with the besanon model ( the motions of which we have scaled appropriately ) . ( 110mm,110mm)pm_signif_ab.eps ( 110mm,110mm)pm_signif_e.eps for the sample discussed below , we selected objects with @xmath46 . note that since our definition of @xmath47 is based on a 95-percent confidence interval , corresponding to two standard deviations for a normal distribution , our criterion could be described as `` motion at the 10-sigma level . '' selecting objects based on the @xmath48 statistic introduces a bias against objects with small proper motions . we investigated the nature of this bias by adding artificial stars with a range of proper motions into our images , analyzing the images as before , and comparing the output properties of the artificial stars to their input values . in figure [ fig : moving_signif ] , we show the fraction of artificial stars which had measured motions of high enough significance to be included in our proper motion sample . for bright stars , the fraction drops to 50% at a total proper motion of about @xmath49 per year . ( 80mm,80mm)moving_completeness_b.eps we found that 110 objects passed this test . however , upon visually inspecting each candidate , we discovered that in five cases , the motions were due to a blend of two nearby stars , or a star mixed with the light of a background galaxy . that left a set of 105 candidates with real motions at a high significance . the median number of epochs of measurement for these objects was 19 , and only 2 stars had fewer than 15 epochs . we show an example of the motions for one of these candidates in figure [ fig : fitted_motion ] . ( 80mm,80mm)candidate_motion_rows.eps ( 80mm,80mm)candidate_motion_cols.eps how accurate are the derived motions ? using our set of artificial stars again , we computed a fractional error @xmath50 based on the one - dimensional motion of stars in row and column directions separately . @xmath51 we included a constant @xmath52 per year to prevent division by zero . figure [ fig : fractional_errors ] shows the median value of @xmath50 as a function of input proper motion for stars of different magnitudes . the fractional error reaches 10% for proper motions of about @xmath49 per year . ( 80mm,80mm)moving_completeness_d.eps let us turn back to the real stars in the sdf . figure [ fig : pm_histogram ] shows that the distribution of proper motions among stars in our sample has a peak at @xmath49 per year , which is ( not coincidentally ) the point at which the efficiency of detecting motion falls to 50% . the largest motions we found are about @xmath53 per year , which is far less than the limit of @xmath23 per year set by our matching procedure . we conclude that our matching requirement that each measurement lie within @xmath22 of its match in the fiducial epoch does not have a strong effect on the resulting proper motions . ( 80mm,80mm)pm_histogram.eps some of the analysis described below requires the color of an object , in order to distinguish different types of star . we chose the @xmath54 color for several reasons : it samples a wide range of wavelengths , avoids @xmath6-band measurements which are hard to make for cool stars , is often used in studies of galactic structure , and is commonly tabulated in models of stellar properties . since we used @xmath8-band images to search for motions , all the candidates had good @xmath8-band magnitudes ; we will convert them to the standard cousins @xmath55 system in the next paragraph . however , many of the candidates grow faint in images taken at shorter wavelengths . we inspected each candidate in deep sdf images ( @xcite ) taken in @xmath6 , @xmath4 and @xmath56 passbands , and compared its appearance in those images to the @xmath6 , @xmath4 and @xmath56 magnitudes listed for each object in the sdf catalogs . in six cases , the @xmath4-band measurement was clearly incorrect , sometimes due to confusion with a brighter object nearby ; the number of improper magnitudes was much larger in @xmath6-band . removing these six objects from our sample , we are left with a set of 99 stars which have well measured proper motions and good magnitudes in @xmath4 , @xmath56 , @xmath8 and @xmath9 . we call this our `` first sample . '' we list these candidates in table [ table : sample ] . l l r r r r r r i d & ra & dec & @xmath4@xmath57 & @xmath7@xmath57 & @xmath8@xmath57 & @xmath9@xmath57 & ra pm & dec pm + sdfpm j132339.4 + 271916 & 200.91424 & 27.32130 & 21.784 & 21.025 & 20.164 & 19.607 & -0.049 @xmath58 0.004 & -0.024 @xmath58 0.004 + sdfpm j132343.5 + 272230 & 200.93138 & 27.37519 & 20.889 & 20.198 & 19.501 & 18.853 & -0.034 @xmath58 0.004 & -0.003 @xmath58 0.006 + sdfpm j132343.6 + 272753 & 200.93170 & 27.46494 & 22.821 & 21.995 & 20.868 & 20.265 & -0.017 @xmath58 0.002 & 0.001 @xmath58 0.004 + sdfpm j132346.9 + 274556 & 200.94582 & 27.76573 & 21.810 & 20.993 & 20.003 & 19.425 & -0.073 @xmath58 0.004 & -0.018 @xmath58 0.004 + sdfpm j132347.5 + 271829 & 200.94827 & 27.30829 & 21.711 & 21.005 & 20.505 & 20.139 & -0.037 @xmath58 0.004 & -0.023 @xmath58 0.004 + sdfpm j132348.0 + 273232 & 200.95033 & 27.54234 & 24.208 & 23.350 & 22.209 & 21.615 & -0.053 @xmath58 0.002 & 0.018 @xmath58 0.002 + sdfpm j132348.0 + 273702 & 200.95029 & 27.61729 & 23.585 & 22.739 & 21.520 & 20.867 & -0.022 @xmath58 0.002 & -0.002 @xmath58 0.002 + sdfpm j132350.2 + 272744 & 200.95944 & 27.46232 & 19.051 & 19.110 & 19.193 & 19.141 & 0.030 @xmath58 0.004 & -0.036 @xmath58 0.004 + sdfpm j132351.0 + 273453 & 200.96275 & 27.58148 & 21.617 & 20.713 & 19.596 & 18.739 & 0.017 @xmath58 0.002 & -0.027 @xmath58 0.004 + sdfpm j132353.1 + 272759 & 200.97157 & 27.46647 & 23.648 & 22.870 & 22.073 & 21.626 & -0.006 @xmath58 0.002 & -0.018 @xmath58 0.002 + sdfpm j132353.4 + 272207 & 200.97253 & 27.36882 & 20.417 & 19.870 & 19.564 & 19.130 & -0.034 @xmath58 0.004 & -0.035 @xmath58 0.002 + sdfpm j132354.0 + 272806 & 200.97518 & 27.46843 & 24.444 & 23.488 & 21.818 & 20.910 & -0.012 @xmath58 0.000 & -0.009 @xmath58 0.002 + sdfpm j132354.3 + 273016 & 200.97647 & 27.50454 & 23.857 & 23.334 & 22.733 & 22.322 & -0.032 @xmath58 0.004 & 0.001 @xmath58 0.004 + sdfpm j132354.5 + 273356 & 200.97722 & 27.56561 & 22.793 & 22.390 & 22.199 & 22.029 & -0.056 @xmath58 0.004 & -0.042 @xmath58 0.002 + sdfpm j132356.7 + 274445 & 200.98637 & 27.74593 & 22.248 & 21.560 & 21.150 & 20.835 & -0.024 @xmath58 0.004 & -0.025 @xmath58 0.002 + sdfpm j132357.5 + 271458 & 200.98977 & 27.24948 & 23.025 & 22.144 & 20.812 & 20.046 & -0.019 @xmath58 0.002 & -0.007 @xmath58 0.002 + sdfpm j132400.0 + 273750 & 201.00007 & 27.63083 & 21.997 & 21.169 & 20.115 & 19.443 & -0.009 @xmath58 0.000 & -0.021 @xmath58 0.004 + sdfpm j132401.1 + 273613 & 201.00494 & 27.60384 & 23.932 & 23.034 & 21.688 & 20.989 & -0.036 @xmath58 0.002 & -0.002 @xmath58 0.002 + sdfpm j132402.5 + 272639 & 201.01074 & 27.44421 & 25.483 & 24.796 & 23.022 & 22.008 & 0.003 @xmath58 0.002 & -0.025 @xmath58 0.002 + sdfpm j132403.6 + 271314 & 201.01540 & 27.22081 & 24.830 & 24.027 & 23.312 & 22.843 & -0.024 @xmath58 0.004 & -0.027 @xmath58 0.006 + sdfpm j132403.6 + 273833 & 201.01540 & 27.64265 & 22.152 & 21.247 & 19.998 & 19.174 & -0.022 @xmath58 0.004 & -0.011 @xmath58 0.002 + sdfpm j132404.5 + 272557 & 201.01911 & 27.43275 & 24.593 & 23.464 & 21.534 & 20.563 & -0.033 @xmath58 0.004 & -0.006 @xmath58 0.002 + sdfpm j132404.5 + 273829 & 201.01897 & 27.64141 & 23.774 & 23.016 & 22.452 & 22.084 & -0.023 @xmath58 0.002 & -0.026 @xmath58 0.002 + sdfpm j132407.4 + 272751 & 201.03110 & 27.46440 & 23.326 & 22.556 & 22.030 & 21.693 & -0.032 @xmath58 0.002 & 0.009 @xmath58 0.002 + sdfpm j132407.4 + 272924 & 201.03107 & 27.49026 & 20.825 & 20.089 & 19.268 & 18.488 & 0.008 @xmath58 0.002 & -0.036 @xmath58 0.004 + sdfpm j132407.8 + 273621 & 201.03275 & 27.60605 & 20.992 & 20.211 & 19.334 & 18.498 & -0.011 @xmath58 0.002 & -0.028 @xmath58 0.002 + sdfpm j132408.0 + 274455 & 201.03344 & 27.74886 & 23.205 & 23.035 & 23.011 & 22.952 & -0.031 @xmath58 0.004 & -0.005 @xmath58 0.002 + sdfpm j132409.7 + 273406 & 201.04050 & 27.56846 & 22.629 & 21.914 & 21.501 & 21.195 & -0.018 @xmath58 0.002 & -0.032 @xmath58 0.002 + sdfpm j132411.1 + 271310 & 201.04634 & 27.21970 & 21.478 & 20.735 & 19.868 & 19.261 & 0.001 @xmath58 0.002 & -0.036 @xmath58 0.004 + sdfpm j132412.6 + 272220 & 201.05261 & 27.37244 & 21.385 & 20.713 & 20.279 & 19.940 & -0.005 @xmath58 0.002 & -0.039 @xmath58 0.002 + sdfpm j132413.0 + 272651 & 201.05433 & 27.44771 & 21.211 & 20.568 & 20.195 & 19.890 & 0.007 @xmath58 0.002 & -0.048 @xmath58 0.002 + sdfpm j132413.2 + 273139 & 201.05511 & 27.52759 & 24.476 & 23.966 & 23.694 & 23.519 & 0.024 @xmath58 0.004 & -0.047 @xmath58 0.002 + sdfpm j132413.5 + 271547 & 201.05657 & 27.26317 & 23.270 & 22.586 & 22.068 & 21.668 & -0.010 @xmath58 0.002 & -0.024 @xmath58 0.002 + sdfpm j132415.4 + 271328 & 201.06428 & 27.22462 & 23.341 & 22.562 & 21.837 & 21.368 & -0.024 @xmath58 0.004 & -0.013 @xmath58 0.002 + sdfpm j132415.9 + 271624 & 201.06654 & 27.27336 & 21.285 & 21.492 & 21.654 & 21.798 & -0.037 @xmath58 0.002 & 0.004 @xmath58 0.004 + sdfpm j132418.0 + 271902 & 201.07534 & 27.31740 & 20.754 & 20.099 & 19.576 & 19.011 & -0.029 @xmath58 0.004 & 0.001 @xmath58 0.004 + sdfpm j132419.4 + 273411 & 201.08115 & 27.56974 & 23.404 & 22.691 & 22.296 & 21.994 & -0.013 @xmath58 0.002 & -0.026 @xmath58 0.004 + sdfpm j132423.6 + 274031 & 201.09851 & 27.67550 & 24.878 & 24.092 & 22.810 & 22.094 & -0.008 @xmath58 0.002 & -0.030 @xmath58 0.002 + sdfpm j132423.7 + 274425 & 201.09904 & 27.74048 & 24.684 & 23.560 & 21.358 & 20.143 & -0.023 @xmath58 0.002 & 0.026 @xmath58 0.004 + sdfpm j132425.8 + 272415 & 201.10777 & 27.40424 & 22.134 & 21.348 & 20.242 & 19.579 & 0.015 @xmath58 0.004 & -0.018 @xmath58 0.002 + sdfpm j132427.4 + 271919 & 201.11456 & 27.32220 & 21.667 & 20.920 & 19.939 & 19.277 & -0.016 @xmath58 0.002 & -0.006 @xmath58 0.002 + sdfpm j132427.5 + 274302 & 201.11464 & 27.71747 & 21.899 & 21.050 & 19.979 & 19.326 & -0.004 @xmath58 0.000 & -0.029 @xmath58 0.004 + sdfpm j132429.2 + 273817 & 201.12177 & 27.63832 & 22.417 & 21.510 & 20.291 & 19.581 & -0.031 @xmath58 0.002 & 0.008 @xmath58 0.002 + sdfpm j132429.7 + 273932 & 201.12378 & 27.65908 & 23.801 & 23.352 & 23.149 & 22.980 & -0.019 @xmath58 0.002 & -0.011 @xmath58 0.002 + sdfpm j132429.8 + 274304 & 201.12447 & 27.71788 & 24.576 & 23.709 & 22.521 & 21.894 & -0.002 @xmath58 0.002 & -0.026 @xmath58 0.002 + sdfpm j132430.6 + 272406 & 201.12789 & 27.40185 & 25.066 & 24.034 & 22.769 & 22.151 & -0.026 @xmath58 0.002 & -0.014 @xmath58 0.004 + sdfpm j132430.9 + 273624 & 201.12897 & 27.60693 & 23.844 & 23.145 & 22.437 & 22.015 & -0.001 @xmath58 0.002 & -0.036 @xmath58 0.002 + sdfpm j132431.3 + 271528 & 201.13047 & 27.25786 & 23.296 & 22.392 & 21.111 & 20.377 & -0.008 @xmath58 0.002 & -0.028 @xmath58 0.002 + sdfpm j132431.9 + 272236 & 201.13305 & 27.37668 & 20.942 & 20.325 & 19.957 & 19.596 & -0.007 @xmath58 0.002 & -0.039 @xmath58 0.004 + sdfpm j132432.0 + 273510 & 201.13343 & 27.58623 & 26.614 & 25.586 & 23.635 & 22.704 & -0.005 @xmath58 0.002 & -0.027 @xmath58 0.002 + sdfpm j132432.9 + 274301 & 201.13713 & 27.71701 & 21.809 & 21.082 & 20.626 & 20.313 & 0.012 @xmath58 0.004 & -0.053 @xmath58 0.002 + sdfpm j132434.5 + 271432 & 201.14385 & 27.24230 & 21.333 & 20.533 & 19.396 & 18.506 & -0.074 @xmath58 0.006 & 0.042 @xmath58 0.006 + sdfpm j132435.0 + 271638 & 201.14597 & 27.27749 & 24.078 & 23.314 & 22.761 & 22.360 & -0.022 @xmath58 0.002 & -0.010 @xmath58 0.002 + sdfpm j132436.5 + 272345 & 201.15221 & 27.39585 & 25.624 & 24.331 & 22.079 & 20.895 & 0.022 @xmath58 0.004 & -0.027 @xmath58 0.002 + sdfpm j132438.0 + 273622 & 201.15867 & 27.60634 & 20.594 & 20.088 & 19.836 & 19.525 & -0.048 @xmath58 0.002 & -0.054 @xmath58 0.002 + sdfpm j132438.4 + 273433 & 201.16001 & 27.57608 & 25.234 & 24.507 & 23.642 & 23.123 & -0.037 @xmath58 0.004 & -0.007 @xmath58 0.006 + sdfpm j132438.8 + 272847 & 201.16168 & 27.47980 & 21.242 & 20.387 & 19.401 & 18.580 & 0.019 @xmath58 0.004 & -0.017 @xmath58 0.002 + sdfpm j132439.0 + 272413 & 201.16284 & 27.40377 & 21.748 & 20.978 & 20.063 & 19.473 & -0.029 @xmath58 0.004 & -0.021 @xmath58 0.004 + sdfpm j132439.2 + 273949 & 201.16367 & 27.66381 & 25.890 & 25.991 & 23.708 & 22.493 & -0.025 @xmath58 0.002 & 0.006 @xmath58 0.002 + sdfpm j132440.4 + 272911 & 201.16870 & 27.48653 & 20.569 & 20.101 & 19.872 & 19.567 & -0.010 @xmath58 0.002 & -0.012 @xmath58 0.002 + sdfpm j132440.6 + 272548 & 201.16954 & 27.43018 & 25.902 & 25.050 & 23.210 & 22.293 & 0.002 @xmath58 0.002 & -0.033 @xmath58 0.004 + sdfpm j132440.7 + 271501 & 201.16992 & 27.25041 & 20.436 & 20.017 & 19.781 & 19.510 & 0.000 @xmath58 0.002 & -0.024 @xmath58 0.002 + sdfpm j132444.4 + 273945 & 201.18514 & 27.66254 & 23.530 & 22.957 & 22.607 & 22.279 & -0.016 @xmath58 0.002 & -0.011 @xmath58 0.002 + sdfpm j132444.9 + 272709 & 201.18741 & 27.45264 & 22.431 & 21.622 & 20.613 & 20.051 & 0.013 @xmath58 0.004 & -0.023 @xmath58 0.002 + sdfpm j132446.0 + 272605 & 201.19208 & 27.43499 & 22.250 & 21.553 & 21.112 & 20.813 & -0.026 @xmath58 0.002 & -0.004 @xmath58 0.002 + sdfpm j132446.8 + 274114 & 201.19519 & 27.68733 & 24.778 & 23.590 & 21.544 & 20.448 & -0.040 @xmath58 0.002 & 0.012 @xmath58 0.004 + sdfpm j132447.0 + 272814 & 201.19610 & 27.47077 & 23.045 & 22.143 & 20.783 & 20.064 & 0.008 @xmath58 0.000 & -0.025 @xmath58 0.002 + sdfpm j132447.8 + 272157 & 201.19930 & 27.36586 & 23.236 & 22.290 & 20.836 & 19.993 & -0.025 @xmath58 0.002 & 0.002 @xmath58 0.002 + sdfpm j132448.1 + 272803 & 201.20046 & 27.46761 & 21.790 & 21.173 & 20.812 & 20.554 & -0.025 @xmath58 0.002 & -0.004 @xmath58 0.002 + sdfpm j132448.6 + 272747 & 201.20255 & 27.46314 & 25.725 & 24.967 & 23.319 & 22.436 & -0.022 @xmath58 0.002 & -0.010 @xmath58 0.002 + sdfpm j132448.8 + 273205 & 201.20361 & 27.53479 & 21.771 & 20.993 & 20.036 & 19.427 & -0.081 @xmath58 0.004 & 0.007 @xmath58 0.002 + sdfpm j132449.7 + 274510 & 201.20714 & 27.75295 & 25.447 & 24.324 & 22.278 & 21.207 & -0.043 @xmath58 0.002 & -0.042 @xmath58 0.002 + sdfpm j132451.7 + 273110 & 201.21556 & 27.51972 & 25.980 & 25.510 & 25.241 & 24.948 & -0.021 @xmath58 0.008 & -0.042 @xmath58 0.004 + sdfpm j132452.1 + 271813 & 201.21743 & 27.30386 & 24.103 & 23.165 & 21.596 & 20.792 & -0.025 @xmath58 0.002 & -0.007 @xmath58 0.004 + sdfpm j132453.1 + 271821 & 201.22154 & 27.30592 & 22.338 & 21.642 & 21.242 & 20.944 & -0.035 @xmath58 0.002 & -0.029 @xmath58 0.004 + sdfpm j132454.0 + 274226 & 201.22508 & 27.70739 & 23.462 & 22.682 & 21.698 & 21.148 & -0.012 @xmath58 0.002 & -0.013 @xmath58 0.002 + sdfpm j132455.3 + 272957 & 201.23056 & 27.49939 & 25.095 & 24.307 & 23.423 & 22.968 & -0.022 @xmath58 0.002 & -0.004 @xmath58 0.004 + sdfpm j132456.0 + 274126 & 201.23345 & 27.69078 & 25.805 & 24.801 & 22.578 & 21.065 & -0.060 @xmath58 0.002 & 0.020 @xmath58 0.002 + sdfpm j132456.1 + 272807 & 201.23412 & 27.46872 & 22.942 & 22.059 & 20.673 & 19.936 & -0.030 @xmath58 0.004 & -0.011 @xmath58 0.000 + sdfpm j132458.1 + 272326 & 201.24241 & 27.39072 & 23.633 & 22.700 & 21.372 & 20.676 & -0.022 @xmath58 0.002 & 0.011 @xmath58 0.002 + sdfpm j132459.8 + 271251 & 201.24932 & 27.21436 & 23.548 & 22.849 & 22.374 & 21.961 & -0.022 @xmath58 0.002 & -0.006 @xmath58 0.006 + sdfpm j132500.3 + 272357 & 201.25147 & 27.39927 & 22.401 & 21.675 & 21.122 & 20.775 & -0.015 @xmath58 0.002 & -0.010 @xmath58 0.002 + sdfpm j132503.8 + 273938 & 201.26620 & 27.66072 & 21.497 & 20.741 & 19.999 & 19.479 & -0.018 @xmath58 0.002 & -0.006 @xmath58 0.002 + sdfpm j132504.6 + 273028 & 201.26917 & 27.50779 & 21.614 & 20.808 & 19.792 & 19.074 & -0.021 @xmath58 0.002 & -0.011 @xmath58 0.000 + sdfpm j132505.4 + 273731 & 201.27253 & 27.62530 & 22.684 & 21.943 & 21.160 & 20.728 & -0.017 @xmath58 0.002 & -0.024 @xmath58 0.002 + sdfpm j132505.3 + 271440 & 201.27212 & 27.24446 & 22.854 & 22.143 & 21.642 & 21.233 & -0.007 @xmath58 0.002 & -0.029 @xmath58 0.002 + sdfpm j132505.5 + 273401 & 201.27318 & 27.56696 & 22.398 & 21.772 & 21.336 & 21.058 & -0.021 @xmath58 0.002 & -0.011 @xmath58 0.002 + sdfpm j132506.1 + 273816 & 201.27567 & 27.63800 & 21.999 & 21.309 & 20.880 & 20.594 & -0.007 @xmath58 0.002 & -0.034 @xmath58 0.002 + sdfpm j132508.4 + 273553 & 201.28501 & 27.59830 & 20.672 & 20.018 & 19.546 & 19.011 & -0.003 @xmath58 0.002 & -0.029 @xmath58 0.004 + sdfpm j132512.6 + 271620 & 201.30271 & 27.27244 & 22.398 & 21.731 & 21.362 & 21.034 & -0.009 @xmath58 0.000 & -0.011 @xmath58 0.000 + sdfpm j132512.9 + 274045 & 201.30410 & 27.67940 & 25.443 & 24.610 & 23.500 & 22.863 & -0.015 @xmath58 0.002 & -0.009 @xmath58 0.002 + sdfpm j132514.3 + 272421 & 201.30981 & 27.40588 & 21.422 & 20.499 & 19.379 & 18.486 & -0.007 @xmath58 0.002 & -0.018 @xmath58 0.002 + sdfpm j132514.7 + 272642 & 201.31146 & 27.44522 & 23.121 & 22.559 & 22.187 & 21.881 & -0.025 @xmath58 0.004 & -0.018 @xmath58 0.006 + sdfpm j132514.7 + 271707 & 201.31146 & 27.28543 & 22.993 & 22.180 & 21.211 & 20.640 & -0.027 @xmath58 0.002 & -0.046 @xmath58 0.004 + sdfpm j132515.3 + 274212 & 201.31400 & 27.70335 & 22.304 & 22.297 & 22.357 & 22.406 & -0.004 @xmath58 0.004 & -0.089 @xmath58 0.002 + sdfpm j132515.7 + 272708 & 201.31560 & 27.45226 & 24.696 & 23.629 & 21.645 & 20.570 & -0.031 @xmath58 0.002 & -0.012 @xmath58 0.002 + sdfpm j132516.9 + 274518 & 201.32065 & 27.75507 & 23.660 & 22.886 & 22.281 & 21.891 & 0.008 @xmath58 0.004 & -0.032 @xmath58 0.004 + sdfpm j132521.1 + 271927 & 201.33814 & 27.32435 & 25.853 & 25.090 & 24.331 & 23.887 & -0.018 @xmath58 0.004 & -0.031 @xmath58 0.006 + sdfpm j132525.4 + 273755 & 201.35591 & 27.63220 & 22.179 & 21.330 & 20.105 & 19.355 & -0.031 @xmath58 0.002 & -0.004 @xmath58 0.002 + sdfpm j132527.7 + 274407 & 201.36559 & 27.73528 & 25.738 & 24.934 & 23.663 & 22.972 & -0.016 @xmath58 0.002 & -0.043 @xmath58 0.004 + sdfpm j132527.7 + 272350 & 201.36576 & 27.39737 & 22.596 & 21.898 & 21.498 & 21.218 & -0.010 @xmath58 0.002 & -0.019 @xmath58 0.002 + sdfpm j132528.3 + 274355 & 201.36796 & 27.73214 & 25.676 & 24.699 & 22.785 & 21.825 & -0.043 @xmath58 0.002 & 0.003 @xmath58 0.002 + sdfpm j132528.3 + 272012 & 201.36811 & 27.33679 & 22.163 & 21.424 & 20.568 & 20.040 & -0.031 @xmath58 0.004 & -0.002 @xmath58 0.002 + sdfpm j132533.6 + 274708 & 201.39036 & 27.78563 & 23.894 & 22.665 & 20.660 & 19.533 & -0.061 @xmath58 0.008 & 0.041 @xmath58 0.006 + sdfpm j132533.9 + 272808 & 201.39152 & 27.46905 & 23.653 & 22.925 & 22.398 & 22.030 & -0.029 @xmath58 0.002 & -0.026 @xmath58 0.002 + before we can compare our measurements to models of galactic structure , we need to convert the suprime - cam @xmath8 magnitudes , which are calibrated on the ab system ( @xcite ; @xcite ; @xcite ) , to @xmath55 magnitudes , which are on the standard johnson - cousins system . we used synthetic photometry to find the relationship between the suprime - cam @xmath59 and johnson - cousins @xmath54 , taking bandpasses from @xcite and @xcite , respectively . we selected main sequence stars , o5v to m6v , from the library of @xcite , stars ranging in metallicity @xmath60 \leq 0 $ ] from models of @xcite , and flux - calibrated spectra of white dwarfs observed by the sdss ( @xcite ) . we convolved each spectrum with the suprime - cam passbands and with the johnson - cousins passbands to compute synthetic magnitudes , and used the spectrum of vega from @xcite to set the zeropoints to the values given in @xcite . we found that the following linear relationship fit the data well , yielding a scatter of less than 0.03 mag across the range of colors @xmath61 : @xmath62 we use this equation to convert the observed colors for stars in the sdf to johnson - cousins @xmath54 when comparing our results to stellar models . we begin with the besanon model . we generated 10 simulated catalogs of objects in the area of the sdf , using the parameters found by @xcite and including stars down to an apparent magnitude of @xmath63 . we applied cuts to the synthetic catalogs to match the combined limits of the @xmath8-band proper motion images and the sdf catalogs . @xmath64 the result should be a set of stars similar to those in the actual sdf , though ten times more numerous . the large size of this synthetic sample will make it easier to delineate sparsely populated regions in the reduced proper motion diagram , to which we now turn . reduced proper motion was introduced by @xcite as a way to separate stars of different luminosities using only the observable apparent magnitude , @xmath65 , and proper motion , @xmath66 , in units of arcseconds per year . we will base our reduced proper motion on @xmath4-band magnitudes , so that @xmath67 it is also possible to express this quantity in terms of a star s absolute magnitude , @xmath68 , and tangential velocity , @xmath69 , expressed in units of km / s , @xmath70 using this version of the formula , we compute @xmath71 for stars in the simulated catalogs produced from the besanon model . figure [ fig : reduced_besancon ] shows the reduced proper motion as a function of @xmath54 color . we assigned objects in the simulation to three populations based on their metallicity and the component of their space velocity in the direction of galactic rotation , which we denote as @xmath72 to avoid confusion with the passband @xmath4 . * if @xmath73 and @xmath74 < -1.20 $ ] , we assign the star to the * halo * * if @xmath75 and @xmath74 > -0.50 $ ] , we assign the star to the * thin disk * * otherwise , we assign the star to the * thick disk * stars from different populations appear in distinct regions in this diagram . we have drawn rough outlines by hand to aid the reader in recognizing the populations . ( 150mm,240mm)reduced_pm_f_mono.eps note that there is a clear `` red edge '' in the distribution of wds , at a color of @xmath76 . theoretical models of cooling wds ( @xcite ; @xcite ) indicate that at an age of about 10 gyr and a temperature of @xmath77 , an increase in opacity due to molecular hydrogen causes the @xmath54 color to shift back to the blue as the star continues to cool . wds with atmospheres dominated by other elements , such as helium or carbon , would continue to grow redder as they cool . objects near the bottom of the wd region are likely to be members of the halo . in figure [ fig : reduced_sdf ] , we present the reduced proper motions for the real stars in our first sample . we must switch to the first form of reduced proper motion , equation 7 , to compute @xmath71 for the observed stars . to facilitate comparison with the besanon model , we include the hand - drawn regions from figure [ fig : reduced_besancon ] as well as all objects from the simulated catalogs as tiny points . note that the saturation of very bright stars @xmath78 in the subaru images , plus our limited ability to measure proper motions @xmath79 per year , combine to eliminate any candidates with reduced proper motions @xmath80 . in the discussion which follows , please recall that the boundaries of the regions drawn in the diagram are only approximations intended to provide rough classifications ; the number of items within each region could change by ten or twenty percent if one shifted the boundaries slightly . ( 150mm,240mm)reduced_pm_g_mono.eps our 99 proper motion ( pm ) candidates divide into four groups : 9 fall inside the wd region , 43 inside the halo region , 23 in the disk region , and 24 lie in an area which had no stars in the besanon model . let us discuss the wd candidates first , and then consider the objects in the `` empty '' area . the pm candidates falling into the wd region are concentrated near the red edge of the region , just as models of wd cooling predict . we take the combined results for ten simulated catalogs generated from the besanon model , correct for completeness as a function of magnitude and proper motion , based on our tests with artificial stars ( see figure [ fig : moving_signif ] ) , and divide by 10 to find predictions of @xmath81 wds in the halo , @xmath82 in the thick disk and @xmath83 in the thin disk , for a total of @xmath84 wds satisfying our selection criteria in the sdf . our sample yields 9 candidates in this region , consistent with the model . note that several candidates lie just outside the wd region ; we must make additional measurements of these objects before we can make any confident claim about the exact number of wds . which of our candidates are most likely to be members of the halo ? as a young wd cools , it slides diagonally down and to the right on this diagram , parallel to the lower envelope of the simulation s objects . due to their high velocities , halo wds should lie near the bottom of the distribution . there are several candidates at some distance below the main locus of simulation objects ; we believe the two at @xmath85 and @xmath86 are the most likely of our candidates to be halo dwars . roughly one - quarter of our pm candidates lie in an `` empty '' region between the simulation s wd and halo stars . since we drew the boundaries by hand , they may certainly be shifted by small amounts ; that would cause a number of the anamolous objects to fall within the halo or wd regions . however , some of our candidates are more than one magnitude away from any population in the simulations . is it possible that they may be ordinary stars reddened by dust ? according to @xcite , dust along the line of sight through the sdf should produce @xmath87 mag ; the corresponding extinction , @xmath88 and @xmath89 , is too small to shift candidates a significant distance in the reduced pm diagram . we suggest that these objects may be metal - poor subdwarfs in the halo , a class of star which is not included in the besanon model . if we make an hr diagram using @xmath90 and @xmath54 , we find that the halo stars in the besanon model cross the disk main sequence in the range @xmath91 , and are slightly more luminous in the redder portion of this range . however , as @xcite show , extremely metal - poor subdwarfs are much fainter than disk stars in this color range , by up to 4 magnitudes . stars with these photometric properties and halo kinematics would lie several magnitudes below the halo region drawn in our diagrams . we therefore tentatively identify as extremely metal - poor halo stars the pm candidates which fall far from the wd and halo regions . our first step will be to acquire spectra of some of our pm candidates to verify their identity as wds and metal - poor subdwarfs . after we have spectra for a good fraction of our candidates , we can assign types to the candidates with more confidence . at that point , we will make a more detailed and quantitative comparison of the wds in our sample with those expected from models of the milky way . we can also look at candidates with motions of somewhat lower significance . in this paper , we examined the 110 stars which had @xmath92 . there were 72 stars with @xmath93 , and they show nearly the same degree of asymmetry in their motions as our first sample . it is likely that a significant number of these stars have real proper motions . however , it will take extra effort to distinguish them from the growing number of false detections , and to check their photometry ( many of them are fainter than the stars in our sample ) . as we sift through this set of stars with less significant motions , we can also try to improve the @xmath6 and @xmath4 measurements of our stars , so that we will not discard so many candidates due to their uncertain colors . finally , we can apply the same techniques to find stars with large proper motions in other deep fields with multiple visits by subaru . our next target will be the area of the subaru / xmm - newton deep survey ( @xcite , @xcite ) , which is roughly in the opposite direction from the sdf . we can use it to check our results in the sdf , but also to look for differences predicted by models of galactic populations . we thank the staff at the subaru telescope for their assistance with the observations used in this project . mwr gratefully acknowledges grant s-03031 from the jsps invitation fellowship for research in japan . this work is supported in part by a jsps core - to - core program `` international research network for dark energy '' and by a jsps research grant ( 17104002 ) . long ago , charles alcock suggested that we use suprime - cam to search for white dwarfs ; we are glad to be putting his idea into practice at last . acknowledges support by the benoziyo center for astrophysics and the william z. and eda bess novick new scientists fund at the weizmann institute . funding for the sloan digital sky survey ( sdss ) and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , and the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , the university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . adelman - mccarthy , j. et al . 2008 , , 175 , 297 bessell , m. s. 1990 , , 102 , 1181 bohlin , r. c. & gilliland , r. l. 2004 , , 127 , 3508 carollo , d. et al . 2006 , , 448 , 579 chabrier , g. et al . 2000 , , 543 , 216 droege , t. f. et al . 2006 , , 118 , 1666 fukugita , m. et al . 1996 , , 111 , 1748 furusawa , h. et al . 2008 , , 176 , 1 ibata , r. et al . 2007 , , 671 , 1591 juric , m. et al . 2008 , apj , 673 , 864 liebert , j. et al . 1988 , apj , 332 , 891 kashikawa , n. et al . 2004 , pasj , 56 , 1011 kalirai , j. s. et al . 2004 , apj , 601 , 277 lejeune , t. , cuisinier , f. , & buser , r. 1997 , , 125 , 229 luyten , w. j. 1922 , , 34 , 54 mendez , r. a. 2002 , , 395 , 779 miyazaki , s. et al . 2002 , , 54 , 833 nagao , t. et al . 2004 , , 613 , 9 nelson , c. a. et al . 2002 , apj , 573 , 644 oppenheimer , b. r. et al . 2001 , science , 292 , 698 ouchi , m. et al . 2004 , , 611 , 660 patel , k. c. & spath , s. r. 2004 , spie , 5487 , 112 pickles , a. j. 1998 , , 110 , 863 poznanski , d. et al . 2007 , , 382 , 1169 press , w. h. , flannery , b. p. & teukolsky , s. a. 1986 , `` numerical recipes in c : the art of scientific computing , '' ( cambridge : cambridge university press ) reid , i. n. 2005 , , 43 , 247 reid , i. n. & gizis , j. e. 1998 , , 116 , 2929 richer , h. b. 2000 , , 529 , 318 robin , a. c. , reyl , c. , derrire , s. , & picaud , s. 2003 , a&a , 409 , 523 richmond , m. w. 2005 , , 57 , 969 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 sekiguchi , k. et al . 2004 , , 205 , 8105 shimasaku , k. et al . 2006 , , 58 , 313 skrutskie , m. f. et al . 2006 , aj , 131 , 1163 stone , r. c. 1989 , , 97 , 1227 treffers , r. r. & richmond , m. w. 1989 , pasp , 101 , 725 yanny , b. et al . 2000 , apj , 540 , 825 yagi , m. et al . 2002 , , 123 , 66

we present new techiques for improving the efficiency of supernova ( sn ) classification at high redshift using 64 candidates observed at gemini north and south during the first year of the supernova legacy survey ( snls ) . the snls is an ongoing five year project with the goal of measuring the equation of state of dark energy by discovering and following over 700 high - redshift sne ia using data from the canada - france - hawaii telescope legacy survey . we achieve an improvement in the sn ia spectroscopic confirmation rate : at gemini 71% of candidates are now confirmed as sne ia , compared to 54% using the methods of previous surveys . this is despite the comparatively high redshift of this sample , where the median sn ia redshift is ( ) . these improvements were realized because we use the unprecedented color coverage and lightcurve sampling of the snls to predict whether a candidate is an sn ia and estimate its redshift , before obtaining a spectrum , using a new technique called the `` sn photo - z . '' in addition , we have improved techniques for galaxy subtraction and sn template fitting , allowing us to identify candidates even when they are only 15% as bright as the host galaxy . the largest impediment to sn identification is found to be host galaxy contamination of the spectrum when the sn was at least as bright as the underlying host galaxy the target was identified more than 90% of the time . however , even sne on bright host galaxies can be easily identified in good seeing conditions . when the image quality was better than 0.55 , the candidate was identified 88% of the time . over the five - year course of the survey , using the selection techniques presented here we will be able to add more confirmed sne ia than would be possible using previous methods .

astro-ph0509195

type ia supernovae ( sne ia ) have been used as standardized candles to discover the acceleration of the universe @xcite . programs are now underway to characterize the dark energy driving this expansion by measuring its average equation of state , @xmath4 . essence aims to discover and follow - up @xmath5 200 sne ia for this purpose @xcite , while the goal of the snls ( supernova legacy survey ) is to obtain 700 well observed sne ia in the redshift range 0.2 to 0.9 @xcite . the snls uses the canada - france - hawaii telescope legacy survey ( cfhtls ) imaging data for sn discoveries and lightcurves . over the course of a year , four fields are imaged every four days ( 23 days rest frame ) during dark and gray time , in four filters ( griz ) . since the same fields are continuously imaged , this is a `` rolling search , '' where new sn candidates are discovered throughout the month as lightcurves are being built . the snls uses the very large telescope ( vlt ) , keck and gemini telescopes to determine sne types and redshifts . the vlt has the shortest setup time , so it handles the bulk of the lower redshift ( @xmath6 ) candidates , as well as some higher redshift targets . gemini - n and gemini - s generally observe the faintest , highest redshift targets ( @xmath7 ) , where the nod - and - shuffle mode provides a reduction of sky line residuals in the red part of the spectrum . finally , keck provides essential observations when the northernmost field can not be observed by the vlt . here we present the methods and spectroscopy from the gemini telescopes for the first year of operation of the snls . separate science analyses on certain spectroscopic features , as well as observations from other telescopes , will be presented in upcoming papers . a note on our candidate names : the classical international astronomical union ( iau ) naming convention for supernovae does not work well for projects of the scale of the snls , where there are more candidates than confirmed sne . under the traditional naming system , the candidates from high - redshift searches have an internal ( unofficial ) name that they were given at the time of discovery , then if they are suspected sne they are given one type of official iau name , and if they are confirmed sne they are given a second type of official iau name . to avoid these problems , we use a unified naming scheme for all snls discoveries , regardless of whether they are confirmed to be sne . an example name is snls-04d3aa the first two digits are the year of discovery , the second two are the field name ( d1 , d2 , d3 , d4 ) , and the final two are identifying letters . we start with `` aa '' for the first candidate discovered in a field in a given year . the second is `` ab , '' then `` ac '' and so on , until the 27th candidate , which is named `` ba . '' historically , only @xmath8 of high - redshift sn candidates observed spectroscopically in the range @xmath9 are confirmed as type ia sne . the rest of the candidates are either identified or unidentified core collapse sne or agn , or are sne ia which have such high host contamination or low signal - to - noise ( s / n ) that they can not be identified . the exact percentage confirmed depends on the design of the search and the redshift range studied ( see section [ photoz ] for a detailed discussion ) . also found that the sn ia confirmation rate was higher for classical searches than it was for `` rolling '' searches . a classical high - z sn search technique @xcite is optimized for the discovery of sne ia . it involves taking a reference image after new moon to be subtracted from a discovery image taken @xmath103 weeks later ( 2 weeks restframe at z=0.5 ) . since the rise time of sne ia is about 18 days , this guarantees that the sne discovered will be either rising or near maximum light . cuts are made to retain only candidates that have increased in brightness by an amount consistent with the rapid rise of an sn ia . in `` rolling searches , '' such as the snls and essence , the field is continuously imaged every few days in several filters . sne are discovered very early , so there is no built - in bias toward the discovery of sne ia ( except that sne ia are usually brighter than core collapse sne at a given redshift ) . rolling searches are typically limited not by the number of candidates discovered , but by the amount of spectoscopic confirmation time available , so techniques must be developed to increase the percentage of targets confirmed as sne ia by spectroscopy . in principle , this should be easier in a rolling search than in a classical search . in the snls , sne are almost almost always discovered well before maximum light , so multicolor information on the rising part of the lightcurve is available before the candidates are observed spectroscopically . here we demonstrate the use of a technique to prioritize candidate spectroscopy . after each photometric data point is taken , a candidate s lightcurve is fit under the assumption that it is an sn ia . the lightcurve shape , phase , and redshift of the candidate are allowed to vary in the fit , while the milky way reddening is fixed to the values of @xcite . host reddening can be solved for or held fixed , depending on the number of data points fit . we do not use a fit until it has detections in at least two epochs in at least two filters , as well as at least one prior epoch of nondetections . in practice , most candidates have detections on several epochs in at least three filters , and dozens of nondetections . significantly deviant objects are lowered in priority ( for example , type ii sne and agn are often too blue at early times ( see * ? ? ? * ; * ? ? ? * ) ) , while rising candidates that are consistent with being an sn ia are given top priority for spectroscopic observation . because the inferred color of the object depends on the redshift , one product of the fitting process is an estimated redshift . we call this a `` supernova photometric redshift , '' or `` sn photo - z , '' since the redshift is estimated from prior assumptions about the intrinsic brightness of sne ia ( 0.2 mag of dispersion is allowed for in the assumed sn ia properties ) . the sn photo - z is independent from galaxy photometric redshifts , and is useful for determing the telescope and instrument setup necessary to observe key sn ia spectroscopic features . note that galaxy photometric redshifts were not used in assigning the follow - up priority for the targets considered in this paper . figures [ iapz ] and [ iipz ] show examples of sn photo - z fits to objects that were eventually confirmed to be type ia and type ii sne , respectively . while we show the entire lightcurve fit here for clarity , it is apparent even before maximum light that the candidate in figure [ iapz ] is likely to be an sn ia , while the candidate in figure [ iipz ] is too blue at early times to be an sn ia . the candidates are usually spectroscopically observed near maximum light , so most sn photo - z fits are done using only the rising part of the lightcurve . since we observe many of our highest redshift ( and therefore spectroscopically expensive ) targets at gemini , we often apply the most strict selection critera on what to observe there . therefore the data from this telescope make an excellent testbed for our selection method . however , at other telescopes we observe a broader range of candidates to ensure that we are not missing unusual sne ia . it is also important to note that the sn photo - z is only used to prioritize follow - up observations for all scientific studies we use the measured spectroscopic redshift . the sn photo - z will be described in more detail in @xcite , and the results of its implementation on our sn ia confirmation rate are given in section [ photoz ] . the observational setup and conditions for each candidate are given in table [ obs - table ] . in this section we discuss the observational modes used in greater detail . all supernovae were observed with the gemini multi object spectrographs ( gmos ) @xcite . longslit spectra of the sn candidates were taken with the gmos r400 grating ( 400 lines per mm ) , using the 0.75 slit , giving a resolution of 6.9 . the detector binning was set at @xmath11 , giving a spatial scale of 0.145 per pixel and a spectral dispersion of 1.34 per binned pixel . the slit angle was chosen to include the host galaxy wherever possible , as the galaxy features provide the most accurate determination of the redshift . the central wavelength used was based on our estimated redshift of each candidate . for candidates at estimated redshifts @xmath12 we used a central wavelength of 680 nm ( with the order blocking filter gg455 ) , covering 4650 to 8900 . for higher redshifts the 720 nm setting ( with the og515 order blocker ) was used , giving coverage from 5100 to 9300 . the observations were executed in queue mode which allows us to specify the desired observing conditions . we require better than 0.75 image quality ( corrected to zenith ) and clear sky and generally those requirements were met . observations were made in either nod - and - shuffle ( n&s ) , electronic nod - and - shuffle , or classical observing modes . during a classical observation , the object and the nearby sky are simultaneously imaged on adjacent portions of the ccd . the sky is fit column - by - column and subtracted , but this can leave systematic residuals near bright and spatially variable sky lines . since this technique requires no additional overhead we use it on the brightest targets ( @xmath13 ) . ( all snls magnitudes reported in this paper are in the ab system . ) the nod - and - shuffle mode @xcite enables more accurate sky subtraction by nodding frequently ( every 60 seconds ) between two positions along the slit . the charge on the ccd detectors is simultaneously shuffled between the illuminated science region of the ccd and the unilluminated ( storage ) region . thus , both the object and its immediate sky background are imaged on both the same pixels and at the same slit position . systematics due to pixel response , fringing in the sky lines , slit irregularities , and any temporal sky variations can be removed by subtracting the n&s acquired sky spectrum from the object spectrum . this method works well towards the red end of the spectrum ( where sky lines are more problematic ) and we found it particularly useful for candidates at magnitudes fainter than @xmath14 . typical nod distances for this method are a few arcseconds . there are a few drawbacks to nod - and - shuffle . first , the noise in sky - subtracted n&s images is higher by a factor of @xmath15 compared to classically - reduced long - slit spectra , because one image is subtracted from another . second , there are increased overheads on any observation , as each nod cycle adds approximately 24 seconds of nod time . for a typical 1800s observation , with 15 ( 60 second ) nod cycles , these overheads add an extra 360s . the extra overhead time can be minimized by choosing a small nod distance , or by employing the electronic n&s mode . the gmos instrument requires the use of an on - instrument wavefront sensor ( oiwfs ) , which provides `` fast guiding '' ( image motion compensation ) and higher order corrections . during a normal n&s observation , this sensor physically moves during each nod cycle . the electronic n&s mode avoids this by electronically changing the position that the oiwfs guides around instead of physically moving the sensor . this decreases the overheads for a n&s observation by nearly 200s ( for an 1800s exposure ) , but is only available for small nod distances ( up to 2 ) . these small nod distances are sometimes not possible when the candidate resides in an extended host galaxy . data reduction was done by collaborators at both university of toronto and oxford university using independent pipelines primarily written in iraf using the gemini software package version 1.6 . here we describe the toronto pipeline , although the reduction methods are similar for both pipelines . the independent reductions have been used to check the consistency of the final output spectra . in gmos the spectra are spread over 3 ccds . first , bias subtraction is performed using a master bias for each chip generated from bias frames covering an entire gmos queue run . next , flat - fielding is done on a chip - by - chip basis using flat - fields taken before and after each observation . the pipeline then branches depending on whether a nod - and - shuffle or classical observing setup was used . a side effect of the n&s observing is that charge - traps , local defects in the gmos detectors , appear as low - level horizontal stripes in the science observations . these are removed using a special n&s dark - frame observation . the dark image reveals the charge - traps , and is used to construct a bad pixel mask ( bpm ) that screens out these areas during image combination . additionally , between exposures the detector is translated by a few pixels to dilute the effect of the charge - traps on any one part of the ccd . the next stage ( for both observing modes ) is to locate cosmic rays in the science frames and to add them to the bpm generated from the n&s dark frame . we use the lacosmic package @xcite , which locates @xmath1699% of cosmic - rays in an image . next , n&s spectra are sky - subtracted using the gnsskysub iraf routine , while classical spectra are sky - subtracted using a spline function fit along the spatial direction with the science object pixels excluded from the fit . the resulting frames for each object are combined using an average combine , rejecting charge - traps and cosmic rays identified in the bpms . at this stage a sky frame median - filtered in the spatial direction is added back to the data to ensure the correct variance weighting can be used in the extraction stage . this leaves cosmetically clean 2-d spectral frames . the spectra are then extracted using the apall task with variance weighting ( using an appropriate effective gain and readnoise ) , subtracting off the median sky added back in the previous step . this produces an error spectrum and a science spectrum essentially free from systematic sky subtraction errors , ensuring that careful @xmath2 fits can be made to the extracted spectra ( discussed in section [ chi2 ] ) . wavelength calibration is then done on the extracted 1-d spectrum using a solution derived from arc lamps taken once per run , and tweaked using the night - sky lines for each observation . the wavelength solution is then applied to the sky - subtracted 2-d data so that galaxy lines may be overplotted ( fig . [ gallines ] ) . the final steps are flux calibration on the extracted 1-d spectrum , combined with a telluric correction ( based on standard star spectra ) , and atmospheric extinction correction derived from the effective airmass of observation and the mauna kea extinction curve of @xcite . the spectra from the 3 chips are then combined into one spectrum , and an error spectrum is simultaneously created . the chip gaps are effectively weighted to zero so that they have no influence on the @xmath2 fitting . when possible , the slit was placed through both the sn and the center of the host galaxy . in these cases the redshift was determined from lines in the host galaxy spectrum . even if the host and the object are completely blended , it is still possible to identify the host galaxy lines as they are narrower than sn features . sky - subtracted , wavelength calibrated , combined two - dimensional spectra were created for each candidate and the positions of common galaxy spectral features were overplotted , as shown in figure [ gallines ] . the lines used to identify the redshift of each host galaxy are identified in the last column of table [ der - table ] . in some cases the host galaxy was too faint and the redshift was determined from the sn spectrum . type ia sne can be identified by a lack of hydrogen in their spectra , combined with broad ( several thousand km s@xmath17 ) p - cyngi lines of elements such as , , , , blends of fe - peak lines , and sometimes , , or . see for a review of sn classification , and @xcite , , @xcite , and @xcite for particular issues and techniques associated with classification of high - redshift sne . high - redshift sn spectra are often blended with their host galaxies , so determining the sn type can be a challenge ( at high redshift the galaxies have a smaller angular size , so a greater fraction of the host light is covered by the sn point - spread function ) . some studies do not attempt to separate sn and host galaxy light , but use a cross - correlation technique to identify sne @xcite . other authors attempt to separate sn and host galaxy light using point source deconvolution . here we use a @xmath2 fitting technique to separate sn from host galaxy light and determine the sn type . it was first developed by @xcite and subsequently used by and @xcite . each sn spectrum was fit using a @xmath2 matching program which compares the observed spectrum to a library of template sne of all types covering a range of epochs . the redshift , amount of host galaxy contamination , and reddening are varied to find the best fit . at a given redshift , the code computes : @xmath19 ^ 2}{\sigma(\lambda)^2}},\ ] ] where @xmath20 is the observed spectrum , @xmath21 is the sn template spectrum , @xmath22 is the host galaxy template spectrum , @xmath23 is the redding law , @xmath24 is the error on the spectrum , and @xmath25 , @xmath26 , and @xmath27 are constants that are varied to find the best fit in host galaxy , template sn , and reddening space . if the redshift was not fixed , this equation is then reevaluated over a range of redshifts to find the minimum @xmath2 in redshift space . we use the reddening law of @xcite and @xmath28 . if the host galaxy spectrum free of sn light could be extracted , then it was used as the galaxy spectrum subtracted in the fitting process . if no galaxy contamination was evident , then no host was subtracted in the fit . for other cases , template galaxies from @xcite and were subtracted . in cases where the hubble type could be estimated from imaging , the galaxy spectral energy distribution , or narrow galaxy lines , the galaxy type was restricted in the fitting procedure . the sn template library includes 184 sn ia spectra , 75 sn ib / c spectra , and 47 sn ii spectra . the spectra were chosen to have good s / n , to span the widest possible range of epochs , to have the widest possible wavelength coverage , and to represent all known sn subtypes . the @xmath2 fitting program produces a list of the best matching sn templates , host galaxy templates , redshifts , and reddening . as in sn detection , it is not yet possible to fully automate this process a human must still inspect the results and make the ultimate determination of the sn type . the type of the sn is estimated and placed into one of six categories reflecting the type and uncertainty in the classification ( section [ ci ] ; fig . [ specfig ] ) . the spectroscopic epoch is determined from the weighted mean of the best 5 epochs ( weighted by the @xmath2 of each fit ) . the average of several epochs was chosen to dilute the effect of outliers , and to help smooth out the diversity in sn ia spectra . the exact number of epochs averaged has little effect on the result , since the average is weighted by the @xmath2 of each fit the best few fits will be the dominant contributors to the average . we find @xmath29 for an error of @xmath30 days on the spectroscopic date determination . " two factors make it harder to classify sne at high redshift as compared to their counterparts at low redshift : they generally have lower signal - to - noise spectra and there is a greater degree of contamination from the host galaxy . while many classifications are obvious , others have a degree of uncertainty associated with them . to quantify this , after examination of its spectrum we give each sn a sn ia confidence index : 1 . * certain ia : * the spectrum shows distinctive features of an sn ia such as or . often 6150 and 5400 are redshifted out of the observed spectral range , so 4000 is used as the key indictor for sne ia . however , at some phases , or for some sn ia subtypes ( e.g. sn 1991 t ) , 4000 would not be expected . in these cases the candidate can be classified as a category 5 if the spectrum is an exact match to the overall spectral energy distribution ( sed ) of an sn ia ( at the phase indicated by the lightcurve ) and no other type of sn matches . * highly probable ia : * the spectrum is a match to a ia , but lacks an unambiguous detection of one of the features that is unique to sne ia ( or ) . other sn types do not match the spectrum well . these candidates usually have another piece of confirming evidence , such as a lightcurve consistent with a sn ia at the measured redshift , or they are found in an e or s0 galaxy . * probable ia : * the spectrum matches an sn ia better than any other sn type , but another sn type ( usually sne ic ) is not ruled out from the spectrum alone . this is either because the spectrum has low s / n or because other sne look similar at the same phase . these candidates have a lightcurve consistent with an sn ia at the measured redshift , and the spectrum has a phase consistent with the lightcurve . these sne are denoted as ia * , following the notation of . * unknown : * the type can not be determined from the spectrum . often the spectrum has low signal - to - noise , or there is too much host galaxy contamination to make a reliable determination . probably not a ia : * the spectrum has features marginally inconsistent with an sn ia , but the type can not be unambiguously determined . * not a ia : * the spectral features are inconsistent with a sn ia . in this case the spectrum can usually be identified as an sn ii , sn ib / c , or an agn ( though note that there are no clear cases of agn from gemini spectra these were screened out in advance ) . the primary means of classification ibfinpstws the sn spectrum , although we use all information available to supplement this including the lightcurves , colors , and the agreement between lightcurve phase and the spectroscopic phase determined from the fitting process . for example , if the type can not be determined from the spectrum it would normally be classified as index 2 . but if the lightcurves are also inconsistent with the lightcurve of an sn ia , then it would be moved into index 1 or 0 . the host galaxy spectrum is never used on its own to classify a candidate , but if the host is clearly an elliptical galaxy , then this information may be used in conjunction with the candidate spectrum and the lightcurve to solidify the status of a candidate as a probable sn ia . we emphasize that the lightcurve alone is never used to classify an sn ia for a candidate to have a classification of sn ia or sn ia * it must have a spectrum matching with the expected spectral energy distribution of an sn ia at the phase determined from the lightcurve . one representative spectrum from each category is shown in figure [ specfig ] . all spectra discussed in this paper are available in the online version of this article . the snls officially started in june 2003 ( there was a presurvey ramp - up to full operations ) , and we began using gemini - n and s to observe sn candidates in august 2003 ( semester 2003b ) . we usually obtain the gemini data within a day or two of the observations and produce `` real - time '' reductions . after the full calibration data is released at the end of a run the data are rereduced . here we report on the final spectroscopic reductions through november 2004 . all of the spectra used in this study are available in the online version of the paper . table [ der - table ] lists properties derived from the observations of each candidate , such as type , redshift , and epoch relative to maximum light . table [ class - host - table ] shows the distribution of sn types with respect to host galaxy type . while it is not always possible to extract the host galaxy separately and examine its sed , galaxy lines are apparent because they are narrower than sn features . therefore we group galaxies into absorption - line galaxies and emission - line galaxies ( if a galaxy has any emission lines it is considered an emission - line galaxy ) . just as at low redshift , in our sample core - collapse sne are never seen in absorption - line ( early type ) galaxies . figure [ hist ] is a histogram of the number of candidates of each different type with redshift . for clarity we group index 0 and 1 sne together as `` not ia '' and index 4 and 5 sne together as `` ia . '' sne with weaker classifications ( index 3 ) are plotted separately as `` ia*. '' candidates that could not be identified ( index 2 ) , but whose redshifts could be determined , are also shown . it is apparent that we targeted the highest redshift observations at gemini the median redshift of sne ia / ia * is 0.81 . as the redshift increases the fraction of less secure identifications rises , because for faint sne it is hard to achieve a high signal - to - noise ratio in a reasonable integration time . furthermore , at the higher redshifts one must rely more on the rest - frame uv light to classify the sne , and template uv observations of all sn types are scarce . core collapse sne cluster at lower redshifts on the histogram because they are usually intrinsically fainter . for example , a candidate with a peak magnitude of @xmath31 could be an sn ia at @xmath32 or could be a type ii sn at a lower redshift . the technique of the sn photo - z is introduced in section [ photozintro ] and described in much greater detail in @xcite . here we report the result of its application on our sn ia confirmation rate . the snls implemented the sn photo - z and improved real - time photometry in march 2004 , although in two cases after this ( 04d1dr and 04d4 ft ) , candidates had to be observed spectroscopically before a photo - z could be obtained . when no photo - z information was available before selecting candidates for spectroscopy , 14/26 ( 54% ) of the candidates were confirmed as sne ia close to previosly published rates . however , when we did have an sn photo - z to guide the decision making , the ia confirmation rate jumped to 27/38 ( 71% ) . there will be always be some unidentified candidates for which it is difficult to estimate a type from spectroscopy these may still be sne ia , but are too buried in a host , or have a spectrum with too low s / n , to be identified . however , where the photo - z excels is in its power to reject candidates that are not sne ia . with no prior photo - z information , 7/26 ( 27% ) of candidates with gemini spectroscopy were found to be certainly not or probably not sne ia . after implementation of the technique , only 3/38 ( 8% ) of the observations were certainly not or probably not sne ia . another benefit of the sn photo - z is that it provides a prediction of the time of maximum light for an sn ia , allowing observations to be scheduled to within a few days of this date for maximum efficiency and minimal host galaxy contamination . figure [ hiepoch ] shows the distribution of the sne ia observed by gemini with respect to maximum light . ninety percent of sne ia were observed within 0.5 mag of maximum light , and over half of the sne ia were observed within 0.1 mag of maximum light . it is clear that the flexibility provided by queue observing plays a large role in optimizing the efficiency of the spectroscopic classification of targets in the snls . note that it is not always possible to schedule observations at maximum light because we require dark time to observe such faint targets , and gmos is not always on the telescope and available in queue mode . figure [ hiepoch ] also shows that the least solid classifications ( index 3 ) , labeled ia * , were usually observed after maximum light . sne before maximum were almost always classified with more certainty . this is partially because after maximum light , especially around + 7 to + 10 days after max , it is often difficult to distinguish between sne ia and sne ic . by this time certain distinguishing sn ia features , such as 4000 , may no longer be apparent , and the sn ia line velocities have decreased to the range more typical of those found in sne ic . the more uncertain classifications ( sne ia * ) often occur near maximum light , because the most difficult sne ( those at the highest redshift or with the most host galaxy contamination ) can only be observed near maximum light . before or after maximum light they are so faint that they would not be placed in the spectroscopic observing queue . an added benefit to obtaining early spectroscopy is that sne ia show the greatest diversity at early times @xcite , when the spectroscopy is probing the outer layers of supernova . figure [ hiepoch ] shows that one compromise would be to target observations at about one week before maximum light . at @xmath33d , a typical ia is only about 0.25 fainter than at peak , near enough to peak brightness to make the observations feasible . at the same time it would provide an opportunity to observe sne ia when they show the greatest diversity and also are the most distinct from sne ic . this window of opportunity is very narrow , however . at @xmath34d a typical sn ia is 0.75 mag fainter than at peak in the restframe b - band still too faint compared to its host galaxy , and by @xmath35d sn ia spectra have lost some of their diversity . figure [ imagpi ] shows the i magnitude at the time of spectroscopy versus the percentage increase in i brightness in a 6 pixel ( 1.12 ) diameter at the time of spectroscopy . the percentage increase is measured relative to the reference image , where there is no supernova light . the i magnitude and percentage increase were measured from cfht images , interpolated to the time of spectroscopy . typically candidates were sent to gemini only if they were in the magnitude range @xmath36 . ( there are some exceptions , especially in the d3 field , which can not be seen by the vlt . ) it is apparent from figure [ imagpi ] that when the sn signal is greater than that of the host galaxy , candidate identification is relatively easy when the percentage increase is greater than 100% the candidates were not identified only 7% ( 2/30 ) of the time . clearly , image quality ( iq ) plays a role in whether or not sne can be successfully identified in the presence of significant host galaxy light , as shown in figure [ iqpi ] . we plot the image quality at the time of spectroscopy ( determined from the gemini acquisition image ) against the percent increase ( determined from cfht images as described above ) . if the iq was better than 0.55@xmath37 , candidates were identified 88% of the time . in good seeing the sn light is more concentrated and can be extracted in a narrow aperture , even in the presence of host contamination . the @xmath2 matching of the sn spectrum against a host and sn spectral template library produces more identifications , more robust identifications , and greater coverage of parameter space than traditional methods ( unaided expert matching by eye ) . this method works optimally with spectra that are free from systematic deviations , and whose errors are well characterized . it works exceptionally well with gmos nod and shuffle data , where systematic effects associated with sky subtraction are almost completely removed . correcting the spectra for telluric features also helps , as does a carefully generated error spectrum . while both the spectra in this paper and in were classified by the same person ( dah ) , here the @xmath2 matching software is upgraded with more template supernova and galaxy spectra . were not able to identify any candidates with a percentage increase below 25% , but in this work we have successfully identified several candidates with percent increases between 15%25% . it is unclear if this is due to software improvements , observing conditions , differences in the spectrographs used , or a better selection of candidates . figure [ date ] also shows that the spectroscopic epoch determined by the fitting program matches well with the epoch determined from the lightcurve . unlike the spectral feature age technique @xcite , it is important to note that this program is not specifically tuned to determine the epoch of an sn ia , and it makes no assumptions that the input spectrum is a ia ( it is possible for it to determine the epoch of an sn ib / c for example , although this has not been extensively tested ) . we find the remarkable result that our spectroscopic fitting technique can determine the epoch to within 2.5 days , despite the low s / n and significant host contamination in this data set . before comparing these confirmation rates to previous work , a few caveats are necessary . first , not all spectroscopic follow - up programs have as their goal the confirmation of candidates as sne ia . one may wish to determine the types of all transients to have a better determination of sn rates . or in some cases the goal may be to spectroscopically identify sne ii , as they also have cosmological utility . while we have pursued these goals at other telescopes , the role of gemini has been largely to confirm sne ia . second , different programs may have different criteria for what is called an sn ia / ia*. for example , we have lightcurve infomation available at the time of classification , which was not always the case in previous studies . still , our criteria are as close as possible to the criteria used in other published works . we adopt similar criteria to that of @xcite , and we are using the the same human classifier ( dah ) and classification program ( albeit slightly upgraded ) as . despite these caveats , it is important to have some comparison to previous work . a comparison to is problematic because of the inhomogeneous nature of the sample and techniques presented there . they present results from six separate searches , some classical , and some rolling , some targeting @xmath38 , and some targeting @xmath39 . the most appropriate comparison to this work is to consider all searches in lidman et al . that targeted @xmath40 . in this case the sn ia / ia * confirmation rate was 62% ( median sn ia @xmath41 ) . a similar sn ia confirmation rate is found if the lidman et al . search 4 ( the rolling search ) is excluded , but sne followed by the scp at all telescopes for @xmath40 are included . since rolling searches find all sne above some magnitude cutoff , the percentage of sne ia in the sample is lower than in a classical search . in their first two years of operation , essence @xcite , had a 43% sn ia / ia ? confirmation rate with a median redshift of 0.43 [ though in the third year the sn ia / ia ? confirmation rate rose to @xmath1060% ( matheson , private communication ) ] . again , a direct comparison to the results presented here may not be appropriate if the two searches have different goals or different criteria for counting sne ia , but does make the point that if the goal is to maximize the yield of sne ia , a method like the sn photo - z is helpful . despite the difficulties in comparing this study to previous ones as noted above , we can draw some broad conclusions from the results . first , to make the most efficient use of 810 m time it is helpful to have as much information as possible before sending a target to the telescope . in a rolling search , more information is available at follow - up time than in a classical search , so new techniques are called for . here we have demonstrated that by fitting the available data before spectroscopy we can improve the sn ia / ia * confirmation rate over our rates when the technique was not applied . our sn ia / ia * confirmation rate ( 71% ) is also an increase over comparable previously published rates ( 43%-62% ) , despite being at a much higher median redshift ( z=0.81 vs. z=0.4 - 0.5 ) . by making it possible to separate likely sne ia from likely core collapse sne before a large amount of telescope time is invested in spectroscopy , the techniques demonstrated here improve the success rate of all follow - up programs , no matter their goal . over the five - year lifetime of the snls , the these techniques will result in @xmath3 more confirmed sne ia . we aim to spectroscopically observe @xmath42 sn candidates during the survey . in the absence of the photo - z this would result in @xmath10 540 sne ia , but with the photo - z we could achieve 710 sne ia . this is equivalent to adding another year to the survey . we have demonstrated several techniques for improving the yield of spectroscopically identified sne ia at high redshift ( @xmath43 ) . the sn photo - z is shown to effectively screen out non - ia candidates before they are observed spectroscopically . when no photometric redshifts were available , 35% of candidates turned out to be probably or certainly not sne ia after spectroscopy ( ci 0 or 1 ) . with photo - z information , the non - ia `` contamination '' rate dropped to 8% . using the photo - z we show that we can schedule sn observations to within a few tenths of a magnitude of maximum light , and that the optimal phase for sn ia identifcation and diversity is @xmath33 days . after spectroscopy , we show that by @xmath2 fitting of template sne we can effectively subtract host light and determine the type of an sn when the sn is only 15% as bright as the host in some cases . for targets where the sn is at least as bright as the underlying host , or when the image quality is exceptional ( better than 0.55 ) , the candidate is identified more than 90% of the time . using @xmath2 fitting we can also obtain an independent measurement of the spectroscopic epoch that agrees well with the phase determined from the lightcurve . these techniques have been developed using the first year s observations of the highest redshift candidates of the snls at gemini north and south . of the candidates observed at gemini , 41/64 are certain or probable sne ia . this is roughly one - third of the spectroscopic follow - up program of the snls . the techniques outlined here will add @xmath3 more confirmed sne ia over the five - year project to discover , confirm , and follow @xmath44 sne ia to measure the equation of state of dark energy . the snls collaboration gratefully acknowledges the assistance of pierre martin and the cfht queued service observations team . jean - charles cuillandre and kanoa withington were also indispensable in making possible real - time data reduction at cfht . we also thank gemini queue observers and support staff , especially inger jrgensen , kathy roth , percy gomez , and marcel bergmann , for both taking the data presented in this paper and making observations available quickly . canadian collaboration members acknowledge support from nserc and ciar ; french collaboration members from cnrs / in2p3 , cnrs / insu and cea ; portuguese collaboration members acknowledge support from fct - fundao para a cincia e tecnologia . snls relies on observations with megacam , a joint project of cfht and cea / dapnia , at the canada - france - hawaii telescope ( cfht ) which is operated by the national research council ( nrc ) of canada , the institut national des science de lunivers of the centre national de la recherche scientifique ( cnrs ) of france , and the university of hawaii . this work is based in part on data products produced at the canadian astronomy data centre as part of the canada - france - hawaii telescope legacy survey , a collaborative project of the national research council of canada and the french centre national de la recherche scientifique . this work is also based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the particle physics and astronomy research council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) and conicet ( argentina ) . this research used observations from gemini program numbers : gn-2004b - q-16 , gs-2004b - q-31 , gn-2004a - q-19 , gs-2004a - q-11 , gn-2003b - q-9 , and gs-2003b - q-8 . , a. l. , matheson , t. , filippenko , a. v. , leonard , d. c. , tonry , j. , riess , a. g. , challis , p. , clocchiatti , a. , garnavich , p. m. , hogan , c. j. , jha , s. , kirshner , r. p. , leibundgut , b. , phillips , m. m. , schmidt , b. p. , schommer , r. a. , smith , r. c. , soderberg , a. m. , spyromilio , j. , stubbs , c. , suntzeff , n. b. , & woudt , p. 2000 , , 544 , l111 , c. , howell , d. a. , folatelli , g. , garavini , g. , nobili , s. , aldering , g. , amanullah , r. , antilogus , p. , astier , p. , blanc , g. , burns , m. s. , conley , a. , deustua , s. e. , doi , m. , ellis , r. , fabbro , s. , fadeyev , v. , gibbons , r. , goldhaber , g. , goobar , a. , groom , d. e. , hook , i. , kashikawa , n. , kim , a. g. , knop , r. a. , lee , b. c. , mendez , j. , morokuma , t. , motohara , k. , nugent , p. e. , pain , r. , perlmutter , s. , prasad , v. , quimby , r. , raux , j. , regnault , n. , ruiz - lapuente , p. , sainton , g. , schaefer , b. e. , schahmaneche , k. , smith , e. , spadafora , a. l. , stanishev , v. , walton , n. a. , wang , l. , wood - vasey , w. m. , & yasuda ( the supernova cosmology project ) , n. 2005 , , 430 , 843 , t. , blondin , s. , foley , r. j. , chornock , r. , filippenko , a. v. , leibundgut , b. , smith , r. c. , sollerman , j. , spyromilio , j. , kirshner , r. p. , clocchiatti , a. , aguilera , c. , barris , b. , becker , a. c. , challis , p. , covarrubias , r. , garnavich , p. , hicken , m. , jha , s. , krisciunas , k. , li , w. , miceli , a. , miknaitis , g. , prieto , j. l. , rest , a. , riess , a. g. , salvo , m. e. , schmidt , b. p. , stubbs , c. w. , suntzeff , n. b. , & tonry , j. l. 2005 , , 129 , 2352 , s. , aldering , g. , goldhaber , g. , knop , r. a. , nugent , p. , castro , p. g. , deustua , s. , fabbro , s. , goobar , a. , groom , d. e. , hook , i. m. , kim , a. g. , kim , m. y. , lee , j. c. , nunes , n. j. , pain , r. , pennypacker , c. r. , quimby , r. , lidman , c. , ellis , r. s. , irwin , m. , mcmahon , r. g. , ruiz - lapuente , p. , walton , n. , schaefer , b. , boyle , b. j. , filippenko , a. v. , matheson , t. , fruchter , a. s. , panagia , n. , newberg , h. j. m. , couch , w. j. , & the supernova cosmology project . 1999 , , 517 , 565 , s. , pennypacker , c. r. , goldhaber , g. , goobar , a. , muller , r. a. , newberg , h. j. m. , desai , j. , kim , a. g. , kim , m. y. , small , i. a. , boyle , b. j. , crawford , c. s. , mcmahon , r. g. , bunclark , p. s. , carter , d. , irwin , m. j. , terlevich , r. j. , ellis , r. s. , glazebrook , k. , couch , w. j. , mould , j. r. , small , t. a. , & abraham , r. g. 1995 , , 440 , l41 , a. g. , filippenko , a. v. , challis , p. , clocchiatti , a. , diercks , a. , garnavich , p. m. , gilliland , r. l. , hogan , c. j. , jha , s. , kirshner , r. p. , leibundgut , b. , phillips , m. m. , reiss , d. , schmidt , b. p. , schommer , r. a. , smith , r. c. , spyromilio , j. , stubbs , c. , suntzeff , n. b. , & tonry , j. 1998 , , 116 , 1009 , a. g. , filippenko , a. v. , leonard , d. c. , schmidt , b. p. , suntzeff , n. , phillips , m. m. , schommer , r. , clocchiatti , a. , kirshner , r. p. , garnavich , p. , challis , p. , leibundgut , b. , spyromilio , j. , & smith , r. c. 1997 , , 114 , 722 , a. g. , strolger , l. , tonry , j. , tsvetanov , z. , casertano , s. , ferguson , h. c. , mobasher , b. , challis , p. , panagia , n. , filippenko , a. v. , li , w. , chornock , r. , kirshner , r. p. , leibundgut , b. , dickinson , m. , koekemoer , a. , grogin , n. a. , & giavalisco , m. 2004 , , 600 , l163 llllrclrrr 03d1as & 02:24:24.520 & -04:21:40.19 & 2003 - 09 - 27 & 6000 & n+s & 720 & 0.41 & 23.96 & 1475 + 03d1ax & 02:24:23.320 & -04:43:14.41 & 2003 - 09 - 29 & 2400 & c & 720 & 0.61 & 23.16 & 65 + 03d1bk & 02:26:27.410 & -04:32:11.99 & 2003 - 09 - 28 & 4800 & n+s & 720 & 0.47 & 23.95 & 83 + 03d1cj & 02:26:25.081 & -04:12:39.89 & 2003 - 10 - 26 & 5400 & n+s & 720 & 0.57 & 24.11 & 76 + 03d1 cm & 02:24:55.288 & -04:23:03.68 & 2003 - 10 - 27 & 5400 & n+s & 720 & 0.72 & 23.77 & 970 + 03d1co & 02:26:16.238 & -04:56:05.76 & 2003 - 11 - 01 & 7200 & n+s & 720 & 0.84 & 23.64 & 188 + 03d1ew & 02:24:14.088 & -04:39:56.98 & 2003 - 12 - 21 & 7200 & n+s & 720 & 0.71 & 23.76 & 1661 + 03d1fp & 02:26:03.073 & -04:08:02.02 & 2003 - 12 - 26 & 7200 & n+s & 720 & 0.55 & 23.86 & 30 + 03d1fq & 02:26:55.683 & -04:18:08.10 & 2003 - 12 - 24 & 5400 & n+s & 720 & 0.48 & 23.59 & 41 + 03d4cj & 22:16:06.660 & -17:42:16.72 & 2003 - 08 - 26 & 2700 & c & 680 & 0.83 & 21.85 & 10000 + 03d4ck & 22:15:08.910 & -17:56:02.17 & 2003 - 08 - 27 & 2400 & c & 680 & 0.46 & 22.78 & 1905 + 03d4cn & 22:16:34.600 & -17:16:13.55 & 2003 - 08 - 27 & 4800 & c & 720 & 0.46 & 23.81 & 49 + 03d4cy & 22:13:40.460 & -17:40:53.90 & 2003 - 09 - 26 & 5400 & n+s & 720 & 0.59 & 24.19 & 149 + 03d4cz & 22:16:41.870 & -17:55:34.54 & 2003 - 09 - 27 & 3600 & c & 720 & 0.41 & 24.41 & 15 + 03d4fd & 22:16:14.471 & -17:23:44.37 & 2003 - 10 - 24 & 3600 & n+s & 720 & 0.67 & 23.61 & 321 + 03d4fe & 22:16:08.844 & -17:55:19.21 & 2003 - 10 - 24 & 3600 & n+s & 720 & 0.48 & 23.58 & 66 + 03d4gl & 22:14:44.177 & -17:31:44.47 & 2003 - 10 - 29 & 3600 & n+s & 720 & 0.63 & 23.43 & 121 + 04d1de & 02:26:35.925 & -04:25:21.65 & 2004 - 08 - 17 & 7200 & n+s & 720 & 0.51 & 23.63 & 56 + 04d1dr & 02:27:23.905 & -04:51:27.43 & 2004 - 08 - 14 & 5400 & n+s & 720 & 0.57 & 24.28 & 72 + 04d1hd & 02:26:08.850 & -04:06:35.22 & 2004 - 09 - 13 & 2400 & c & 680 & 0.85 & 22.16 & 1254 + 04d1ho & 02:24:44.856 & -04:39:15.55 & 2004 - 09 - 16 & 3600 & c & 720 & 0.59 & 23.13 & 24 + 04d1hy & 02:24:08.678 & -04:49:52.22 & 2004 - 09 - 11 & 5400 & n+s & 720 & 0.73 & 23.53 & 996 + 04d1jf & 02:25:18.914 & -04:49:09.05 & 2004 - 10 - 13 & 2400 & c & 680 & 0.92 & 22.94 & 19 + 04d1ln & 02:25:53.482 & -04:27:03.75 & 2004 - 10 - 17 & 2400 & c & 680 & 0.62 & 22.80 & 29 + 04d1ow & 02:26:42.708 & -04:18:22.55 & 2004 - 11 - 08 & 5400 & n+s & 720 & 0.69 & 23.87 & 4201 + 04d2aa & 10:02:02.100 & + 02:40:51.76 & 2004 - 01 - 23 & 7320 & n+s & 720 & 0.58 & 24.19 & + 04d2ad & 10:00:08.093 & + 02:39:01.40 & 2004 - 01 - 22 & 7320 & n+s & 720 & 0.54 & 24.17 & + 04d2ae & 10:01:52.414 & + 02:13:21.11 & 2004 - 01 - 21 & 5490 & n+s & 720 & 0.75 & 23.64 & + 04d3aa & 14:16:49.935 & + 52:45:31.12 & 2004 - 01 - 30 & 5400 & n+s & 720 & 1.02 & 24.05 & 12 + 04d3ae & 14:22:21.569 & + 52:21:39.21 & 2004 - 01 - 25 & 2400 & c & 680 & 0.94 & 23.10 & 196 + 04d3ax & 14:22:39.072 & + 52:51:52.57 & 2004 - 01 - 28 & 5400 & n+s & 720 & 1.30 & 24.95 & 246 + 04d3bf & 14:17:45.096 & + 52:28:04.31 & 2004 - 02 - 17 & 2700 & c & 680 & & 23.16 & + 04d3dd & 14:17:48.431 & + 52:28:14.72 & 2004 - 04 - 25 & 5400 & n+s & 720 & 0.59 & 24.06 & 375 + 04d3de & 14:22:13.503 & + 52:17:09.71 & 2004 - 04 - 27 & 7200 & n+s & 720 & 0.57 & 24.01 & 651 + 04d3fj & 14:19:50.703 & + 52:41:31.84 & 2004 - 04 - 28 & 7200 & n+s & 720 & 0.72 & 24.19 & 108 + 04d3fq & 14:16:57.906 & + 52:22:46.53 & 2004 - 04 - 26 & 5400 & n+s & 720 & 0.97 & 23.04 & 412 + 04d3gu & 14:22:07.359 & + 52:38:54.60 & 2004 - 05 - 22 & 4800 & c & 720 & 1.03 & 22.60 & 6 + 04d3gx & 14:20:13.678 & + 52:16:58.60 & 2004 - 05 - 21 & 7200 & n+s & 720 & 0.40 & 24.38 & 2439 + 04d3hn & 14:22:06.878 & + 52:13:43.46 & 2004 - 05 - 22 & 4800 & c & 720 & 0.47 & 22.94 & 14 + 04d3kr & 14:16:35.937 & + 52:28:44.20 & 2004 - 06 - 16 & 2400 & c & 680 & 0.82 & 21.60 & 156 + 04d3lp & 14:19:50.927 & + 52:30:11.85 & 2004 - 05 - 27 & 5400 & n+s & 720 & 0.52 & 24.45 & 78 + 04d3lu & 14:21:08.009 & + 52:58:29.74 & 2004 - 06 - 23 & 3600 & n+s & 720 & 0.84 & 23.49 & 17 + 04d3mk & 14:19:25.830 & + 53:09:49.56 & 2004 - 06 - 19 & 4320 & n+s & 720 & 1.25 & 23.27 & 86 + 04d3ml & 14:16:39.107 & + 53:05:35.66 & 2004 - 06 - 20 & 3600 & n+s & 720 & 0.41 & 23.90 & 2252 + 04d3nc & 14:16:18.224 & + 52:16:26.09 & 2004 - 07 - 13 & 2400 & c & 720 & 0.66 & 23.39 & 149 + 04d3nh & 14:22:26.729 & + 52:20:00.92 & 2004 - 06 - 23 & 1800 & c & 680 & 0.80 & 21.66 & 199 + 04d3nq & 14:20:19.193 & + 53:09:15.90 & 2004 - 07 - 14 & 1500 & c & 680 & 0.61 & 21.03 & 10000 + 04d3nr & 14:22:38.526 & + 52:38:55.89 & 2004 - 07 - 15 & 7200 & n+s & 720 & 0.69 & 24.40 & 2300 + 04d3ny & 14:18:56.332 & + 52:11:15.06 & 2004 - 07 - 10 & 5400 & n+s & 720 & 0.79 & 23.23 & 203 + 04d3oe & 14:19:39.381 & + 52:33:14.21 & 2004 - 07 - 11 & 3600 & n+s & 720 & 0.62 & 23.21 & 28 + 04d3og & 14:20:39.748 & + 53:01:15.02 & 2004 - 07 - 19 & 2700 & c & 720 & 1.10 & 21.81 & 151 + 04d3pd & 14:22:33.506 & + 52:13:47.77 & 2004 - 07 - 18 & 3600 & n+s & 720 & 0.83 & 23.34 & 72 + 04d4 dm & 22:15:25.470 & -17:14:42.71 & 2004 - 07 - 18 & 3600 & n+s & 720 & 0.96 & 23.55 & 123 + 04d4ec & 22:16:29.286 & -18:11:04.13 & 2004 - 07 - 19 & 3600 & n+s & 720 & 0.70 & 23.08 & 37 + 04d4 ft & 22:14:31.097 & -17:40:19.74 & 2004 - 08 - 12 & 3600 & c & 720 & 0.67 & 23.21 & 78 + 04d4gg & 22:16:09.268 & -17:17:39.98 & 2004 - 08 - 16 & 3600 & c & 720 & 0.81 & 23.03 & 30 + 04d4hu & 22:15:36.193 & -17:50:19.81 & 2004 - 09 - 18 & 5400 & n+s & 720 & 0.52 & 23.45 & 84 + 04d4hx & 22:13:40.587 & -17:23:03.35 & 2004 - 09 - 16 & 5400 & n+s & 720 & 0.73 & 24.15 & 25 + 04d4ic & 22:14:21.841 & -17:56:36.43 & 2004 - 09 - 12 & 5160 & n+s & 720 & 0.69 & 23.34 & 231 + 04d4ih & 22:17:17.041 & -17:40:38.74 & 2004 - 10 - 07 & 5400 & n+s & 720 & 0.52 & 23.95 & 41 + 04d4ii & 22:15:55.645 & -17:39:27.09 & 2004 - 09 - 15 & 7200 & n+s & 720 & 0.45 & 23.86 & 57 + 04d4im & 22:15:00.885 & -17:23:45.84 & 2004 - 10 - 10 & 7200 & n+s & 720 & 0.44 & 23.23 & 18 + 04d4jy & 22:13:51.605 & -17:24:18.13 & 2004 - 10 - 14 & 8881 & n+s & 720 & 0.77 & 24.08 & 1779 + 04d4kn & 22:15:04.324 & -17:19:45.05 & 2004 - 10 - 19 & 5400 & n+s & 720 & 0.60 & 23.74 & 97 + llllcrrrll 03d1as & 0.872 & 0.001 & sn : & 2 & & & & : + 03d1ax & 0.496 & 0.001 & sn ia & 5 & -3.0 & 0.1 & -3.3 & h&k + 03d1bk & 0.8650 & 0.0005 & sn ia & 5 & -6.3 & 0.1 & -2.1 & h&k , h@xmath45 , h@xmath46 + 03d1cj & 0.364 & 0.001 & sn ib / c : & 1 & & & & h@xmath47 , + 03d1 cm & 0.87 & 0.02 & sn ia & 4 & -5.0 & 0.3 & 2.4 & sn + 03d1co & 0.68 & 0.01 & sn ia & 5 & -4.8 & 0.3 & -1.2 & sn + 03d1ew & 0.868 & 0.001 & sn ia & 5 & 0.9 & 0.4 & -0.2 & + 03d1fp & 0.270 & 0.001 & sn iib & 0 & & & & h@xmath47 , h@xmath45 , , , + 03d1fq & 0.80 & 0.02 & sn ia & 4 & -1.6 & 0.3 & 2.2 & sn + 03d4cj & 0.27 & 0.01 & sn ia & 5 & -9.7 & 0.1 & -6.9 & sn + 03d4ck & 0.189 & 0.001 & sn iin & 0 & & & & sn + 03d4cn & 0.818 & 0.001 & sn ia & 4 & -0.5 & 0.6 & 2.8 & , , h@xmath45 + 03d4cy & 0.9271 & 0.0005 & sn ia & 4 & 5.3 & 0.3 & 7.7 & + 03d4cz & 0.695 & 0.001 & sn ia & 4 & 9.4 & 0.1 & 8.7 & h&k , g - band + 03d4fd & 0.791 & 0.003 & sn ia & 5 & -1.6 & 0.3 & -1.5 & + 03d4fe & ? & & sn : & 1 & & & & + 03d4gl & 0.56 & 0.01 & sn ia & 4 & -9.0 & 0.2 & -6.0 & sn + 04d1de & 0.7677 & 0.0002 & sn ia * & 3 & -6.8 & 0.1 & -4.5 & , , h@xmath45 , poss h&k + 04d1dr & 0.6414 & 0.0003 & sn : & 2 & & & & , , h@xmath45 , h@xmath46 + 04d1hd & 0.3685 & 0.0005 & sn ia & 5 & -4.8 & 0.1 & -4.0 & , weak + 04d1ho & 0.7012 & 0.0004 & sn & 2 & & & & , , h&k , h@xmath45 , h@xmath46 + 04d1hy & 0.85 & 0.02 & sn ia & 5 & -3.5 & 0.3 & -3.2 & sn + 04d1jf & 0.3800 & 0.0002 & sn : & 2 & & & & , , h@xmath45 + 04d1ln & 0.2072 & 0.0002 & sn ii - p & 0 & & & & h@xmath47 , , + 04d1ow & 0.93 & 0.02 & sn ia & 4 & 2.6 & 0.2 & -0.8 & sn + 04d2aa & ? & & sn : & 2 & & & & + 04d2ad & 0.6802 & 0.0002 & sn : & 2 & & & & , , h@xmath45 + 04d2ae & 0.843 & 0.001 & sn ia & 4 & 0.0 & 0.1 & -1.5 & h&k + 04d3aa & 0.2045 & 0.0002 & sn ii & 0 & & & & h@xmath47 , h@xmath45 , , + 04d3ae & 0.217 & 0.001 & sn ii & 0 & & & & h@xmath47 , + 04d3ax & 0.3558 & 0.0002 & sn ii : & 1 & & & & h@xmath47 , h@xmath45 , + 04d3bf & 0.1560 & 0.0005 & sn ia & 5 & & & 15.1 & , , , h@xmath47 , h@xmath45 + 04d3dd & 1.01 & 0.02 & sn ia & 4 & 2.9 & 0.3 & -1.4 & sn + 04d3de & ? & & sn ii - p & 0 & & & & + 04d3fj & ? & & sn : & 1 & & & & + 04d3fq & 0.73 & 0.01 & sn ia & 4 & 0.8 & 0.3 & 3.6 & sn + 04d3gu & 0.748 & 0.001 & sn : & 2 & & & & h&k , balmer , weak + 04d3gx & 0.91 & 0.02 & sn ia * & 3 & 10.8 & 0.2 & 8.3 & sn + 04d3hn & 0.5516 & 0.0003 & sn ia & 5 & 7.2 & 0.1 & 4.9 & h&k , balmer + 04d3kr & 0.3373 & 0.0002 & sn ia & 5 & 4.4 & 0.1 & 1.7 & , , h@xmath45 + 04d3lp & 0.983 & 0.001 & sn ia * & 3 & 1.0 & 0.2 & -1.2 & + 04d3lu & 0.8218 & 0.0002 & sn ia & 4 & 5.5 & 0.1 & 5.6 & h&k + 04d3mk & 0.813 & 0.001 & sn ia & 5 & -2.4 & 0.1 & 1.2 & , h&k + 04d3ml & 0.95 & 0.02 & sn ia & 4 & -1.8 & 0.3 & 1.1 & sn + 04d3nc & 0.817 & 0.001 & sn ia * & 3 & 7.2 & 0.2 & & poss + 04d3nh & 0.3402 & 0.0002 & sn ia & 5 & 3.5 & 0.1 & 3.7 & h@xmath47 , h@xmath45 , , poss h&k + 04d3nq & 0.22 & 0.01 & sn ia & 5 & 8.1 & 0.1 & 8.8 & sn + 04d3nr & 0.96 & 0.02 & sn ia * & 3 & 8.1 & 0.3 & 3.9 & sn + 04d3ny & 0.81 & 0.02 & sn ia & 5 & 1.6 & 0.2 & 2.5 & sn + 04d3oe & 0.756 & 0.001 & sn ia & 4 & 1.4 & 0.1 & 1.8 & h&k + 04d3og & 0.352 & 0.001 & sn & 2 & & & & h@xmath47 , h@xmath45 , , , + 04d3pd & 0.760 & 0.001 & sn : & 2 & & & & h@xmath45 , , + 04d4 dm & 0.811 & 0.001 & sn ia & 4 & 2.7 & 0.2 & -0.7 & , + 04d4ec & 0.593 & 0.001 & sn : & 2 & & & & , , h@xmath45 , h@xmath46 + 04d4 ft & 0.2666 & 0.0002 & sn & 2 & & & & , h@xmath47 , h@xmath45 + 04d4gg & 0.4238 & 0.0004 & sn ia & 5 & -9.7 & 0.1 & -10.0 & , , h@xmath47 , h@xmath45 , h@xmath46 + 04d4hu & 0.7027 & 0.0003 & sn ia & 5 & 5.2 & 0.5 & 5.0 & , , h@xmath45 + 04d4hx & 0.545 & 0.005 & sn : & 2 & & & & 4000 break , poss h&k + 04d4ic & 0.68 & 0.02 & sn ia & 4 & 2.5 & 0.3 & 4.8 & sn + 04d4ih & 0.934 & 0.001 & sn ia * & 3 & 8.7 & 0.2 & 5.6 & , h&k , some balmer + 04d4ii & 0.866 & 0.001 & sn ia & 5 & -4.6 & 0.2 & -3.5 & + 04d4im & 0.7510 & 0.0005 & sn ia & 4 & 0.7 & 0.2 & 0.2 & h&k , h@xmath48 + 04d4jy & 0.93 & 0.02 & sn ia * & 3 & 2.0 & 0.4 & 0.0 & sn + 04d4kn & 0.9095 & 0.0005 & sn : & 2 & & & & , h&k , balmer +

over the past two decades , helioseismology has revolutionized our understanding of the interior structure and dynamics of the sun . asteroseismology will soon place this knowledge into a broader context by providing structural data for hundreds of sun - like stars . solar - like oscillations have already been detected from the ground in several stars , and nasa s kepler mission is poised to unleash a flood of stellar pulsation data . deriving reliable asteroseismic information from these observations demands a significant improvement in our analysis methods . in this paper we report the initial results of our efforts to develop an objective stellar model - fitting pipeline for asteroseismic data . the cornerstone of our automated approach is an optimization method using a parallel genetic algorithm . we describe the details of the pipeline and we present the initial application to sun - as - a - star data , yielding an optimal model that accurately reproduces the known solar properties . # 1]*#1 ] *

0903.0616

most of what we can learn about stars comes from observations of their outermost surface layers . we are left to infer the properties of the interior based on our best current understanding of the constitutive physics . the exception to this general rule arises from observations of pulsating stars , where seismic waves probe deep through the interior and bring information to the surface in the form of light and radial velocity variations . the most dramatic example is the sun , where such observations have led to the identification of millions of unique pulsation modes , each sampling the solar interior in a slightly different and complementary way . the radial profile of the sound speed inferred from inversion of these data led to such precise constraints on the standard solar model that , before the recent controversy over heavy element abundances @xcite , the observations and theory agreed to better than a few parts per thousand over 90% of the solar radius @xcite . although such detailed inversions are not currently possible for other stars , pulsation data do allow us to determine the global properties and to probe the gross internal composition and structure , providing valuable independent tests of stellar evolution theory . the reason for this qualitative difference is simple : if we could move the sun to the distance of the nearest star , many of the pulsation modes would no longer be detectable . without spatial resolution , only those modes with the lowest spherical degree ( @xmath0 ) would produce significant variations in the total integrated light or the spectral line profiles . these are also the modes that probe deepest into the stellar interior , collectively sampling the physical conditions from the core to the photosphere . the solar oscillations are excited by the acoustic noise generated by the near - surface convection , with nearly sonic speed . thus we expect to find similar oscillations in all stars with vigorous outer convection zones . a first unambiguous detection of solar - like oscillations in another star was in fact made by @xcite . recent improvements in our ability to make high - precision radial velocity measurements from the ground have been driven largely by efforts to detect extra - solar planets . this has led to the detection of solar - like oscillations in more than a dozen main - sequence and subgiant stars ; in addition , large - scale ground - based photometric campaigns have shown evidence for solar - like oscillations in numerous giants ( for a recent review see * ? ? ? * ) . the oscillation amplitudes and the frequency of maximum power in these stars agree reasonably well with our theoretical expectations but in some cases the mode lifetimes appear to be significantly shorter than expected @xcite , suggesting that our knowledge of the convective driving and damping physics is incomplete . scintillation in the earth s atmosphere severely limits our ability to detect the parts - per - million light variations due to these pulsations in main - sequence stars . however , space - based photometric programs designed to detect extra - solar planet transits also have the sensitivity to document the stellar pulsation signals . thus , the french - led corot mission @xcite is providing extensive data of very high quality on oscillations in a broad range of stars . the field is progressing very rapidly and many new observations will become available in the next few years , particularly after the launch of nasa s kepler mission in 2009 . with this in mind , it will be a distinct advantage to have in place the computational methods that will allow us to maximize the science return of these data . this will lead us quickly to a deeper understanding of the solar oscillations in the context of similar pulsations in other stars , and will yield unprecedented constraints on the formation and evolution of stellar and planetary systems . the likely excitation mechanism by turbulent convection near the surface creates a broad envelope of power with a peak that scales with the acoustic cutoff frequency ( see * ? ? ? * ) . within this envelope a large fraction of the predicted low - degree pulsation modes are excited to detectable amplitudes , leading to readily identifiable patterns of peaks in the power spectrum characterized by the _ large separation _ ( @xmath1 ) , and the _ small separation _ ( @xmath2 ) , where @xmath3 is the radial order and @xmath4 is the spherical degree of the modes with oscillation frequency @xmath5 . without any detailed modeling , these overall patterns immediately lead to an estimate of the mean density of the star and can indicate the presence of interior chemical gradients that are a good proxy for the stellar age . but a full analysis must include a detailed comparison of the individual frequencies with theoretical models . one complication with such a comparison is the existence of so - called _ surface effects _ , which appear as systematic differences between the observed and calculated oscillation frequencies that grow larger towards the acoustic cutoff frequency . surface effects arise primarily due to incomplete modeling of the near - surface layers of the star where convection plays a major role @xcite . addressing this inherent deficiency in our 1d models would require ( among other things ) that we substitute the results of extensive 3d calculations for the parameterized mixing - length treatment of convection that is currently used in nearly all stellar evolution codes . alternately , we can make an empirical correction to the calculated frequencies following @xcite , who recently devised a method for calibrating the functional form of surface effects using solar data , and then obtaining corrections for other stars based on the frequencies of a reference model , scaled to correct for the different mean density ( see appendix [ appa ] ) . we have developed an objective and automated method of fitting stellar models to the asteroseismic data soon expected to emerge from nasa s kepler mission . this will lead to reliable determinations of stellar radii to help characterize the extra - solar planetary systems discovered by the mission , and stellar ages to reveal how such systems evolve over time . for the asteroseismic targets that do not contain planetary companions it will allow a uniform determination of fundamental physical properties for hundreds of solar - type stars , providing a new window on stellar structure and evolution . in [ sec2 ] we provide an overview of the stellar models as well as the optimization and analysis techniques that establish the foundation of our fitting method . we describe our initial numerical experiments to calibrate the method with sun - as - a - star data in [ sec3 ] , and we outline our plans for further validation in [ sec4 ] . the kepler mission will soon yield precise high - cadence time - series photometry of hundreds of pulsating stars every few months for at least 3.5 years @xcite . we will then face the challenge of determining the fundamental properties of these stars from the data , by attempting to match them with the output of computer models . the traditional approach to this task is to make informed guesses for each of the model parameters , and then to adjust them iteratively until an adequate match is found . the volume of asteroseismic data that will emerge from the kepler mission calls for a more automated approach to modeling that initially explores a broad range of model parameters in an objective manner . the cornerstone of our model - fitting approach is a global optimization method using a parallel genetic algorithm . the result of the global search provides the starting point for a local analysis using a levenberg - marquardt algorithm with singular value decomposition , which also allows us to explore the information content of the observables and the impact of including additional observational constraints @xcite . we have recently adapted the aarhus stellar evolution code ( astec ; * ? ? ? * ) and adiabatic pulsation code ( adipls ; * ? ? ? * ) to interface with the parallel genetic algorithm . these are essentially the same models that were developed for the analysis of helioseismic data , and are the source of model s of @xcite , which has been used extensively as a reference model for solar inversions . using these models for the analysis of pulsations in solar - type stars will provide some internal consistency in our understanding of solar - like oscillations . briefly , these stellar models use the opal 2005 equation of state ( see * ? ? ? * ) and the most recent opal opacities ( see * ? ? ? * ) , supplemented by kurucz opacities at low temperatures . the nuclear reaction rates come from @xcite , convection is described by the mixing - length theory of @xcite , and we included the effects of helium settling as described by @xcite . each model evaluation involves the computation of a stellar evolution track from the zero - age main sequence through a mass - dependent number of internal time steps , terminating prior to the beginning of the red - giant stage . rather than calculate the pulsation frequencies for each of the 200300 models along the track , we exploit the fact that the average frequency spacing of consecutive radial overtones @xmath6 in most cases is a monotonically decreasing function of age @xcite . once the evolution track is complete , we start with a pulsation analysis of the model at the middle time step and then use a binary decision tree comparing the observed and calculated values of @xmath6to select older or younger models along the track . in practice , this recipe allows us to interpolate the age between the two nearest time steps by running the pulsation code on just 8 models from each stellar evolution track . since we are interested in developing a general - purpose modeling tool for asteroseismic data from the kepler mission , we need to select a global method for optimizing the match between our model output and the available observations of any given star . using only observations and the constitutive physics of the model to restrict the range of possible values for each parameter , a genetic algorithm ( ga ; * ; * ? ? ? * ) can provide a relatively efficient means of searching globally for the optimal model . although it is more difficult for a ga to find _ precise _ values for the optimal set of parameters efficiently , it is well suited to search for the _ region _ of parameter space that contains the global minimum . in this sense , the ga is an objective means of obtaining a good first guess for a more traditional local analysis method , which can narrow in on the precise values and uncertainties of the optimal model parameters . @xcite developed a fully parallel and distributed implementation of the pikaia genetic algorithm that was originally written by @xcite . @xcite used this modeling tool in the context of white dwarf asteroseismology , which ultimately led to a number of interesting physical results , including : ( 1 ) a precise estimate of the astrophysically important ( @xmath7c + @xmath8he @xmath9 @xmath10o ) nuclear reaction rate @xcite , ( 2 ) the first unambiguous detection of a crystallized core in a massive pulsating white dwarf @xcite , and ( 3 ) asteroseismic confirmation of a key prediction of diffusion theory in white dwarf envelopes @xcite . the impact of this method on the analysis of pulsating white dwarfs suggests that seismological modeling of other types of stars could also benefit from this approach . our implementation of the ga optimizes four adjustable model parameters ; these are the stellar mass ( @xmath11 ) from 0.75 to 1.75 @xmath12 , the metallicity ( @xmath13 ) from 0.002 to 0.05 ( equally spaced in @xmath14 ) , the initial helium mass fraction ( @xmath15 ) from 0.22 to 0.32 , and the mixing - length parameter ( @xmath16 ) from 1 to 3 . the stellar age ( @xmath17 ) is optimized internally during each model evaluation by matching the observed value of @xmath6 ( see [ sec2.1 ] ) . the ga uses two - digit decimal encoding , so there are 100 possible values for each parameter within the ranges specified above . each run of the ga evolves a population of 128 models through 200 generations to find the optimal set of parameters , and we execute 4 independent runs with different random initialization to ensure that the best model identified is truly the global solution . this method requires about @xmath18 model evaluations , compared to @xmath19 models for a complete grid at the same sampling density , making the ga nearly 1000 times more efficient than a complete grid ( currently 1 week of computing time , compared to many years for a grid ) . of course , a grid could in principle be applied to hundreds of observational data sets without calculating additional models but the ga approach also gives us the flexibility to improve the physical ingredients in the future , while the physics of a grid would be fixed . once the ga brings us close enough to the global solution , we can switch to a local optimization method . we implement a modified levenberg - marquardt ( lm ) algorithm that uses singular value decomposition ( svd ) on the calculated design matrices to filter the least important information from the observables ( some of which may be dominated by noise ) . this then provides an effective local inversion technique . lm is relatively fast and stable , and convergence typically occurs within 34 iterations . we treat the local analysis as a @xmath20-minimization problem , where one seeks to find the set of parameters * p * that minimizes @xmath21 here @xmath22 and @xmath23 are the @xmath24 measurements and errors , while the @xmath25 are the calculated model observables resulting from * p*. the lm+svd method requires an initial guess of * p * , and these are taken to be the results from the global search by the ga . the lm+svd analysis subsequently uses derivative information from the model at the current parameter values to calculate suggested parameter changes @xmath26 that will bring the model observables closer to the observations . we have three main motivations for implementing a local optimization method at the end of the global search . first , the ga has a limited resolution for each parameter , and the values that match the observations best are most likely between the fixed sample points . for example , the parameter @xmath13 near solar values is tuned by the ga along sample points spaced 0.0006 apart , corresponding to a precision of roughly 3% . considering that the search resolution is limited for _ all _ of the model parameters , this significantly limits the precision of the global optimization . the resolution of the local analysis is limited only by the precision of the stellar evolution and pulsation codes , so we use it to adjust the models below the resolution of the ga search . our second motivation for the local analysis is to quantify the final parameter uncertainties and correlations , and to probe the information content of the observables . we do this by calculating the derivatives of the model observables with respect to each of the fitted parameters , and then dividing each vector by the corresponding measurement error . this matrix can be referred to simply as the design matrix * d*. we subsequently calculate the singular value decomposition of * d * for some of its very useful properties ( see appendix [ appb ] ) . the final motivation for implementing a local analysis is to explore the effects of using different physical descriptions of the stellar interior @xcite . when the changes to the underlying physics are relatively subtle , we can assume that the global search by the ga using one set of assumptions will also provide a good starting point for a local analysis under slightly perturbed conditions . by applying the techniques explained in @xcite , such an analysis could demonstrate that the observational data contain enough information to distinguish clearly between different choices for the equation of state , for example . ultimately , this technique could reveal discrepancies in the observables that indicate which physical description is most suitable for the star under investigation . we apply the local analysis to each of the sets of optimal parameters identified by the four independent runs of the ga . the local analysis involves the gradual refinement of the optimal parameters through the iterative application of several steps : 1 . scan the @xmath13 parameter within @xmath270.002 of its original value , and perform a new minimization for each starting value of @xmath13 . we retain the set of parameters * p@xmath28 * that result in the lowest value of the reduced @xmath20 . 2 . starting with * p@xmath28 * , scan the @xmath15 parameter within @xmath270.01 of its initial value , and perform a new minimization for each starting value of @xmath15 . again we retain the * p@xmath28 * that yields the lowest reduced @xmath20 . 3 . rescan the @xmath13 parameter beginning with * p@xmath28*. we adopt as the optimal model the set of parameters that yield the lowest reduced @xmath20 from this final iteration . the reason for scanning @xmath15 and @xmath13 becomes clear when we begin to understand the intrinsic parameter correlations , which are enhanced by the limited set of observations ( see [ sec3.2 ] ) . in most cases the best model found by the global search leads to the final best model after the local optimization . however , sometimes a model that appears to be marginally worse at the end of the global search is improved more substantially during the local analysis . we take our final solution to be the best match from the four independent analyses , and for clarity we report the results of the corresponding global search even when it is not the best of the four models identified by the ga ( cf . table [ tab1 ] ) . the final uncertainties in the parameter values ( @xmath29 ) and in the model observables ( @xmath30 ) are calculated using svd ( see appendix [ appb ] ) . the overall goal of our model - fitting pipeline is to take a range of oscillation frequencies and other constraints as input , to identify and refine the model that best matches these observations , and to produce the optimal values of several parameters and other characteristics of the model as output . to ensure that our pipeline yields reliable results , we must begin by applying it to data where the basic stellar properties are already known . the most obvious choice is the sun , where high - quality sun - as - a - star observations are available from multiple experiments . if we feed the pipeline a stellar - like set of solar data , we can judge the experiment a success if the pipeline returns the known solar properties within acceptable tolerances . to reach this goal , we must first optimize the efficiency of the search method by passing synthetic data through the model - fitting procedure ( [ sec3.1 ] ) , calibrate the differences between our models and real observations due to near - surface effects and ensure that the resulting empirical correction does not introduce any large systematic errors in the final model parameters ( [ sec3.2 ] ) , and finally quantify any differences in the derived model parameters due to the source and error properties of the solar data ( [ sec3.3 ] ) . the results of these experiments are listed in table [ tab1 ] , and described in the following subsections . we will eventually want to validate this model - fitting pipeline using other stars that differ from the sun , where the physical properties are known with lower precision . however , because of the computation - intensive nature of our method ( which currently requires about 1 week on 512 processors ) , we will consider here only the initial validation using solar data . although genetic algorithms are often more efficient than other comparably global optimization methods , they are still quite demanding computationally . fortunately , the procedure is inherently parallelizable ; we need to calculate many models , and each one of them is independent of the others . so the number of available processors determines the number of models that can be calculated in parallel . also , there is very little communication overhead ; parameter values are sent to each processor , and they return either a list of observables or just a goodness - of - fit measure if the predictions have already been compared to the observations . the parallel version of the pikaia genetic algorithm is perfectly general , and did not require any structural modifications to interface with our stellar evolution and pulsation codes . the efficiency of genetic - algorithm - based optimization can be defined as the number of model evaluations required to yield the global solution , relative to the number of models that would be required for a complete grid at the same sampling density . in practice , a ga is usually hundreds or even thousands of times more efficient than a complete grid , and its performance is fairly insensitive to the few internal parameters that control its operation . we initially set these internal parameters ( e.g. population size , run length , crossover and mutation rates ) based on our experience with white dwarf models , but we also ran synthetic data through the optimization procedure in a series of `` hare & hound '' ( h&h ) exercises to ensure that the input parameters were recovered faithfully . the basic procedure for an h&h exercise is for one team member to calculate the theoretical oscillation frequencies and other observables for a specific set of model parameters ( @xmath29 ) . a subset of the predictions , typical of whatever is available from actual observations , is then given to another team member who does not know the source parameters . the data are passed through the complete optimization method in an attempt to recover the source parameters without any additional information . the results of such exercises are used to quantify the success rate of the optimization method ( the fraction of independent runs that lead to the known source parameters ) , and to improve its efficiency ( minimize the number of model evaluations ) if possible . for our first h&h exercise ( with source parameters listed in the row labeled `` h&h1 '' in table [ tab1 ] ) , we assume that typical asteroseismic data from the kepler mission will include twelve frequencies for each of the radial ( @xmath31 ) , dipole ( @xmath32 ) , and quadrupole ( @xmath33 ) modes , with consecutive radial orders in the range @xmath34 is necessary for the operation of our pipeline . information about the value of @xmath3 is only provided for completeness . ] . thus , we allowed the ga to fit a total of 36 oscillation frequencies . we assigned statistical uncertainties to each frequency by scaling up the errors on the corresponding modes in bison data @xcite by a factor of 10 , which is roughly what we expect from kepler data ( @xmath35hz ) . we complemented this synthetic asteroseismic information with artificial data on the effective temperature and luminosity , with errors comparable to what is expected for stars in the kepler input catalog ( * ? ? ? * @xmath36 k , @xmath37 ) . this is the simplest conceivable test of the pipeline if we generate a set of model data and then use the optimization method to match those data using exactly the same models , how quickly will the ga find the true values of the model parameters ? the results of such a test verify the basic functionality of the algorithm , and tell us the approximate number of iterations ( or `` generations '' of the ga ) that we need to execute before stopping the search . of the best solution in the population as a function of generation number for four independent runs of the genetic algorithm on the h&h1 data set . all four runs converge to the global solution within 150 generations.[fig1 ] ] the results of h&h1 for four independent runs of the ga are illustrated in figure [ fig1 ] . each convergence curve shows the reduced @xmath20 ( hereafter @xmath38 where @xmath39 is the number of constraints and @xmath40 is the number of parameters ) for the best model in the ga population as a function of generation number . all of the runs converge in less than @xmath41150 generations , in each case leading to the same mass as the source model and values for the other parameters offset by less than @xmath411% from the source values ( the best model from the ga is listed in the row labeled `` global '' under h&h1 in table [ tab1 ] ) . since we expect the algorithm to converge more quickly in this highly idealized case , we decided to continue running the ga for 200 generations for the more difficult tests that follow . the local analysis using lm+svd brings the final value of the metallicity ( @xmath13 ) closer to the source model , while retaining comparable accuracy for the other parameters ( see the row labeled `` local '' under h&h1 in table [ tab1 ] ) . note that the unusually small values of @xmath42 in table [ tab1 ] are a consequence of adopting scaled - up bison errors for the model frequencies , which vastly overestimate the true theoretical uncertainties in this case . the biggest challenge to comparing the oscillation frequencies from theoretical models with those actually observed in solar - type stars are the systematic errors due to _ surface effects_. the mixing - length parameterization of convection that is used in most stellar models is insufficient to describe the near - surface layers , and this leads to a systematic difference of several @xmath43hz ( up to about 0.3% for a solar model ) between the observed and calculated frequencies ( see figure [ fig2 ] ) . the offset is nearly independent of the spherical degree ( @xmath4 ) of the mode and grows larger towards the acoustic cutoff frequency . the 3d simulations of convection that might in principle reduce this discrepancy for individual stars are far too computationally expensive for the model - fitting approach that we are developing . instead , we adopt the method for empirical correction of surface effects described by @xcite , which uses the discrepancies between model s and golf data for the sun @xcite to calibrate the empirical surface correction . the models that we have adopted for the pipeline include a slightly different set of physical ingredients than what was used to produce model s from @xcite . to characterize the surface effects , we need a model that uses the same physics as the other models in the present investigation , while matching the model s frequencies as closely as possible . to find such a model ( hereafter model ) , we used our pipeline to fit the computed frequencies of model s. since this involves the comparison of two sets of model frequencies both of which include a mixing - length parameterization of convection there are no surface effects to consider . again we allowed the ga to fit 36 frequencies for modes with @xmath34 and @xmath44 and we used the same scaled bison errors and the same constraints on the effective temperature and luminosity ( see [ sec3.1 ] ) . the resulting optimal models from the global search and the local analysis had the parameter values listed in the rows below `` model '' in table [ tab1 ] . following @xcite , we fit a power law to the differences between the frequencies of the radial modes of this model and the corresponding frequencies from bison data to characterize the surface effects ( see figure [ fig2 ] and appendix [ appa ] ) . we found a power law exponent @xmath45 , slightly lower than the value ( @xmath46 ) derived by @xcite using data from the golf experiment . with this exponent fixed , the recipe of @xcite describes how to predict the surface effect for any other set of calculated oscillation data , allowing us to apply this empirical correction to each of our models before comparing them to observations . ] if our strategy of making this empirical correction to each of our models is to succeed , it must not only work _ well _ for models in a certain region of the search space it must work _ best _ for the model that simultaneously matches all of the independent observational constraints within their uncertainties . our second h&h exercise ( with parameters listed in the row labeled `` h&h2 '' in table [ tab1 ] ) was designed to test the behavior of the pipeline with the surface correction included . the source model and other constraints were identical to those used for h&h1 , but the surface effect was first calculated from eq.([eq.fobs - fref ] ) and then applied to each model frequency prior to fitting so that the h&h2 data would mimic real observations . we fixed the value of @xmath47 to 4.82 ( as determined above ) and we calculated the value of @xmath48 using eq.([eq.fobs - fref ] ) with the scaling factor @xmath49 determined using bison data as the reference set of observations . the fitting procedure then included the empirical surface correction attempting to remove the systematic frequency errors that we artificially introduced as detailed in appendix [ appa ] . again this test reveals how long we must run the ga for it to converge to the global solution , but it also quantifies any systematic errors on the derived parameter values that arise from our implementation of the empirical surface correction . as expected , the ga converges to the global solution more slowly with the inclusion of the surface correction . within 150200 generations , two of the four runs converged to the parameters listed in the row labeled `` global '' under h&h2 in table [ tab1 ] , while the other two remained in a nearby local minimum . if we had continued the search , these two runs would eventually also have found the global solution . the largest systematic error appears in the value of @xmath13 , which is about 6% ( two sample points ) higher in the global solution than in the source model . there are also smaller errors ( @xmath411% ) on the derived values of @xmath15 and @xmath17 . the results of the local analysis ( listed in the row labeled `` local '' under h&h2 in table [ tab1 ] ) bring both @xmath13 and @xmath17 closer to the source values while preserving the accuracy of the other parameters . the remaining errors are not a deficiency in the ability of the method to find the true solution the identified model actually results in a lower @xmath42 than the source parameters . the reason for this counter - intuitive result is that we can only generate a surface correction when there is a reference set of observed oscillation frequencies ( see appendix [ appa ] ) . our source model used the bison data for reference , and these data have a slightly different value of the large frequency separation ( @xmath50hz ) than the resulting source model ( @xmath51hz ) . since this quantity is used by the binary decision tree to fit the age of the model , the small difference leads to a slightly offset derived age which subsequently modifies the optimal composition through intrinsic parameter correlations . in other words , when generating a model with the surface correction included from a given set of parameters , those same parameters will generally not provide the best fit to the resulting observables . this actually highlights our inability to generate realistic data with surface effects , rather than an inherent limitation in our fitting method . in any case , this exercise demonstrates that the systematic errors resulting from our surface correction are small , and that running the ga for 200 generations should be sufficient for real observations . ultimately , our model - fitting pipeline can only be judged a success if it leads to accurate estimates of the stellar properties for the star that we know best : the sun . up to this point , we have essentially been fitting models to synthetic data using the solar data from bison only to calibrate the empirical surface correction and to provide realistic errors . there are many other ingredients in our models that could in principle be insufficient descriptions of the actual conditions inside of real stars deficiencies that could easily lead to systematic errors in our determinations of the optimal model parameters for a given set of oscillation data . for example , we initially tried to use models that employed the simpler eff equation of state @xcite for computational expediency , but this led to estimates of the stellar mass about 10% too high for the sun , and unacceptably large systematic errors on many of the other stellar properties . even attempting to ignore the effects of helium settling proved to be too coarse an approximation , leading to 5% errors on the mass . the only potential ingredient that we omitted without serious consequences was heavy element settling . this is not to say that simpler stellar models can not be used in the analysis of asteroseismic data , but rather that some of the more sophisticated ingredients are required to obtain accurate results from a _ global _ search of the parameter space . having demonstrated the effectiveness of the method by fitting our models to synthetic data , and after calibrating the empirical surface correction using the differences between model and the bison data , we finally applied our model - fitting pipeline to solar data from the bison and golf experiments . the oscillation frequencies from these two sources are identical to each other within the observational uncertainties , but their noise properties are slightly different allowing us to quantify any systematic errors that might arise from subtle effects in the data acquisition and analysis methods . in both cases we used the same set of modes referenced in the earlier experiments ( @xmath34 , @xmath44 ) with the respective errors again scaled up by a factor of 10 , and the same constraints on the effective temperature and luminosity ( see [ sec3.1 ] ) . the two sets of input data differed only in the absolute values of the oscillation frequencies ( yielding distinct values of @xmath6 for fitting the stellar age ) , and in the statistical uncertainties assigned to each mode ( leading to subtle differences in the weighting of the fit ) . in general , the four independent runs of the ga for each data set led to slightly different results after 200 generations , so we list only the solution leading to the best model in the rows below `` bison '' and `` golf '' in table [ tab1 ] . we found that if we continued running the ga for up to 300 generations , the same optimal solution was identified in 23 of the independent runs . to reduce the total computing time required for future experiments , we decided to stop the ga earlier when 12 of the runs would still reliably identify the global solution . both data sets lead to identical values of the mass and metallicity from the global search , with slight variations in the values of the other parameters . these minor differences largely disappear after the local analysis . note that because we multiplied the true observational errors by a factor of 10 for the fitting , the resulting values of @xmath42 are @xmath410.1 . although the fits used a limited range of frequencies and did not include @xmath52 modes , the optimal models also match the modes with lower frequencies and higher degree ( see the bison fit in figure [ fig3 ] ) and accurately reproduce the known solar properties . and plot them against the oscillation frequency , along with the optimal model from our asteroseismic modeling pipeline ( open points ) . note that the pipeline only used the @xmath44 frequencies between the dashed lines for the fit , but the resulting optimal model also matches the @xmath4=3 modes and frequencies outside of the fitting range.[fig3 ] ] rccccccccc h&h1 & 1.000 & 0.0191 & 0.271 & 2.06 & 4.54 & 5780 & 1.001 & 1.000 & @xmath53 + global : & 1.000 & 0.0197 & 0.273 & 2.04 & 4.51 & 5763 & 0.990 & 1.000 & 0.052 + local : & 1.001 & 0.0193 & 0.269 & 2.03 & 4.58 & 5755 & 0.986 & 1.000 & 0.004 + error : & 0.009 & 0.0014 & 0.012 & 0.07 & 0.16 & 61 & 0.040 & 0.004 & @xmath53 + model & @xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53 + global : & 0.990 & 0.0191 & 0.277 & 2.04 & 4.56 & 5778 & 0.993 & 0.996 & 0.062 + local : & 0.993 & 0.0188 & 0.275 & 2.05 & 4.52 & 5787 & 1.001 & 0.997 & 0.039 + error : & 0.009 & 0.0013 & 0.010 & 0.07 & 0.16 & 60 & 0.042 & 0.003 & @xmath53 + h&h2 & 1.000 & 0.0191 & 0.271 & 2.06 & 4.54 & 5780 & 1.001 & 1.000 & @xmath53 + global : & 1.010 & 0.0210 & 0.272 & 2.12 & 4.67 & 5773 & 1.005 & 1.004 & 0.042 + local : & 1.010 & 0.0197 & 0.268 & 2.09 & 4.55 & 5776 & 1.006 & 1.003 & 0.007 + error : & 0.010 & 0.0016 & 0.012 & 0.07 & 0.16 & 61 & 0.041 & 0.004 & @xmath53 + bison & @xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53 + global : & 1.010 & 0.0217 & 0.278 & 2.14 & 4.52 & 5793 & 1.018 & 1.004 & 0.197 + local : & 1.012 & 0.0200 & 0.267 & 2.12 & 4.65 & 5779 & 1.010 & 1.004 & 0.146 + error : & 0.008 & 0.0014 & 0.010 & 0.06 & 0.17 & 55 & 0.039 & 0.003 & @xmath53 + golf & @xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53&@xmath53 + global : & 1.010 & 0.0217 & 0.273 & 2.10 & 4.71 & 5752 & 0.990 & 1.004 & 0.183 + local : & 1.011 & 0.0199 & 0.267 & 2.11 & 4.66 & 5779 & 1.008 & 1.004 & 0.124 + error : & 0.014 & 0.0017 & 0.017 & 0.07 & 0.21 & 67 & 0.043 & 0.005 & @xmath53 with the successful validation of our model - fitting pipeline using solar data , we now need to ensure that our adopted treatment of surface effects yields reasonable optimal models when applied to other stellar data . while numerous ground - based and space - based data sets on other stars are already available , the computation - intensive nature of our method necessarily limits the present investigation to the single case of the sun . the obvious next step is to use archival ground - based data on several well studied solar - type stars to validate the pipeline for various stellar masses ( e.g. @xmath16 cen a & b , near @xmath411.1 and @xmath410.9 @xmath12 respectively ) and for different evolutionary stages ( e.g. the `` future sun '' @xmath54 hyi , at @xmath417 gyr ) . although @xcite demonstrated their method by applying it to models of these three stars in addition to the sun , our experience using it with solar data suggests that a _ global _ exploration of the models may present additional challenges . since our stellar models and the empirical correction for surface effects have both been calibrated using a main - sequence star at 1.0 @xmath12 , the @xmath16 cen system will help validate the models with interior physical conditions that differ slightly from those of the sun . its proximity and multiple nature make it an excellent second test of our pipeline , because it has very well determined properties including stellar radii from interferometry @xcite . there are also strong constraints on the component metallicities and effective temperatures @xcite , while the initial composition and age of the two stars are presumably identical . in the next phase of this project , we plan to use the published oscillation frequencies of @xmath16 cen a @xcite and @xmath16 cen b @xcite with the additional constraints from interferometry , spectroscopy , and the binary nature of the system to further validate our pipeline and the empirical correction for surface effects . @xcite successfully applied their recipe to a set of stellar models that broadly resemble the components of the @xmath16 cen system , so we have good reason to believe that our implementation will also succeed but this remains to be demonstrated . the g2 subgiant @xmath54 hyi has long been studied as a `` future sun '' , with an age near 7 gyr . it has been characterized almost as extensively as the @xmath16 cen system , including recent interferometric measurements of its diameter @xcite and dual - site asteroseismic observations that determined its mean density with an accuracy of 0.6% @xcite . these data included the detection of several @xmath32 modes that deviate from the asymptotic frequency spacing , suggesting that they are `` mixed modes '' behaving like g - modes in the core and p - modes in the envelope . this is expected for evolved stars like @xmath54 hyi because as they expand and cool the p - mode frequencies decrease , while the g - mode frequencies increase as the star becomes more centrally condensed . this leads to a range of frequencies where these modes can overlap and exchange their character , manifested as so - called _ avoided crossings_. this behavior changes very quickly with stellar age , and propagates from one mode to the next as a star continues to evolve . consequently , the particular mode affected yields a very strong constraint on the age of the star ( see * ? ? ? * ) . in subsequent work , we plan to use the published oscillation data for @xmath54 hyi @xcite along with the constraints from interferometry and spectroscopy to validate our pipeline and the empirical treatment of surface effects for stars that are significantly more evolved than the sun . this will require an automated method to recognize mixed modes in the data set and to incorporate them into the optimization of stellar age along each track . once we have validated the model - fitting pipeline with additional stars that sample a range of masses and evolutionary stages , we can begin to consider additional observables and parameters that are not constrained by currently available data sets . high - quality asteroseismic data are soon expected from the kepler mission , spanning sufficiently long periods of time that the effects of rotation @xcite and magnetic activity cycles @xcite should be detectable . the kepler mission is designed to discover earth - sized habitable planets , and our model - fitting pipeline will be able to characterize the planet - hosting stars with asteroseismology . this is essential to convert precise transit photometry into an absolute radius for the planetary body . in addition , accurate rotation rates and ages will provide clues about the formation and evolution of the planet - hosting systems . the determination of accurate stellar properties for a broad array of solar - type stars will give us a new window on stellar structure and evolution , and will provide a broader context for our understanding of the sun and our own solar system . we hope to facilitate this process by applying our stellar model - fitting pipeline to the data that will soon emerge from the kepler mission . the authors wish to thank tim brown and margarida cunha for helpful discussions during the early phases of this project . this work was supported in part by an nsf astronomy & astrophysics fellowship ( to t.s.m . ) under award ast-0401441 , by a newkirk graduate fellowship ( to o.l.c . ) at the high altitude observatory and by the european helio- and asteroseismology network ( helas ) a major international collaboration funded by the european commission s sixth framework programme , by the danish natural science research council , and by nasa grant nnx09ae59 g . computer time was provided by nsf mri grants cns-0421498 , cns-0420873 , cns-0420985 , the university of colorado , and a grant from the ibm shared university research ( sur ) program . the national center for atmospheric research is a federally funded research and development center sponsored by the u.s . national science foundation . asplund , m. , grevesse , n. , sauval , a. j. , allende prieto , c. , & kiselman , d. 2004 , , 417 , 751 baglin , a. , michel , e. , auvergne , m. and the corot team , 2006 , in proc . soho 18 / gong 2006 / helas i conf . beyond the spherical sun , k. fletcher , esa sp-624 ( noordwijk , the netherlands : esa publications division ) bahcall , j. n. , & pinsonneault , m. h. 1995 , rev . phys . , 64 , 885 ballot , j. , garca , r. a. , & lambert , p. 2006 , , 369 , 1281 ballot , j. , appourchaux , t. , toutain , t. , & guittet , m. 2008 , , 486 , 867 bazot , m. , bouchy , f. , kjeldsen , h. , charpinet , s. , laymand , m. , & vauclair , s. 2007 , , 470 , 295 bedding , t. r. , kjeldsen , h. , butler , r. p. , mccarthy , c. , marcy , g. w. , otoole , s. j. , tinney , c. g. , & wright , j. t. 2004 , , 614 , 380 bedding , t. r. , et al . 2007 , , 663 , 1315 bedding , t. r. , & kjeldsen , h. 2007 , in aip conf . 948 , unsolved problems in stellar physics , eds r. j. stancliffe , j. dewi , g. houdek , r. g. martin , c. a. tout ( melville : american institute of physics ) , p. 117 bhm - vitense , e. 1958 , , 46 , 108 bouchy , f. & carrier , f. 2002 , , 390 , 205 brown , t. m. , & gilliland , r. l. 1994 , , 32 , 37 brown , t. m. , christensen - dalsgaard , j. , weibel - mihalas , b. , & gilliland , r. l. 1994 , , 427 , 1013 caffau , e. , ludwig , h .- g . , steffen , m. , ayres , t. r. , bonifacio , p. , cayrel , r. , freytag , b. , & plez , b. 2008 , , 488 , 1031 carrier , f. & bourban , g. 2003 , , 406 , l23 chaplin , w. j. , elsworth , y. , isaak , g. r. , miller , b. a. & new , r. 1999 , , 308 , 424 chaplin , w. j. , elsworth , y. , houdek , g. , & new , r. 2007 , , 377 , 17 charbonneau , p. 1995 , , 101 , 309 christensen - dalsgaard , j. 1993 , in proc . gong 1992 : seismic investigation of the sun and stars , ed . t. m. brown , asp conf . , 42 , 347 christensen - dalsgaard , j. 2002 , rev . phys . , 74 , 1073 christensen - dalsgaard , j. 2004 , , 220 , 137 christensen - dalsgaard , j. 2008a , , 316 , 13 christensen - dalsgaard , j. 2008b , , 316 , 113 christensen - dalsgaard , j. , & gough , d. o. , 1980 , , 288 , 544 christensen - dalsgaard , j. , & thompson , m. j. 1997 , , 284 , 527 christensen - dalsgaard , j. , et al . 1996 , science , 272 , 1286 christensen - dalsgaard , j. , arentoft , t. , brown , t. m. , gilliland , r. l. , kjeldsen , h. , borucki , w. j. , & koch , d. 2007 , communications in asteroseismology , 150 , 350 creevey , o. l. , monteiro , m. j. p. f. g. , metcalfe , t. s. , brown , t. m. , jimnez - reyes , s. j. & belmonte , j. a. 2007 , , 659 , 616 creevey , o. l. 2008 , phd thesis , university of la laguna . creevey , o. l. 2009 , asp conf . , in press ( arxiv:0810.2442 ) eggleton , p. p. , faulkner , j. , & flannery , b. p. 1973 , , 23 , 325 gizon , l. , & solanki , s. k. 2003 , , 589 , 1009 iglesias , c. a. , & rogers , f. j. 1996 , , 464 , 943 kennelly , e. j. , et al . 1998 , , 495 , 440 kervella , p. , et al . 2003 , , 404 , 1087 kjeldsen , h. , bedding , t. r. , viskum , m. , & frandsen , s. 1995 , , 109 , 1313 kjeldsen , h. , et al . 2005 , , 635 , 1281 kjeldsen , h. , bedding , t. r. , & christensen - dalsgaard , j. 2008 , , 683 , l175 latham , d. w. , brown , t. m. , monet , d. g. , everett , m. , esquerdo , g. a. , & hergenrother , c. w. 2005 , , 37 , 1340 lazrek , m. , et al . 1997 , , 175 , 227 metcalfe , t. s. 2003 , , 587 , l43 metcalfe , t. s. 2005 , , 363 , l86 metcalfe , t. s. , & charbonneau , p. 2003 , j. computat . phys . , 185 , 176 metcalfe , t. s. , nather , r. e. , & winget , d. e. 2000 , , 545 , 974 metcalfe , t. s. , montgomery , m. h. , & kanaan , a. 2004 , , 605 , l133 metcalfe , t. s. , dziembowski , w. a. , judge , p. g. , & snow , m. 2007 , , 379 , l16 michaud , g. , & proffitt , c. r. 1993 , in proc . iau colloq . 137 : inside the stars , eds a. baglin , & w. w. weiss , asp conf . , 40 , 246 north , j. r. , et al . 2007 , , 380 , l80 porto de mello , g. f. , lyra , w. , & keller , g. r. 2008 , , 488 , 653 rogers , f. j. , swenson , f. j. , & iglesias , c. a. 1996 , , 456 , 902 stello , d. , kjeldsen , h. , bedding , t. r. , de ridder , j. , aerts , c. , carrier , f. , & frandsen , s. 2004 , , 220 , 207 for convenience we summarize the analysis of @xcite , to correct for the near - surface frequency effects . this is based on analyzing the large frequency separation @xmath55 ( cf . 1 ) and the individual frequencies . since the offset from incorrect modeling of the near - surface layers is not a strong function of @xmath4 , we can derive the correction by considering only the radial modes ( @xmath31 ) and then apply it to all of the modes . suppose we have a set of observed frequencies for radial modes @xmath56 , where @xmath3 is the radial order . let @xmath57 be the frequencies of the model that we seek the one that best describes the parameters and internal structure of the star , but which still fails to model correctly the surface layers . in the solar case , this model is defined by the inversion of the frequencies over a broad range of available modes . here the difference between the observed and best model frequencies is found to be well fitted by a power law , which has the convenient property of being free of a frequency scale : @xmath58 ( see also * ? ? ? * ) , where @xmath59 is a suitably chosen reference frequency , and @xmath48 and @xmath47 are parameters to be determined . in practice we do not know the best model . thus , we must determine the correction from a reference model , with frequencies @xmath60 , which is assumed to be close to the best model . in that case we have to a good approximation from homology scaling @xmath61 where the scaling factor @xmath49 is related to the mean densities of the best and reference models . substituting eq.([eq.fbest.fref ] ) into eq.([eq.fobs - fref ] ) and differentiating with respect to @xmath3 gives @xmath62 from which we finally obtain @xmath63 and @xmath64 ( for details , see * ? ? ? if we know @xmath47 then we can calculate @xmath49 using eq.([eq.r ] ) , or _ vice versa _ using eq.([eq.b ] ) . we can then obtain @xmath48 using equations ( [ eq.fobs-fref ] ) and ( [ eq.fbest.fref ] ) . applying this method to the sun , we used the frequencies of model , which were obtained by fitting the frequencies of model s from @xcite , but using slightly modified physics ( see [ sec3.2 ] ) . we assumed this to be the `` best '' solar model , which means that we can set @xmath65 ( see eq . [ eq.fbest.fref ] ) . the observed solar frequencies were obtained from the bison experiment @xcite . from the procedure described above , choosing @xmath66hz , setting @xmath65 and using the data for the dozen radial modes with @xmath34 , we obtained a value of @xmath45 . @xcite repeated a similar analysis for different numbers of modes ( all values from 7 to 13 ) , and found the derived value of @xmath47 to range from 4.4 to 5.25 . clearly , the frequency differences do not exactly follow a power law , and so the exponent in the power - law fit depends substantially on the frequency range . however , the value for @xmath48 varied by less than 0.1@xmath43hz in all cases . if we repeat our analysis using golf data @xcite for the same set of modes , we derive a value of @xmath67 . both of our values are comparable to the value @xmath46 derived by @xcite . after calibrating the value of the exponent @xmath47 using solar data , the prescription of @xcite describes how to predict the surface effect for any other stellar model being compared to real observations . the value of @xmath49 is determined by applying eq.([eq.r ] ) to each trial model through a comparison of the calculated frequencies with the observations . the value of @xmath48 is then determined using equations ( [ eq.fobs-fref ] ) and ( [ eq.fbest.fref ] ) , and subsequently used in conjunction with @xmath47 to predict the surface effect for each frequency of the model . this recipe is applied to every model calculated during the global search by the ga , and during the local analysis using lm+svd . effectively we are adjusting the predicted surface effect for each model based on the difference between the mean densities implied by the two sets of frequencies . in addition , the magnitude of the surface effect reduces the weight of the mode in the final fit , giving more weight to low frequencies but still attempting to optimize the match at high frequencies . the final weight for each mode is determined by the quadratic sum of the ( statistical ) observational error and half the ( systematic ) surface correction . as discussed in detail by @xcite , the analysis of the linearized relation between observables and parameters is conveniently carried out in terms of singular value decomposition ( svd ) because it has some very useful properties when applied to a matrix . suppose we have a _ model _ that when given a set of input parameters * p * ( we normally do not know * p * and we wish to retrieve them ) produces an output set of expected model observables . generally the model- ( or parameter- ) fitting problem consists of matching a set of observations * o * to the output model observables , by making parameter changes @xmath26 to * p * , until we find the model that fits the observations best . once we find * p * , to exploit svd we calculate the partial derivatives of each of the observables with respect to each of the model parameters * p * , and subsequently divide by the measurement errors . the result is what we call the design matrix * d*. the svd of * d * (= * uwv@xmath68 * ) neatly describes the relationship between the model observables and the stellar parameters through a set of linear transformation vectors . here the matrices * u * and * v * are of order @xmath69 and @xmath70 respectively , where @xmath39 is the number of observables and @xmath40 is the number of parameters ; * w * is a diagonal matrix of order @xmath70 , with diagonal elements consisting of the singular values @xmath71 . * u * and * v * consist of orthonormal vectors that span the observable and parameter spaces respectively , while the diagonal elements of * w * assign a relative importance to each vector . each column vector of * u * describes the content of each observable , the singular value @xmath71 quantifies the importance of this vector ( i.e. , it organizes the observables by amount and type of information ) , while each column vector of * v * describes how the information from each observable vector is distributed among the parameters ( i.e. , which observables are important for constraining certain parameters ) . we will see this more clearly below . if the models were linear then the solution @xmath26 , in a least - squares sense , could be found simply from the discrepancies @xmath26 between the observations and the model observables , normalized by the errors . in terms of svd , the result can be written as @xmath26 = * vw@xmath72u@xmath68 * @xmath26 , which shows clearly that the parameter changes required to reconcile * p * with the observables comes from a product of ( 1 ) the discrepancies between the observations and the observables , and ( 2 ) a set of linear transformation vectors that are scaled by the inverse of the singular values . below we will explore how we can use svd to illucidate both the information contained in the set of observables and the available constraints on the parameters . and @xmath73 orthonormal matrices from the svd of the design matrix , evaluated for the model that best fits the bison data . each row in @xmath74 shows how the set of observables are responsible for each of the parameters given in each row of @xmath73 , e.g. , row 1 contains mostly the parameter of @xmath13 , and this is constrained mainly by the 3 frequencies that appear in row 1 of @xmath74 . if we scale the rows in @xmath73 by the inverse of the singular values we obtain the parameter correlations , which describe how to define the 5-dimensional error boundary.[fig4],width=528 ] figure [ fig4 ] illustrates part of the orthonormal matrix * u * ( left ) and the full orthonormal matrix * v*. because we have @xmath75 parameters in the problem , each of the matrices consists of 5 _ column vectors _ ; thus we plot @xmath74 and @xmath73 , ordered such that the horizontal vectors corresponding to the largest singular values @xmath71 are at the top . each column vector of * u * has @xmath76 elements , which is the number of observations that are used to determine the solution , but we only include the first 5 elements for illustration . the elements of the matrix can only take values between @xmath77 and + 1 , and each number is represented by a triangle whose size and direction are proportional to the magnitude and sign of each element . the column vectors are labeled 15 : 1 represents the first column vector , which is associated with the largest singular value , 2 is the second largest singular value , while 5 is the smallest . thus , the vectors are effectively _ ranked _ according to their information content , where the least important information is given in the vectors corresponding to the smallest @xmath71 . for example , if we were to perform the matrix multiplication of * uwv@xmath68 * using only the top vectors , we would be able to recover most of the input matrix * d*. essentially this is equivalent to zeroing the values of * w * that are below a certain threshold before doing the multiplication . clearly the observables that appear among the top vectors are most important for constraining the parameter solution , while the parameters that appear among the top vectors are the most constrained . inspecting the left panel of figure [ fig4 ] , by scanning down the fourth column we see that the observable @xmath78 has components mainly in the second and fourth vectors . the luminosity @xmath79 is present in vectors 24 , while the individual frequencies are present mostly in vectors 1 and 2 . the most important information comes from the first vector , and we can see clearly that in this case the seismic information dominates . we also see that @xmath79 and @xmath78 usually appear in the same vectors , implying that these observables work together to constrain the parameter solution . the right panel shows how the information from the observables is distributed among the parameters . the top vector has mainly one large component , @xmath13 . this implies that @xmath13 is most constrained by the set of observables , and in particular by the observables that appear in the top row of the @xmath74 matrix , i.e. the frequencies . similarly , the mixing - length parameter ( @xmath16 ) appears predominantly in the fourth vector , and is constrained mainly by @xmath78 and @xmath79 . returning to the linear parameter - fitting problem , we saw that calculating @xmath26 depended strongly on the size of the singular value associated with the transformation vectors ( as well as the discrepancies ) . because @xmath13 appears in the first vector and the inverse of the first singular value is the smallest , this means that @xmath13 is not allowed to deviate much from its initial value . so if we find that we are in a local minimum that is not the global solution , it is unlikely that we will find a better solution that deviates much in @xmath13 . for this reason we choose to explore a range of @xmath13 values during the local analysis . the vectors * v*@xmath80 are the basis vectors , and describe the normalized changes that we can make to each of the parameters while respecting the constraints imposed by the observables . the inverse of the @xmath71 describe the magnitude of each of the vectors that keeps the parameters within 1@xmath81 of the final answer . since the vectors corresponding to larger @xmath82 have smaller @xmath71 ( larger inverses ) , the parameters associated with these vectors will extend further into the parameter space , i.e. they are less constrained by the observations . and @xmath15 , as described by svd , for the final fit model to the bison data . the arrows are the 2-element directional vectors given by the projections of @xmath83 and @xmath84 which dominate the determination of @xmath11 and @xmath15 ( cf . [ fig4 ] ) , and the ellipse is defined by these vectors.[fig5],width=336 ] inspecting the matrix @xmath73 in figure [ fig4 ] , note that vector 2 reveals both @xmath11 and @xmath15 to be positive , indicating that they should be increased or decreased together to satisfy the constraints imposed by the observables . however , vector 3 implies that if @xmath11 increases , @xmath15 should decrease . when both of these vectors are projected onto the @xmath11-@xmath15 plane and scaled according to 1/@xmath85 and 1/@xmath86 , vector 3 is the longer of the two . these scaled vectors are shown as arrows in figure [ fig5 ] , defining the semi - major and minor axes of the projected 1@xmath81 error ellipse . the error ellipse defines the boundary of the allowed parameter values that will change the @xmath20 value by less than @xmath87 , where @xmath87 is the value that defines the 1@xmath81 error for the relevant number of degrees of freedom . this figure shows clearly that there is a very strong inverse correlation between these parameters , and hints at the size of the uncertainty in each one or @xmath15 tends to increase the luminosity , so if one of these parameters is increased the other must be decreased to satisfy the luminosity contraint while continuing to match the other observables . ] . though not as clearly visible in figure [ fig4 ] , @xmath13 and @xmath15 have a strong direct correlation , so for the local analysis we explore a range of @xmath15 as well as @xmath13 . finally , the _ covariance matrix _ can be expressed in a very compact form using svd ; the covariance between parameters @xmath88 and @xmath89 is : @xmath90 in particular , the formal uncertainties @xmath91 on each of the parameters are obtained from @xmath92 . given this very clear form for describing the correlations between the parameters and their uncertainties , we can also determine the uncertainties on the model observables ( such as @xmath93 and @xmath94 ) or some other properties ( such as the core hydrogen content ) by calculating the models at the boundaries of the error ellipse . equation ( [ eqn : covariance ] ) can also be used to investigate _ how _ to reduce the uncertainties in each of the parameters with what precision do we need to measure the individual observables ? what other information may be useful to help break the degeneracies between some parameters ? and are there limits to what we can learn from the available data ?

the ward - takahashi identities are central to the gauge invariance of the photoproduction amplitude . here we demonstrate that unitarity and in particular the inclusion of both the and thresholds on equal footing yields a photoproduction amplitude that satisfies both two - body unitarity and the generalized ward - takahashi identities . the final amplitude is a solution of a set of coupled channel integral equations for the reactions and .

nucl-th9407038

pion photoproduction has a long history going back to the early 50 s @xcite . in recent years the interest in this reaction has been revived as a result of new experimental facilities which allow more careful studies of this reaction over a much wider energy range , and the advent of quantum chromodynamics ( qcd ) as a fundamental theory of strong interaction which should yield the internal structure of the nucleon . in particular the study of the reaction @xmath4 is motivated by : ( i ) the fact that this is the simplest nuclear system one can examine to understand the mechanism of pion photoproduction . furthermore the amplitude for this reaction can be used as input into the calculation of pion photoproduction off heavier nuclei @xcite , with the expectation of gaining information about nuclear structure . ( ii ) the recent interest in chiral symmetry @xcite and its possible violation in this reaction @xcite , which have raised questions regarding the role of unitarity in this reaction @xcite , and the use of pion photoproduction to test chiral models of qcd , e.g. chiral perturbation theory @xcite . ( iii ) the new facilities at cebaf , and other facilities which will open the way to examine the structure of the nucleon and the resonances observed in @xmath0 scattering and pion photoproduction . these results could shed light on the need to introduce quark - gluon degrees of freedom , and in particular determine the energy at which these new degrees of freedom become more important than the traditional meson - baryon degrees of freedom . in addition , this data could be used to test models of qcd . to examine the structure of resonances observed in @xmath0 scattering using pion photoproduction , we need to include the important thresholds for all possible reactions at that resonance energy . to achieve this we may need to include in our formulation those unitarity cuts , and therefore thresholds , important for that resonance @xcite . on the other hand the incident photon will interact with the electromagnetic ( e.m . ) charge and current distribution in the target , and to that extent it is essential to satisfy charge conservation or @xmath5 gauge invariance . the fact that the proton has an internal structure , that could be described in terms of quarks at short distances and mesons at large distances , suggests that these degrees of freedom should be taken into consideration in determining the electromagnetic ( e.m . ) charge and current distribution . the meson degrees of freedom , which include nucleon dressing and the distortion of the final outgoing pion wave are part of the mechanism needed to include unitarity and therefore the proper thresholds . thus under ideal conditions one should satisfy both unitarity and @xmath5 gauge invariance in a consistent manner . in addition , the quark dynamics should be included in the determination of the charge and current distribution even though they do not contribute directly to any thresholds since the quarks are confined . historically , early calculations of pion photoproduction were made to satisfy unitarity by imposing the watson theorem @xcite . this basically consisted of representing the final distortion in the @xmath0 channel by the on - shell pion - nucleon amplitude or @xmath0 phase shift . in this way the cross - section for pion photoproduction consisted of two parts , the born amplitude which included the interaction of the photon with the nucleon and meson currents ( see fig . 1 ) , and the on - shell @xmath0 amplitude which was included to satisfy two - body unitarity . the two ingredients were considered to be independent . thus for the born amplitude , the photon coupled to a nucleon with an internal structure represented by an on - mass - shell form factor extracted from elastic electron - proton scattering , while the pion coupling to a finite nucleon ( see fig . 1 ) involved a @xmath6 form factor which was constrained by the @xmath5 gauge invariance to be the same as the e.m . form factor @xcite . here we should note that there is no consistency between the @xmath6 form factor used and the e.m . form factor other than the overall gauge invariance of the born amplitude . on the other hand , the @xmath0 amplitude , which also has the @xmath7 form factor , was determined solely by the @xmath0 scattering data . in other words there was no consistency between the born amplitude for pion photoproduction and the @xmath0 amplitude that determined the dynamics of pion - nucleon scattering . more recently @xcite , extensions of this procedure have been developed with considerable success , that extend the watson theorem by employing the off - shell @xmath0 amplitude generated by either a separable potential @xcite or a chiral lagrangian @xcite to determine the final distortion of the outgoing pion . even in these calculations no attempt has been made to have any consistency between the @xmath0 dynamics and the born amplitude for pion photoproduction . the present investigation is an attempt at a reformulation of pion photoproduction that satisfies both unitarity and @xmath5 gauge invariance . the @xmath0 dynamics required for unitarity will determine the charge and current distribution . in this way we maintain consistency between the pion dynamics and the requirement of e.m . gauge invariance . to include gauge invariance and the dynamics that determine the underlying structure of the nucleon , we need to gauge the lagrangian for the underlying strong interaction dynamics . ideally this means we need to start from the qcd lagrangian and include the e.m . interaction . however at this stage we have no practical procedure for projecting meson - baryon degrees of freedom without resorting to models of qcd . for example , if we assume that the cloudy bag model ( cbm ) @xcite determines the dynamics of the quarks and pions , then we should first gauge the cbm lagrangian @xcite to include the e.m . interaction . to substitute the quark degrees of freedom by the baryonic degrees of freedom , we need to take the second step of implementing the procedure adopted in the cbm by effectively integrating out the quark degrees of freedom . this two step process of getting an effective meson - baryon lagrangian that includes the electromagnetic and meson interaction consistently , is illustrated in fig . [ fig.2 ] . unfortunately , to get the correct current for the coupling of the photon to the baryon , we need to construct , in the cbm , a state that is an eigenstate of the total four - momentum that can be boosted from one inertial frame to another @xcite . this in practice is not possible in a bag model because the bag boundary condition can not be simply boosted . however , this procedure could be followed in other chiral soliton models @xcite with some difficulty , or we can use a model such as the njl lagrangian @xcite or the global colour model ( gcm ) @xcite which can in principle give a translationally invariant state @xcite . the resultant lagrangian would have gauge invariance , and we could then proceed to implement unitarity , taking as a starting point the lagrangian in the space of baryons , mesons and photons . the effect of quarks would be to have form factors for all vertices , but now we would have a self - consistency between the different form factors in the lagrangian . in particular , the form factor associated with the nucleon e.m . current and the form factor in the @xmath6 vertex would be consistent . furthermore , this @xmath7 form factor goes into the dynamics that generates the @xmath0 interaction . since the quarks are confined , and therefore do not contribute to unitarity , an alternative procedure that would satisfy gauge invariance and unitarity , see fig . [ fig.2 ] , would be to integrate the quark degrees of freedom in favour of baryon degrees of freedom . in this case , we would have form factors associated with the meson - baryon lagrangian , and these form factors are related to the quark structure of the nucleon . we then can implement ohta s @xcite approach of minimal e.m . coupling to gauge the resultant lagrangian with form factors . the contribution to the nucleon propagator and @xmath6 form factor from meson dressing will then be included explicitly , while maintaining both unitarity and gauge invariance . in sec . [ sec.2 ] we employ ohta s minimum electromagnetic coupling to generate a lagrangian that corresponds to a nucleon with internal structure and that is gauge invariant . at this stage we will assume that this internal structure is due to quark degrees of freedom . furthermore , this lagrangian in lowest order gives a photoproduction amplitude that is gauge invariant . this is basically a summary of ohta s results , and sets the stage for the next step of including the contribution to unitarity from the pionic degrees of freedom . this will be detailed in sec . [ sec.3 ] where we derive a set of coupled integral equations for both @xmath0 elastic scattering and @xmath8 . since the threshold energies for @xmath0 scattering and @xmath9 are the same , we include both of these thresholds in our coupled channel formulation . to establish the gauge invariance of the photoproduction amplitude resulting from the coupled channel approach , we proceed in sec . [ sec.4 ] to make use of the ward - takahashi @xcite identities and the procedure proposed by kazes @xcite to derive the amplitude for this reaction from the one pion irreducible @xmath6 three - point function derived in sec . [ sec.3 ] . we then show in sec . [ sec.5 ] that the amplitude resulting from the solution of the coupled channel is in fact identical to that derived from the application of the ward - takahashi identities . in this way we establish that the solution of the coupled channels problem satisfies both unitarity and gauge invariance . finally in sec . [ sec.6 ] we present some concluding remarks . our main motivation in the present investigation is to set up a unitary and gauge invariant formulation of pion photoproduction . since the quarks and gluons are confined , they do not contribute to unitarity , and we can include this sub - structure by introducing form factors into our lagrangian . in other words we can consider a lagrangian for the system of nucleons and pions as @xmath10 the lagrangians for the baryon and meson with internal structure are taken to be @xcite [ eq:2.2 ] @xmath11\psi(x ) \label{eq:2.2a}\ ] ] @xmath12 where the form factors @xmath13 , @xmath14 , and @xmath15 are to be determined from the underlying quark - gluon structure , e.g. we could write the corresponding propagators for the nucleon and pion as [ eq:2.3 ] @xmath16^{-1 } \label{eq:2.3a}\ ] ] @xmath17^{-1}\ , \label{eq:2.3b}\ ] ] where @xmath18 and @xmath19 could be written in terms of their quark structure as illustrated in fig . [ fig.3 ] . it should be pointed out at this stage that the qcd model used in evaluating the the diagrams in fig . [ fig.3 ] should not give rise to unitarity cuts since the quarks are confined . the mass shifts in both the baryon and the meson propagators can then be written in terms of the functions in the lagrangian , e.g. by taking @xmath20 we can write the fourier transform of the functions @xmath13 and @xmath14 in terms of @xmath21 and @xmath22 as @xmath23 and @xmath24 . thus the nucleon propagator is given as @xmath25 the conditions that this propagator have a simple pole at the nucleon mass , @xmath26 , and the residue at this pole be one , impose the following conditions on the functions @xmath27 and @xmath28 , i.e. [ eq:2.5 ] @xmath29 and @xmath30 with @xmath31 in a similar manner we can determine @xmath32 in terms of @xmath19 with the requirement that the pion propagator has a pole at @xmath33 with unit residue . in this way the underlying quark - gluon degrees of freedom have been included , and the gauging of these propagators will give us the coupling of the photon to the nucleon and the pion that have internal structure . at this stage we should point out that any pionic dressing of the nucleon will shift the mass @xmath26 , and to that extent the nucleon mass in eq . ( [ eq:2.5 ] ) may need to be considered as the bare nucleon mass . this will also apply to the pion mass , if it is to get any further dressing . the interaction terms in the lagrangian in eq . ( [ eq:2.1 ] ) can be written as @xmath34 here again the vertex function @xmath35 could be determined from an underlying qcd model , e.g. the cbm gives the @xmath6 form factor in terms of the bag radius and the wave function of the quark in the bag . the non - locality in this lagrangian can be turned into momentum dependence as demonstrated by ohta @xcite . this in turn allows us to gauge the lagrangian using the minimal substitution rule . to achieve this we follow ohta s @xcite procedure of converting the momenta into operators , i.e. @xmath36 , and then introducing the minimal substitution rule @xmath37 where @xmath38 is the charge operator given in terms of the nucleon isospin operator by @xmath39 and @xmath40 is the photon field . thus for the nucleon propagator we have , using this procedure , @xmath41 where after writing the propagator in terms of the form factors @xmath28 and @xmath27 , we have @xmath42 \nonumber \\ & & \quad -mh\left((\hat{p}-qa)^{2}\right ) \ . \label{eq:2.10}\end{aligned}\ ] ] clearly the presence of the photon field in the form factors @xmath13 and @xmath14 does not allow a simple determination of the @xmath43 vertex . as such , the photon field factors must be extracted from @xmath13 and @xmath14 . to this end , it is assumed that there exists a taylor series expansion for both @xmath13 and @xmath14 of the form @xmath44 restricting our discussion to @xmath13 only and applying the minimal substitution to equation ( [ eq:2.11 ] ) gives @xmath45 since the charge operator @xmath38 , is directly proportional to the electromagnetic coupling constant @xmath46 , eq . ( [ eq:2.12 ] ) can be expanded in powers of @xmath38 . retaining only those terms linear in @xmath38 gives , @xmath47\biggr\}\ . \label{eq:2.13}\end{aligned}\ ] ] in the context of perturbation theory , @xmath48 can be replaced by the corresponding eigenvalue depending upon whether it acts before or after the photon field @xmath49 . since the photon gives momentum to the nucleon , as illustrated in fig . [ fig.4 ] , we can use the following substitution @xmath50 proposed by ohta @xcite to reduce equation ( [ eq:2.13 ] ) to the form @xmath51 this is the relation governing how @xmath13 behaves under gauging ; there is a corresponding relation for @xmath14 . with these results in hand , equation ( [ eq:2.9 ] ) , after a little algebra involving the use of equation ( [ eq:2.14 ] ) , can be written to first order in @xmath49 as @xmath52 where @xmath53 is the photon momentum and @xmath54 \nonumber \\ & & \qquad + q\left(p'+p\right)_{\mu } \frac{s^{-1}_0(p')-s^{-1}_0(p)}{p'^{2}-p^{2}}\ , \label{eq:2.17}\end{aligned}\ ] ] is the resulting electromagnetic current for the nucleon . in the event that @xmath55 , i.e. the nucleon has no internal structure , then @xmath56 which is the standard dirac current . it is now a straight forward matter to check that @xmath57 , given in eq . ( [ eq:2.17 ] ) , satisfies the required ward - takahashi identity @xcite , i.e.@xmath58 thus we have the gauging of the nucleon propagator to be @xmath59 where perturbation theory has been used to replace @xmath48 by the corresponding eigenvalue . in this way we have derived the nucleon current with a form factor defined by the underlying quark model , and which satisfies the requirement of gauge invariance to the extent that the @xmath43 vertex satisfies the ward - takahashi identity in a similar manner we can proceed to gauge the pion propagator @xmath60 where @xmath61 is the pion momentum operator and @xmath19 contains terms which dress the pion in terms of its quark - gluon content , e.g. @xmath62 . defining @xmath63 as the pion charge operator , we make use of the minimal substitution prescription @xmath64 to write the gauged pion propagator as @xmath65 where we have dropped the @xmath66 index in the photon field for convenience . following the procedure developed above for the nucleon , we write a taylor series expansion for @xmath67 , and expand the resultant expression to first order in the photon field . this gives us @xmath68 where @xmath69 has been used in analogy with equation ( [ eq:2.14 ] ) . the gauged pion propagator , to first order in @xmath49 , is now given by @xmath70 where @xmath53 is the photon momentum , and @xmath71 is the pion electromagnetic current , which satisfies the ward - takahashi identity @xmath72 in the limit of a structureless pion , @xmath73 , eq . ( [ eq:2.25 ] ) reduces to the standard current for a point pion , i.e.@xmath74 in analogy with eq . ( [ eq:2.19 ] ) we can determine the behavior of the pion propagator @xmath75 under gauge transformation to be @xmath76 having established the gauging of the nucleon and pion propagators , we turn our attention to the gauging of the interaction lagrangian , and in particular the pion production vertex . since we have not specified the qcd model to be used in determining the @xmath6 vertex , we will consider the most general structure this vertex can have off - mass - shell @xcite , i.e.@xmath77 where @xmath78 are the form factors , with @xmath48 and @xmath79 the nucleon four - momenta , and @xmath61 the pion four - momentum . we have taken the momenta to be operators since we will be employing ohta s approach to gauge this vertex . since we have a non - local vertex , we can attach the photon before or after the pion has been emitted , and to maintain charge conservation for the overall vertex within an isospin formalism , we need to introduce the following definitions , [ eq:2.30 ] @xmath80 and @xmath81=ie\epsilon_{3ij}\,\tau_{j}\ , \label{eq:2.30b}\ ] ] which enables the relevant gauge transformations to be written as [ eq:2.31 ] @xmath82 and @xmath83 since we are considering pion emission . applying these gauge transformations to the above general @xmath6 vertex , and proceeding with the same steps used in gauging the propagator , i.e. assuming a taylor series expansion for the form factors @xmath84 , we get @xmath85 in writing eq ( [ eq:2.32 ] ) we have made use of perturbation theory to replace the various momenta operators by their corresponding eigenvalues , which for photon absorption and pion emission are given by [ eq:2.33 ] @xmath86 and @xmath87 having established the procedure for gauging an individual form factor , eq . ( [ eq:2.32 ] ) , we can generate the contact term for @xmath88 as a result of the gauging of the @xmath6 form factor to be @xmath89 where @xmath90 \nonumber \\ & & -e_{n}\tau_{i}\frac{(2p'-k)_{\mu}}{p'^{2}-(p'-k)^{2 } } \left[\lambda^{\dagger}_5(q , p',p)-\lambda^{\dagger}_5(q , p'-k , p)\right ] \nonumber \\ & & -\tau_{i}e_{n}\frac{(2p+k)_{\mu}}{p^{2}-(p+k)^{2 } } \left[\lambda^{\dagger}_5(q , p',p)-\lambda^{\dagger}_5(q , p',p+k)\right ] \nonumber \\ & & -i\tau_{i}e_{n}\left\{g_{2}(q^{2},p'^{2},(p+k)^{2 } ) + g_{4}(q^{2},p'^{2},(p+k)^{2})(\not{\ ! p}'-m)\right\ } \nonumber \\ & & \hskip 4 cm \times\ \gamma_{5}\left\{\gamma_{\mu } + \frac{(2p+k)_{\mu}}{p^{2}-(p+k)^{2}}\not{\ ! k}\right\ } \nonumber \\ & & -ie_{n}\tau_{i}\left\{\gamma_{\mu } -\frac{(2p'-k)_{\mu}}{p'^{2}-(p'-k)^{2}}\not{\ ! k}\right\ } \gamma_{5}\{g_{3}(q^{2},(p'-k)^{2},p^{2 } ) \nonumber \\ & & \hskip 3 cm + g_{4}(q^{2},(p'-k)^{2},p^{2})(\not{\ ! p}-m)\ } \ . \label{eq:2.35}\end{aligned}\ ] ] in writing this expression for the contact term , we have restricted ourselves to terms linear in the e.m . coupling , and have made use of the relations in eq . ( [ eq:2.30 ] ) to simplify the isospin . the resultant ward - takahashi identity for the contact term is then given by @xmath91 which is in agreement with the results of kazes @xcite and naus and koch @xcite . in the present investigation these form factors will play the role of introducing cut - offs that maintain gauge invariance and will allow us to incorporate the pionic dressing of the nucleon without having to resort to other renormalization procedures . from a practical point of view , the above procedure will allow us to introduce qcd based parameters that provide consistency between the different form factors . although the above lagrangian is not the most general we can envisage , the procedures followed and the general conclusions can be extended to more general forms for the meson - baryon lagrangian . in particular we could have included a @xmath0 interaction that under gauging would give us a term for the process @xmath92 . the detail form of such an interaction will depend on the model considered , e.g. in some chiral lagrangians such an interaction would arise from @xmath93 and @xmath94 exchange @xcite . having established the form of the lagrangian , and therefore the corresponding hamiltonian , we now proceed to include the pionic contribution to the currents and the amplitude for pion photoproduction . since the pionic dressing contributes to both the thresholds for @xmath0 scattering and modifies the e.m . currents , we need to include pionic contributions while preserving two - body unitarity and gauge invariance . this can be achieved by deriving coupled integral equations for pion - nucleon elastic scattering and pion photoproduction that include the @xmath0 , @xmath95 as well as the @xmath1 thresholds . although this latter threshold is not required for two - body unitarity to be satisfied , the fact that it is at the same energy as the @xmath0 threshold suggests that we need to include it for consistency . however , it turns out that the inclusion of the @xmath1 threshold is essential if we are to preserve gauge invariance at the operator level for the pion photoproduction amplitude . the method employed for the derivation of these coupled integral equations is based on the classification of the feynman diagrams that contribute to a given amplitude , in perturbation theory , according to their irreducibility . in other words , we take all feynman diagrams that contribute to a given amplitude and classify these diagrams into classes according to their irreducibility . then with the help of taylors @xcite last - cut lemma , we can write the diagrams that belong to a given class , and of a given irreducibility , in terms of amplitudes for the same process or a related process . this procedure generates coupled integral equations for all related reactions . this method has been used to derive unitary equations for the @xmath96 system @xcite , the @xmath97 system @xcite and the @xmath98 system @xcite . although taylor s @xcite original classification scheme suffers from double counting problems when applied to covariant perturbation theory , these problems have been recently overcome @xcite at the cost of imposing irreducibility constraints on sub - amplitudes in channels other than the @xmath99-channel , and the requirement that certain subtractions be included in the final equations . since this double counting does not arise for two - body unitarity in @xmath0 scattering and pion photoproduction @xcite , we have chosen to use the simplified version of the classification of diagrams that is often used in time ordered perturbation theory , and has been applied to this reaction previously @xcite . since the last - cut lemma plays a central role in our derivation of integral equations for pion photoproduction , we will very briefly state the basic definition of a cut and the last - cut lemma as applied in time ordered perturbation theory . we define a @xmath53-cut as an arc that separates the initial state from the final state in a given diagram , and cuts @xmath53-particle lines with at least one line being an internal line . an amplitude is @xmath13-particle irreducible if all diagrams that contribute to this amplitude will not admit any @xmath53-cut with @xmath100 . with these two definitions , we can introduce the last - cut lemma which states that for a given amplitude that is @xmath101-particle irreducible , there is a unique way of obtaining an @xmath13-particle cut closest to the final ( initial ) state for all diagrams that contribute to the amplitude . by virtue of this lemma , we can expose one- , two- and three - particle intermediate states and the corresponding unitarity cuts and in this way derive equations for the amplitude that satisfy unitarity . from the above statement of the lemma , it is clear that we need to expose the @xmath102-particle unitarity cut before the @xmath103-particle unitarity cut . with this lemma we can now expose the @xmath0 and @xmath1 unitary cuts and derive a set of integral equations governing pion - nucleon scattering and pion photoproduction . however , before we set to derive integral equations for @xmath0 elastic scattering and pion photoproduction , it would be helpful to our discussion in this section if we symbolically write our lagrangian , derived in sec . [ sec.2 ] , as @xmath104 the additional terms in this lagrangian , over those in eq . ( [ eq:2.1 ] ) , are the result of gauging the lagrangian in eq . ( [ eq:2.1 ] ) . in addition we have included terms that give rise to @xmath105 , @xmath106 and @xmath107 , @xmath108 . these additional terms are included in the event that we need to include @xmath94 and @xmath93 exchange which are often needed in chiral lagrangians that describe @xmath0 scattering @xcite . here we will not examine the detailed form of such a @xmath0 interaction and the associated gauging as the result would then be model dependent . however , we will assume that the procedure detailed in the last section could be applied to such an interaction maintaining gauge invariance . all terms in this lagrangian have form factors that are consistent with ward - takahashi identities as demonstrated in sec . [ sec.2 ] . restricting the analysis of the pion nucleon and pion photoproduction amplitudes to lowest order in the electromagnetic coupling , @xmath46 , leads to the result that the two amplitudes can be considered separately as there are no radiative corrections to the @xmath109-@xmath110 amplitude . the anaylsis of the @xmath109-@xmath110 amplitude under these conditions has been carried out previously @xcite and a summary of the two - body results is presented here . at the two - body level the @xmath109-@xmath110 amplitude is given by @xcite @xmath111 where @xmath112 is the @xmath113 amplitude , @xmath114 is the @xmath115 amplitude and @xmath116 is the dressed nucleon propagator . the superscript gives the irreducibility of the amplitudes . the one particle irreducible @xmath117 amplitude , @xmath118 , satisfies the two - body equation @xmath119 with @xmath120 being the @xmath0 propagator . since we are restricting our analysis to the inclusion of the @xmath0 and @xmath1 thresholds only , we can choose our @xmath0 propagator so that the nucleon propagator does not include pionic dressing , i.e. @xmath121 where @xmath122 is the `` bare '' nucleon propagator defined in sec . [ sec.2 ] , eq . ( [ eq:2.3a ] ) . however the substitution @xmath123 , if carried through in the @xmath0 propagator @xmath124 , will require that @xmath122 have a simple pole at the physical nucleon mass . the input to eq . ( [ eq:3.3 ] ) is the two particle irreducible @xmath0 amplitude , @xmath125 . application of the last - cut lemma to this amplitude will expose states with two or more pions and a nucleon , i.e. the @xmath126 and higher thresholds . if we are not to include these thresholds into our final equation , then we can assume that @xmath125 is some @xmath0 potential whose parameters can be adjusted so that the full @xmath0 amplitude @xmath127 reproduces the experimental phase shifts . on the other hand , if we are to use the lagrangian defined in eq . ( [ eq:3.1 ] ) for our input , then we need to expose the @xmath126 intermediate states by the application of the last - cut lemma to @xmath125 . this gives us @xmath128 where the second term is the cross diagram . here again we can make the substitution @xmath123 , with the condition that @xmath122 has a pole at the physical nucleon mass , if we want to include two - body unitarity only . since the lagrangian in eq . ( [ eq:3.1 ] ) does not include any term for two pion production or absorption , @xmath129 , the three particle irreducible amplitude @xmath130 is just the @xmath0 potential resulting from @xmath131 and could include @xmath94 and @xmath93 exchange . if we were to examine the full @xmath126 part of the hilbert space , then the @xmath132 amplitude in eq . ( [ eq:3.4 ] ) gets dressed @xcite to the extent that @xmath133 , and the @xmath132 amplitude in the crossed digram becomes the same as that in eq . ( [ eq:3.2 ] ) for the pole diagram . we now turn to the one particle irreducible @xmath134 amplitude , @xmath114 . using the last - cut lemma , we can write . @xmath135 here again since the lagrangian under consideration has no component that gives rise to two pion production or absorption , i.e. @xmath136 , then the two particle irreducible @xmath137 amplitude @xmath138 has no intermediate states and is just the @xmath6 vertex in the lagrangian , i.e. @xmath139 . the dressed nucleon propagator in eq . ( [ eq:3.2 ] ) is given by @xmath140 where , with the help of the last - cut lemma , we can write the mass shift due to pionic dressing as @xmath141 since the lagrangian has no term for @xmath142 transition , then @xmath143 , while the `` bare '' nucleon propagator @xmath144 in eq . ( [ eq:3.6 ] ) is given by @xmath145 where @xmath146 is the bare nucleon mass , and includes the quark - gluon contribution to the mass . from this point on we assume this ` bare ' mass has no momentum dependence as the quark - gluon contribution does not introduce any unitarity thresholds and the mass shift due to quark - gluon structure has no threshold . in this case the dressed propagator @xmath116 should have a pole at the physical nucleon mass . an integral equation for the @xmath109-@xmath110 amplitude , @xmath127 , can now be derived @xcite . to achieve this , we first substitute equations ( [ eq:3.3 ] ) and ( [ eq:3.5 ] ) into ( [ eq:3.2 ] ) resulting in @xmath147 we now make use of the fact that @xmath148 and noting the definition of @xmath127 , obtain @xmath149 here the born amplitude for @xmath0 scattering @xmath150 is given by @xmath151 where for the present lagrangian @xmath152 where @xmath35 is given in eq . ( [ eq:2.29 ] ) in terms of form factor that could in principle be extracted from a qcd model and @xmath130 could be any @xmath0 potential we may need to introduce . this born amplitude , or @xmath153 potential is illustrated in fig . [ fig.5 ] . we now turn to the amplitude for single pion photoproduction , @xmath154 . this has been considered in detail previously by araki and afnan ( aa ) @xcite . for the sake of completeness we present here a summary of their results . applying the last - cut lemma to the pion photoproduction amplitude , @xmath155 , we first expose the @xmath0 unitary cut . here we will restrict the analysis to first order in the electromagnetic coupling @xmath46 . we now can divide the diagrams that contribute to this amplitude into two classes : the diagrams that are one - particle irreducible we sum to get the one - particle irreducible amplitude @xmath156 . the rest of the diagrams are one particle reducible , and when summed give the nucleon pole contribution to the full amplitude . the resultant decomposition for the full amplitude is then given by @xmath157 where @xmath158 is the one - particle irreducible @xmath159 amplitude . following the procedure presented above for the @xmath0 amplitude , we employ the last - cut lemma to write an integral equation for the one - particle irreducible photoproduction amplitude , @xmath156 . this is given by @xmath160 where we have retained only those terms linear in @xmath46 . finally , applying the last - cut lemma to expose the @xmath0 unitarity cut in the @xmath159 amplitude , @xmath158 , we get @xmath161 in this way we have exposed all two body unitarity cuts in the photoproduction amplitude . we now write this photoproduction amplitude , @xmath155 , as the solution to a two - body integral equation , and in this way define the born amplitude for the reaction @xmath162 . this involves similar steps to those used to derive the equivalent equation for @xmath0 elastic scattering , i.e.eqs ( [ eq:3.9 ] ) to ( [ eq:3.12 ] ) . this results in the integral equation for @xmath155 being @xmath163 with the born amplitude for pion photoproduction given by @xmath164 where we have made use of the fact that @xmath165 at this stage it is premature to link the photoproduction born amplitude @xmath166 to the underlying lagrangian since the last - cut lemma can be further applied to both @xmath167 and @xmath168 to reveal the contribution of the @xmath1 cut . as we will observe , when we expose the @xmath1 unitarity cut , @xmath166 is not the usual born amplitude for pion photoproduction . it will include additional terms that give rise to the dressing of some of the vertices . to examine the diagrams that contribute to @xmath166 and in this way determine their physical importance , the last - cut lemma is applied to both @xmath167 and @xmath168 to reveal the @xmath169 unitarity cut . the @xmath169 branch point occurs at the same energy as the @xmath0 branch point but at a lower energy than the start of the @xmath126 cut as illustrated in fig . [ fig.6 ] . as a result , in this approach , the @xmath1 cut will be treated on an equal footing with the @xmath0 cut but differently from the @xmath170 cut , which has been truncated out of our analysis since we are considering only two - body unitarity as far as the pion dynamics is concerned . in the approach of aa @xcite , the @xmath126 and @xmath1 cuts were exposed simultaneously since their truncation was carried out with respect to the number of particles present in intermediate states in contrast with the present approach of truncation on the basis of the position of the threshold in the energy plane . to include the @xmath1 threshold in our pion photoproduction amplitude , we first consider the diagrams that contribute to the two - particle irreducible @xmath159 amplitude , @xmath167 . these diagrams can be divided into two classes : those that do nt have @xmath1 intermediate states , the sum of which we denote by @xmath171 . the second class of diagrams are those that have @xmath1 intermediate states . to these we apply the last - cut lemma and thus expose the corresponding threshold . we denote the sum of the diagrams in this class by @xmath172 , where the subscript @xmath173 denotes that we include only the connected diagrams in the sum . here , the sub - amplitudes @xmath174 for @xmath175 and @xmath176 for @xmath177 do not have to be the sum of connected diagrams provided diagrams that contribute to @xmath159 amplitude @xmath167 are connected . finally , @xmath178 is the the product of the propagators for the nucleon , pion and photon . since we have a pion in @xmath178 , the nucleon propagator in @xmath178 need not include any pionic dressing , but the mass in this propagator needs to be the physical nucleon mass . thus exposing the @xmath1 threshold in the @xmath159 amplitude allows us to write this amplitude as @xmath179 in the absence of a term in the lagrangian for the @xmath142 transition , the amplitude @xmath171 is nothing more than the nucleon e.m . current as given in eq . ( [ eq:2.17 ] ) , i.e. @xmath180 . in a similar manner , the absence of a two pion production or absorption term in the lagrangian allows us to take the @xmath175 amplitude , @xmath174 to be basically that resulting from the gauging of the @xmath7 vertex , and given in eq . ( [ eq:2.35 ] ) , i.e. @xmath181 . finally , taking into consideration the fact that we are restricting all our results to the inclusion of amplitudes in lowest order in the e.m . coupling , the amplitude @xmath176 for the process @xmath177 , has to be disconnected and of the form @xmath182 where @xmath183 is the photon propagator . this allows us to write the two - particle irreducible @xmath159 amplitude as @xmath184 where for the present lagrangian we have @xmath185 and @xmath186 with @xmath187 and @xmath188 resulting from the gauging @xmath189 and @xmath190 in our basic meson - baryon lagrangian respectively . we now turn to the two - particle irreducible @xmath191 amplitude , @xmath168 . here again we divide all diagrams that contribute to this amplitude into two classes : those with @xmath1 intermediate states . these diagrams are summed , with the help of the last - cut lemma , to produce the amplitude in which the @xmath1 cut is exposed . the sum of all the diagrams in this class is given by @xmath192 , where @xmath193 is the two - particle irreducible @xmath194 amplitude . here again the diagrams that contribute to the sub - amplitude can be disconnected provided the diagrams that are summed to give the full amplitude are the sum of connected diagrams . the second class of diagrams are those with no @xmath1 intermediate states . the sum of these diagrams we denote as @xmath195 . the subscript @xmath196 indicates that three - particle intermediate states have only pions , i.e. @xmath126 intermediate states . thus the two - particle irreducible @xmath197 amplitude can be written as @xmath198 at this stage we could parameterize @xmath195 as an energy independent amplitude without affecting the two - body unitarity or the inclusion of the @xmath1 threshold of our final amplitude . however , if we want to relate our input to the underlying lagrangian given in eq . ( [ eq:3.1 ] ) , we need to examine the structure of @xmath195 . in particular , we need to divide the diagrams that contribute to this amplitude into two classes . those with no @xmath126 intermediate states , and therefore three - particle irreducible . these we denote by @xmath199 , and for the lagrangian under consideration reduce to the contribution of just the basic term in the lagrangian , @xmath200 . the second class of diagrams has @xmath126 intermediate states , and need to be included if three - particle unitarity is to be included in the final photoproduction amplitude . however , since we are neglecting three - body unitarity we have a choice of either neglecting the contribution of these diagrams to the @xmath197 amplitude @xmath195 , or examining the structure of these diagrams with the help of the last - cut lemma in conjunction with the modified version of taylor s classification of diagrams @xcite . here we should point out that the contribution to @xmath195 from diagrams which admit a @xmath201-cut that cut initial , final and internal lines will be included in @xmath202 . that leaves diagrams in which the @xmath201-cut can intersect initial and internal lines only , or final and internal lines only . because we are including the e.m.interaction to first order , and have excluded any direct coupling between a @xmath203 intermediate state and the @xmath95 initial state , there are no diagrams in this class . this leaves the diagrams that admit @xmath204-cuts that intersect initial and internal lines only . these will involve the connected @xmath205 diagrams which have the contribution of three - body unitarity . for the present investigation we have chosen to neglect this contribution to three - particle unitarity and have taken @xmath206 , and therefore @xmath207 we now turn our attention to the contribution from the @xmath1 threshold to @xmath168 , i.e. @xmath208 . since the e.m . coupling is included to first order in the present analysis , we have that the @xmath209 amplitude @xmath176 has to be disconnected and of the form @xmath210 on the other hand the three - particle irreducible @xmath211 amplitude @xmath212 can be written in terms of a connected and disconnected part as @xmath213 since the photon is absorbed in this process , the disconnected amplitude @xmath214 has the form @xmath215 where @xmath216 is the two - particle irreducible @xmath217 amplitude . since we already have a spectator pion , and to avoid the problem of non - linearity of our final integral equations , we could take @xmath218 where @xmath187 is given in eq . ( [ eq:2.17 ] ) . since there is no direct coupling in the lagrangian between the final @xmath0 state and any state with four particles ( e.g. @xmath203 or @xmath219 ) , we take @xmath220 to be the @xmath211 interaction in the lagrangian , i.e.@xmath221 ) , @xmath220 can also have the last - cut lemma further applied before contact with the underlying lagrangian is made . at this stage this will be ignored since we are only interested in the _ minimal _ requirements of two - body unitarity and so @xmath125 will be considered as a background @xmath0 potential . we now make use of eqs . ( [ eq:3.27 ] ) and ( [ eq:3.29 ] ) in eq . ( [ eq:2.26 ] ) to write the two - particle irreducible amplitude for @xmath222 as @xmath223\ , g\,\lambda^{(1)\dagger}_5 \ . \label{eq:3.32}\end{aligned}\ ] ] this basically consists of the ` seagull ' term ( @xmath199 ) , the nucleon emitting a pion with the photon being absorbed on the @xmath0 interaction . the details of this @xmath211 amplitude will be determined by the gauging of the @xmath0 interaction . the two final terms on the right hand side of eq . ( [ eq:3.32 ] ) correspond to photon absorption on the pion and the crossed diagrams for pion photoproduction . we now can write the born amplitude for pion photoproduction @xmath166 given in eq . ( [ eq:3.18 ] ) as @xmath224\,g\ , \lambda^{(1)\dagger}_5 + \lambda_5^{(2)\dagger}\,s_0\,\gamma^{(2)}\ .\label{eq:3.33}\ ] ] for the present , the ` seagull ' term is the contact term in the lagrangian , i.e.@xmath225 while the two - particle irreducible pion current @xmath226 is given in eq . ( [ eq:2.25 ] ) . for the minimal requirement of two - body unitarity and the inclusion of the @xmath0 and @xmath1 thresholds , we could replace the reducibility of the @xmath227 amplitude @xmath167 and the @xmath228 amplitude @xmath229 by an amplitude that is of higher reducibility , and which is directly related to the basic terms in our lagrangian , depending on the diagrams these amplitudes contribute to . we will come back to this point when we consider the gauge invariance of our overall amplitude . having derived a set of coupled integral equations for the pion photoproduction amplitude , we proceed in this section to derive an expression for this amplitude with the help of the ward - takahashi identities @xcite as employed by kazes @xcite . we then can compare the resultant amplitude , which is gauge invariant , to the unitary amplitude derived in the last section . the starting point for deriving a gauge invariant pion photoproduction amplitude is the @xmath6 three - point function or green s function for @xmath230 , given by @xmath231| 0 \rangle \nonumber \\ & = & s(p')\,\delta(q)\,\lambda^{\dagger}_5(q , p',p)\ , \vec{\tau}\cdot\vec{\phi}\,s(p ) \ , \label{eq:4.1}\end{aligned}\ ] ] where @xmath232 is the pion momentum and @xmath233 ( @xmath234 ) is the momentum of the final ( initial ) nucleon . here @xmath116 and @xmath235 are the nucleon and pion propagators respectively . in eq . ( [ eq:4.1 ] ) we have taken @xmath236 to be the pauli isospin matrix and @xmath237 to be the pion field . to gauge this amplitude and thus generate a gauge invariant pion photoproduction amplitude , we make use of the procedure of coupling the photon to all propagators and vertices . this is equivalent to the substitutions @xcite [ eq:4.2 ] @xmath238 in the green s function @xmath239 . to maintain unitarity we need to include the pionic dressing of both the propagators and vertices . in particular , we will find that the minimal requirement of two - body unitarity will impose a constraint on the level of dressing in both the propagator @xmath240 and the vertex @xmath241 . with the above gauging procedure , our @xmath6 green s function can be gauged to give @xmath242 the resultant pion photoproduction amplitude @xmath243 is now given as the sum of four diagrams , i.e.@xmath244 here @xmath232 is the pion momentum , @xmath53 the photon momentum while @xmath233 and @xmath234 are the nucleon momenta in the final and initial states respectively . the superscripts @xmath245 and @xmath246 in @xmath247 and @xmath248 indicate the isospin index , which we have now explicitly included . at this stage the pionic dressing in the @xmath230 amplitude @xmath249 , the nucleon current @xmath250 , the pion current @xmath251 and ` seagull ' term @xmath252 are included to all orders . these four diagrams are illustrated in fig . [ fig.7 ] and at this stage give the most general form for the @xmath197 amplitude if we ignore the irreducibility of the sub - amplitudes in the figure @xcite . using the ward - takahashi identities @xcite for the nucleon propagator , the pion propagator and the @xmath6 vertex , we can establish the gauge invariance of the amplitude @xmath247 . in particular we have that @xmath253 reduces to @xmath254 which is in agreement with the results of kazes @xcite . taking matrix elements of equation ( [ eq:4.5 ] ) between on - mass shell initial and final states gives @xmath255 since the inverse propagators are zero on - mass shell . in this way we have established the fact that @xmath247 is indeed gauge invariant and the requirement of current conservation is satisfied . at this stage the @xmath6 green s function we have considered has the fully dressed propagators and vertex , and to get these quantities will require a full solution to the underlying field theory . however , to satisfy the minimum requirement of two - body unitarity , we need not include the pionic dressing of the final nucleon propagator , while the initial nucleon propagator and the @xmath137 amplitude need only include the minimal pionic dressing to include the @xmath0 threshold . in this case the green s function we need to consider is given by @xmath256 then this green s function can be gauged to give a corresponding amplitude for pion photoproduction that satisfies the _ minimum _ requirement of two - body unitarity . here the final nucleon propagator is gauged by the transformation given in eq . ( [ eq:2.19 ] ) , while the gauging of the @xmath230 amplitude @xmath257 and the dressed nucleon propagator @xmath258 will be derived using the results of the previous section . the result of gauging the green s function in eq . ( [ eq:4.7 ] ) is @xmath259 where the @xmath197 amplitude @xmath260 is given by @xmath261 this result differs from that of eq . ( [ eq:4.4 ] ) to the extent that the nucleon in the final state did not require pionic dressing for two - body unitarity to be satisfied , and as a result the corresponding current , @xmath262 does not include any pionic corrections . on the other hand the current resulting from gauging the initial nucleon has the necessary pionic corrections to satisfy two - body unitarity . the proof that this amplitude satisfies gauge invariance follows the same procedure considered above , with the difference being the difference in the ward - takahashi identities for the bare and dressed nucleon propagators @xmath122 and @xmath116 . in this case the gauge invariance of the amplitude @xmath263 is given by @xmath264 which on - mass shell gives a conserved current , provided both the bare and dressed nucleon propagators have poles at the physical nucleon mass . thus we have constructed a gauge invariant amplitude for pion photoproduction starting with the green s function for the process @xmath137 in which the final nucleon propagator has less pionic dressing than the propagator in the initial state . we proceed in the next section to demonstrate that this gauge invariant amplitude is identical to that resulting from the solution of the coupled equations , and that it satisfies two - body unitarity . in the previous section we demonstrated how one may construct a gauge invariant amplitude for pion photoproduction given the @xmath137 green s function and the transformation of the pion propagator , nucleon propagator and the @xmath137 amplitude under the gauge symmetry . in sec . [ sec.2 ] these gauge transformations were derived for the different terms in the basic meson - baryon lagrangian . here we are going to use the gauging of these basic terms in the lagrangian to derive the results of gauging the pionic dressed nucleon propagator @xmath240 , the pion propagator @xmath75 and the one - particle irreducible @xmath137 amplitude @xmath265 . from this point on we will drop the superscript @xmath266 on the dressed nucleon propagator assuming that @xmath240 includes the necessary pionic dressing to include two - body unitarity . these results when used in the definition of the three - point function will give us the amplitude for pion photoproduction with the application of lsz reduction to the resultant green s function . the amplitude resulting from this gauging procedure is identical to that resulting from the solution of the coupled set of integral equations derived in sec . [ sec.3 ] . in this way we establish the gauge invariance of the pion photoproduction amplitude resulting from the solution of the coupled integral equations , which already satisfy unitarity . since the nucleon propagator @xmath267 need not include pionic dressing , but might include the quark - gluon structure , we may use the result of sec . [ sec.2 ] to write @xmath268 if we assume that our ` bare ' nucleon has the quark - gluon structure , and this structure does not contribute to unitarity , then the nucleon current is @xmath269 with @xmath250 given in eq . ( [ eq:2.17 ] ) . note that in the absence of the quark - gluon structure , this current will reduce to the standard dirac current for a point fermion . in a similar manner , the gauge prescription for the basic input into our coupled integral equations , which are also the terms in the lagrangian , are given as @xmath270 here , in addition to the gauging of the @xmath6 vertex , we have included the gauging of the two - particle irreducible @xmath271 amplitude . for the pion propagator we take @xmath272 in our meson - baryon lagrangian the only dressing the pion can have is via nucleon antinucleon loops @xcite . if we ignore these loop corrections to the pion dressing , the only dressing we could have are the result of the underlying quark - gluon structure . the pion current in this case can be written in terms of the gauge invariant current given in eq . ( [ eq:2.25 ] ) , i.e.@xmath273 . to gauge the one - particle irreducible @xmath0 amplitude @xmath118 , we need to first gauge the @xmath0 propagator @xmath124 . since only two - body unitarity need be considered , the nucleon propagator in @xmath124 can be taken to be the bare propagator @xmath122 , and the @xmath0 propagator reduces to @xmath274 . to first order in the e.m . coupling the gauge transformation of the @xmath0 propagator reduces to @xmath275 with the help of the two - body equation for @xmath118 , and the transformation properties of the @xmath0 amplitude @xmath125 and @xmath124 , the gauging of the one particle irreducible @xmath0 amplitude @xmath118 reduces to @xmath276 where to first order in the e.m . coupling @xmath277 with these results in hand , we can now determine how @xmath265 behaves under gauging . given that the one - particle irreducible amplitude for @xmath137 is given , using eq . ( [ eq:3.5 ] ) , by the relation @xmath278 we can make use of eqs . ( [ eq:5.2 ] ) , ( [ eq:5.6 ] ) and ( [ eq:5.7 ] ) to write the gauge transformation for @xmath257 as @xmath279 where to first order in e.m . coupling the @xmath197 amplitude @xmath280 is given by @xmath281 all that is left to determine is how the nucleon propagator @xmath116 behaves under gauging when the @xmath0 unitarity cut has been exposed . from eqs . ( [ eq:3.6 ] ) and ( [ eq:3.7 ] ) we have that @xmath282^{-1}\nonumber \\ & = & s_0 + s_0\,\sigma^{(1)}\,s\ , \label{eq:5.12}\end{aligned}\ ] ] where @xmath283 with the help of eqs . ( [ eq:5.1 ] ) , ( [ eq:5.3 ] ) , ( [ eq:5.6 ] ) and ( [ eq:5.10 ] ) , the dressed nucleon propagator can be gauged to give @xmath284 where the dressed @xmath159 amplitude @xmath285 is given , to first order in the e.m . coupling , by @xmath286\nonumber \\ & = & \gamma^{(2 ) } + \lambda^{(1)}\,g\,\left [ \gamma^{ct } + \tilde{f}^{(3)}_3\,g\,\lambda^{(1)\dagger}_5\right ] \ . \label{eq:5.15}\end{aligned}\ ] ] in writing the second line in eq . ( [ eq:5.15 ] ) , we have made use of eqs . ( [ eq:3.22 ] ) and ( [ eq:5.11 ] ) . if we now compare the dressed nucleon current resulting from the gauging of the dressed nucleon propagator , and make use of eqs . ( [ eq:3.16 ] ) and ( [ eq:3.32 ] ) , we find that this current is in fact identical to the one - particle irreducible current derived in sec . [ sec.3 ] , i.e.@xmath287 therefore the nucleon propagator @xmath116 behaves , to first order in the electromagnetic coupling , under gauging as @xmath288 thus exposing the @xmath0 unitarity cut in @xmath116 , and gauging , leads to the same results as exposing the @xmath1 unitarity cut which is what one expects . to complete the proof of the equivalence of the two formulations of secs . [ sec.3 ] and [ sec.4 ] we are required to show that @xmath289 where @xmath290 results from substituting eqs . ( [ eq:5.17 ] ) , ( [ eq:5.10 ] ) , ( [ eq:5.5 ] ) and ( [ eq:5.1 ] ) into eq . ( [ eq:4.7 ] ) with the result that gauging the three - point function for the process @xmath137 is given by @xmath291 and the corresponding pion photoproduction amplitude @xmath290 is of the form @xmath292 in writing eq . ( [ eq:5.19 ] ) , we have restricted ourselves to first order in the electromagnetic coupling . substituting for @xmath280 from eq . ( [ eq:5.11 ] ) and making use of the definition of the two - particle irreducible @xmath197 amplitude @xmath168 given in eq . ( [ eq:3.32 ] ) , we get @xmath293 in writing the second line we have made use of eq . ( [ eq:3.15 ] ) to relate the one - particle and two - particle irreducible @xmath197 amplitudes . making use of the adjoint of eq . ( [ eq:5.3 ] ) we can write @xmath290 as @xmath294 \nonumber \\ & = & m^{(2 ) } + \lambda^{(2)\dagger}_5\,s_0 \left[\gamma^{(2 ) } + \lambda^{(2)}_5\,g\,m^{(0)}\right ] + t^{(2)}\,g\,m^{(0)}\ . \label{eq:5.22}\end{aligned}\ ] ] the second line of eq . ( [ eq:5.22 ] ) is derived by making use of eq . ( [ eq:3.19 ] ) to write @xmath295 in terms of @xmath155 , and eq . ( [ eq:3.14 ] ) to get the last term on the right hand side . we now employ the definition of the born amplitude for pion photoproduction @xmath166 , eq . ( [ eq:3.33 ] ) , and the @xmath0 potential @xmath150 given in eq . ( [ eq:3.12 ] ) to write eq . ( [ eq:5.22 ] ) as @xmath296 if we compare this result with eq . ( [ eq:3.17 ] ) , we observe that the right hand side of both equations are identical . this proves the fact that the amplitude which is a solution of the coupled equations that include the @xmath95 , @xmath0 and @xmath1 thresholds is identical to that derived using the gauging of the @xmath137 green s function with the help of the ward - takahashi identities . thus we have that @xmath297 and in this way we have established the fact that the solution of the coupled integral equation , eq . ( [ eq:3.17 ] ) , gives us an amplitude that satisfies two - body unitarity , and is gauge invariant . to establish the connection with previous results in the literature , we iterate eq . ( [ eq:5.23 ] ) with the result @xmath298 where @xmath299 is the born term illustrated in fig . [ fig.1 ] and has often been used as a gauge invariant term on its own . the factor of @xmath300 gives the distortion in the @xmath0 channel . the second term , illustrated in fig . [ fig.8 ] , is required to maintain gauge invariance at the operator level for @xmath155 , and results from including the @xmath1 channel into our integral equations . this inclusion of the @xmath1 channel effectively allows us to couple the photon not only to the initial nucleon in the @xmath137 vertex , but also to the nucleon in the final state . in this way we have coupled the nucleon to every propagator and vertex in the three - point green s function for @xmath301 to employ pion photoproduction at medium energies to examine nucleon structure and in particular the resonances observed in @xmath0 scattering , we need a theory that incorporates charged current conservation and the conservation of probability , i.e. gauge invariance and unitarity . if in addition we would like to test current models of qcd based on quark - gluon degrees of freedom , we need to include form factors in which we have consistency between the @xmath6 and @xmath43 form factors resulting from qcd . in the present investigation , we considered a lagrangian written in terms of the meson - baryon degrees of freedom . this lagrangian has form factors which in principle could be derived from a qcd model . we then gauged this lagrangian with the help of the procedure proposed by ohta @xcite . this gauging procedure could have been circumvented had we considered a specific qcd model . however , to avoid the problem of working within a specific model , we chose to maintain generality and restrict our gauging to the approach proposed by ohta . within the framework of this lagrangian which includes not only the meson - baryon degrees of freedom but also the coupling of the photon to these degrees of freedom , we derived a set of coupled integral equations that satisfy two - particle unitarity . this was achieved by exposing the @xmath0 , @xmath95 and @xmath302 thresholds following the analysis of taylor @xcite . to prove that the solution of these coupled equations is also gauge invariant , we employed the ward - takahashi identities , derived for the basic terms in the lagrangian , to gauge the one particle irreducible three point function for the process @xmath137 . since we demanded that our amplitude satisfy _ only _ two - body unitarity , we assumed that the final nucleon propagator in the @xmath137 green s function did not have any pionic dressing , and the initial nucleon propagator had the minimal dressing to include the @xmath0 threshold , while the @xmath6 vertex in the three - point function was one - particle irreducible . this gauging procedure gave us an amplitude that is identical to that resulting from a solution of coupled integral equations . we found it essential that we include the @xmath1 threshold into our coupled integral equation . the need for the inclusion of the @xmath1 threshold to satisfy gauge invariance at the operator level , raises some questions regarding the application of watson s theorem @xcite or its off - shell modification @xcite to unitarise the gauge invariant born term represented in fig . [ fig.1 ] , and to maintain gauge invariance in the final result . in fact when we cast our amplitude into a form of watson s theorem , i.e. a distorted wave in the @xmath0 channel , we found that we had an additional term that has not been included in most calculations in the past . these additional terms are required in order to preserve gauge invariance of the photoproduction amplitude at the operator level . we are presently considering a simple model to determine the contribution of these additional terms required for maintaining gauge invariance and unitarity in the final amplitude . the authors would like to thank the australian research grant scheme for their financial support and b. c. pearce for some stimulating discussions . one of the authors ( ira ) would like to thank professor j. speth and the hospitality of ikp at forschungszentrum jlich , where part of this work was done . if we include the @xmath94 and @xmath93 exchange @xmath0 potential into our basic lagrangian , then we have introduced @xmath303 and @xmath304 vertices into the theory and these in turn will contribute to pionic dressing . for the present , since we have not specified the @xmath0 interaction , we will ignore the dressing of the pion propagator .

we examine the accretion properties in a sample of 42 hard ( 3 - 60kev ) x - ray selected nearby broad - line agns . the energy range in the sample is harder than that usually used in the similar previous studies . these agns are mainly complied from the rxte all sky survey ( xss ) , and complemented by the released _ integral agn catalog . the black hole masses , bolometric luminosities of agn , and eddington ratios are derived from their optical spectra in terms of the broad h emission line . the tight correlation between the hard x - ray ( 3 - 20kev ) and bolometric / line luminosity is well identified in our sample . also identified is a strong inverse baldwin relationship of the h emission line . in addition , all these hard x - ray agns are biased toward luminous objects with high eddington ratio ( mostly between 0.01 to 0.1 ) and low column density ( ) , which is most likely due to the selection effect of the surveys . the hard x - ray luminosity is consequently found to be strongly correlated with the black hole mass . we believe the sample completeness will be improved in the next few years by the ongoing _ swift and _ integral missions , and by the next advanced missions , such as nustar , simbol - x , and next . finally , the correlation between rfe ( = optical /h ) and disk temperature as assessed by leads us to suggest that the strength of the emission is mainly determined by the shape of the ionizing spectrum . _ _ _

0811.3255

it is now generally believed that the power of active galactic nuclei ( agns ) is extracted through the accretion of gas onto a central supermassive black hole ( smbh ) . such energy mechanism means that agns are characterized by their strong hard x - ray emission ( @xmath3kev ) , which is widely used as direct evidence suggesting the existence of a nuclear accretion activity ( @xmath4 , e.g. , silverman et al . 2005 ; brusa et al . 2007 ; hasinger et al . the commonly accepted model is that the hard x - ray emission from agns is primarily produced by the inverse compton scattering of the uv / soft x - ray photons emitted from the accretion disk ( e.g , zdziarski et al . 1995 , 2000 ; haardt & maraschi 1991 ; kawaguchi et al . the absorption - corrected x - ray photon spectra within the energy band 2 - 10kev could be best described as a cut - off powerlaw with index @xmath5 ( e.g. , zdziarski et al . 1995 ; reeves & turner 2000 ; piconcelli et al . 2005 ; dadina 2008 ; panessa et al . the synthesis spectra become flat beyond 10kev because of the compton reflection caused by the ionized surface of the accretion disk ( e.g. , george & fabian 1991 ) . because the hard x - ray emission from central engine can penetrate the obscuration material much more easily than lower energy emission , it possesses particular importance in testing the traditional unified model ( antonucci 1993 ) in seyfert 2 galaxies ( e.g. , moran et al . 2002 ; cardamone et al . the studies of the _ chandra and _ xmm - newton observatories showed that the cosmic x - ray background ( cxrb ) at 2 - 30kev might be contributed by many unknown obscured agns which are predicted by the cxrb models ( e.g. , comastri et al . 1995 ; de luca & molendi 2004 ; worsley et al . 2005 ; gilli et al . 2007 ; severgnini et al . 2003 ; levenson et al . taking into account of the issue of co - evolution of agn and bulge of its host galaxy ( e.g. , heckman et al . 2004 ; kauffmann et al . 2003 ; wang et al . 2006 ; wang & wei 2008 and references therein ) , the hard x - ray emission from agns is also an important tool in detecting and separating agn s contribution from circumnuclear star formation activity . the x - ray luminosities of known most x - ray luminous starforming and elliptical galaxies are not higher than @xmath6 ( zezas et al . 2003 ; lira et al . 2002a , 2002b ; osullivan et al . _ heckman et al . ( 2005 ) identified a very tight correlation between the hard x - ray ( 3 - 20kev ) and [ ] luminosities in a sample of hard x - ray selected agns when they performed a comparison between the hard x - ray selected and [ ] emission - line selected agns . in 2 - 10kev bandpass , similar correlations were identified in the palomar optically selected agns by panessa et al . ( 2006 ) . on the contrary , a very weak @xmath7}}$]-@xmath8 correlation was identified in the agns selected by their bright [ ] emission lines ( heckman et al . 2005 ) . the result suggests that many agns might be missed in the hard x - ray survey . moreover , netzer et al . ( 2006 ) claimed that the @xmath7}}$]/@xmath8 ratio depends on the x - ray luminosity . the questions are therefore naturally raised : why is the @xmath7}}$]-@xmath8 correlation broken in some kind of agns ? which parameters ( or what are the physical reasons that ) determine the correlation ? are the hard x - ray selected agns particular in some parameters ? both black hole mass ( @xmath9 ) and eddington ratio ( @xmath10 ) are two key parameters determining the observed properties of agns . in addition , with the development of the technology in hard x - ray detection , a major advance in studying agn hard x - ray emission will be achieved in the next a few years due to the launch of new missions with enhanced hard x - ray capability ( in sensitivity and imaging ) , such as nustar , simbol - x , and next ( e.g. , takahashi et al . 2008 ; ferrando et al . the study on the existent surveys certainly prepares the ground for the future surveys . in this paper , we examine the optical spectral properties of a sample of 42 hard x - ray selected broad - line agns , which allows us to investigate the properties of @xmath9 and @xmath10 in these hard x - ray agns . the sample is mainly compiled from the agn catalog of the rxte 3 - 20kev all sky survey ( xss , sazonov & revnivtsev 2004 ) , and complemented by the agn catalog released by the _ integral all - sky hard x - ray survey ( bassani et al . 2006a and references therein ) . note that the energy range used in this work is extended to the compton reflection region , and is harder than the range ( i.e. , 0.5 - 10 kev ) usually used in the similar previous studies . the paper is organized as follows . section 2 and 3 shows the sample selection and spectroscopic observations , respectively . the data reduction is described in the next section . section 5 presents the results and discussions . the cosmology with @xmath11 , @xmath12 , and @xmath13 ( bennett et al . 2003 ) is adopted in our calculations throughout the paper . _ our sample is mainly compiled from the rxte 3 - 20kev all - sky survey ( xss , revnivtsev et al . sazonov & revnivtsev ( 2004 , hereafter sr04 ) identified 95 nearby agns with @xmath14 detected in xss . the survey covers about 90% of the sky at @xmath15 . the sensitivity of the survey in the energy band 3 - 20 kev is better than @xmath16 . sr04 therefore provides a nearly complete sample for bright hard x - ray selected agns , although the sample is still biased against sources with absorption column @xmath17 . the ibis telescope ( ubertini et al . 2003 ) onboard the _ integral observatory ( winkler et al . 2003 ) provides a better capability for heavily obscured objects because of its good sensitivity beyond 20 kev . after about three years _ integral observation , their all - sky survey project allows bassani et al . ( 2006a ) to identify 62 agns in the energy bandpass 20 - 100kev above a flux of @xmath18 by analyzing 11,300 _ integral points ( see also in bassani et al . 2006b ; bird et al . 2006 ; beckmann et al . _ _ _ both catalogs are then combined to enlarge our sample content , and to attempt to alleviate the bias against the obscuration . we further limit the objects to the broad - line agns according to the identification provided by previous optical spectroscopy . the bat instrument ( barthelmy et al . 2005 ) onboard the _ swift satellite ( gehrels et al . 2004 ) provides an all - sky hard x - ray survey with a sensitivity down to a few @xmath19 in higher energy band 14 - 195 kev . markwardt et al . ( 2005 ) identified 44 previous known agns in the _ swift survey . however , considering the spectroscopic observation condition ( see below ) , only three objects are not listed in our sample . it is worthy noting that the ongoing _ swift / bat and _ integral survey could provide a larger sample of @xmath20 hard x - ray selected agns at the end of our program ( i.e. , krivonos et al . 2006 ; sazonov et al . 2007 ; tueller et al . 2008 ) . we will perform a subsequent spectroscopic study on these agns in the next work . _ _ _ _ figure 1 shows the column density ( @xmath21 ) distribution for our sample . @xmath21 is complied from sr04 , and from sazonov et al . ( 2007 ) for the objects detected by _ integral only . briefly , these authors determined @xmath21 by fitting the available spectra from different x - ray instruments by an absorbed powerlaw . values of @xmath21 less than @xmath22 were ignored in the fitting of sr04 and sazonov et al . ( 2007 ) . as shown in the figure , the current sample is still strongly biased against agns with large column density . _ both because of the constraint of the observatory site and because of the instrumental capability , the spectroscopic observations are only carried out for the bright ( @xmath23mag ) objects located in the northern sky with declination @xmath24 . in total , there are 53 objects fulfilling the selection criterion . six of these objects are not observed because of the poor weather conditions . among the remaining objects , five objects ( i.e. , xssj02151 - 0033 , xssj11570 + 5514 , xssj22363 - 1230 , ngc788 , igrj21247 + 5058 ) are excluded in subsequent spectral modeling because of the weakness or absence of broad h@xmath25 and/or h@xmath0 emission lines emission . in addition , the continuum is dominated by the contribution from starlight of host galaxy rather than agn in xssj02151 - 0033 and xssj11570 + 5514 , which precludes the measurements of the complex . we include ngc788 in our observation program because multi - classification is shown in the ned ( see also in beckmann et al . our spectrum indicates a seyfert 2 nucleus in the object . in fact , the object is classified as a seyfert 2 galaxy with polarized broad emission lines in the catalogue of quasars and active nuclei ( veron - cetty & veron 2006 ) . our spectrum taken in dec 20 , 2006 shows that the spectrum of igrj21247 + 5058 is dominated by a typical late type star , which is consistent with masetti et al . ( 2004 ) . by identifying the broad emission bump around 6700 as h@xmath25 , these author argued that the background agn is aligned with a f- or early g - type star by chance . ] the total 42 high ( or intermediate high ) quality optical spectra were obtained by using the national astronomical observatories , chinese academy of science ( naoc ) , 2.16 m telescope in xinglong observatory during several observing runs carried out from november 2005 to mar 2008 . the spectra were taken by the omr spectrograph equipped with a back - illuminated spec10 1340@xmath26400 ccd as detector . a grating of 300@xmath27 and a slit of 2.0 oriented in south - north direction were used in our observations . this setup results in a final spectral resolution @xmath28 as measured from both comparison spectra and night sky emission lines . the blazed wavelength was fixed to be 6000 , which provides a wavelength coverage of 3800 - 8300 in observer frame . this attempt covers both h@xmath25 and h@xmath0 region in most spectra because of their small redshifts . each object was observed successively twice . the two exposures were combined prior to extraction to enhance the s / n ratio and eliminate the contamination of cosmic - rays easily . the exposure time for each frame is generally between 900 and 3600 seconds depending on the brightness of the object . the wavelength calibration associated with each object was carried out by the helium - neon - argon comparison arcs taken between the two successive frames . the arcs were obtained at the position being nearly identical to that of the specified object . two or three kitt peak national observatory ( kpno ) standard stars ( massey et al . 1988 ) were observed per night for both flux calibration and removal of the atmospheric absorption features . all the objects were observed as close to meridian as possible . table 1 shows the log of observations of the 42 objects listed in our sample . the standard procedures using the iraf package are adopted by us to reduce the unprocessed frames . the ccd reductions include bias subtraction , flat - field correction , and cosmic - rays removal before the extraction of signal . one - dimensional sky - subtracted spectra are then wavelength and flux calibrated . the uncertainties of the wavelength and flux calibrations are no more than 1 and 20% , respectively . the galactic extinctions are corrected by the color excess , the parameter @xmath29 taken from the nasa / iapc extragalactic database ( ned ) , assuming an @xmath30 extinction law of milk way ( cardelli et al . 1989 ) . the spectra are then transformed to the rest frame , along with a @xmath31-correction , according to the narrow peak of h@xmath0 . in order to determine the strength of the optical complex and to reliably profile the emission lines in h@xmath0 region , the complex should be modeled and then removed from each spectrum at first . we simply adopt the empirical template of the complex introduced in boroson & green ( 1992 , hereafter bg92 ) . for each object , the template is first broadened to the fwhm of the h@xmath0 broad component by convolving with a gaussian profile ( bg92 ) . the broadened template and a broken powerlaw are combined to model the continuum by a @xmath32 minimization for individual object . the minimization is performed over the rest frame wavelength range from 4300 to 7000 , except for the regions around the strong emission lines ( i.e. , h@xmath25 + [ ] , [ ] , h@xmath0 , h@xmath33 , [ ] , ) . the fittings are illustrated in the left column of figure 2 for three typical cases . the flux of the complex is integrated between the rest wavelength 4434 and 4686 . the complex is not measured in xssj11067 + 7234 because its continuum is dominated by the starlight from the host galaxy . after removing the blends and continuum , the isolated agn emission lines at h@xmath0 region are modeled by the specfit task ( kriss 1994 ) in iraf package . in each object , each of the [ ] @xmath344959,5007 doublet is modeled by a single gaussian component . the flux ratio of the doublet is fixed to be 3 ( e.g. , dimitrijevic et al . 2007 ) . in principle , each h@xmath0 profile is modeled by a set of two gaussian components : a narrow and a broad gaussian component ( denoted as h@xmath35 and h@xmath36 , respectively ) . the profile modelings are illustrated for the three cases in the right column in figure 2 . the deblendings of the h@xmath0 emission line are , however , difficult in 15 objects because no obvious transition between the narrow and broad components is observed . in these 15 cases , the reported h@xmath36 measurements include both components , and the contribution of h@xmath35 is usually expected to be less than 3% for typical broad - line agns as suggested by bg92 . the described modeling works very well in most objects , except in three ones . in xssj22539 - 1735 , a very broad h@xmath0 component ( @xmath37 ) is required to properly reproduce the observed profile . such component has been reported in h@xmath0 and highly ionized emission lines in previous studies ( e.g. , sulentic et al . 2000b ; wang et al . 2005 ; veron - cetty et al . 2007 ; marziani et al . 2003 ; mullaney & ward 2008 ; hu et al . a double peaked and a boxy h@xmath0 profile is observed in xssj18408 + 7947 and xssj23073 + 0447 , respectively . in all of the three cases , each h@xmath0 line is modeled by three gaussian components , i.e. , a narrow and two broad components . the modeled narrow component is then subtracted from the observed spectrum to derive a residual profile which is then used to measure fwhm and integrated flux . table 2 lists the following items measured from the spectra . column ( 2 ) and ( 3 ) lists the equivalent widths ( ews ) of h@xmath38 and h@xmath39 , respectively . all the ews refer to the continuum level determined from the continuum fitting at the position of wavelength @xmath404861 . the line ratio rfe ( = /h@xmath41 ) is shown in column ( 4 ) . column ( 5 ) lists the luminosity of the h@xmath0 broad component , and column ( 6 ) the fwhm of the broad h@xmath0 in units of @xmath42 . all quoted widths are not corrected for the intrinsic resolution , since the widths are much broader than the resolution . column ( 7 ) shows the luminosity of the [ ] emission line . both [ ] and h@xmath41 luminosities are calculated from the cosmology - corrected luminosity distance given by ned . the next two columns show the absorption - corrected hard x - ray luminosities in the bandpass 3 - 20 kev ( xss ) and in 17 - 60 kev ( _ integral ) . _ six out of the 42 agns are detected by _ integral only . in order to estimate their x - ray luminosities in the bandpass 3 - 20 kev from the luminosities in 17 - 60 kev , a transformation is derived in terms of the 18 objects detected by both surveys . the left panel of figure 3 shows the relationship between the two sets of luminosity . a unweighted fitting yields a relationship @xmath43 with a standard deviation of 0.25 . after transform the luminosities in the bandpass 17 - 60 kev to 3 - 20 kev for the six objects , the strong correlation between the [ ] and 3 - 20 kev luminosity is shown in the right panel of figure 3 for the total 42 objects . the correlation is highly consistent with that derived in heckman et al . ( 2005 ) , which firmly demonstrates the accuracy of our observations and calibrations . _ the detection of hard x - ray emission from a nucleus is regarded as strong evidence of accretion activity occurring around the central smbh . the correlations between x - ray luminosity and luminosities of optical emission lines ( e.g. , h@xmath25 , h@xmath0 , [ ] ) have been extensively established in the previous studies ( e.g. , elvis et al . 1984 ; ward et al . 1988 ; mulchaey et al . 1994 ; heckman et al . 2005 ; panessa et al . however , the correlations seem to depend on sample selection ( heckman et al . 2005 ) and on x - ray luminosity itself ( netzer et al . 2006 ) . in this paper , we estimate the luminosity of agn at rest wavelength 5100 from the h@xmath41 component according to the calibration given by greene & ho ( 2005 ) : @xmath44 the bolometric luminosity is then derived by adopting the usually used bolometric correction @xmath45 ( e.g. , kaspi et al . the significant correlation between @xmath46 and @xmath47 is shown in the left panel in figure 4 . a unweighted least square fitting gives a relationship : @xmath48 or @xmath49 for simplification . the correlation strongly indicates a close link between the hard x - ray emission and ionizing radiation emitted from the accretion disk . note that the bolometric luminosity spans four orders of magnitude , down to @xmath50 , which is close to the formal definition of low luminous agns ( i.e. , @xmath51 , ho 2003 ) . these faint galaxies generally radiate at low state with eddington ratio marginally exceeds 0.01 . it is now generally believed that black hole mass ( @xmath9 ) and specific accretion rate ( @xmath10 ) are two basic parameters determining the properties of agns . the role of @xmath10 in driving the eigenvector i ( ei ) space , the ei space has been subsequently extended to ultraviolet and soft x - ray bands ( i.e. , fwhm ( ) and @xmath52 , e.g. , wang et al . 1996 ; xu et al . 2003 ; grupe 2004 ; sulentic et al . 2007 ; see sulentic et al . 2000a for a review ) . ] has been extensively investigated in numerous previous studies ( e.g. , boroson 2002 ; sulentic et al . 2006 ; xu et al . 2003 ; grupe 2004 ; zamanov & marziani 2002 ; marziani et al . 2001 ) , because of the great progress in the calibration of the @xmath53 relationship ( e.g , kaspi et al . 2000 , 2005 , 2007 ; mclure & jarvis 2004 ; vestergaard & peterson 2006 ; peterson et al . 2004 ; bentz et al . 2006 ) due to the recent great advance in the reverberation mapping ( e.g. , kapsi et al . 2000 ; peterson & bentz 2006 ) . we refer the reads to mcgill et al . ( 2008 , and references therein ) for a summary of the existing formula used to calculate @xmath9 basing upon `` single - epoch '' observation . in this work , the black hole mass is estimated by the width and luminosity of h@xmath41 in individual object , according to the scaling law obtained by greene & ho ( 2005 ) . @xmath54 for each object , the estimated @xmath9 and @xmath10 is listed in column ( 10 ) and ( 11 ) in table 2 , respectively . the bolometric luminosity @xmath46 is estimated from the h@xmath41 component as described above . the calculations of both @xmath46 and @xmath10 allow us to find that the current sample is strongly biased against sub - luminous agns usually with low @xmath10 . in fact , the left panel of figure 4 shows the lack of agns with @xmath51 . figure 5 displays the distributions of the @xmath10 ( _ left panel ) and @xmath9 ( _ middle panel ) for the total 42 agns . as shown in the middle panel , the @xmath9 is sampled within a wide range ( @xmath55 dex ) from @xmath56 to @xmath57 with a peak at @xmath58 . in contrast , the left panel shows that @xmath10 distributes in a relatively narrow range as compared with the previous studies . the total range of our @xmath10 spans from 0.01 to 1 . in particular , about @xmath59% objects listed in the sample have @xmath10 between 0.01 and 0.1 . however , the @xmath10 of bright local agns usually spans at least three orders of magnitude from 1 to 0.001 ( e.g. , woo & urry 2002 ; boroson 2002 ) . basing upon the 2 - 10kev x - ray luminosity , panessa et al . ( 2006 ) indicated that the @xmath10 of the palomar optically selected agns ranges from @xmath60 to 0.1 . as an additional test , the right panel of figure 4 shows the @xmath9 vs. @xmath47 plot . @xmath47 as a function of @xmath9 is over - plotted as dashed lines for three different @xmath10 ( i.e. , 1 , 0.1 , and 0.01 ) . as shown in the plot , a majority of our objects are located below the line with @xmath61 and above the line with @xmath62 . _ _ in summary , the hard x - ray selected agns listed in our sample are luminous agns with a wide range of @xmath9 but a nearly constant accretion activity ( i.e. , @xmath10 ) , which is likely due to the selection effect of the survey that is biased toward x - ray luminous objects . in active ( or luminous ) agn , the optical / uv ionizing radiation is believed to be emitted from a standard geometric thin disk ( e.g. , shakura & sunyaev 1973 ) . the low energy photon is comptonized by a hot corona to produce the hard x - ray emission below 10kev ( haardt & maraschi 1991 ) . a compton recoil of the soft photon is required to occur on the ionized surface of the accretion disk to produce the emission spectrum beyond 10kev . the main observation feature of the reflection is a bump peaked at about 30kev ( e.g. , george & fabian 1991 ; zycki et al . 1994 ; ross & fabian 2002 , 2005 ) . the bias towards active agns could be possibly caused by the fact that either intensive compton reflection takes place only in agns at high state with large @xmath10 or the surveys are biased against the compton - thick objects . in the first case , the theory of the compton reflection predicts that the x - ray emission contributed by the reflection depends primarily on the x - ray ionizing parameter , and secondarily on the uv radiation produced by the dissipation inside the accretion disk . in the second case , our analysis implies a possible connection between the less x - ray absorption and high @xmath10 in broad - line agns . in fact , we selected the objects with regardless of their x - ray spectral properties . figure 1 shows the lack of objects with large column density in the sample . we believe that the sample completeness would be improved by including the ongoing _ swift / bat survey with large effective collecting area and harder energy bandpass ( 14 - 195 kev ) in the future studies . _ figure 4 shows that the hard x - ray luminosity is strongly correlated with the @xmath9 in our sample . in fact , the correlation is naturally expected given the tight @xmath47 vs. @xmath46 correlation and nearly constant @xmath10 . the @xmath63 vs. @xmath9 correlation provides us a potential estimate of black hole mass for luminous agns within @xmath64 . the following relationship is obtained by us through a least square fitting : @xmath65 . our results conflicts with panessa et al . ( 2006 ) and pellegrini ( 2005 ) who did not find the correlation , but is in agreement with kiuchi et al . panessa et al . ( 2006 ) investigated a sample of 47 nearby seyfert galaxies selected from the palomar spectroscopy ( ho et al . the luminosity obtained by different x - ray instruments is down to @xmath66 . the studies in pellegrini ( 2005 ) are based on the _ chandra observations down to a luminosity @xmath66 . kiuchi et al . ( 2006 ) used the broad - line agn sample detected by the _ asca large sky survey ( alss ) and _ asca medium sensitivity survey in the northern sky ( amssn ) with a detection limit of a few @xmath67 in 2 - 10kev bandpass . as discussed above , it is worthy noting that the correlation most likely does not reflect the physics but a selection effect of the surveys . _ _ _ as a key parameter in the ei space , the ratio ( rfe ) is defined as the ratio of the optical complex to h@xmath0 . although the total emission increases by fourfold with reasonable ionization parameter in agn ( korista et al . 1997 ) , traditional photoionization models can not explain the strong emission in optical and uv bands ( e.g , netzer & wills 1983 ; joly 1987 ; collin - souffrin et al . 1988 ) . at present , the problem is only slightly alleviated by the major improvements in the atomic data and by the improved treatment of the line excitation process ( sigut & pradhan 2003 ; baldwin et al . we refer the reads to collin & joly ( 2000 ) for a summary of the mechanisms that can enhance the emission . bg92 put forward a picture in which the rfe is determined by the vertical structure of the accretion disk . the vertical structure is governed by @xmath10 . a large @xmath10 leads to a large x - ray heated volume that could generate large emission . sulentic et al . ( 2000b ) developed a semi - analysis model in which the rfe depends on @xmath10 as @xmath68 . although the rfe is found to generally increase with @xmath10 in optical bright quasars , netzer et al . ( 2004 ) found that a number of high - z quasars deviate the trend , i.e. , with very small rfe but large @xmath10 ( see also in netzer & trakhtenbrot 2007 ) . using the large database provided by the sloan digital sky survey , netzer & trakhtenbrot ( 2007 ) recently suggested that the enhanced rfe is mainly caused by increased metal abundance . in the current sample , the distribution of rfe is shown in the right panel of figure 5 . rfe uniformly ranges from @xmath69 to 1 . on the contrary , @xmath10 distributes in a quite narrow range as describe above . in fact , no correlation between rfe and @xmath10 is found in our sample ( see the upper panel in figure 6 ) , which motivates us to suspect that rfe does not depend on @xmath10 only . rfe is plotted against the characteristic disk temperature @xmath70 in the bottom panel in figure 6 . the temperature scales with @xmath10 and @xmath9 as predicted by the standard geometric thin disk model ( e.g. , shakura & sunyaev 1973 ) . the exact formula of disk temperature depends on various accretion disk models . for a rapidly rotating kerr hole , with a spin parameter @xmath71 and efficiency of 0.31 , we have @xmath72 . the diagram indicates an obvious correlation between the two parameters . a spearman rank - order test calculated by survival analysis yields a formal correlation coefficient @xmath73 ( @xmath74 , where @xmath75 is the probability of null correlation ) . the correlation is not highly significant probably because of the small sample size . the estimated temperature spans about one order of magnitude ( @xmath76 ) , corresponding to a factor of @xmath77 . the correlation then suggests a trend of more intensive emission for higher disk temperature . a marginal dependence of the continuum shape of qsos on @xmath70 was recently identified by bonning et al . ( 2007 ) who compared the observations of sdss with the nlte models of accretion disk . the current result means that the strength of the emission is likely controlled by the spectral shape of the ionizing continuum . the physical reason of the absence of the baldwin relationship ( baldwin 1977 ) for low ionization emission lines is still an open question . in fact , a weak inverse baldwin relationship for h@xmath0 has been demonstrated by recent studies basing upon large agn samples ( e.g. , croom et al . 2002 ; greene & ho 2005 ) . we identify a tight , positive correlation between @xmath63 and ew(h@xmath41 ) in our hard x - ray selected agns ( i.e. , an inverse baldwin relationship ) . figure 7 presents the correlation with correlation coefficient @xmath78 ( @xmath79 ) estimated by the spearman rank - order analysis . ew(h@xmath41 ) roughly scales with hard x - ray luminosity as @xmath80 . wilkes et al . ( 1999 ) identified a marginal baldwin effect in h@xmath0 line . noted that they examined only the luminous local quasars with @xmath81 although many models are developed to explain the baldwin effect for high ionization emission lines ( e.g. , , wandel 1999 ; korista et al . 1998 ; shields et al . 1995 ; wills et al . 1999 ; baskin & loar 2004 ; bachev et al . 2004 ) , these models can not explain the difference between and h@xmath0 . croom et al . ( 2002 ) suggested that the inverse baldwin relationship could be explained if the longer wavelength continuum contains emission from other components ( e.g. , thermal dust emission , non - thermal radio emission , starlight ) . we estimate the possible contribution of the unknown sources as follows . we start from the relationship @xmath82 , and re - write @xmath83 , where @xmath84 and @xmath85 is the agn luminosity and luminosity of other unknown sources at the h@xmath0 wavelength , respectively . the x - ray luminosity could be replaced by @xmath84 given eq . ( 2 ) and the bolometric correction factor of 9 . replacing @xmath86 as @xmath84 given eq.(1 ) finally yields a relationship @xmath87 . considering the typical case with @xmath88 , about 40% of the observed continuum at h@xmath0 wavelength is estimated to be contributed by the unknown sources . the properties ( @xmath10 and @xmath9 ) of accretion onto smbh are examined in a sample of 42 hard x - ray selected ( 3 - 60kev ) broad - line agns in terms of their optical spectra taken by us . the energy range is harder than that usually used in the similar previous studies . these agns are mainly compiled from the rxte all sky survey ( sazonov & revnivtsev 2004 ) , and are complemented by the released _ integral agn sample ( bassani et al . the statistical analysis allows us to draw the following conclusions : _ 1 . we confirm the tight correlation between the hard x - ray and optical emission line luminosities ( and bolometric luminosity ) in our sample , which suggests a close link between the hard x - ray emission reflected by the ionized surface of the accretion disk and uv / optical radiation . using the hard x - ray luminosity , a strong inverse baldwin relationship of the h@xmath0 emission line is identified in the sample . 2 . the hard x - ray selected broad - line agns listed in the sample are found to be strongly biased toward luminous agns with high @xmath10 and low column density . since @xmath10 is constant ( mostly between 0.01 and 0.1 ) in a first order approximation , the hard x - ray luminosity is strongly correlated with the black hole mass in our sample , which is most likely due to the selection effect of the surveys . although the rfe parameter is independent on @xmath10 in our sample , it is found to be correlated with the accretion disk temperature as assessed by @xmath89 . this result implies that the strength of the emission is determined by the shape of the ionizing spectrum . finally , it should be mentioned that a new era in agn hard x - ray study will be opened in next a few years due to the launch of new missions with enhanced hard x - ray detection capability in not only sensitivity , but also imaging , such as simbol - x , next and nustar . these missions will provide larger , and more complete samples to study the present open issues . we would like to thank the anonymous referee for his / her valuable comments that help to improve the paper . the authors are grateful to todd a. boroson and richard f. green for providing us the template . special thanks go to the staff at xinglong observatory as a part of national astronomical observatories , china academy of science for their instrumental and observational help . this search is supported by the nfs of china under grant 10503005 . antonucci , r. , 1993 , , 31 , 473 bachev , r. , et al . , 2004 , , 617 , 171 baldwin , j. a. , 1977 , , 214 , 679 baldwin , j. a. , ferland , g. j. , korista , k. t. , et al . , 2004 , , 615 , 610 barthelmy , s. d. , barbier , l. m. , cummings , j. r. , et al . , 2005 , ssrv , 120 , 143 baskin , a. , & loar , a. , 2004 , , 350 , l31 bassani , l. , malizia , a. , stephen , j. b. , et al . , 2006b , arxiv : astro - ph/10455 bassani , l. , molina , m. , malizia , a. , et al . , 2006a , , 636 , l65 beckmann , v. , gehrels , n. , shrader , c. r. , et al . , 2006 , , 638 , 642 bennett , c. l. , halpern , m. , hinshaw , g. , et al . , 2003 , , 148 , 1 bentz , m. c. , peterson , b. m. , pogge , r. w. , et al . , 2006 , , 644 , 133 bird , a. j. , barlow , e. j. , bassani , l. , et al , 2006 , , 636 , 765 bonning , e. w. , cheng , l. , shields , g. a. , et al . , 2007 , , 659 , 211 boroson , t. a. , 2002 , , 565 , 78 boroson , t. a. , & green , r. f. , 1992 , , 80 , 109 ( bg92 ) brusa , m. , zamorani , g. , comastri , a. , et al . , 2007 , , 172 , 353 cardamone , c. n. , moran , e. c. , & kay , l. e. , 2007 , , 134 , 1264 cardelli , j. a. , clayton , g. c. , & mathis , j. s. , 1989 , , 345 , 245 comastri , a. , setti , g. , zamorani , g. , et al . , 1995 , , 296 , 1 collin , s. , hameury , j. m. , & joly , m. , , 205 , 19 collin , s. , & joly , m. , 2000 , newar , 44 , 531 croom , s. m. , smith , r. j. , boly , b. j. , et al . , 2001 , , 322 , l29 dadina , m. , 2008 , astro - ph / arxiv0801.4338 , accetped by de luca , a. , & molendi , s. , 2004 , , 419 , 837 dimitrijevic . m. s. , popovic , l. c. , kovacevic , l. , et al . , 2007 , , 374 , 1181 elvis , m. , soltan , a. , & keel , w. c. , 1984 , , 283 , 479 ferrando , p. , arnaud , m. , cordier , b. , et al . , 2003 , arxiv : astro - ph/0309424 gehrels , n. , chincarini , g. , giommi , p. , et al . , 2004 , , 611 , 1005 george , i. m. , & fabian , a. c. , 1991 , , 249 , 352 gilli , r. , comastri , a. , & hasinger , g. , 2007 , , 463 , 79 greene , j. e. , & ho , l. c. , 2005 , , 630 , 122 grupe , d. , 2004 , , 127 , 1799 haardt , f. , & maraschi , l. , 1991 , , 1991 , 380 , l51 hasinger , g. , miyaji , t. , & schmidt , m. , 2005 , , 441 , 417 heckman , t. m. , kauffmann , g. , brinchmann , j. , et al . , 2005 , , 613 , 109 heckman , t. m. , ptak , a. , hornschemeier , a. , et al . , 2005 , , 634 , 161 ho , l. c. , 2005 , active galactic nuclei : from central engine to host galaxy , aspc , ser , 290 , 379 ho , l. c. , filippenko , a. v. , & sargent , w. l. , 1997 , , 112 , 315 hu , c. , wang , j. m. , ho , l. c. , et al . , 2008 , astro - ph / arxiv:0807.2060 , accepted by apjl joly , m. , 1987 , , 184 , 33 kaspi , s. , smith , p. s. , netzer , h. , et al . , 2000 , , 533 , 631 kaspi , s. , smith , p. s. , netzer , h. , et al . , 2005 , , 629 , 61 kaspi , s. , brandt , w. n. , maoz , d. , et al , 2007 , , 659 , 997 kauffmann , g. , heckman , t. m. , tremonti , c. , et al . , 2003 , , 346 , 1055 kawaguchi , t. , shimura , t. & mineshige , s. , 2001 , , 546 , 966 kiuchi , g. , ohta , k. , akiyama , m. , et al . , 2006 , , 647 , 892 korista , k. , baldwin , j. , & ferland , g. , 1998 , , 507 , 24 korista , k. , baldwin , j. , ferland , g. , et al . , 1997 , , 108 , 401 kriss , g. , 1994 , adass , 3 , 437 krivonos , r. , revnivtsev , m. , lutovinov , a. , et al . , 2007 , , 475 , 775 levenson , n. a. , heckman , t. m. , krolik , j. h. , et al . , 2006 , 648 , 111 lira , p. , ward , m. j. , zezas , a. , et al . , 2002 , , 333 , 709 lira , p. ; ward , m. j. , zezas , a. , et al . , 2002 , , 330 , 259 markwardt , c. b. , tuller , j. , skinner , g. k. , et al . , 2005 , , 633 , l77 marziani , p. , sulentic , j. w. , zamanov , r. , et al . , 2003 , mmsai , 74 , 490 marziani , p. , sulentic , j. w. , zwitter , t. , et al , 2001 , , 558 , 533 masetti , n. , palazzi , e. , bassani , l. , et al . , 2004 , , 426 , l41 massey , p. , strobel , k. , barnes , j. v. , et al . , 1988 , , 328 , 315 mcgill , k. l. , woo , j. , treu , t. , et al . , 2008 , , 673 , 703 mclure , r. j. , & jarvis , m. j. , 2004 , , 353 , l45 moran , e. c. , filippenko , a. v. , & chornock , r. , 2002 , , 579 , l71 mulchaey , j. s. , koratkar , a. , ward , m. j. , et al . , 1994 , , 436 , 586 mullaney , j. r. , & ward , m. j. , 2008 , , 385 , 53 netzer , h. , mainieri , v. , rosati , p. , et al . , 2006 , , 453 , 525 netzer , h. , shemmer , o. , maiolino , r. , et al . , 2004 , , 614 , 558 netzer , h. , & trakhtenbrot , b. , 2007 , , 654 , 754 netzer , h. , & wills , b. j. , 1983 , , 275 , 445 osullivan , e. , forbes , d. a. , & ponman , t. j. , 2001 , , 328 , 461 panessa , f. , bassani , l. , cappi , m. , et al . , 2006 , , 455 , 173 panessa , f. , bassani , l. , de rosa , a. , et al . , 2008 , , 483 , 151 piconcelli , e. , jimenez - bailon , e. , guainazzi , m. , et al , 2005 , , 423 , 15 pellegrini , s. , 2005 , , 624 , 155 peterson , b. m. , ferrarese , l. , gilbert , k. m. , et al . , 2004 , , 613 , 682 peterson , b. m. , & bentz , m. c. , 2006 , newar , 50 , 796 reeves , j. n. , & turner , m. j. l. , 2000 , , 316 , 234 revnivtsev , m. , sazonov , s. , jahoda , k. , et al . , 2004 , , 418 , 927 ross , r. r. , & fabian , a. c. , 2002 , , 336 , 315 ross , r. r. , & fabian , a. c. , 2005 , , 358 , 211 shakura , n. i. , & sunyaev , r. a. , 1973 , , 24 , 337 sazonov , s. y. , & revnivtsev , m. g. , 2004 , , 423 , 469 sazonov , s. , revnivtsev , m. , krivonos , r. , et al . , 2007 , , 462 , 57 severgnini , p. , caccianiga , a. , braito , v. , et al . , 2003 , , 406 , 483 shields , j. c. , ferland , g. j. , & peterson , b. m. , 1995 , , 411 , 507 sigut , t. a. a. , & pradhan , a. k. , 2003 , , 145 , 15 silverman , j. d. , green , p. j. , barkhouse , w. a. , et al . , 2005 , , 618 , 123 sulentic , j. w. , bachev , r. , marziani , p. , et al . , 2007 , , 666 , 757 sulentic , j. w. , marziani , p. , & dultzin - hacyan , d. , 2000a , , 38 , 521 sulentic , j. w. , marziani , p. , & zwitter , t. , et al . , 2000b , , 545 , l15 sulentic , j. w. , repetto , p. , stirpe , g. m. , et al . , 2006 , , 456 , 929 takahashi , t. , kelley , r. , mitsuda , k. , et al . , 2008 , astro - ph / arxiv:0807.2007 tueller , j. , mushotzky , r. f. , barthelmy , s. , et al . , 2008 , , 681 , 113 ubertini , p. , lebrun , f. , di cocco , g. , et al . , 2003 , , 411 , l131 veron - cetty , m .- p . , & veron , p. , 2006 , , 455 , 773 veron - cetty , m .- p . , veron , p. , joly , m. , et al . , 2007 , , 475 , 487 vestergaard , m. , & peterson , b. m. , 2006 , , 641 , 689 wandel , a. , 1999 , , 527 , 649 wang , j. , wei , j. y. , & he , x. t. , 2005 , newa , 10 , 353 wang , j. , wei , j. y. , & he , x. t. , 2006 , , 638 , 106 wang , j. , & wei , j. y. , 2008 , , 679 , 86 wang , t. , brinkmann , w. , & bergeron , j. , 1996 , , 309 , 81 ward , m. j. , done , c. , fabian , a. c. , et al . , 1988 , , 324 , 767 wilkes , b. j. , kuraszkiewicz , j. , green , p. j. , et al . , 1999 , , 513 , 76 wills , b. j. , bortherton , m. s. , fang , d. , et al . , 1993 , , 415 , 563 winkler , c. , courvoisier , t. j. -l . , di cocco , g. , et al . , 2003 , , 411 , l1 worsley , m. a. , fabian , a. c. , bauer , f. e. , et al . , 2005 , , 357 , 1281 xu , d. w. , komossa , s. , wei , j. y. , et al . , 2003 , , 590 , 73 zamanov , r. , & marziani , p. , 2002 , , 571 , l77 zdziarski , a. a. , johnson , w. n. , & done , c. , et al . , 1995 , , 438 , l63 zdziarski , a. a. , poutanen , j. , & johnson , w. n. , 2000 , , 542 , 703 zezas , a. , hernquist , l. , fabbiano , g. , et al . , 2003 , , 699 , l73 zycki , p. t. , krolik , j. h. , zdziarski , a. a. , et al . , 1994 , , 437 , 597 ccccccccc name & other name & r.a . & decl . & z & date & exposure ( s ) & @xmath90 & agn type + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) + xssj00368 + 4557 & cgcg535 - 012 & 00 36 21.0 & 45 39 54 & 0.048 & 2400 & 2006 aug 15 & 15.45 & s1 + qso0241 + 622 & & 02 44 57.7 & 62 28 07 & 0.044 & 3600 & 2006 dec 17 & 16.1 & s1 + 3c111 & & 04 18 21.3 & 38 01 36 & 0.0485 & 3600 & 2007 nov 03 & 18.0 & s1 + xssj04331 + 0520 & 3c120 & 04 33 11.1 & 05 21 16 & 0.033 & 1800 & 2007 feb 16 & 14.2 & blrg + xssj05103 + 1640 & iras05078 + 1626 & 05 10 45.5 & 16 29 56 & 0.018 & 1800 & 2007 feb 16 & 15.6 & s1 + xssj05162 - 0008 & akn120 & 05 16 11.4 & -00 08 59 & 0.033 & 1200 & 2006 nov 18 & 14.1 & s1 + xssj05552 + 4617 & mcg+8 - 11 - 11 & 05 54 53.6 & 46 26 22 & 0.02 & 1800 & 2007 feb 17 & 15.0 & s1 + mark6 & & 06 52 12.2 & 74 25 37 & 0.01881 & 3600 & 2007 feb 16 & 15.0 & s1.5 + xssj07434 + 4945 & mrk79 & 07 42 32.8 & 49 48 35 & 0.022 & 1800 & 2005 nov 29 & 14.9 & s1 + xssj08117 + 7600 & pg0804 + 761 & 08 10 58.6 & 76 02 42 & 0.1 & 2400 & 2007 nov 16 & 15.15 & s1 + xssj09204 + 1608 & mrk704 & 09 18 26.0 & 16 18 19 & 0.029 & 1800 & 2006 nov 17 & 15.38 & s1 + xssj09261 + 5204 & mrk110 & 09 25 12.9 & 52 17 11 & 0.036 & 1800 & 2005 nov 30 & 15.6 & nls1 + xssj10231 + 1950 & ngc3227 & 10 23 30.6 & 19 51 54 & 0.0038 & 1800 & 2006 jan 31 & 11.1 & s1 + xssj11067 + 7234 & ngc3516 & 11 06 47.5 & 72 34 07 & 0.008836 & 900 & 2006 dec 18 & 12.5 & s1 + xssj11417 + 5910 & sbs1136 + 594 & 11 39 08.9 & 59 11 55 & 0.06 & 2400 & 2007 jan 26 & 15.5 & s1 + xssj12032 + 4424 & ngc4051 & 12 03 09.6 & 44 31 53 & 0.0023 & 1200 & 2006 nov 16 & 10.83 & nls1 + xssj12106 + 3927 & ngc4151 & 12 10 32.6 & 39 24 21 & 0.0033 & 1800 & 2006 jan 29 & 11.5 & s1 + xssj12164 + 1427 & pg1211 + 143 & 12 14 17.7 & 14 03 13 & 0.081 & 3600 & 2006 feb 04 & 14.63 & nls1 + ngc4253 & & 12 18 26.5 & 29 48 46 & 0.01293 & 1200 & 2006 dec 21 & 13.7 & s1.5 + xssj12206 + 7509 & mrk205 & 12 21 44.0 & 75 18 38 & 0.07 & 2400 & 2007 jan 27 & 14.5 & s1 + xssj12288 + 0200 & 3c273 & 12 29 06.7 & 02 03 08 & 0.15834 & 900 & 2006 dec 21 & 12.86 & rlq + xssj12408 - 0516 & ngc4593 & 12 39 39.4 & -05 20 39 & 0.009 & 1200 & 2007 jan 26 & 14.67 & s1 + xssj13420 - 1432 & npm1g-14.0512 & 13 41 12.9 & -14 38 41 & 0.042 & 2400 & 2008 mar 13 & 15.50 & nls1 + xssj13530 + 6916 & mrk279 & 13 53 03.4 & 69 18 30 & 0.031 & 1200 & 2007 feb 16 & 14.57 & s1 + xssj14181 + 2514 & ngc5548 & 14 17 59.5 & 25 08 12 & 0.017 & 1800 & 2008 mar 03 & 13.3 & s1 + xssj15042 + 1046 & mrk841 & 15 04 01.2 & 10 26 16 & 0.036 & 1200 & 2007 feb 16 & 14.0 & s1 + xssj15348 + 5750 & mrk290 & 15 35 52.3 & 57 54 09 & 0.030 & 2400 & 2008 mar 13 & 15.1 & s1 + xssj17276 - 1359 & pds456 & 17 28 19.8 & -14 15 56 & 0.184 & 1800 & 2007 sep 07 & 14.69 & rqq + xssj17413 + 1851 & 4c+18.51 & 17 42 07.0 & 18 27 21 & 0.19 & 2400 & 2007 sep 07 & 16.43 & rlq + xssj18196 + 6454 & h1821 + 643 & 18 21 57.3 & 64 20 36 & 0.297 & 1800 & 2006 aug 16 & 14.1 & rqq + xssj18348 + 3238 & 3c382 & 18 35 02.1 & 32 41 50 & 0.059 & 1800 & 2006 aug 16 & 15.5 & blrg + xssj18408 + 7947 & 3c390.3 & 18 42 09.0 & 79 46 17 & 0.056 & 1800 & 2006 aug 15 & 14.37 & blrg + 1h1934 - 063a & & 19 37 33.0 & -06 13 05 & 0.01059 & 1800 & 2006 aug 16 & 14.09 & s1 + ngc6814 & & 19 42 40.6 & -10 19 25 & 0.00521 & 1800 & 2007 nov 04 & 12.06 & s1.5 + xssj20404 + 7521 & 4c+74.26 & 20 42 37.3 & 75 08 02 & 0.1 & 1800 & 2006 sep 15 & 15.13 & rlq + xssj20441 - 1042 & mrk509 & 20 44 09.7 & -10 43 25 & 0.034 & 1800 & 2006 sep 15 & 13.0 & s1 + xssj21128 + 8216 & s52116 + 81 & 21 14 01.2 & 82 04 48 & 0.084 & 2400 & 2006 aug 15 & 15.7 & blrg + xssj22423 + 2958 & akn564 & 22 42 39.3 & 29 43 31 & 0.025 & 1800 & 2006 nov 15 & 14.55 & nls1 + xssj22539 - 1735 & mr2251 - 178 & 22 54 05.8 & -17 34 55 & 0.064 & 1800 & 2006 nov 15 & 14.36 & rqq + xssj23033 + 0858 & ngc7469 & 23 03 15.6 & 08 52 26 & 0.016 & 1800 & 2006 aug 15 & 13.0 & s1 + xssj23040 - 0834 & mrk926 & 23 04 43.5 & -08 41 09 & 0.047 & 1800 & 2007 nov 03 & 14.6 & s1 + xssj23073 + 0447 & pg2304 + 042 & 23 07 02.9 & 04 32 57 & 0.042 & 2400 & 2007 nov 02 & 15.44 & s1 + ccccccccccc xssj00368 + 4557 & & 82.6 & 0.35 & 42.1 & 2555 & 41.3 & 43.48 & & 7.42 & 0.23 + qso0241 + 622 & & 43.6 & 0.26 & 42.1 & 9401 & 41.8 & 44.34 & 44.28 & 8.54 & 0.02 + 3c111 & & 119.9 & 0.00 & 43.3 & 3445 & 42.8 & 44.53 & 44.59 & 8.33 & 0.30 + xssj04331 + 0520 & 10.5 & 82.7 & 0.24 & 42.2 & 5367 & 42.1 & 44.16 & & 8.11 & 0.06 + xssj05103 + 1640 & 9.0 & 74.5 & @xmath91 & 41.3 & 4855 & 41.5 & 43.47 & 43.58 & 7.55 & 0.04 + xssj05162 - 0008 & 2.3 & 72.7 & 0.55 & 42.5 & 6595 & 41.5 & 44 . & 44.26 & 8.48 & 0.05 + xssj05552 + 4617 & 53.5 & 103.2 & 0.30 & 41.2 & 5761 & 41.7 & 43.56 & 43.67 & 7.60 & 0.02 + mark6 & 6.0 & 37.4 & 0.42 & 41.4 & 5627 & 41.7 & 43.32 & 43.41 & 7.74 & 0.03 + xssj07434 + 4945 & 8.6 & 65.8 & 0.40 & 41.6 & 3815 & 41.7 & 43.45 & & 7.47 & 0.07 + xssj08117 + 7600 & & 74.7 & 0.52 & 43.4 & 3146 & 42.3 & 44.47 & 44.43 & 8.32 & 0.39 + xssj09204 + 1608 & 6.9 & 91.8 & 0.29 & 41.9 & 6421 & 41.6 & 43.65 & & 8.11 & 0.03 + xssj09261 + 5204 & & 118.6 & 0.08 & 42.1 & 1798 & 41.8 & 43.81 & 44.17 & 7.12 & 0.47 + xssj10231 + 1950 & 5.1 & 39.3 & 0.007 & 40.5 & 4556 & 40.6 & 42.4 & 42.67 & 7.01 & 0.02 + xssj11067 + 7234 & & 31.9 & & 41.1 & 5020 & 40.7 & 43.04 & & 7.46 & 0.03 + xssj11417 + 5910 & 7.3 & 73.8 & 0.13 & 42.0 & 4608 & 41.9 & 43.82 & & 7.91 & 0.07 + xssj12032 + 4424 & & 21.5 & 0.61 & 39.9 & 1202 & 40.0 & 42.04 & 41.87 & 5.53 & 0.20 + xssj12106 + 3927 & 37.2 & 100.0 & 0.12 & 40.8 & 6354 & 41.3 & 43.17 & 43.37 & 7.49 & 0.01 + xssj12164 + 1427 & & 120.3 & 0.40 & 42.5 & 1804 & 41.7 & 44.22 & & 7.32 & 0.60 + ngc4253 & 11.3 & 32.1 & 0.72 & 40.8 & 2153 & 41.3 & 42.59 & 42.7 & 6.57 & 0.13 + xssj12206 + 7509 & & 95.6 & 0.21 & 42.5 & 3037 & 41.8 & 43.88 & & 7.83 & 0.23 + xssj12288 + 0200 & & 71.3 & 0.51 & 44.1 & 2970 & 43.2 & 45.91 & 45.92 & 8.66 & 0.74 + xssj12408 - 0516 & & 49.8 & 0.54 & 41.0 & 3654 & 40.4 & 42.57 & 43.04 & 7.12 & 0.05 + xssj13420 - 1432 & 15.3 & 64.6 & 0.66 & 41.8 & 4112 & 41.8 & 43.57 & & 7.67 & 0.07 + xssj13530 + 6916 & 5.7 & 59.6 & 0.44 & 42.2 & 5147 & 41.8 & 43.94 & & 8.07 & 0.06 + xssj14181 + 2514 & 7.8 & 62.7 & 0.04 & 41.3 & 10312 & 41.5 & 43.66 & 43.08 & 8.20 & 0.008 + xssj15042 + 1046 & 9.4 & 129.6 & 0.33 & 42.5 & 5709 & 42.2 & 43.67 & & 8.33 & 0.061 + xssj15348 + 5750 & 4.2 & 23.5 & 0.71 & 41.1 & 4740 & 41.6 & 43.43 & & 7.41 & 0.03 + xssj17276 - 1359 & & 74.6 & 3.58 & 44.1 & 2957 & @xmath92 & 44.93 & & 8.66 & 0.75 + xssj17413 + 1851 & 1.8 & 115.5 & 0.04 & 43.0 & 13506 & 42.2 & 45.45 & & 9.35 & 0.02 + xssj18196 + 6454 & 2.9 & 65.4 & 0.18 & 44.3 & 5797 & 43.7 & 45.61 & & 9.36 & 0.23 + xssj18348 + 3238 & & 111.4 & 0.19 & 42.8 & 8340 & 41.8 & 44.58 & 44.45 & 8.84 & 0.04 + xssj18408 + 7947 & 3.2 & 118.8 & 0.15 & 42.5 & 13799 & 42.0 & 44.39 & 44.61 & 9.13 & 0.01 + 1h1934 - 063a & & 55.2 & 0.94 & 41.1 & 1354 & 41.0 & 42.48 & 42.59 & 6.29 & 0.38 + ngc6814 & & 38.8 & 0.75 & 40.2 & 5182 & 40.2 & 42.35 & 42.47 & 6.95 & 0.01 + xssj20404 + 7521 & & 85.4 & @xmath91 & 43.6 & 9418 & 42.5 & 44.73 & 44.98 & 9.37 & 0.05 + xssj20441 - 1042 & 10.7 & 111.8 & 0.21 & 42.4 & 2261 & 42.1 & 44.2 & 44.13 & 7.47 & 0.36 + xssj21128 + 8216 & 6.3 & 150.0 & 0.14 & 42.6 & 4123 & 42.3 & 44.25 & 44.65 & 8.12 & 0.13 + xssj22423 + 2958 & & 56.6 & 0.74 & 41.7 & 3649 & 41.7 & 43.42 & & 7.52 & 0.09 + xssj22539 - 1735 & 5.7 & 132.5 & 0.02 & 42.9 & 4093 & 42.6 & 44.63 & 44.62 & 8.28 & 0.16 + xssj23033 + 0858 & 13.7 & 45.3 & 0.30 & 41.5 & 2819 & 41.6 & 43.27 & 43.40 & 7.17 & 0.12 + xssj23040 - 0834 & 3.4 & 65.1 & @xmath91 & 42.5 & 7083 & 42.2 & 44.23 & 44.21 & 8.51 & 0.04 + xssj23073 + 0447 & 2.7 & 42.0 & @xmath91 & 41.2 & 7322 & 41.2 & 43.67 & & 7.85 & 0.01 +

a multi - domain spectral method for computing very high precision 3-d stellar models is presented . the boundary of each domain is chosen in order to coincide with a physical discontinuity ( e.g. the star s surface ) . in addition , a regularization procedure is introduced to deal with the infinite derivatives on the boundary that may appear in the density field when stiff equations of state are used . consequently all the physical fields are smooth functions on each domain and the spectral method is absolutely free of any gibbs phenomenon , which yields to a very high precision . the power of this method is demonstrated by direct comparison with analytical solutions such as maclaurin spheroids and roche ellipsoids . the relative numerical error reveals to be of the order of . this approach has been developed for the study of relativistic inspiralling binaries . it may be applied to a wider class of astrophysical problems such as the study of relativistic rotating stars too . # 1#2 # 1#2 ^ 2#1 ^ 2 # 1to 0pt#1

astro-ph9803086

one of the most promising source of gravitational waves is the coalescence of inspiralling compact binaries . the recent development of interferometric gravitational waves detectors ( _ e.g . geo600 , ligo , tama _ and _ virgo _ ) gives an important motivation for studying this problem . such a study requires a relativistic formalism to derive the equations of motion and then an accurate and tricky method to solve the resulting system of partial differential equations . we have recently @xcite proposed a relativistic formalism able to tackle the problem of co - rotating _ as well as _ counter - rotating binaries system , these latter being more relevant from the astrophysical point of view . we present now a very accurate approach based on multi - domain spectral method that circumvents the gibbs phenomenon to numerically solve this problem and which can be applied to a wide class of other astrophysical situations . various astrophysical applications of spectral methods have been developed in our group ( for a review , see @xcite ) , including 3-d gravitational collapse of stellar core @xcite , neutron star collapse into a black hole @xcite , tidal disruption of a star near a massive black hole @xcite , rapidly rotating neutron stars @xcite , magnetized neutron stars @xcite and their resulting gravitational radiation @xcite , spontaneous symmetry breaking of rapidly rotating neutron stars @xcite , proto - neutron stars evolution @xcite . in computational fluid dynamics , spectral methods are known for their very high accuracy @xcite ; indeed for a @xmath1 function , the numerical error decreases as @xmath2 ( _ evanescent error _ ) , where @xmath3 is the number of coefficients involved in the spectral expansion , or equivalently the number of grid points in the physical domain . this is much faster than the error decay of finite - difference methods , which behaves as @xmath4 , with @xmath5 generally not larger than @xmath6 . for this reason , spectral methods are particularly interesting for treating 3-d problems such as binary configurations a situation in which the number of grid points is still severely limited by the capability of present and next generation computers . spectral methods lose much of their accuracy when non - smooth functions are treated because of the so - called _ gibbs phenomenon_. this phenomenon is well known from the most familiar spectral method , namely the theory of fourier series : the fourier coefficients @xmath7 of a function @xmath8 which is of class @xmath9 but not @xmath10 decrease as @xmath11 only . in particular , if the function has some discontinuity , its approximation by a fourier series does not converge towards @xmath8 at the discontinuity point : there remains a gap which is of the order 10% . the multi - domain spectral method described in this paper circumvents the gibbs phenomenon . the basic idea is to divide the space into domains chosen so that the physical discontinuities are located onto the boundaries between the domains ( sect . [ s : mapping ] ) . the simplest example is the case of a perfect fluid star , where two domains may be distinguished : the interior and the exterior of the star . the boundary is then simply the surface of the star . the second ingredient of the technique is a mapping between the domains defined in this way and some simple mathematical domains , which are cross products of intervals : @xmath12 \times [ b_1,b_2 ] \times [ c_1,c_2]$ ] . the spectral expansion is then performed with respect to functions of the coordinates spanning these intervals ( sect . [ s : multi ] ) . the method of resolution of a basic equation , namely the poisson equation , is exposed in sect . [ s : resol_poisson ] . for stiff equations of state , the above procedure is not sufficient to ensure the smoothness of all the functions . indeed , for a polytrope with an adiabatic index greater than 2 , the density field has an infinite derivative on the surface of the star . we present in sect . [ s : regul ] a method for regularizing the density and recover the spectral precision . the power of the multi - domain spectral method is illustrated in sect . [ s : illustr ] , where comparisons are performed between numerical solutions obtained by an implementation of the method and analytical solutions ( ellipsoidal configurations of incompressible fluids ) . finally sect . [ s : conclu ] concludes the article by discussing the great advantages of the multi - domain spectral method for dealing with relativistic binary neutron stars . in order to treat problems involving poisson equations with non - compact sources as they appear in relativistic gravitation we take for the physical domain where the computation must be performed the whole three - dimensional space @xmath13 . in doing so , we know the physical boundary conditions we have to impose in order to solve the poisson equations . these boundary conditions can be easily set at infinity . we divide @xmath13 into @xmath14 domains @xmath15 ( @xmath16 ) . in the present work , these domains are taken to be _ star - like _ ( in the mathematical sense ) with respect to a some origin @xmath17 , which means that for every point @xmath18 in the domain @xmath19 , the segment @xmath20 is entirely included in @xmath21 ( see fig . [ f : domaine ] ) . the multi - domain spectral method we are going to describe can be extended to more general domains , at the price of a greater technical ( but non conceptual ) difficulty . however , for stellar configurations , the star - like hypothesis is sufficient for most applications . let us denote by @xmath22 the boundary surface between the domains @xmath19 and @xmath23 . @xmath24 is simply connected and its boundary is @xmath25 ; we call it the _ nucleus_. for @xmath26 , @xmath19 s inner boundary is @xmath27 and outer boundary @xmath22 . the last domain , @xmath28 , has @xmath29 as inner boundary and extends to infinity ( cf . [ f : domaine ] ) . let us choose some cartesian frame of @xmath13 centered at @xmath17 and let us call @xmath30 the associated spherical coordinates : @xmath31 , @xmath32 $ ] and @xmath33 . the mapping of each domain onto the cross product of intervals @xmath12 \times [ b_1,b_2 ] \times [ c_1,c_2]$ ] will be defined with respect to the spherical coordinates @xmath30 . since each domain @xmath34 is star - like with respect to @xmath17 , the equation of the boundaries @xmath22 can be written under the form [ e : s_l : def ] r = s_l ( , ) , where @xmath35 is a smooth function on @xmath36\times [ 0,2\pi[$ ] . the basic idea is to introduce a mapping [ e : map : gen ] rcl [ 0,1][0,2 [ & & d_0 + ( , , ) & & ( r , , ) so that the origin @xmath17 corresponds to @xmath37 and the boundary @xmath25 to @xmath38 . using the fact that @xmath24 is star - like , a simple form of the mapping ( [ e : map : gen ] ) can be chosen as @xmath39 where @xmath40 is a smooth function subject to regularity properties which are discussed below . thanks to eqs . ( [ e : theta])-([e : phi ] ) , we will make no distinction between @xmath41 and @xmath42 , as well as between @xmath43 and @xmath44 , i.e. we will abandon the primes on the angles . the fact that @xmath24 s boundary coincides with @xmath45 translates as [ e : r0=surf ] r_0(1 , , ) = s_0 ( , ) . beside eq . ( [ e : r0=surf ] ) , the function @xmath40 must satisfy some _ regularity properties _ induced by the singular behavior of spherical coordinates at @xmath46 , @xmath47 and @xmath48 . we define a function @xmath49 to be _ regular _ if it can be expressed as a polynomial of the cartesian coordinates @xmath50 we will assume that all the physical fields are regular functions on each domain @xmath19 ( the domains @xmath19 are in fact chosen in this manner ) with respect to the previous definition . it is easy to see that any regular function is expandable as [ e : devel , regul ] f(r , , ) = _ m =- m^m _ = |m|^l r^t(r^2 ) ^|m| p_-|m| ( ) e^im , where @xmath51 and @xmath18 are positive integers , @xmath52 , @xmath53 is some polynomial of degree @xmath54 and @xmath55 is some even polynomial . a simple form of the mapping ( [ e : r = r0])-([e : phi ] ) has been already introduced in the literature @xcite , namely @xmath56 , where @xmath57 is the equation of the star s surface [ eq . ( [ e : s_l : def ] ) ] . however for such a mapping the regularity condition ( [ e : devel , regul ] ) would be quite complicated when expressed in terms of @xmath58 . we choose instead the mapping defined by [ e : r0=afbg ] r_0 ( , , ) = _ 0 [ + a_0 ( ) f_0 ( , ) + b_0 ( ) g_0 ( , ) ] , where @xmath59 and @xmath60 are the following even and odd polynomials @xmath61 and the constant @xmath62 as well as the two functions @xmath63 and @xmath64 are such that ( i ) the fourier expansion of @xmath63 ( resp . @xmath64 ) with respect to @xmath43 contains only even ( resp . odd ) harmonics and ( ii ) the equation of the surface @xmath25 can be written _ 0 [ 1 + f_0 ( , ) + g_0 ( , ) ] = s_0 ( , ) [ e : f0g0=s0 ] . the polynomials @xmath59 and @xmath60 defined by ( [ e : a0:def])-([e : b0:def ] ) are such that @xmath65 the properties ( [ e : f0g0=s0 ] ) and ( [ e : a0b0=1 ] ) ensure that eq . ( [ e : r0=surf ] ) is satisfied , i.e. that the mapping ( [ e : r0=afbg ] ) is from @xmath66\times[0,\pi]\times[0,2\pi[$ ] to @xmath24 . the polynomials @xmath59 and @xmath60 are chosen in order to satisfy at the minimum level the regularity conditions mentioned above . in particular , near the origin @xmath67 behaves as @xmath68 and is independent of @xmath69 , which would not have been the case of the mapping @xmath70 introduced in refs . @xcite . the jacobian of the transformation @xmath71 is @xmath72\end{aligned}\ ] ] since @xmath73 and @xmath74 for any @xmath75 $ ] , the mapping could become singular if @xmath63 or @xmath64 is negative and has an important amplitude . we can not control the sign of the function @xmath63 because it contains only odd harmonics of @xmath43 . however , by a suitable choice of @xmath62 , one can impose [ e : g0>0 ] g_0 ( , ) 0 , as we shall see below . we have found that this condition was sufficient to ensure that @xmath76 , i.e. that the mapping is regular , for all the astrophysically relevant situations we have encountered . the equation for the surface of the domain @xmath24 being given , under the form ( [ e : s_l : def ] ) , @xmath77 , the procedure which leads to @xmath62 , @xmath63 and @xmath64 , i.e. to the full determination of the function @xmath78 , is the following one . first let us choose a point @xmath79 on the surface @xmath25 . equation ( [ e : f0g0=s0 ] ) implies the following relation [ e : s0(*):1 ] _ 0 [ 1 + f_0 ( _ * , _ * ) + g_0 ( _ * , _ * ) ] = s_0 ( _ * , _ * ) . let us introduce the following auxiliaries quantities @xmath80 \\ & & \tilde f_0(\theta,\varphi ) : = \alpha_0 f_0(\theta,\varphi ) \label{e : tf0:def } \\ & & \tilde g_0(\theta,\varphi ) : = \alpha_0 g_0(\theta,\varphi ) - \mu \label{e : tg0:def } \ .\end{aligned}\ ] ] equation ( [ e : s0(*):1 ] ) translates then in [ e : a+l+m = s ] _ 0 + = s_0 ( _ * , _ * ) , whereas eq . ( [ e : f0g0=s0 ] ) becomes f_0 ( , ) + g_0 ( , ) = s_0 ( , ) - s_0 ( _ * , _ * ) . having expanded @xmath81 into fourier series with respect to @xmath43 , one deduces the functions @xmath82 ( resp . @xmath83 ) by taking only the odd ( resp . even ) harmonics of this fourier expansion . @xmath84 is then computed as = - \ { g_0 ( , ) , ( , ) [ 0,2 [ } . in doing so , condition ( [ e : g0>0 ] ) will automatically be fulfilled . the value of the coefficient @xmath62 is deduced from the above value of @xmath84 via eq . ( [ e : a+l+m = s ] ) . finally the functions @xmath63 and @xmath64 are computed from eqs . ( [ e : tf0:def ] ) and ( [ e : tg0:def ] ) . for @xmath85 , we introduce the mapping [ e : map : interm ] rcl [ -1,1][0,2 [ & & d_l + ( , , ) & & ( r , , ) under the form @xmath86 where @xmath87 is a smooth function which satisfies @xmath88 which means that the inner ( resp . outer ) boundary of @xmath19 is defined by @xmath89 ( resp . @xmath90 ) . we choose @xmath91 as @xmath92 \nonumber \\ & & + \beta_l \ , \label{e : rl = afbg}\end{aligned}\ ] ] where @xmath93 and @xmath94 are the following polynomials @xmath95 and the constants @xmath96 and @xmath97 and the two functions @xmath98 and @xmath99 are defined from the equations of the surfaces @xmath27 and @xmath22 by @xmath100 + \beta_l = s_{l-1}(\theta,\varphi ) \label{e : flglm1=slm1 } \\ & & \alpha_l \l [ + 1 + g_l(\theta,\varphi ) \r ] + \beta_l = s_l(\theta,\varphi ) \label{e : flgl = sl } \\ & & f_l(\theta,\varphi ) \leq 0 \label{e : fl<0 } \\ & & g_l(\theta,\varphi ) \geq 0 \label{e : gl>0}\end{aligned}\ ] ] note that the polynomials @xmath93 and @xmath94 defined by ( [ e : al : def])-([e : bl : def ] ) are such that @xmath101 the properties ( [ e : flglm1=slm1 ] ) and ( [ e : albl(0 ) ] ) [ resp . ( [ e : flgl = sl ] ) and ( [ e : albl(1 ) ] ) ] ensure that eq . ( [ e : rlm1=surf ] ) [ resp . ( [ e : rlp1=surf ] ) ] is satisfied , i.e. that the mapping ( [ e : rl = afbg ] ) is from @xmath102\times[0,\pi]\times[0,2\pi[$ ] to @xmath19 . the conditions ( [ e : fl<0 ] ) and ( [ e : gl>0 ] ) ensure that this mapping is not singular , by the same argument as that presented for @xmath40 in sect . [ s : map_nucl ] , the sign of @xmath98 being opposite to that of @xmath99 because @xmath93 is a decreasing function of @xmath103 , whereas @xmath94 is an increasing function of @xmath103 . the equation for the inner and outer boundaries of the domain @xmath19 being given , under the form ( [ e : s_l : def ] ) : @xmath104 ( inner boundary ) and @xmath105 ( outer boundary ) , the procedure which leads to @xmath96 , @xmath97 , @xmath98 and @xmath99 , i.e. to the full determination of the function @xmath91 , is the following one . first let us choose a point @xmath79 on the surface @xmath27 along with the corresponding point @xmath79 on the surface @xmath22 . equations ( [ e : flglm1=slm1 ] ) and ( [ e : flgl = sl ] ) imply the following relations @xmath106 + \beta_l = s_{l-1}(\theta_*,\varphi _ * ) \label{e : slm1(*):1 } \\ & & \alpha_l \l [ 1 + g_l(\theta_*,\varphi _ * ) \r ] + \beta_l = s_l(\theta_*,\varphi _ * ) \label{e : sl(*):1 } \ .\end{aligned}\ ] ] let us introduce the following auxiliaries quantities @xmath107 equations ( [ e : slm1(*):1 ] ) and ( [ e : sl(*):1 ] ) then translate in @xmath108 whereas eqs . ( [ e : flglm1=slm1 ] ) and ( [ e : flgl = sl ] ) become @xmath109 from the values of @xmath110 and @xmath111 obtained above , @xmath112 and @xmath84 are computed as @xmath113\times [ 0,2\pi [ \r\ } \\ & & \mu = - \min \l\ { \tilde g_l(\theta,\varphi ) , \ ( \theta,\varphi ) \in [ 0,\pi]\times [ 0,2\pi [ \r\ } \ . \end{aligned}\ ] ] in doing so , the conditions ( [ e : fl<0 ] ) and ( [ e : gl>0 ] ) will automatically be fulfilled . the value of the constants @xmath96 and @xmath97 are deduced from the above values of @xmath112 and @xmath84 via eqs . ( [ e : slm1(*):2 ] ) and ( [ e : sl(*):2 ] ) . finally the functions @xmath98 and @xmath99 are computed from eqs . ( [ e : tfl : def ] ) and ( [ e : tgl : def ] ) . in the case where the external domain @xmath114 extends to infinity we introduce the mapping [ e : map : ext ] rcl [ -1,1][0,2 [ & & d_ext + ( , , ) & & ( r , , ) under the form @xmath115 where @xmath116 is a smooth function which satisfies @xmath117 where @xmath118 . the above two equations show that the inner boundary of @xmath119 is defined by @xmath89 , whereas @xmath90 corresponds to the infinity . we have already introduced such a compactification of the infinite domain in ref . @xcite , in the case of a spherical inner boundary . we choose the function @xmath120 as u ( , , ) = _ ext [ + a_ext ( ) f_ext ( , ) - 1 ] , [ e : u = af ] where @xmath121 is the same polynomial of @xmath103 as that defined in eq . ( [ e : al : def ] ) , and the constant @xmath122 and the function @xmath123 are defined from the equations of the surface @xmath29 by @xmath124 = s_{\rm ext}(\theta,\varphi)^{-1 } \label{e : fext = s } \\ & & f_{\rm ext}(\theta,\varphi ) \leq 0 \label{e : fext<0 } \ .\end{aligned}\ ] ] the condition ( [ e : fext<0 ] ) ensures that @xmath125 i.e. that the mapping ( [ e : u = af ] ) is not singular . the equation for the inner boundary of the domain @xmath119 being given , under the form ( [ e : s_l : def ] ) : @xmath126 , the procedure which leads to @xmath122 and @xmath123 , i.e. to the full determination of the function @xmath120 , is the following one . first let us choose a point @xmath79 on the surface @xmath127 . equation ( [ e : fext = s ] ) implies the following relation _ ext [ -2 + f_ext ( _ * , _ * ) ] = s_ext(_*,_*)^-1 . by introducing the auxiliaries quantities @xmath128 this equation translates in [ e : a+l = s : ext ] - 2 _ ext + = s_ext(_*,_*)^-1 , whereas eq . ( [ e : fext = s ] ) becomes f_ext ( , ) = s_ext(,)^-1 - s_ext(_*,_*)^-1 . from the above value of @xmath129 , @xmath112 is computed according to = - \ { f_ext ( , ) , ( , ) [ 0,2 [ } . in doing so , the condition ( [ e : fext<0 ] ) will automatically be fulfilled ( recall that @xmath130 ) . the value of @xmath122 is deduced from the above value of @xmath112 via eq . ( [ e : a+l = s : ext ] ) . finally the function @xmath123 is computed from eq . ( [ e : tfext : def ] ) . the spirit of the multi - domain spectral method is to perform spectral expansions on each domain @xmath19 , and with respect to the coordinates @xmath58 instead of the physical coordinates @xmath30 . we shall take as basis functions _ separable _ functions of @xmath58 , i.e. functions that can be put under the form @xmath131 . the variable @xmath43 being periodic , it is natural to use fourier series in @xmath43 , i.e. to choose _ k ( ) = e^ik , where @xmath132 is an even integer that we will call the _ number of degrees of freedom in @xmath43_. the associated collocation points ( `` grid points '' ) are _ k = 2 k / n _ 0 k n_- 1 . as concerns @xmath133 one must use functions that are compatible with the expansion ( [ e : devel , regul ] ) of any regular scalar field @xmath8 . we shall not use @xmath134 , as suggested by eq . ( [ e : devel , regul ] ) , but a wider set , namely the functions @xmath135 where @xmath136 is an integer that we will call the _ number of degrees of freedom in @xmath41 _ and @xmath137 is the degree of the harmonic in the fourier series with respect to @xmath43 : @xmath138 in the present case . the advantages of this choice are to allow the use of fast - fourier - transform algorithms for computing the coefficients , as well as very simple matrices for the usual differential operators @xcite . the associated collocation points are _ j = j / ( n_-1 ) 0 j n_- 1 . as concerns the variable @xmath103 , we also choose a set wider than merely @xmath139 , namely : @xmath140 where @xmath141 is an integer that we will call the _ number of degrees of freedom in @xmath67 _ and @xmath142 denotes the @xmath143 degree chebyshev polynomial . the associated collocation points are [ e : xi : coll : nucl ] _ i = ( 2 in_r -1 ) 0 i n_r - 1 . the above choice concerns the nucleus @xmath24 only . for the intermediate and external domains , we choose instead x_kji ( ) = t_i ( ) , along with the collocation points _ i = - ( i / ( n_r -1 ) ) 0 i n_r - 1 . note that for the nucleus the above choice is the same as that presented in @xcite , once @xmath103 is replaced by @xmath67 . we report the interested reader to that paper for a more detailed discussion about this choice ( see also appendices a , b and d of @xcite ) . when a symmetry is present , we use different bases , in order to take the symmetry into account . for instance , an often existing symmetry is the symmetry with respect to the equatorial plane , i.e. the plane @xmath144 . in this case , we use , instead of ( [ e : cosjt])-([e : sinjt ] ) , @xmath145 the associated collocation points span only @xmath146 $ ] , instead of @xmath36 $ ] : [ e : theta : col : sym ] _ j = 2 j n_-1 0 j n_- 1 . another usual symmetry is the above equatorial symmetry augmented by the symmetry under the transformation @xmath147 . this is the case of a triaxial ellipsoid , or of an axisymmetric star perturbed by even @xmath137 modes . in this case , the @xmath43-basis functions are _ k ( ) = e^2ik . the associated collocation points span @xmath148 , instead of @xmath149 : _ k = k / n _ 0 k n_- 1 . the basis in @xmath41 become _ kj ( ) = ( 2j ) , instead of ( [ e : cos2jt])-([e : sin2jt ] ) , the collocation points in @xmath41 remaining those given by eq . ( [ e : theta : col : sym ] ) . in this case , the basis for @xmath103 in the nucleus contain only even polynomials : x_kji ( ) = t_2i ( ) , the collocation points remaining those given by eq . ( [ e : xi : coll : nucl ] ) . in this section , we present how a first order differential operator , the gradient , and a second order one , the laplacian , both applied to a scalar field , are expressed in term of the coordinates system described above . the computation of any other kind of operator is straightforward . the components of the gradient of a scalar field @xmath8 in an orthonormal basis associated with the spherical coordinates @xmath30 are @xmath150 where the following abbreviations have been introduced : @xmath151 note that we have re - introduced the primes on @xmath41 and @xmath43 [ cf . ( [ e : theta])-([e : phi ] ) ] to avoid any confusion between the partial derivatives . the partial derivatives that appear in the quantities @xmath152 are computed by ( i ) a ( banded ) matrix multiplication on the coefficients of the spectral expansion of the functions @xmath98 and @xmath99 and ( ii ) analytically for the polynomials @xmath153 and @xmath154 . in the nucleus , @xmath155 is re - expressed as j_2 = ( 3 ^ 3 -2 ^ 5 ) f_0/ + 12 ( 5 ^ 2 - 3 ^4 ) g_0/ 1 + ( 3 ^ 3 -2 ^ 5 ) f_0 + ( 5 ^ 2 - 3 ^4 ) g_0 , in order to avoid any division by a vanishing quantity at @xmath37 . the same thing is done for @xmath156 . the above expressions are valid for the nucleus and the intermediate domains , i.e. for @xmath157 . for the compactified domain @xmath119 , the quantity to be considered is @xmath158 instead @xmath159 . indeed , gradients in the compactified domain are used in the computation of non - linear terms in the relativistic gravitational field equations ( scalar products of gradients of the metric potentials ) . we shall see below that the source of the poisson equation on @xmath119 is to be multiplied by @xmath160 , so that if each gradient is multiplied by @xmath161 , this multiplication by @xmath160 is automatically performed . the orthonormal components of @xmath158 on @xmath119 are @xmath162 the laplacian of a scalar field @xmath8 reads @xmath163 \der{f}{\xi } \bigg\ } \ , \label{e : laplacien}\end{aligned}\ ] ] where the primes on @xmath41 and @xmath43 have been abandoned again and the following abbreviation has been introduced : [ e : lapang ] _ : = ^2 ^2 + 1 + 1 ^ 2 ^2 ^2 . for many astrophysical applications , one has to solve the poisson - like equation f = ( f ) , [ e : poisson ] for some ` potential ' @xmath8 . note that for relativistic computations , @xmath164 is not compactly supported ( see e.g. @xcite ) and generally decreases as @xmath165 when @xmath166 . when expressed in terms of the variables @xmath58 , the laplacian takes the complicated form ( [ e : laplacien ] ) , for which it is not obvious to find eigenfunctions . therefore , we introduce , in each domain @xmath19 , a new radial coordinate : = _ l , where @xmath96 and @xmath97 are the same constants as in eqs . ( [ e : r0=afbg ] ) and ( [ e : rl = afbg ] ) ( in the nucleus : @xmath167 ) . in the exterior domain , we introduce : = _ ext ( - 1 ) , where @xmath122 is the same constant as in eq . ( [ e : u = af ] ) . we may then split the laplacian operator @xmath168 into a _ pseudo - laplacian _ @xmath169 and a part which would vanish if the domains @xmath19 were exactly spherical ( in this case , the coordinates @xmath170 and @xmath171 introduced here above would coincide with the physical coordinates @xmath67 and @xmath172 respectively ) . by _ pseudo - laplacian _ , we mean the operator which once expressed in terms of @xmath173 has the same structure than the laplacian operator in spherical coordinates : [ e : tilde_delta ] : = ^2 ^2 + 2 + 1 ^ 2 _ , where @xmath174 is defined by eq . ( [ e : lapang ] ) . in the exterior domain , the _ pseudo - laplacian _ is defined instead by : = ^2 ^2 + 1 ^ 2 _ . it is much easier to invert the operator @xmath169 than the operator @xmath168 : using spherical harmonics in @xmath69 , the problem reduces to a system of second order ordinary differential equations with respect to the variable @xmath170 . moreover , the junction conditions between the various domains are easily imposed , as explained below . the poisson equation ( [ e : poisson ] ) becomes [ e : a_tilde_poisson ] a f = ( f ) + r(f ) , where a : = _ l^2 j_1 ^ -2 ( 1 + j_2 ^ 2 + j_3 ^ 2 ) , @xmath175 { 2\ov j_1 r_l } \der{f}{\xi } + \l [ j_1^{-2 } { r_l^2\ov ( \xi + \beta_l/\alpha_l)^2 } ( 1 + j_2 ^ 2 + j_3 ^ 2)- 1 \r ] { 1\ov r_l^2 } \delta_{\theta\varphi } f \nonumber \\ & & + j_1^{-1 } \bigg\ { 2 \bigg ( { j_2\ov r_l } { \partial^2 f\ov \partial \theta\partial \xi } + { j_3\ov r_l \sin\theta } { \partial^2 f \ov \partial \varphi\partial \xi } \bigg ) \nonumber \\ & & + \bigg [ { 1\ov r_l^2 } \delta_{\theta\varphi } r_l + j_1^{-1 } \bigg ( j_1^{-1 } ( 1 + j_2 ^ 2 + j_3 ^ 2 ) { \partial^2 r_l \ov\partial \xi^2 } - 2 \big ( { j_2 \ov r_l } { \partial^2 r_l\ov \partial \theta\partial \xi } + { j_3\ov r_l \sin\theta } { \partial^2 r_l \ov \partial \varphi\partial \xi } \big ) \bigg ) \bigg ] \der{f}{\xi } \bigg\ } \ . \label{e : reste}\end{aligned}\ ] ] in order to let appear only the operator @xmath176 in the left - hand - side of eq . ( [ e : a_tilde_poisson ] ) , we introduce a_l : = _ d_l a , and recast eq . ( [ e : a_tilde_poisson ] ) into [ e : tilde_poisson ] f = 1a_l [ ( f ) + r(f ) + ( a_l - a ) f ] . since @xmath8 appears on the right - hand - side of this equation , we solve it by iteration . besides , we introduce some relaxation in the computation of the term @xmath177 in the right - hand - side of eq . ( [ e : tilde_poisson ] ) . more specifically , we solve at each step of the iterative scheme the following equation [ e : tilde_poisson_res ] f^j+1 = ^j , where the index @xmath178 denotes the step at which the quantities are taken and @xmath179 is the following source , computed from the value of @xmath8 at the step @xmath178 : @xmath180 \big\ } \ .\end{aligned}\ ] ] in this expression , @xmath112 is a relaxation parameter ( a typical value is @xmath181 ) and @xmath182 is to be computed according to eq . ( [ e : reste ] ) . for the first step ( @xmath183 ) , @xmath184 , @xmath185 and @xmath186 are set to zero or to their value at a previous step in an evolutionary scheme . we have exposed the method of resolution of eq . ( [ e : tilde_poisson_res ] ) elsewhere @xcite . let us simply mention that we first perform a transformation from the bases in @xmath69 described in sect . [ s : expansion ] ( chebyshev polynomials in @xmath187 fourier expansion in @xmath43 ) to spherical harmonics @xmath188 , by means of a matrix multiplication onto the coefficients of the @xmath41 expansion . for each value of @xmath189 , eq . ( [ e : tilde_poisson_res ] ) gives then a second order ordinary differential equation with respect to @xmath170 , the solution of which amounts to invert a banded matrix . two solutions of the homogeneous equation ( @xmath190 ) are then added in order to connect the solution and its first derivative across the boundaries between the @xmath191 domains . more precisely , the global boundary condition , generally @xmath192 when @xmath166 , is imposed by setting the value of @xmath8 at the exterior boundary of the external domain , which is exactly @xmath193 as explained in sect . [ s : map_externe ] . the matching between the various domains amounts then to the resolution of a simple system of @xmath194 linear equations for the coefficients of the homogeneous solutions to be added in each domain . note that this matching is performed for each value of @xmath189 . the analytical properties of the source of the gravitational field at the boundary of the star depend on the equation of state ( eos ) . for a polytrope of adiabatic index @xmath195 ( @xmath196 ) , the matter density @xmath197 behaves as @xmath198 where @xmath199 is the specific enthalpy . consequently , for @xmath200 , the derivative @xmath201 has an infinite value for @xmath202 , i.e. at the surface of the star and @xmath203 diverges at surface of the star . for values of @xmath204 only derivatives of higher order diverge ( actually there exists some value of @xmath195 , e.g. @xmath205 , for which all derivatives vanish or have a finite value at the surface of the star ) . in a steady state configuration @xmath199 is taylor expandable at the neighbourhood of the star s surface ( this can be easily seen on eq . ( [ e : integ_prem ] ) below ) . therefore @xmath199 vanishes as @xmath206 where @xmath207 is the equation of the star s surface and @xmath197 behaves as @xmath208^{1/(\gamma -1)}$ ] ( this analysis remains valid even for eos more general than the polytropic one if @xmath209 is defined as @xmath210 ) . consequently @xmath197 is generaly not a @xmath1 function . this singular behaviour implies that the @xmath211 truncation error of the spectral approximation is no more evanescent and moreover that gibbs phenomenon is present . this fact is especialy awkward when studying the stability of equilibrium configurations or looking for bifurcation points because high accuracy is required . in practice @xmath195 can not be larger than @xmath6 @xcite . note that in the litterature the potential in spherical coordinates is often computed by expanding the source in spherical harmonics @xmath212 and by computing the radial part with a finite difference method . in this case the gibbs phenomenon will appear in the angular part of the solution . the situation is even worse if the radial part of the potential is computed with a spectral method . a method to recover spectral accuracy in such cases is as follows . we first introduce a known potential @xmath213 such that @xmath214 as the same pathological behaviour as @xmath197 and such that @xmath215 is a regular function ( at least more regular then @xmath197 ) and numericaly solve _ regu = n - n_div where @xmath216 . consider for instance [ e : phi_div ] _ div= f ( , , ) ( 1-^2)^(+2 ) where @xmath217 , @xmath218 is an arbitrary regular function and @xmath103 is a new radial variable such that @xmath45 at the surface of the star ( see sect . [ s : mapping ] ) . it is easy to see that @xmath219 has a term vanishing at the surface as @xmath220 ( i.e. with the same pathological behaviour as @xmath197 ) . we have indeed @xmath221 f \label{e : lap } \end{aligned}\ ] ] where @xmath222 is the laplacian computed with respect to ( @xmath223 ) ( cf . ( [ e : tilde_delta ] ) ) . the choice of the factor of @xmath224 is done in order that @xmath213 has the required regularity properties at @xmath37 and the required behaviour at the boundary of the star . the choice of @xmath225 is arbitrary . if we choose for @xmath225 an harmonic function , @xmath226 , the first term of the r.h.s . of the eq . ( [ e : lap ] ) vanishes . this is an advantage because this term can be quite large and , consequently , give rise to a large error in computing @xmath227 . we write @xmath228 , where _ lm : = ^l ( 1-^2)^(+2 ) y_l^m ( , ) and where @xmath229 are some numerical coefficients to be determined . we then obtain @xmath230 , with c_l ( ) : = ( + 2 ) . we have now to determine the values of the coefficients @xmath229 which give the most regular function @xmath231 . the criterion which seems to give the best results is the following one . we expand @xmath197 and @xmath232 as truncated series of spherical harmonics @xmath212 and chebyshev polynomial @xmath233 : n ( , , ) = _ i , l , m=0^i , l , m n_ilm t_i ( ) y_l^m ( , ) and each of the functions @xmath234 in a chebyshev series : c_l ( ) = _ i^i c_li t_i ( ) . the value of @xmath229 is computed in such a way that the @xmath235 coefficient of the truncated series of @xmath236 vanishes : a_lm = n_ilm / c_li . by means of the above procedure , we eliminate in @xmath197 the pathological term vanishing as @xmath220 but we introduce another pathological term @xmath237 . however the divergence occurs in a higher order derivative of this term so that it has a much weaker effect on the accuracy of the result . the method can be improved by taking [ e : def_k ] _ div = f ( , , ) ( 1-^2)^+2 . instead of ( [ e : phi_div ] ) . the coefficients @xmath238 are chosen in a such a way that the @xmath239 , @xmath240 derivatives of @xmath236 vanish at @xmath45 . let us call @xmath241 the _ regularization degree _ of the procedure . note that , since @xmath213 and @xmath242 vanish at the surface of the star , the boundary condition one has to impose to solve @xmath243 is the same than that for @xmath244 . we want to point out that , the above regularization technique can be used , _ mutatis mutandis _ , also when a finite difference method is used . consider two polytropic eos of adiabatic index @xmath245 and @xmath246 with a spherically symmetric distribution of the enthalpy @xmath247 . the corresponding sources density are @xmath248 and @xmath249 . figure [ f : gam30 ] shows the mass distributions @xmath197 and @xmath236 for various values of the regularization degree @xmath241 [ eq . ( [ e : def_k ] ) ] . note that in the case of @xmath250 the procedure improves considerably the behaviour of the source @xmath236 even for @xmath251 . the method can be tested in the case of @xmath245 by direct comparison with the analytical solution . in this case the gravitational field @xmath252 reads g = _ 0^r(1-u^2)^1/2u^2 du = + r ( 1-r^2)^1/2 - 2r(1-r^2)^3/2 figure [ f : erreur_regu ] shows the relative @xmath253 error @xmath254 on @xmath255 as a function of the number of degrees of freedom @xmath141 . the error @xmath254 follows approximately a power law @xmath256 . the dependence of the exponent @xmath257 with respect to the regularization degree @xmath241 is shown in figure [ f : pente ] . a value as high as of @xmath258 can be achieved with only @xmath259 . note that the relation @xmath260 is only an approximate law . this means that the error tends to become evanescent when the regularization degree increases . in this section , we sketch the general structure of a code for computing single star stationary configurations under the influence of rotation and a tidal potential . for simplicity we present only the newtonian case , the relativistic one showing no new qualitative feature but simply involving more equations . the equilibrium configuration of a cold star rotating rigidly at the angular velocity @xmath261 with respect to some inertial frame and embedded in a tidal potential @xmath262 is governed by the following three equations [ e : poisson_grav ] _ grav = 4 g , [ e : integ_prem ] h + _ grav -12 ( r ) ^2 + _ tide = const , [ e : eos ] = ( h ) . equation ( [ e : poisson_grav ] ) is the poisson equation linking the gravitational potential @xmath263 to the mass density @xmath264 . equation ( [ e : integ_prem ] ) is the first integral which can be derived from the euler equation governing the ( perfect ) fluid velocity under the stationarity assumption ; this equation relates the specific enthalpy of the fluid @xmath199 to the internal and external potentials . finally eq . ( [ e : eos ] ) is the matter equation of state in the zero - temperature approximation . the number of domains used for solving this problem is @xmath265 : one domain @xmath24 for the star ( the nucleus , cf . [ s : map_nucl ] ) , one intermediate domain @xmath266 ( cf . [ s : map_interm ] ) , the outer boundary of which is spherical and the external domain @xmath267 ( cf . sect . [ s : map_externe ] ) . in fact , if one would have to treat only the newtonian case , one domain would be sufficient ( i.e. the nucleus ) because eq . ( [ e : poisson_grav ] ) has a compact support , which is no more true in the relativistic case . a solution is specified by the central value of @xmath199 ( or @xmath264 ) , @xmath268 say , the value of @xmath261 and the expression of @xmath262 . these quantities being given , the iterative method of resolution is as follows . the @xmath191 domains are first taken to be exactly spherical . one starts from a very crude density profile , for instance @xmath269 in @xmath24 . solving the poisson equation ( [ e : poisson_grav ] ) by means of the method presented in sect . [ s : resol_poisson ] leads to the gravitational potential @xmath263 . inserting its value into eq . ( [ e : integ_prem ] ) give a new profile for the specific enthalpy @xmath199 ( the constant on the right - hand - side of eq . ( [ e : integ_prem ] ) is fully determined by the requirement that @xmath270 at the centre of the star ) . the surface of the star being defined by @xmath202 , its equation @xmath77 [ using the notation of eq . ( [ e : s_l : def ] ) ] is found by searching for the equipotential @xmath202 in the newly determined @xmath199 field . this defines a new domain @xmath24 . the corresponding mapping @xmath78 , i.e. the value of the constant @xmath62 and the functions @xmath63 and @xmath64 [ cf . ( [ e : r0=afbg ] ) ] is computed according to the procedure described in sect . [ s : map_nucl ] . the new intermediate domain @xmath266 is defined by the new inner boundary @xmath25 ( the surface of the star ) and the unchanged spherical outer boundary @xmath271 . the corresponding mapping @xmath272 is computed according to the procedure described in sect . [ s : map_interm ] . the external domain @xmath267 remains unchanged . the physical location @xmath30 of the collocation points @xmath273 ( where @xmath274 is the domain index ) corresponding to these new mappings is a priori different from that of the previous mappings , where all the fields were known . therefore , one has to compute the values of the fields at the new collocation points . in the present case , it is sufficient to do so only for the specific enthalpy @xmath199 . in the domain no . @xmath274 , the collocation point @xmath275 has the physical radial coordinate r = r_l^j(_i,_j,_k ) , where the superscript @xmath178 refers to the step in the iterative procedure : @xmath276 is the current value of the mapping of the domain @xmath19 , whereas @xmath277 is the previous value . let us denote by @xmath278 the inverse mapping at the previous step . this inverse mapping is computed by searching for the zero of the function @xmath279 . the values of @xmath199 at the collocation points of the new mapping are given by @xmath280 where @xmath281 . the value of @xmath199 on the right - hand - side is to be taken at a point which a priori does not coincide with a collocation point in @xmath103 . it is computed by a direct summation , by means the spectral expansion of @xmath199 . using the notations of sect . [ s : expansion ] , it writes @xmath282 \phi_k(\varphi ) \ , \label{e : interpol}\end{aligned}\ ] ] where the @xmath283 are the coefficients of @xmath199 is the domain no . @xmath274 . note that from the computational point of view , this summation is the most expensive operation of the method : it scales indeed as @xmath284 . it may be possible to replace the whole summation ( [ e : interpol ] ) by a truncated one or by some interpolation from the values of @xmath199 at the collocation points , in order to reduce the computational cost . the main advantage of the summation ( [ e : interpol ] ) is that it does not introduced any additional error in the method : the right - hand - side of eq . ( [ e : interpol ] ) is the value of @xmath199 at the specified point within the spectral accuracy . once @xmath199 is computed at the collocation points of the new mapping by means of eq . ( [ e : h_new ] ) , the equation of state ( [ e : eos ] ) is used to find the values of the mass density @xmath264 at the collocation points . a new iteration may then begin . in all the computations we have made , we have found that this procedure converges . for stationary rotating stars in general relativity , a rigorous proof of convergence of such iterative method ( except for the re - mapping of the physical space at each step ) has been given by schaudt & pfister @xcite . the multi - domain spectral method can handle constant density ( incompressible matter ) rotating bodies without any gibbs phenomenon . with classical spectral methods , the gibbs phenomenon would have been very severe since the density itself , and not some of its derivatives , is discontinuous across the stellar surface for incompressible fluids . this gives us the opportunity to quantify the accuracy of the method since exact analytical solutions are known for incompressible bodies : the so - called _ ellipsoidal figures of equilibrium _ ( see e.g. @xcite ) . note that an ellipsoid is not a particular case for the mapping ( [ e : r0=afbg ] ) : all the coefficients of the expansion of @xmath63 and @xmath64 onto the bases described in sect . [ s : expansion ] are non - zero . in this respect , the ellipsoidal figures constitute a strong test of the method . for single rigidly rotating objects in the newtonian regime , the more simple ellipsoidal solutions are constituted by the family of _ maclaurin spheroids _ , which are axisymmetric about their rotation axis . we have computed them by means of the procedure presented in sect . [ s : code ] , setting @xmath285 and the equation of state ( [ e : eos ] ) to be simply @xmath269 . the axisymmetry allows to employ @xmath286 . the code converges towards ellipsoidal configurations and we measure the error by comparing the eccentricity @xmath287 ( where @xmath288 and @xmath289 are respectively the polar and equatorial radii ) of the numerical solution with that of the analytical solution . the result of this comparison is presented in fig . [ f : virielml ] for a maclaurin spheroid located on the maclaurin sequence at the point where the jacobi and dedekind sequences branch off : the eccentricity is @xmath290 , which corresponds to the ratio @xmath291 . shown in fig . [ f : virielml ] is the relative error on the eccentricity as a function of the number of coefficients in the @xmath41 expansion . for these calculations , the number of coefficients in the @xmath103 expansion in each domain is @xmath292 . the straight line behaviour of the left side of fig . [ f : virielml ] shows that the error is _ evanescent _ , i.e. that it decreases as @xmath293 . for @xmath294 , the error saturates at the level of @xmath295 . this is due to the round - off errors in the computation , which is performed with a 15-digit accuracy . it is instructive to compare this result with that obtained with a classical spectral method , i.e. with fixed spherical domains , as exposed in @xcite . for instance , the fig . [ f : virielml ] can be directly compared with fig . 7 of @xcite : this latter shows an power - law error decay only ( of the type @xmath296 ) , due to the gibbs phenomenon at the star s surface . moreover , the error saturates at the level of @xmath297 . note that this result was obtained with a polytropic equation of state ( adiabatic index @xmath298 ) , for which the density is continuous across the surface of star ; the fixed - spherical - domain spectral method presented in @xcite was not able to treat incompressible fluid . also shown in fig . [ f : virielml ] is the relative accuracy with which the 3-d virial theorem is satisfied . the 3-d newtonian virial theorem states that for a stationary configuration @xmath299 , where @xmath300 is the total kinetic energy ( with respect to the inertial frame ) , @xmath301 is the integral of the pressure throughout the star and @xmath302 is the gravitational potential energy . we have computed each of these three integrals for the numerical solution and evaluated the quantity : = | 1 - 2 t + 3p |w| | . for an exact solution , @xmath303 . the triangles plotted in fig . [ f : virielml ] depict the value of @xmath304 for the numerical models . figure [ f : virielml ] shows that the virial error is very well correlated with the error evaluated by a direct comparison with the analytical solution . this gives us a strong confidence when using the virial error to evaluate the numerical error in more general cases , when no analytical solution is available . roche ellipsoids are equilibrium solutions for incompressible fluid bodies in a synchronized binary system , within the approximation of taking only the second order term in the expansion ( around the center of mass of one star ) of the gravitational potential of the companion . they are obtained by setting _ tide = - g m_comp |a| ( 1 + xa + 2x^2 - y^2 -z^2a^2 ) in eq . ( [ e : integ_prem ] ) , where @xmath305 is the abscissa of the center of mass of the companion in the cartesian frame @xmath306 centered at the center of mass of the star under consideration . note that in eq . ( [ e : integ_prem ] ) , @xmath67 must now be the distance to the center of mass of the binary system and @xmath41 the angle with respect to the rotation axis of the system . moreover , @xmath261 must be chosen so that @xmath307 , in order that the linear term in @xmath308 which appear in eq . ( [ e : integ_prem ] ) vanishes and one is left with an ellipsoidal solution . the analytical solutions for roche ellipsoid are given in the classical book by chandrasekhar @xcite . however they are given with an accuracy of five digit only ( table xvi in ref . @xcite ) , which is not sufficient for our comparison project : the accuracy achieved by the numerical code is far better than @xmath297 as we shall see below . therefore , we have written a small mathematica @xcite program to compute chandrasekhar s `` index symbol '' @xmath309 , @xmath310 and @xmath311 and obtain roche solutions with an arbitrary number of digits . figure [ f : virielro ] present the results of the comparison between the numerical solution obtained by means of the method described in sect . [ s : code ] and the analytical solution . let us recall that ellipsoidal shapes are not privileged in our formalism , so that this type of comparison constitute a strong test of our method . the comparison is conducted at fixed values of @xmath312 and the mass ratio @xmath313 . two global errors can be then defined : ( i ) the error on the axis ratio @xmath314 and ( ii ) the error on the axis ratio @xmath315 , @xmath316 being the major axis of the triaxial ellipsoid ( directed along the line of the two centers of mass ) , @xmath317 being the orthogonal axis in the orbital plane , and @xmath318 being the axis perpendicular to the orbital plane . these two errors are shown in figure [ f : virielro ] for a roche ellipsoid with @xmath319 and @xmath320 . the corresponding axis ratios are @xmath321 and @xmath322 . the numerical solution is depicted in fig . [ f : coupe_roche ] by three plane sections obtained with the following ( small ) numbers of coefficients : @xmath323 , @xmath324 and @xmath325 . also show in this figure is the numerical grid ( collocation points ) used in the problem ( only the domain @xmath24 and a part of @xmath266 are represented in the figure ) . even with such a small number of points , the relative error is of order @xmath326 ( cf . [ f : virielro ] ) ! this explains why despite the numerical grid is quite coarse , the iso - enthalpy surfaces shown in fig . [ f : coupe_roche ] are so smooth . figure [ f : virielro ] gives the two global errors as a functions of the number of coefficients in the @xmath41 expansions , @xmath136 . the number of coefficients employed in the other directions are @xmath292 and @xmath327 . as in fig . [ f : virielml ] , the exponential decay of the error for @xmath328 means that the error is evanescent . for @xmath329 , the error saturates at the level of a few @xmath0 due to the round - off errors in the computation , this latter being performed with a 15-digit accuracy . the cost in cpu time for different numbers of degrees of freedom is shown in table [ t : prix_a_payer ] . . [ t : prix_a_payer ] cpu time cost on a r4400/150mhz processor as a function of the number of degrees of freedom for the calculation of the roche ellipsoid configuration corresponding to fig . [ f : virielro ] . the iteration is halted when the relative discrepancy between two successive steps reaches @xmath330 . [ cols="^,^,^,^,^ " , ] also shown in fig . [ f : virielro ] is the relative accuracy with which the 3-d virial theorem is satisfied . this error estimator is defined in the same way as in sect . [ s : maclaurin ] . as in the axisymmetric case ( maclaurin ellipsoids ) , we find a high correlation between the virial error and the errors obtained by direct comparison with the analytical solution . we have presented a new numerical approach capable of handling the surface discontinuities of stellar configurations , provided these discontinuities are star - like , which covers a wide range of astrophysically relevant situations . when used along with spectral methods this adaptive - domain technique ensures that no gibbs phenomenon can appear . this results in a very high precision ( evanescent error ) , as demonstrated in sect . [ s : illustr ] by comparison with exact analytical solutions . the relative error for 3-d configurations can reach @xmath0 with a relatively small number of degrees of freedom ( @xmath331 in each domain ) . let us recall that very high accuracy is required for a lot of astrophysical problems such as numerical stability analysis . among these problems let us mention the study of symmetry breaking of rapidly rotating stars and the determination of the orbital frequency of the last stable orbit of a neutron stars binary system . the multi - domain spectral method is particularly well adapted to the computation of relativistic binary neutron star system . three sets of domains can be used in this problem ( see fig . [ f : m - grille ] ) : two sets of ( three or more ) domains centered on each star and a third set of ( two or more ) domains centered at the intersection between the rotation axis and the orbital plane . this latter set of domains which reaches spatial infinity is required to compute the gravitational field of relativistic configurations . when needed , the quantities computed on one of the three domain sets are evaluated at the collocation points of another set by means of the method presented in sect . [ s : code ] . we are currently applying this numerical method to the computation of steady - state configurations of relativistic counter - rotating ( i.e. irrotational with respect to an inertial frame ) neutron star binaries , following the formulation developed in @xcite . we will report on the astrophysical results in a forthcoming paper . an interesting byproduct of the present technical paper is the following one . in a previous work @xcite , we had been able to demonstrate that the virial error is representative of the true error ( measured by direct comparison with analytical solutions ) only in the spherically symmetric case . we had inferred that this remains valid in the axisymmetric and 3-d cases . in the present work , we have confirmed this conjecture , thanks to the ability of the present method to treat incompressible fluids , for which 3-d analytical solutions are available .

we detected a ring - like distribution of far - infrared emission in the direction of the center of the virgo cluster . we studied this feature in the fir , radio , and optical domains , and deduced that the dust within the feature reddens the galaxies in the direction of the virgo cluster but does not affect stars within the milky way . this is likely to be a dusty feature in the foreground of the virgo cluster , presumably in the galactic halo . the hi distribution follows the morphology of the fir emission and shows peculiar kinematic behavior . we propose that a highly supersonic past collision between an hi cloud and the galactic hi formed a shock that heated the interface gas to soft x - ray temperatures . hi remnants from the projectile and from the shocked galactic hi rain down onto the disk as intermediate velocity gas . our finding emphasizes that extragalactic astronomy must consider the possibility of extinction by dust at high galactic latitude and far from the galactic plane , which may show structure on one - degree and smaller scales . this is particularly important for studies of the virgo cluster , for example in the determination of the hubble constant from cepheids in cluster galaxies . # 1#2#3#4#5#6#7 to#2 ' '' ''

astro-ph9905295

the nature of non - luminous matter that is not part of detected and catalogued galaxies remains unsolved by modern astrophysics . as mentioned in a recent thesis , low surface brightness ( lsb ) objects may prove to be the `` icebergs '' of the extragalactic world ( de blok 1997 ) . some searches for non - luminous matter have been successful , the detection of a giant hi ring around the small group of galaxies in leo centered on m96 ( schneider 1983 ) , extended hi emission in the m81 group ( lo & sargent 1979 ) , hi companions to dwarf galaxies ( for @xmath025% of the cases : taylor 1996 ) , and a large neutral hydrogen cloud in the southern outskirts of the virgo cluster ( hi 1225 + 01 : giovanelli & haynes 1989 ) . along with hi clouds , a few large lsb galaxies have been identified : malin-1 ( bothun 1987 ) , f568 - 6 ( bothun 1990 ) , and 1226 + 0105 ( sprayberry 1993 ) . their typical star formation rates are @xmath00.1 m@xmath1/yr and the metallicities are @xmath01/3 solar . the hi rotation curves , measured by de block ( 1997 ) and by pickering ( 1997 ) , indicate that their gaseous component is dynamically significant at all radii and that the galaxies are fully dark - matter dominated ; their detected baryonic component is less than 4% of the total mass . this last conclusion is valid at least as long as we do not accept any of the more exotic theories of gravitation . the lsb galaxies lack bulges , bars , and nuclear activity , as well as co or ir emission ( have no molecules or dust ) . there have also been a few intriguing reports of presumably intergalactic dust clouds . a cloud with 0.5 - 1.2 mag of extinction was identified in microscopium by hoffmeister ( 1962 ) . three other similar objects were listed by rudnicki ( 1986 ) ; they extinguish background objects by 0.57 to 1.2 mag . in all reports the main point of contention was the actual distance to the cloud , which could put it in extragalactic space but could also locate it in the halo of the milky way ( mw ) . sometimes , the argument for an extragalactic location was based on a comparison of the properties of objects whose distance could be estimated and which were located behind the cloud with those of similar objects clearly not within the cloud limits ( rr lyrae stars ; murawski 1983 ) . the extragalactic nature is only fairly confidently established for the abadi - edmunds cloud at @xmath03 mpc ( abadi & edmunds 1978 ) . hi 21 cm line emission was detected from this object , whereas in other cases it was not . however , in other cases far - infrared ( fir ) emission was detected and could be identified ( on morphological and positional criteria ) with the obscuring clouds . fir and hi emission were clearly detected in the case of the okroy cloud ( wszolek 1988a , 1989 ) . fir emission was only marginally detected from the rudnicki - baranowska cloud ( wszolek 1988b ) . this indicates that the physical conditions in this kind of objects are far from being uniform . more such examples must be identified and their properties examined . it is possible that the phenomenon of intergalactic hydrogen clouds could be related to the high - velocity cloud ( hvc ) complexes . these are hi structures whose radial velocities deviate by several 100 km s@xmath2 from the conventional galactic rotation . a recent review of hvcs is by wakker & van woerden ( 1997 ) . their table 2 lists a few cloud complexes at distances @xmath325 kpc ; some of these may not belong at all to the mw . iras searches for fir emission of hvc were negative ( wakker & boulanger 1986 ) , indicating that either the hvcs are dust - free or that their dust grains are much cooler than could be detected with iras . in this context we also mention the proposition by blitz ( 1999 ) that the hvcs make up the missing mass by being essentially dark halos with low velocity dispersions . we report here results from a study of a diffuse ring - like fir feature at high galactic latitude , which we interpret as `` local '' , not extragalactic , despite first indications to the contrary . the region toward which this feature is located is the center of the virgo cluster of galaxies . this part of the sky has been studied in exquisite detail , yet new studies always detect interesting features . for example , katsiyannis ( 1998 ) produced a very deep image of the central regions of the cluster from a combination of 13 deep kodak techpan films obtained with the uk schmidt telescope . the image shows large variations in the brightness of the intra - cluster medium , with the brightest regions north of the cluster center . m87 is fairly central in the region of enhanced brightness , close to the upper left corner of the `` very high contrast image '' in their fig . previous deep imaging of the central vc region ( weil 1997 ) revealed a diffuse extension of ( presumably stellar ) material extending @xmath0100 kpc to the se of m87 . intergalactic red giant stars were apparently discovered near m87 by ferguson ( 1998 ) . it is therefore relevant to search for , and to try and explain , any extended feature one may detect in the direction of the center of the cluster . in this context , we mention the study of haikala ( 1995 ) who examined the uv emission detected in the direction of a dust globule close to the north galactic pole , slightly north of the virgo cluster ( vc ) . any material that could produce extinction needs to be accounted for . to the best of our knowledge , nobody attempted to study the obscuration and fir emission by ism or igm in the direction of a rich , nearby cluster of galaxies . this is particularly important for the vc , which serves as one of the key stones in the distance ladder leading up to the determination of the hubble constant ( van den bergh 1996 ) . the hst key project on the extragalactic distance scale , where the required accuracy of the determination of h@xmath4 is 10% , could be affected significantly by unaccounted extinction . until now , seven galaxies within 10@xmath5 of the virgo center have been observed for cepheids in this context ( macri 1999 ) . the plan of the paper is as follows : we first describe the fir observations , which revealed the feature , and present confirmatory evidence of its reality . we then attempt to derive additional properties of the feature , which has an approximate ring shape , using data in the optical and radio domains . we show that the dust in the feature does not seem to affect the stars in the milky way but that it apparently reddens galaxies in the vc and beyond . the full data set is discussed in the last section of the paper , in which we also derive some properties of the dust grains in the feature . far infrared ( fir ) observations from the cobe satellite , specifically with the dirbe instrument , reveal non - uniform fir emission from the center of the vc . the dirbe instrument mapped the entire sky at ten wavelength bands from 1.25 to 240@xmath6 m and operated from november 1989 to december 1993 ( cryo - cooling was available only for ten months , restricting the availability of the fir channels ) . an important feature of dirbe was that the measurements were performed against an internal calibrator source , with proper accounting for instrumental offsets and interplanetary fir emission . for the present analysis we used the annual average sky maps ( aasm : hauser 1997 ) , which provide a single , ten - month averaged intensity value per pixel in each of the dirbe bands . note that the zodiacal light contribution was not subtracted from the dirbe counts . this is because we do not estimate the zodiacal contribution to the fir bands to be significant or to show features on the angular scales relevant here . we conducted a number of studies of galaxies in the virgo cluster in which we studied various photometric indices for entire objects as well as for localized regions in each galaxy ( almoznino & brosch 1998 , heller 1998 ) . the possibility that these programs could be affected by foreground dust imposed our selection of the virgo cluster as the initial target for combined fir and other spectral band interpretations . we detected a ring - like structure of fir emission in cobe / dirbe maps of the vc , which is centered approximately on m87 . the ring is approximately centered on ( 1950 ) 12@xmath731@xmath8 ; + 13@xmath5 ( [email protected] , b=75@xmath5 ; j2000 ) and its diameter is @xmath9 . the width of the fir emission in the rim of the ring is @xmath10 . the detection was made originally on the cobe / dirbe maps , but the existence of the feature was established also on iras maps ( see below ) . the m87 galaxy ( [email protected] , [email protected] ) is normally taken as the center of the virgo cluster ( vc ) and one could imagine scenarios by which some sort of fir - emitting matter could be distributed around it . for this reason , we decided to follow the fir detection of the feature , which we call here `` the virgo ring '' ( vr ) , and investigate it further . the detection was made on the aasm , which have noise levels of 3 10@xmath13 mjy sr@xmath2 at 100@xmath6 m , 0.6 mjy sr@xmath2 at 140@xmath6 m , and 0.3 mjy sr@xmath2 at 240@xmath6 m ( kashlinsky 1999 ) . the ring is visible even by superficial inspection of these cobe / dirbe gray scale maps . no traces of the ring can be seen on 60@xmath6 m or shorter wavelength maps . to obtain detailed insight into the structure of the vr we produced isophotal maps at @xmath14=100 and 240@xmath6 m using the original [email protected] square pixels , which are shown as isophote plots in figure 1 . the 100@xmath6 m map shows a region of depressed fir flux where [email protected] mjy sr@xmath2 . this is surrounded by regions of enhanced fir emission , which reach f@xmath1510 mjy sr@xmath2 . the 240@xmath6 m map indicates that the region of reduced fir emission has [email protected] mjy sr@xmath2 while the surrounding regions have f@xmath165 mjy sr@xmath2 . it is clear that ( a ) the dirbe data indicate a region of low fir emission surrounded by enhanced emission , and ( b ) the feature is real , because it appears on more than one dirbe map . the lowest values of the fir flux originate presumably from the zodiacal light that was not subtracted from the aasms and from the cosmic fir background . as both these components are much smoother than the feature we describe here , there is no need to model them in detail . the peculiar fir features detected by cobe / dirbe are confirmed by iras measurements . the iras mission mapped the sky in four wavelength bands from january 1983 to november 1983 . the primary goal of the iras survey was the detection of point sources , but a catalog of extended sources has also been produced , as well as sky brightness images in each of the four bands with 2 pixels and 4-6 resolution ( beichman 1988 ) . iras 60 and 100@xmath6 m extended emission data in the 16@xmath17 square fields no . 83 and 84 were used to confirm the existence of the ring and to exclude the possibility of instrumental artefacts produced by the cobe / dirbe instrument . we created maps at these two spectral bands with a 4@xmath184 beam . the vr is clearly visible on the 100@xmath6 m map shown in fig . a similar 100@xmath6 m map based on iras observations , and where this feature is also visible , was reproduced already by leggett ( 1987 ) as their plate 2 . the enhanced iras resolution relative to cobe / dirbe allows a good morphological evaluation of the fir feature . in addition to the north - westerly extension of the fir emission , along the iras scan direction , one sees an arc - like distribution of emission , which could be interpreted as forming an elliptical ring . note that the feature is visible only on the 100@xmath6 m map ( shown in figure 1 ) and is not seen on the 60@xmath6 m map , or on those at even shorter wavelengths ( not shown here ) . although the low resolution cobe / dirbe maps seem to indicate that the fir emission is arranged in a ring , with low fir at the center and high emission on its perimeter , the higher resolution iras maps show that this is not the case . the fir emission is distributed in an open configuration , with a region of low emission centered on @xmath19 , + [email protected] . the fir emission could best be described as a fork , or a two - arc shape limited to @xmath20=185@xmath21 . the eastern side of the feature shows a small region of enhanced fir emission centered on @xmath20=185@xmath5 and @[email protected] that stands out over its surroundings and to which we refer as the `` main blob '' ( mb ) . the dust revealed by the fir observations may ( a ) extinguish and ( b ) redden stars behind it . the first effect is a consequence of the `` total extinction '' property , whereas the second is the result of `` wavelength - selective extinction '' . the relative importance of the two effects is linked through the parameter @xmath23 , which is determined to first order by the size of the dust grains . we tested two assumptions , one of extinction within the milky way ( mw ) that would affect some of the stars but not others , and a second that the vr is extragalactic and is located between the mw and the vc . in the second case it would affect the vc galaxies , but none of the mw stars . for testing the possibility that the dust is `` local '' one requires a large number of stars with magnitudes and colors . these were extracted from the usno - a2.0 catalog , which includes blue and red magnitudes for each star . the usno - a2.0 catalog contains @xmath245 10@xmath25 objects ( @xmath012,750 per square degree ) and is based on scans of the palomar sky survey ( pss ) plates produced with the precision measuring machine ( pmm ) . the catalog is an improvement over the version 1.0 both in astrometric accuracy and in photometric precision . the photometric accuracy is probably not better than @xmath00.15 mag , but the depth of the catalog is considerable , as it reaches 20 - 22 mag ( color - dependent ) . it can , therefore , serve as a source of stellar objects with which one can test the assumption of foreground extinction . we extracted objects in a number of @xmath26 regions from the usno - a2.0 catalog . the extraction locations are listed in table 1 and correspond to some fir - bright regions ( where we expect a higher density of extinguishing dust ) or to some fir - faint regions ( which should be @xmath0transparent ) . we produced wolf diagrams for each location , and show these in figure 2 . the wolf diagrams plot the cumulative distribution of stellar magnitudes against magnitude , and the signature of total extinction in such a plot is a step - like deviation , to fainter magnitudes of the cumulative star counts , from the pattern set by the brighter ( and closer , on average ) stars . the diagrams do not show such a step - like trend for regions in the direction of stronger fir emission when compared with the behavior of the cumulative distribution in regions with lower fir emission . it is also possible to compare the measured behavior of the cumulative star counts with that `` predicted '' in absence of localized extinction effects by using a model for the stellar distribution in the galaxy for the same milky way locations as sampled here . a very successful and intuitively simple stellar distribution model was produced by bahcall & soneira ( 1984 ) and is available on - linejnb / html / galaxy.html ] . we calculated predicted star counts for the locations of the extracted data from the usno - a2.0 catalog using the version of the model retrieved in december 1998 . the locations are listed in table 1 . we compared the predicted cumulative star counts with the actual star counts . the comparisons are shown in figure 3 and show no significant deviations from the predicted behavior . the exercises shown in figs . 2 and 3 indicate that the stellar distributions are not influenced by the material producing the fir emission . the conclusion is , therefore , that this material is either extremely nearby , so that all the stars are affected in the same manner , or that it is very distant , beyond the more distant stars listed in the usno - a2.0 catalog . if the dust observed in the fir does not affect stars in our galaxy , it may be located far from the mw and could affect only objects seen behind it . testing the assumption of a dust cloud distant from the mw requires a sample of background objects with relatively high surface density , as well as brightness and color information . in the virgo region , the `` standard '' extragalactic catalog has been for a number of years the binggeli ( 1985 ) virgo cluster catalog ( vcc ) . the vcc covers @xmath0140 square degrees and contains 2096 galaxies . the surface density of galaxies is , therefore , @xmath015 galaxies / square degree , on average . while this may appear sufficient , the photometry is not adequate because the galaxy magnitudes in the vcc are eye estimates and may have significant deviations . in addition , no colors are available for most vcc galaxies . we decided therefore to rely on a more recent galaxy compilation , which reaches deeper in brightness and is thus denser than the vcc , has better photometry , and contains color information for the objects . currie & young ( 1998 , hereafter vpc ) produced an extensive three - color photometric catalog of galaxies in the central regions of the vc . the catalog is based on cosmos scans of one u plate , two b@xmath27 plates , and one r@xmath28 plate , all obtained with the uk schmidt telescope . the plates were photometrically calibrated and objects were extracted automatically , with stars and galaxies separated by an automatic algorithm . the vpc provides an impartial survey of galaxies in the region of interest for the present study , which reaches to b@xmath2919 mag , thus it is somewhat shallower but comparable in depth with the stars from the usno - a2.0 catalog . the area covered by the vpc covers 23 square degrees and is centered on ( 1950 ) 12:26 + 13:08 . the average galaxy surface density is therefore 49 galaxies / square degree , considerably more than that of the vcc . we attempted to detect total extinction effects on the vpc galaxies by limiting the analysis to regions with high fir emission and comparing these with similar analyses in the direction of regions with lower fir emission . four parallerogram - shaped fields were selected , marked a , b , c , and d on figure 4 . fields c and d are used to determine the nature of the galaxy population in the general region of the vr . field a could also be used for this purpose , but we caution that a background cluster of galaxies ( abell 1552 at [email protected] ) is located in this field and thus region a may not be representative . this galaxy cluster is presumably part of a background sheet - like complex , which includes also abell 1526 at a very similar redshift . field a was selected to offer insight on how the presence of background galaxies disturbs the results . the enhancements of the galaxy background may distort the wolf diagrams of galaxies ( figure 5 ) , and indeed some fir enhancements could be in the direction of these galaxy clusters . however , searches for dust in clusters of galaxies have , so far , been negative ( maoz 1995 ) . thus , we may tentatively discount the fir enhancements in the direction of background clusters of galaxies as chance superpositions . field b is considered not to be affected by absorption / extinction and is in the direction of the low fir emission of the vr . attempts to detect the presence of dust as a `` total extinction '' effect , which modifies the cumulative galaxy counts between the different regions , were not successful . the differences were not significant and indicate that if dust is present , it may cause at most a small amount of total extinction : [email protected] mag . we therefore checked for the presence of color - dependent extinction by studying the distribution of the ( u r@xmath28 ) color index in one square degree areas over the central part of the virgo cluster . the data used for this test , and an extensive description of the method and results , are given in the appendix to this paper . here we emphasize that the results show that the galaxies in the direction of the virgo ring ( vr ) part with the lowest fir emission appear slightly bluer than those in the direction of the two regions with higher fir emission . the difference is significant to @xmath395% . interpreted as dust extinction , this difference in average ( u r@xmath28 ) color index indicates a possible wavelength - dependent extinction of @xmath32(u r@xmath28)@xmath330.3 mag between areas with high fir emission and areas with less dust , a total extinction [email protected] for a typical milky way extinction law , although this was not checked here . here we show that x - ray observations of the region indicate a two - component makeup for the hot gas , and that the morphology and kinematics of the hi are peculiar . bhringer ( 1995 ) mapped the x - ray emission from the immediate vicinity of m87 ; this is the region of interest of the present study . their findings show the presence of thermal x - ray emission from cooler gas than the intracluster medium . a rosat map of the general region , larger than the one analyzed in the 1995 paper , was presented by bhringer ( 1994 ) and shows a ridge of x - ray emission which approximately coincides with the fir emission ridge to the west of m87 . they mention , in particular , the sharp drop in x - ray intensity on the western side of m87 . bhringer ( 1994 ) subtracted a model distribution of x - rays from m87 from the rosat map and derived a residual map ( their fig . 2 ) which shows the background cluster a1552 at 12:30 + 11:30 and a long filament , which is elongated @xmath0north - south at @xmath3512@xmath730@xmath8 and from @xmath36 + 15 to + 6 . this filament curves around m87 on its westerly side and seems to follow the contours of the 100@xmath6 m emission . it is tempting to speculate on a possible link between the x - ray and fir emission presented above , but we caution that this may not be real . one possible factor affecting the morphology of x - ray emiting gas is the amount of foreground hi , which modifies mainly the low energy end of the x - ray spectrum . shadows in the x - ray background caused by foreground hi clouds have been detected mainly in soft x - rays by egger & aschenbach ( 1995 ) . however , the feature detected in the rosat maps by bhringer ( 1994 ) is seen in the hard energy band ( 0.42.4 kev ) , and is thus difficult to attribute it to gas absorption . euve observations of the virgo cluster ( vc ) center ( lieu 1996 ) show the presence of gas at @xmath00.5 10@xmath37 k near m87 . this matter forms an additional component of the intra - cluster material ( icm ) in virgo , as follows from their analysis , and can not be the same hot gas which is responsible for the x - ray emission detected by rosat . in order to confirm the existence of this second icm component of the vc , lieu ( 1996 ) performed hi 21 cm observations with the 43-m green bank telescope ( angular resolution 21 ) . the region surveyed by them was centered on m87 , had an extent of 2@xmath38 , and the grid of hi measurements was spaced every 8@xmath331/3 of a resolution element . a comparison of the hi map of lieu ( 1996 ) with the fir distributions ( see figure 1 ) demonstrates that the fir emission follows the total hi column density . although lieu do not mention the velocity range over which the green bank observations were performed , we assumed these to be at @xmath00 km s@xmath2 because they are supposedly of `` galactic hi '' origin . while some vc galaxies do have negative heliocentric velocities ( c.f . binggeli 1985 ) , they mostly concentrate at 1,0002,000 km s@xmath2 . for this reason , we think it likely that the hi detected by lieu ( 1996 ) does indeed belong to the mw and , by inference , so does the material producing the fir emission . we note at this point that the center of the low n(hi ) region , at 12:28 + 12:45 ( 1950 ) and only @xmath0half a degree away from m87 , has n(hi)@xmath331.8 10@xmath39 atoms @xmath40 . this is coincident with the low fir emission region . the ridges with the higher n(hi ) values correspond to enhanced fir emission regions . we produced n(hi ) plots for the region using data from the leiden - dwingeloo hi survey ( hartmann & burton 1997 , lds ) in order to confirm the hi distribution measured by lieu ( 1996 ) . the lds was conducted with the 25-m radio telescope at dwingeloo and the data we used cover the velocity range 459@xmath41 + 415 km s@xmath2 with a resolution of 1.03 km s@xmath2 . the 25-m radio telescope has a 36 half - power beam and the survey was performed with [email protected] spacings . we used the file total_hi.fit from the cd - rom supplied with the printed atlas to extract the proper sky region . the data were transformed from galactic to equatorial coordinates , accounting for the change of scale from one side of the image to the other . this was done by dividing each pixel value by its _ cos(b ) _ , to yield consistent units over the field . the hi total column density from the lds is shown in figure 6 together with the iras 100@xmath6 m map and confirms the general impression from the lieu ( 1996 ) map . the hi distribution has a region with lower n(hi ) at the center of the virgo ring ( vr ) and ridges of higher hi emission on both sides of the vr . we also produced position - velocity ( pv ) plots using the channel data from the lds , limiting these to l=276.0 , the galactic longitude of the hi peak which coincides with the main fir emission blob in area a of fig . 6 ( the more intense of the fir peaks ) , to the center of the vr ( l=283.0 ) , and to the second highest fir peak at l=290.5 . the pv plots are shown in figure 7 and indicate that at the position of the vr there is a significant disturbance of the hi , with a strong extension to negative velocities appearing in the pv plots of the high - fir region . the sheet - like hi distribution , which links hi at low latitudes with gas near the galactic pole and has a slightly negative lsr velocity , appears disturbed at b@xmath3075@xmath5 . the velocity plot through the peak emission at l=276.0 at this latitude ( figure 8) shows three peaks separated by @xmath020 km s@xmath2 . the strongest has the most negative velocity , approximately 30 km s@xmath2 and a fwhp of @xmath011 km s@xmath2 . the weakest peak at this location is near + 4 km s@xmath2 ( lsr ) . the pv in the low fir region at the center of the vr ( l=283@xmath5 , b@xmath3075@xmath5 ) shows a single strong peak at @xmath07 km s@xmath2 ( lsr ) , with a fwhp of 12.5 km s@xmath2 and a low shoulder extending to more negative velocities , down to the velocity of the strong peak at the location of the main blob ( 30 km s@xmath2 ) . the third pv , at ( [email protected] , b@xmath3075@xmath5 ) is narrow with a fwhp of @xmath0 5 km s@xmath2 and is centered at 7 km s@xmath2 ( lsr ) . we identified a ring - like feature of fir emission at high galactic latitude , which is distant from the main body of the galaxy and extinguishes light from galaxies in the central part of the virgo cluster ( vc ) . there is no way to establish a distance to the extinguishing cloud with the data we presented above , except to note that it is probably @xmath241 kpc . a nearby dust feature , observed by haikala ( 1995 ) in the far - uv , has been located at @xmath0120 pc using the distribution of e@xmath42 color excesses . this dust cloudlet produces a visual extinction [email protected] mag and is located at ( l=251.1 , b=+73.3 ) ; this location is very similar to what we found for the virgo ring ( vr ) and may indicate that either our distance evaluation is wrong , or that the location technique of the haikala feature did not use a sufficient number of more distant stars . indications that the dust cloud can not be a nearby feature originate mainly from the lack of influence on the distribution of stars . supporting evidence to the same comes from the reddening study of knude ( 1996 ) . he used uvbyh@xmath44 measurements of a3-g0 stars with [email protected] mag and @xmath46b@xmath47 to determine the distribution of extinction . his results for e@xmath42 , broken by galactic latitude and by longitude quadrants , are of particular interest . the area of interest for our study is located between the 3rd and 4th quadrants at b@xmath48 ; the reddening to this region is small , e@xmath49 , which translates into [email protected] . the stars studied by knude ( 1996 ) are closer than 1.5 kpc ( for main sequence a stars brighter than 11.5 mag ) , thus the color - dependent extinction of the vc galaxies we detected , which is equivalent to [email protected] mag , should be produced by material more distant than 1.5 kpc . if the cloud would be in the vc itself , its physical size would be @xmath01.5 mpc , very large indeed ! the issue of possible diffuse dust in clusters of galaxies has been studied by ferguson ( 1993 ) . he concluded , from the lack of a difference between cluster and field galaxies in the correlation of the mg@xmath51 index and ( b v ) , that dust is not present in the virgo cluster ( upper limit e(b v)@xmath520.06 mag . ) . a similar conclusion for a large number of abell clusters , based on the ( v i ) color indices of radio quasars seen in their background , was reached by maoz ( 1995 ) . not accounting for foreground dust may affect adversely some key observations . our finding confirms the supposition of zonn ( 1957 ) and zonn & stodolkiewicz ( 1958 ) , that because of the patchy structure of the interstellar dust _ it is not enough to correct for extinction assuming that the dust is localized in a narrow slab near the galactic equator , but the detailed distribution of dust must be investigated to account properly for the extinction . _ in particular , many observations of the virgo cluster and of objects within ( the hst cepheid key project : graham 1998 , macri 1999 ) may carry significant errors because of improper extinction corrections . in this section we estimate the dust temperature and dust - to - gas ratio . to evaluate the temperature of the dust in vr we subtracted from the map intensities the minimal value for the central part of the vr , near m87 , in the 100 and 240@xmath6 m cobe / dirbe bands ( 8 and 3.5 mjy sr@xmath2 respectively ) . to determine the color temperature for the dust in the vr cloud we assumed that the dust particles are in thermal equilibrium and that the foreground galactic ir radiation and ir emission from all point sources in the region have been subtracted accurately . these assumptions may not be necessarily fulfilled in our case . the subtraction of the intensity of the inner part of the ring would be accurate only if the distribution of the foreground galactic radiation is fairly smooth ; this is not the case even at high galactic latitudes . some galactic features may add non - negligible fir contributions to the foreground . the radiation from very cold dust grains ( t@xmath303k ) could not be detected by the means used here . we could not rule out the possibility that transient heating of dust grains takes place , and that only occasional excitation by energetic photons or particles causes them to emit brief pulses of the radiation measured by cobe / dirbe and iras . with all these caveats in mind , we calculated the temperature of two regions with maximal fir intensities within regions delimited by : @xmath20 : 185@xmath5186@xmath5 , @xmath22 : 13@xmath514@xmath5 ( approximately the main blob = mb region ) and @xmath20 : 189@xmath5190@xmath5 , @xmath22 : 12@xmath513@xmath5 ( slightly off the secondary fir peak ) . the temperature was calculated with the relation : @xmath53 from schlegel ( 1998 ) , where r=@xmath54 is the ratio of the fir intensities in the cobe / dirbe bands . the results are 22 and 20k for the first and the second region , respectively . these values can be accepted as upper limits for the dust temperature in vr due to both effects mentioned above . the temperatures do not differ significantly from those usually adopted for interstellar dust clouds and are at the high end of the range for the `` warm '' cirrus component ( lagache 1998 ) . the difference between the two regions , with the mb being `` hotter '' than the secondary fir blob , is marginally significant , because the formal error in the derivation of temperatures is 1 - 2k . a study of the optical depth effects of high latitude clouds ( chlewicki & laureijs 1988 ) found that typically @xmath55=0.7 mjy ster@xmath2/10@xmath39 atoms @xmath56 . deul & burton ( 1990 ) studied the hi content and gas kinematics of seven cirrus clouds detected in the iras maps . they established that the fir to hi ratio varies between 0.9 and 3.0 mjy ster@xmath2/10@xmath39 atoms @xmath40 . the color temperature varies much less , between 0.20 and 0.27 for @xmath57 . unfortunately , we did not detect 60@xmath6 m emission from the vr , thus a direct comparison with the color temperature derived by deul & burton ( 1990 ) is not possible . however , the ratio @xmath55 we find for the fir ridges ( @xmath01.5 mjy ster@xmath2/10@xmath39 atoms @xmath40 ) is in the range of that measured for cirrus clouds . we determined n(hi ) from the brightness temperature using data from fig . 9 and applying the conversion @xmath58 assuming that the hi is optically thin . the vr is located close to the northernmost point of the north polar spur ( nps ) and could , in principle , be part of it . the nps is presumably the remnant of a @xmath015 myr old explosive event in the direction of the galactic center , which released 10@xmath59 ergs and could be the outcome of @xmath60 supernovae produced in a small region within one million years . the giant loop is detected in a number of wavebands , from the x - ray to the radio continuum ( sofue 1994 ) . the possibility that the vr is part of a small structure , extending to lower galactic longitudes from near the northernmost apex of the large nps feature in sofue s fig . 2 ( plate l11 ) , can not be excluded . it is also clear that the virgo ring ( vr ) is not a known high velocity cloud , because the compilation of deul & van woerden ( 1990 ) lists as nearest object hvc 2 at 12:09 + 15:32 and v@xmath61=100.7 km s@xmath2 , covering 3.2 square degrees of the sky . wakker & van woerden ( 1997 ) list a possible small hvc near the vr location . this is an hi feature associated with optical absorption lines observed in sn 1994d , located at [email protected] ; [email protected] and @xmath0240 km s@xmath2 lsr . this sn occured in ngc 4526 , and the absorbing gas ( with five absorption systems ) was found by ho & filippenko ( 1996 ) at the same velocity as the hi measured by kumar & thonnard ( 1983 ) . ho & filippenko mention other similar absorption features in the spectrum of sn 1991 t ( at l=292.59 ; b=65.18 , in ngc 4527 ) , which is @xmath62 away from sn1994d . it seems that there is material in the general direction of the vr with lsr velocities a few 100 km s@xmath2 , and that it contains at least ca ii and na i. in the vicinity of the vr , velocities in excess of 100 km s@xmath2 are required in order to qualify an hi feature as a hvc . our analysis of the hi distribution and kinematics , using the lds survey data , indicates that there is a significant disturbance of the hydrogen at the location of the vr . in particular , the three - peak structure of the velocity - column density is peculiar . there are a number of possible explanations for this hi peculiarity , starting from an expanding supernova ( sn ) shell , a collision between galactic hi and a high - velocity cloud , an ejection of a number of shells from a red ( super)giant , etc . each of these possibilities imposes some constrains on the problem . of all these possibilities , we consider the most likely to be that of a past collision between a small hi cloud and galactic hydrogen . the hi remnant of this event could be the feature seen at 30 km s@xmath2 in the high - fir area and the galactic hi would be the prevalent emission at 7 to 9 km s@xmath2 , which shows up in all plots . at a velocity difference of @xmath020 km s@xmath2 the collision would be highly supersonic . we note that the immediate vicinity of the vr has been studied in exactly this context by stark ( 1994 ) . the region most likely to be part of the same complex is the bx field ( see table 1 of stark 1994 ) , where the x feature peaks near 8 km s@xmath2 ( lsr ) but there are parts of the b feature where velocities up to 40 km s@xmath2 are observed . stark ( 1994 ) interpreted the `` intermediate negative velocity '' ( inv ) gas , with v@xmath4520 km s@xmath2 , as material at local velocity displaced by a shock following a collision with a high velocity cloud . the difference in lsr velocity between the bx features and candidate hvcs in this vicinity ( at a few 100 km s@xmath2 positive ) forced stark to propose scenarios for the dissipation of the impinging hvc . they asserted that this cloud has evaporated and is now present as high - temperature gas . collisions of hvcs with the galactic disk have been studied by kerp ( 1996 ) . they show that a collision with a differential velocity of 25 km s@xmath2 may increase the temperature of the post - shock material from 10@xmath63 to 10@xmath64k ; such temperatures are consistent with the euve measurements ( lieu 1996 ) . it is possible to estimate the parameters of the main blob ( mb ) using the hi information from the ldss . with the t@xmath65 value from 30 to 20 km s@xmath2 we find @xmath66 this implies a gas - to - dust ratio for the mb of @xmath675 10@xmath39 @xmath40 mag@xmath2 , different from the canonical value for the galaxy ( @xmath68=5.4 10@xmath69 @xmath40 mag@xmath2 ; bohlin 1975 ) . the mb appears , therefore , to be dust - rich ( or hi - poor ) . its projected angular size is @xmath70 ; this translates to a size of 10(@xmath71 ) pc and a total hydrogen mass of @xmath72 with an average ( volume ) density of @xmath04 @xmath73 cm@xmath13 assuming a spherical configuration . the cloudlet carries significant kinetic energy , considering its velocity relative to the galactic hi ( assumed to be @xmath75 km s@xmath2 ) ; e@xmath763 10@xmath77 ergs . also , the hi profile observed in its direction ( see top panel of fig . 9 ) is significantly wider than that of the second blob ( bottom panel of fig . this could be an indication of higher turbulence within the mb relative to other hi entities in the same vicinty but at less negative lsr velocities . it is significant that the mb , an infalling hi gas cloudlet , contains dust grains as evidenced by the extinction it produces . this indicates that any shocks that the material in mb might have encountered did not heat up its material to temperatures high enough to completely destroy the grains . however , its highly supersonic interaction with the ambient hi distribution caused a wider 21 cm profile . the mb produces significant extinction and is fairly distant from the mw plane while having an intermediate lsr velocity . this is different from the findings of knude & h@xmath78 g ( 1999 ) , who concluded that intermediate - velocity hi clouds show no extinction . the detection of hi condensations on sub - degree scales and high above the galactic plane is relevant for the studies of galaxies in the virgo cluster and beyond . a spectroscopic survey of objects in the background clusters at [email protected] could reveal absorption lines produced by material within the mb cloudlet . this , along with a measure of the ionization presumably taking place at the interface between the mb and the ambient hi , could reveal its composition and location . the essential finding is , however , the small scale on which significant extinction variations are encountered at high @xmath46b@xmath46 . this shows that the derivation of `` average extinction dependences '' , with galactic latitude ( knude 1996 ) , is not very relevant to the determination of cosmologically - important parameters from studies of individual vc galaxies . our study shows that dust can affect the light of background galaxies in the vc and , through this , the determination of h@xmath4 by using cepheid photometry . this hst key project relies on a canned approach to deredden individual stars and is based on the `` standard '' mw extinction law with r=@xmath79=3.3 ( c.f . madore & freedman 1991 ) . it remains to be proven that this relation applies for high galactic latitude dust clouds , such as those studied here . we detected a dusty hi cloud seen in projection against the central regions of the virgo cluster . we showed that the cloud , which appears ring - like in low - resolution fir and total hi maps , has a complex fir morphology when examined with higher resolution . the hi kinematics indicate a substantial disturbance at this location . the virgo ring ( vr ) is located beyond the main body of the galaxy , because the distributions and colors of stars in its direction are not affected by it . the cloud could not be more distant than the vc , because it influences the colors of galaxies within the cluster . the hi evidence argues strongly for a relatively nearby location . the connection of the ring with the north galactic spur may be accidental , but the peculiar morphology and kinematics of the hi associated with the fir emission make a connection , perhaps through a collision between an hi cloud and galactic hi , more likely . such a collision could perhaps explain also the extreme - uv ( or soft x - ray ) emission observed from the neighborhood of m87 . the importance of our finding lies in the possibility that many key observations done in the direction of the virgo cluster may have been adversely affected by extinguishing clouds , which are not homogeneous and where the total v - band extinction may reach up to a few tenths of a magnitude . nb acknowledges support from the israel science foundation . ea is supported by a grant from the israel ministry of science to develop tauvex , a uv imaging telescope . kr and bw were supported by the polish kbn grant number c76/98 . we are grateful to drs . c.k . young and m.j . currie for supplying electronic versions of the vpc data tables in advance of publication . nb is grateful to mike hauser for discussions on the reality of cobe / dirbe features , to john bahcall for allowing public use of his galaxy model , and to sara beck and sasha kashlinsky for critical readings of one of the first drafts . classical wolf diagrams for galaxies ( zonn & rudnicki 1965 ) were made for the regions labelled a through d in fig . 4 and are shown in figure 5 . only curve c could be interpreted as shaped by extinction , and this only by a large amount of leeway . curve a , as already cautioned , has a completely different shape . both a and d curves are higher than the comparison curve b , and are presumably shaped by the inhomogeneous distribution of galaxies in the vc area . if , however , we would adopt curve c as representative for the background galaxies ( no vc subclusters or background galaxy clusters are evident in this area ) , then the obscuring cloud would have to be relatively nearby , as no parts of the c and b curves overlap . to prevent distortion of the results by possible subclustering , we decided to study the color distribution of the background galaxies . the advantage of using color is that this parameter is distance - independent , provided the objects of study are not `` too distant '' , in the sense of requiring a k - correction . we assume that the objects in the vpc conform to this constraint and do not attempt to correct the color for redshift ( which is not known , in most cases ) . we also caution about `` edge effects '' , which could arise because of a smaller population of the sub - areas located near the edges of the vpc coverage ; such areas cover less than one square degree , have less than the average number of galaxies , and would be less representative . we analyzed the color distribution of all galaxies in the entire vpc . we selected objects only by their position ; specifically , we included all the objects listed in the catalog with their magnitudes and colors , without separating them by morphological type , apparent blue magnitude , angular size , being members or non - members of the cluster , etc . the area enclosed within 184@xmath80 and 10@xmath81 was divided into square degree areas at integer @xmath20 and @xmath22 values . in each sub - area we calculated the statistics of the color distribution for the galaxies . the total number of galaxies per cell does not appear to show a specific pattern , apart for the edge effects mentioned above . we checked the distribution of photometric indices over the sub - areas of the vpc against the locations of the enhanced fir emission . in each one square degree area we counted the galaxies , calculated their mean b@xmath27 magnitude and the average ( u b@xmath27 ) , ( b@xmath27r@xmath28 ) , and ( u r@xmath28 ) color indices , and studied the statistical properties of the distributions of the color indices in each area . the color index ( u r@xmath28 ) has been calculated as ( u - b@xmath27)-(b@xmath27-r@xmath28 ) . the results are presented in tables a1 and a2 for ( u r@xmath28 ) , where each cell contains these parameters for each one square degree region . the results for the other color indices were similar and are not shown . each sub - area in table a1 is represented by one cell in the table , where the rows are the declination of the sub - areas and the columns are the right ascension , both in decimal degrees . the distribution of the ( u r@xmath28 ) color is represented by the mean , the standard error of the mean ( in round brackets ) , and the number of galaxies in the sub - area [ in square brackets ] . the standard error of the mean @xmath82 is defined as : @xmath83 here @xmath84 is the standard deviation of the values in sub - area _ a _ , and @xmath85 is the number of galaxies in the same sub - area . table a2 shows the properties of the distribution of ( u r@xmath28 ) in each of the one square degree areas . specifically , we list the kurtosis with its standard error , and the skewness with its standard error , for the distribution of galaxies within the square degree area . these are listed in order to demonstrate how much are the distributions of ( u r@xmath28 ) values in individual cells similar ( or different ) . the significance of possible differences among the cells was considered using the mean and its standard error . a student s _ t - test _ between the means , by which one estimates the variable @xmath86 was performed . here @xmath87 is the mean of variable _ i _ , and se@xmath88 is its standard error of the mean ( brandt 1970 ) . a cursory perusal of table a1 shows that the only sub - areas where ( u r@xmath28 ) is small , and which are not on the borders of the vpc ( currie & young 1998 ) , are the two areas at ( @xmath20=187.5 , @xmath22=13.5 ) , which we call `` area ga '' , and at ( 187.5 , 14.5 ) which we call `` area gb '' . these regions are central in the virgo ring ( vr ) and this finding , if significant , could indicate less wavelength - dependent extinction in this direction ( which shows reduced fir emission ) than in neighboring areas . the comparison is done with two regions located in the direction of enhanced fir emission , at ( 186.5 , 12.5 ) which we call `` area gc '' , and at ( 188.5 , 12.5 ) which we call `` area gd '' . the latter area * is * located at the edge of the vpc but does not appear to be depleted in galaxies , thus the comparison with it is valid . using the t - test described above , the difference in means of the ( u r@xmath28 ) color index between areas ga and gc yields t=1.97 ( significant at the 95% level ) , and between areas ga and gd t=2.72 ( signficant at the 99.5% level ) . a similar significance level was found for the difference between areas gb and gc , with t=2.76 . the results of the t - test indicate that the galaxies in the vpc , which are located behind the low fir emission area , are slightly bluer than others which are located behind areas of high fir emission . this impression is supported by an examination of the higher moments of the distribution ( table a2 ) . the skewness of the ( u r@xmath28 ) values in areas ga and gb is very similar , but is very different from the skewness of the distribution in areas gc and gd . abadi , h.j . & edmunds , m.g . 1978 , a&a , 70 , 189 almoznino , e. & brosch , n. 1998 , mnras , 298 , 920 bahcall , j.n . & soneira , r. 1984 , apjs , 55 , 67 beichman , c.a . , neugebauer , g. , habing , h.j . , clegg , p.e . & chester , t.j . 1988 _ iras catalogs and atlases : vol . 1 ( explanatory supplement ) _ , nasa rp-1190 binggeli , b. , sandage , a. & tammann , g.a . 1985 , aj , 90 , 1681 blitz , l. , spergel , d.n . , teuben , p.j . & hartmann , d. 1999 , astro - ph/9901307 bohlin , r.c . 1975 , apj , 200 , 402 bhringer , h. , nulsen , p.e.j . , braun , r. & fabian , a.c . 1995 , mnras , 274 , l67 bhringer , h. , briel , u.g . , schwarz , a. , voges , v. , hartner , g. & trumper , j. 1994 , nature , 368 , 828 bothun , g. , impey , c.d . , malin , d.f . & mould , j.r . 1987 , aj , 94 , 23 bothun , g. , schombert , j.m . , impey , c.d . & schneider , s.e . 1990 , apj , 360 , 427 brandt , s. 1970 _ statistical and computational methods in data analysis _ , amsterdam : north - holland publishing co. , pp . 125 _ et seq . _ currie , c.k . & young , m.j . 1998 , a&as , 127 , 367 de blok , w.j.g . 1997 , phd thesis , university of groningen de blok , w.j.g . , van der hulst , j.m . & bothun , g.d . 1995 , mnras , 274 , 235 deul , f.r . & burton , w.b . 1990 , a&a , 230 , 153 egger , r.j . & aschenbach , b. 1995 , a&a , 294 , l25 ferguson , h.c . 1993 , mnras , 263 , 343 ferguson , h.c . , tanvir , n.r . & von hippel , t. 1998 , nature , 391 , 461 giovanelli , r. & haynes , m. 1989 , apj , 346 , l5 graham , j.a . , ferrarese , l. , freedman , w.l . 1998 , baas , 192 , 66.12 haikala , l.k . , mattila , k. , bowyer , s. , sasseen , t.p . , lampton , m. & knude , j. 1995 , apj , 443 , l33 hartmann , d. & burton , w.b . 1997 , _ atlas of galactic neutral hydrogen _ , cambridge ( uk ) : cambridge university press hauser , m.g . , kelsall , t. , leisawitz , d. & weiland , j. 1997 , _ cobe diffuse infrared background experiment ( dirbe ) explanatory supplement _ , cobe reference document ref.no . 97a heller , a. , almoznino , e. & brosch , n. 1998 , mnras , in press ho , l.c . & filippenko , a.v . 1996 , apj , 463 , 818 hoffmeister , c. 1962 , z.f.astrophys . , 55 , 40 kashlinsky , a. 1999 , private communication katsiyannis , a.c . , kemp , s.n . , berry , d.s . & meaburn , j. 1998 , a&as , 132 , 387 kerp , j. , mack , k .- h . , egger , r. , pietz , j. , zimmer , f. , mebold , u. , burton , w.b . & hartmann , d. 1996 , a&a , 312 , 67 knude , j. & h@xmath78 g , e. 1999 , a&a , 341 , 451 knude , j. 1996 , a&a , 306 , 108 kumar , k.c . & thonnard , n. 1983 , aj , 88 , 260 lagache , g. , abergel , a. , boulanger , f. & puget , j .- l . 1998 , a&a , 333 , 709 leggett , s.k . , clowes , r.g . , kalafi , m. , macgillivray , h.t . , puxley , p.j . , savage , a. & wolstencroft , r.d . 1987 , mnras , 227 , 563 lieu , r. , mittaz , j.p.d . , bowyer , s. , lockman , f.j . , hwang , c .- y . & schmitt , j.h.m.m . 1996 , apjl , 458 , 5 lo , k.y . & sargent , w.l.w . 1979 , apj , 227 , 756 macri , l.m . 1999 , astro - ph/9901332 madore , b. & freedman , w.l . 1991 , pasp , 103 , 933 maoz , d. 1995 , apj , 455 , l115 murawski , w. 1983 , acta cosmologica , 12 , 7 okroy , r. 1965 , astron . , 320 , 4 pickering , t.e . , impey , c.d . , van gorkom , j.h . & bothun , g.d . 1997 , aj , 114 , 1858 rudnicki , k. 1986 , _ gamov cosmology _ ( f.melchiorri & r.ruffini , eds . ) , lxxxvi corso of the soc . it . fisica , bologna , p. 480 rudnicki , k. & baranowska , m. 1966 , acta astron . , 16 , 65 schneider , s.e . , helou , g. , salpeter , e.e . & terzian , y. 1983 , apj , 273 , l1 schlegel , d.j . , finkbeiner , d.p . & davis , m. 1998 , apj , 500 , 525 sofue , y. 1994 , apj , 431 , l91 sprayberry , d. , impey , c.d . , irwin , m.j . , mcmahon , r.g . & bothun , g.d . 1993 , apj , 417 , 114 stark , r. , dickey , j.m . , burton , w.b . & wennmacher , a. 1994 , a&a , 281 , 199 taylor , c.l . , thomas , d.l . , brinks , e. & skillman , e.d . 1996 , apjs , 107 , 143 van den bergh , s. 1996 , pasp , 108 , 1091 wakker , b.p & boulanger , f. 1986 , a&a , 170 , 84 wakker , b.p . & van woerden , h. 1991 , a&a , 250 , 509 wakker , b.p . & van woerden , h. 1997 , araa , 35 , 217 weil , m.l . , bland - hawthorn , j. & malin , d.f . 1997 , apj , 490 , 664 wszolek , b. , rudnicki , k. , de bernardis , p. , masi , s. & salvi , a. 1988a , ap.space sci . 152 , 29 wszolek , b. , rudnicki , k. , de bernardis , p. & masi , s. 1988b , in _ large scale structures in the universe _ , ( seitter , w.c . , duerbeck , h.w . & tacke , m. , eds . ) lecture notes in physics 310 , 223 wszolek , b. , rudnicki , k. , de bernardis , p. & masi , s. 1989 , _ the world of galaxies _ ( corvin , h.g . bottinelli , c. , eds . ) p. 499 zonn , w. 1957 , bull . polonaise des sciences , vol . v , no . 1 , 47 zonn , w. & stodolkiewicz , j. 1958 , bull . acad . polonaise des sciences , ser . 3 , 185 zonn , w. & rudnicki , k. , 1965 , in _ stellar astronomy _ , washington dc , 168 figure 1 : fir maps of the vr region . the top right panel shows the iras 100@xmath6 m map . the cobe / dirbe 100@xmath6 m ( top left panel ) and 240@xmath6 m emission ( lower left panel ) show the vr , but have a lower angular resolution than the iras 100@xmath6 m map . the contours of the fir emission are in mjy ster@xmath2 . for orientation purposes , the location of m87 is ( 187.07 , + 12.67 ) . figure 2 : wolf diagram for usno - a2.0 stars in five selected regions , one corresponding to the lowest fir - emitting region in fig . 1 and the other four coinciding with peaks of the fir emission . the regions are defined in table 1 and the cumulative distributions are depicted as follows : usno1222 + 1330 as a dotted line , usno1226 + 1430 as a dashed line , usno1230 + 1130 as a long - dashed line , usno1230 + 1330 as a solid line with open squares , and usno1234 + 1230 as a dot - dashed line . figure 3 : comparison of predicted star counts ( using the bahcall - soneira model ) and the actual star counts from the usno - a2.0 catalog , for the five regions listed in table 3 . each region is one degree square ; the actual star counts are represented by open squares and the model prediction is shown as the solid line . figure 4 : selected areas to check for total extinction effects on galaxies . the contours refer to the iras 100@xmath6 m map and the heavy - lined parallelograms marked a through d are the selected areas for testing the number density of galaxies . figure 5 : wolf diagram for galaxies . the three distribution curves a , c , d correspond to regions within higher fir emission of fig . 1 and curve b to the region of low fir emission . for explanation of the designations a , b , c , d see fig . the data from each of the regions is depicted as follows : region a - solid line , region b - dotted line , region c - dashed line , and region d - long - dashed line . figure 6 : total hi column density from the lds ( left panel ) compared with the iras 100@xmath6 m emission ( right panel ) . darker shades in the hi plot correspond to higher column densities . lighter shades in the fir plot correspond to more intense emission . figure 7 : galactic latitude - velocity plots near b@xmath3075@xmath5 for l=276.0 ( top panel ) , l=283.0 ( middle panel ) , and l=290.5 ( bottom panel ) . the first plot corresponds to the main blob ( more intense fir emission ) , the second to the `` hole '' , center of the vr , and the third to the second blob ( next intense fir emission ) . the horizontal band marked with heavy lines in each panel indicates the region for which data were used to plot fig . figure 8 : cuts through the position - velocity plots of fig . 7 , averaging the brightness temperature for [email protected] , and b=75@xmath5 ( three latitude bands ) . the order of the plots is like in fig . the average hi column density shows the velocity distribution of hi clouds in this region . three peaks stand out clearly , with the most intense one at the highest negative velocities for the main blob . cccccc usno1222 + 1330 & 12:22:00 + 13:30:00 & 185.5 + 13.5 & 277.3 + 74.7 & 1804 usno1226 + 1430 & 12:26:00 + 14:30:00 & 186.5 + 14.5 & 279.3 + 76.0 & 1593 usno1230 + 1130 & 12:30:00 + 11:30:00 & 187.5 + 11.5 & 286.4 + 73.5 & 2126 usno1230 + 1330 & 12:30:00 + 13:30:00 & 187.5 + 13.5 & 284.4 + 75.4 & 1696 usno1234 + 1230 & 12:34:00 + 12:30:00 & 188.5 + 12.5 & 289.0 + 74.7 & 1669 15.5 & 1.485(.109)[08 ] & 1.300(.103)[04 ] & 1.360(.186)[06 ] & 1.678(.162)[08 ] & 1.751(.118)[12 ] 14.5 & 1.347(.079)[27 ] & 1.476(.060)[48 ] & 1.547(.061)[46 ] & 1.380(.096)[26 ] & 1.526(.083)[30 ] 13.5 & 1.573(.053)[68 ] & 1.508(.057)[56 ] & 1.545(.074)[39 ] & 1.390(.094)[26 ] & 1.445(.090)[22 ] 12.5 & 1.363(.070)[36 ] & 1.602(.047)[57 ] & 1.608(.058)[51 ] & 1.593(.070)[45 ] & 1.693(.060)[35 ] 11.5 & 1.274(.075)[36 ] & 1.582(.068)[32 ] & 1.658(.054)[72 ] & 1.514(.066)[45 ] & 1.420(.101)[26 ] 10.5 & 1.296(.070)[11 ] & 1.347(.077)[22 ] & 1.530(.099)[23 ] & 1.596(.104)[20 ] & 1.054(.114)[14 ] 15.5 & -.37(1.48 ) .11(.75 ) & 1.89(2.62 ) 1.35(1.01 ) & .83(1.74 ) .16(.85 ) & .74(1.48 ) -.46(.75 ) & .18(1.23 ) -.64(.64 ) 14.5 & -.31(.87 ) .58(.45 ) & -1.38(.67 ) -.05(.34 ) & -1.10(.69 ) -.36(.35 ) & -.96(.89 ) .11(.46 ) & .35(.83 ) .00(.43 ) 13.5 & .83(.57 ) -.86(.29 ) & -.32(.63 ) -.07(.32 ) & -.03(.74 ) .45(.38 ) & -.09(.89 ) .53(.47 ) & 1.89(.95 ) -.91(.49 ) 12.5 & -.43(.77 ) -.33(.39 ) & 1.38(.62 ) -.71(.32 ) & .43(.67 ) -.47(.33 ) & .35(.70 ) -.64(.35 ) & -.93(.78 ) -.43(.40 ) 11.5 & -.47(.77 ) -.08(.39 ) & -1.06(.81 ) .10(.41 ) & -.21(.56 ) -.10(.28 ) & -.10(.70 ) .38(.35 ) & -.86(.89 ) -.02(.46 ) 10.5 & .59(1.28 ) -.56(.66 ) & -1.04(.95 ) -.55(.49 ) & -.58(.94 ) -.55(.48 ) & .14(.99 ) .57(.51 ) & -.41(1.15 ) .11(.60 )

in this paper we investigate the asymptotics of forward - start options and the forward implied volatility smile in the heston model as the maturity approaches zero . we prove that the forward smile for out - of - the - money options explodes and compute a closed - form high - order expansion detailing the rate of the explosion . furthermore the result shows that the square - root behaviour of the variance process induces a singularity such that for certain parameter configurations one can not obtain high - order out - of - the - money forward smile asymptotics . in the at - the - money case a separate model - independent analysis shows that the small - maturity limit is well defined for any it diffusion . the proofs rely on the theory of sharp large deviations ( and refinements ) and incidentally we provide an example of degenerate large deviations behaviour .

1303.4268

consider an asset price process @xmath0 with @xmath1 , paying no dividend , defined on a complete filtered probability space @xmath2 with a given risk - neutral measure @xmath3 , and assume that interest rates are zero . in the black - scholes - merton ( bsm ) model , the dynamics of the logarithm of the asset price are given by @xmath4 , where @xmath5 represents the instantaneous volatility and @xmath6 is a standard brownian motion . the no - arbitrage price of the call option at time zero is then given by the famous bsm formula @xcite : @xmath7 , with @xmath8 , where @xmath9 is the standard normal distribution function . for a given market price @xmath10 of the option at strike @xmath11 and maturity @xmath12 , the spot implied volatility @xmath13 is the unique solution to the equation @xmath14 . for any @xmath15 and @xmath16 , we define as in @xcite a forward - start option with forward - start date @xmath17 , maturity @xmath12 and strike @xmath11 as a european option with payoff @xmath18 where we define @xmath19 pathwise . by the stationary increment property , its value is simply @xmath20 in the bsm model . for a given market price @xmath21 of the option at strike @xmath11 , forward - start date @xmath17 and maturity @xmath12 , the forward implied volatility smile @xmath22 is then defined ( see also @xcite ) as the unique solution to @xmath23 . the forward smile is a generalisation of the spot implied volatility smile , and the two are equal when @xmath24 . asymptotics of the spot implied volatility surface have received a large amount of attention over the past decade . these results have helped shape calibration methodologies based on arbitrage - free approximation of the spot smile in a large variety of models . small - maturity asymptotics have been studied by berestycki - busca - florent @xcite using pde methods for continuous time diffusions and by henry - labordre @xcite using heat kernel expansions . forde et al . @xcite and jacquier et al . @xcite derived small and large - maturity asymptotics in the heston model and affine stochastic volatility models with jumps using large deviations and saddlepoint methods . wing asymptotics ( as @xmath25 ) have been studied by lee @xcite and further extended by benaim and friz @xcite and in @xcite . fouque et al . @xcite have used perturbation techniques in order to study slow and fast mean - reverting stochastic volatility models . small - noise expansions have been studied by osajima @xcite and takahashi @xcite using watanabe expansions and in deuschel et al . @xcite using laplace method on wiener space . see @xcite for a general overview of implied volatility asymptotics in stochastic volatility models . in exponential lvy models implied volatilities of out - of - the - money options explode as the maturity tends to zero , while the implied volatility for at - the - money options converges to the volatility of the diffusion component as the maturity tends to zero . small - maturity asymptotics for models with jumps ( including lvy processes ) have been investigated in @xcite . however , these asymptotics do not provide any information on the forward smile or forward - start - based payoffs , such as cliquets and forward - start options and the literature on this topic is sparse . glasserman and wu @xcite introduced different notions of forward volatilities for forecasting purposes , keller - ressel @xcite studies a very specific type of asymptotic ( when the forward - start date @xmath17 becomes large ) , and empirical results have been carried out in @xcite . in @xcite the authors empirically studied the forward smile in sato models and ran comparisons with a suite of models including heston and local volatility models for forward smile sensitive products such as cliquets . recently , small and large - maturity forward smile asymptotics were derived in @xcite for a general class of models including the heston model . however , the results in @xcite only apply to the so - called diagonal small - maturity regime , i.e. the behaviour ( as @xmath26 tends to zero ) of the process @xmath27 . the conjecture , stated there , is that for fixed @xmath28 the heston forward smile ( corresponding to @xmath29 ) explodes to infinity ( except at - the - money ) as @xmath12 tends to zero . in this paper we confirm this conjecture and give a high - order expansion for the forward smile . the main result ( theorem [ theorem : fwdsmilesmalltime ] ) is that the small - maturity heston forward smile explodes according to the following asymptotic : @xmath30 for @xmath31 and @xmath28 as @xmath12 tends to zero . here @xmath32 and @xmath33 are even continuous functions ( over @xmath34 ) with @xmath35 and independent of the heston correlation . in the at - the - money case ( @xmath36 ) a separate model - independent analysis ( lemma [ lem : modelindatm ] and theorem [ theorem : fwdatmsmalltime ] ) shows that the small - maturity limit is well defined and @xmath37 holds for any well - behaved diffusion where @xmath38 is the instantaneous variance at time @xmath17 . this exploding nature is consistent with empirical observations in @xcite and the diagonal small - maturity asymptotic from @xcite . the paper is structured as follows . in section [ sec : fwdtimescales ] we introduce the notion of a forward time - scale and characterise it in the heston model . in section [ sec : smfwdstartasymp ] we state the main result on small - maturity asymptotics of forward - start options in the heston model . section [ sec : smfwdsmileasymp ] tackles the forward implied volatility asymptotics : section [ sec : otmasymp ] translates the results of section [ sec : smfwdstartasymp ] into out - of - the - money forward smile asymptotics , and section [ sec : atmfwdsmile ] presents a model - independent result for the at - the - money forward implied volatility . section [ sec : smhnumerics ] provides numerical evidence supporting the asymptotics derived in the paper and the main proofs are gathered in section [ sec : proofssmalltau ] . * notations : * @xmath39 shall always denote expectation under a risk - neutral measure @xmath40 given a priori . we shall refer to the standard ( as opposed to the forward ) implied volatility as the spot smile and denote it @xmath41 . the forward implied volatility will be denoted @xmath42 as above . in this paper we will always assume that forward - start date is greater than zero ( @xmath28 ) unless otherwise stated . in this section we introduce the notion of a forward time - scale and characterise it in the heston model . in the bsm model the time - scale is given by @xmath43 , which is related to the quadratic variation of the driving brownian motion . for diffusions ( such as heston ) the ( spot ) time - scale is the same , which implies that the spot smile has a finite ( non - zero ) small - maturity limit . in the forward case , this however no longer remains true . stochastic volatility models ( eg . heston ) exhibit different time - scales to the bsm model leading to different asymptotic regimes for the forward smile relative to the spot smile . as we will show below ( lemma [ lemma : hestfwdtimescale ] ) , all re - scalings of the heston model lead to limiting logarithmic moment generating functions ( lmgf s ) that are all zero on their domains of definition . but the forward time - scale is the only choice that leads to the limiting lmgf being zero on a bounded domain . this is one of the key properties that allows us to derive sharp large deviation results even though at first sight this zero limit appears trivial and non - consequential . we define the forward lmgf by @xmath44 where @xmath45 . with this definition the domain @xmath46 will depend on @xmath47 , but it will be clear from the context which choice of @xmath47 we are using . [ def : fwdtimescale ] we define a ( small - maturity ) _ forward time - scale _ as a continuous function @xmath48 such that @xmath49 and @xmath50 produces a non - trivial pointwise limit . we shall say that a ( pointwise ) limit is trivial if it is null on @xmath34 or null at the origin and infinite on @xmath51 . a. in the bsm model the forward time - scale is @xmath52 . b. a forward time - scale may not exist for a model . for example , consider exponential lvy models with bounded domain for the lvy exponent . the only non - trivial limit occurs when @xmath53 , which does not satisfy definition [ def : fwdtimescale ] and so a forward - time scale does not exist . c. if @xmath54 satisfies a large deviations principle ( * ? ? ? * section 1.2 ) with speed @xmath55 and assuming further some tail condition ( see ( * ? ? ? * theorem 4.3.1 ) ) , then @xmath55 is the forward time - scale for the model by varadhan s lemma . d. diffusion models have the same spot time - scale ( @xmath24 ) as the bsm model , namely @xmath56 ( see for example @xcite ) . this is not necessarily true in the forward case as we will shortly see . in the heston model , the ( log ) stock price process is the unique strong solution to the following sdes : @xmath57 with @xmath58 , @xmath59 , @xmath60 , @xmath61 and @xmath62 and @xmath63 are two standard brownian motions . the feller sde for the variance process has a unique strong solution by the yamada - watanabe conditions ( * ? ? ? * proposition 2.13 , page 291 ) ) . the @xmath64 process is a stochastic integral of the @xmath65 process and is therefore well defined . the feller condition , @xmath66 , ensures that the origin is unattainable . otherwise the origin is regular ( hence attainable ) and strongly reflecting ( see ( * ? ? ? * chapter 15 ) ) . we do not require the feller condition in our analysis since we work with the forward lmgf of @xmath64 which is always well defined . in order to characterise the heston forward time - scale we require the following lemma , proved in section [ sec : hestfwdtime ] . [ lemma : hestfwdtimescale ] let @xmath67 be a continuous function such that @xmath49 and @xmath68 . the following limits hold for the heston forward lmgf as @xmath12 tends to zero with @xmath69 defined in : a. if @xmath70 then @xmath71 for all @xmath72 and is infinite otherwise ; b. if @xmath73 then @xmath71 for @xmath74 and is infinite otherwise ; c. if @xmath75 then @xmath76 , for all @xmath77 . as it turns out all limits are zero on their domains of definition , but using @xmath78 produces the only ( up to a constant multiplicative factor ) non - trivial zero limit . it follows that @xmath79 is the heston forward time - scale . let now @xmath80 be the pointwise limit ( with @xmath81 ) from lemma [ lemma : hestfwdtimescale ] , i.e. satisfying @xmath82 for @xmath83 and infinity otherwise . further we define the function @xmath84 as the fenchel - legendre transform of @xmath85 : @xmath86 [ lemma : u*characterisation ] the function @xmath87 defined in is characterised explicitly as @xmath88 for all @xmath16 . clearly @xmath89 . now suppose that @xmath90 . then @xmath91 . a similar result holds for @xmath92 and the result follows . in this section we state the main result on small - maturity forward - start option asymptotics . first we need to define a number of functions . all functions below are real - valued and defined on @xmath51 . we recall that @xmath93 if @xmath94 and -1 otherwise . @xmath95 , & \end{array } \right.\ ] ] where @xmath96 @xmath97 @xmath98 @xmath99\big ) , \\ \psi_1(k ) & : = \displaystyle -4 a_0(k ) v \beta _ t { \mathrm{e}}^{-\kappa t}/e_0 ^ 4(k)\\ \psi_2(k ) & : = \displaystyle \frac { v { \mathrm{e}}^{-\kappa t}}{2 e_0 ^ 4(k ) } \big(4 a_0(k ) \beta _ t [ 3 a_0(k ) e_1(k ) - 4 a_1(k ) e_0(k)]-5 e_0 ^ 2(k)\big ) , \\ \psi_3(k ) & : = \displaystyle 8 v \beta _ t { \mathrm{e}}^{-\kappa t}/e_0 ^ 5(k),\\ \psi_4(k ) & : = \displaystyle \frac{v{\mathrm{e}}^{-\kappa t}}{2 e_0 ^ 3(k)}\left ( \frac{e_1 ^ 2(k)-e_0(k ) e_2(k)}{\beta_t } - 2 a_0(k ) a_1(k ) e_0(k ) e_1(k)+2 e_0 ^ 2(k ) r(k)\right ) , \\ \end{array } \right.\ ] ] @xmath100 @xmath101 @xmath102 we now state the main result of the section , i.e. an asymptotic expansion formula for forward - start option prices as the remaining maturity tends to zero . the proof is given in section [ sec : optpricefwdsmile ] . [ theorem : fwdstartoptionssmalltime ] the following expansion holds for forward - start option prices for all @xmath31 as @xmath12 tends to zero : @xmath103 = \left(1-{\mathrm{e}}^{k}\right){1\hspace{-2.1mm}{1}}_{\{k<0\ } } \\ & + \exp{\left({-\frac{\lambda^*(k)}{\sqrt{\tau}}+\frac{c_0(k)}{\tau^{1/4}}+c_1(k)}+k\right ) } \frac{\beta_t \tau^{\left(7/8-\theta\kappa/(2\xi^2)\right)}c_2(k)}{\zeta(k ) \sqrt{2\pi}}\left(1+c_3(k)\tau^{1/4}+o\left(\tau^{1/4}\right)\right),\end{aligned}\ ] ] where @xmath87 is characterised in lemma [ lemma : u*characterisation ] , @xmath104 in , @xmath105 in and @xmath69 in . a. we have @xmath106 and @xmath107 for all @xmath31 . also note that @xmath87 is linear as opposed to being strictly convex in the bsm model , see lemma [ lemma : bsasympsmalltau ] below . b. the forward time - scale @xmath108 results in out - of - the - money forward - start options decaying as @xmath12 tends to zero at leading order with a rate of @xmath109 as opposed to a rate of @xmath110 in the bsm model . c. the fact that the limiting forward lmgf is non - steep ( trivially zero on a bounded interval ) results in a different asymptotic regime for higher order terms compared to the bsm model . in particular we have a @xmath111 dependence as opposed to a @xmath12 dependence in the bsm model and the introduction of the parameter dependent term @xmath112 . the implications of this parameter dependent term for forward - smile asymptotics will be discussed further in remark [ rem : smmatfwdsmileasymp](vii ) . d. the asymptotic expansion is given in closed - form and can in principle be extended to arbitrary order using the methods given in the proof . as an immediate consequence of theorem [ theorem : fwdstartoptionssmalltime ] we have the following corollary , which provides an example of a family of random variables for which the limiting logarithmic moment generating function is zero ( on its effective domain ) but a large deviation principle still holds . this is to be compared to the grtner - ellis theorem ( * ? ? ? * theorem 2.3.6 ) which requires the limiting lmgf to be at least steep at the boundaries of its effective domain for an ldp to hold . @xmath113 satisfies an ldp with speed @xmath108 and good rate function @xmath87 as @xmath12 tends to zero . the proof of theorem [ theorem : fwdstartoptionssmalltime ] holds with only minor modifications for digital options , which are equivalent to probabilities of the form @xmath114 or @xmath115 . one can then show that @xmath116 . note that of course this infimum is null whenever @xmath90 . consider now an open interval of the real line of the form @xmath117 . since @xmath118 $ ] , then by continuity of the function @xmath87 and its properties given in lemma [ lemma : u*characterisation ] , we immediately obtain that @xmath119 since any borel set of the real line can be written as a ( countable ) union / intersection of open intervals , the corollary follows from the definition of the large deviations principle , see ( * ? ? ? * section 1.2 ) . in order to translate the forward - start option results into forward smile asymptotics we require a similar expansion for the bsm model . the following lemma is a direct consequence of ( * ? ? ? * corollary 2.9 ) and the proof is therefore omitted . [ lemma : bsasympsmalltau ] in the bsm model the following expansion holds for all @xmath31 as @xmath12 tends to zero : @xmath120 = \left(1-{\mathrm{e}}^{k}\right){1\hspace{-2.1mm}{1}}_{\{k<0\}}+ \frac{{\mathrm{e}}^{k/2-k^2/(2\sigma^2 \tau)}\left(\sigma^2\tau\right)^{3/2}}{k^2\sqrt{2\pi}}\left[1-\left(\frac{3}{k^2}+\frac{1}{8}\right)\sigma^2\tau+o(\tau)\right].\ ] ] we now translate the small - maturity forward - start option asymptotics into forward smile asymptotics . define the functions @xmath121 ( @xmath122 ) by @xmath123 with @xmath87 characterised in lemma [ lemma : u*characterisation ] , @xmath104 in , @xmath105 in and @xmath69 in . on @xmath51 , @xmath106 and so @xmath124 . further @xmath107 and so @xmath125 . also @xmath126 and @xmath127 so that @xmath128 is a well defined real - valued function . the following theorem proved in section [ sec : optpricefwdsmile]is the main result of the section . [ theorem : fwdsmilesmalltime ] the following expansion holds for the forward smile for all @xmath31 as @xmath12 tends to zero : @xmath129 [ rem : smmatfwdsmileasymp ] a. note that @xmath130 and @xmath131 are strictly positive for all @xmath31 , so that the heston forward smile blows up to infinity ( except atm ) as @xmath12 tends to zero . b. both @xmath32 and @xmath33 are even functions and correlation - independent quantities so that for small maturities the heston forward smile becomes symmetric ( in log - strikes ) around the atm point . consequently , if one believes that the small - maturity forward smile should be downward sloping ( similar to the spot smile ) then the heston model should not be chosen . this small - maturity u - shaped effect for the heston forward smile has been mentioned qualitatively by practitioners ; see @xcite . c. we use the notation @xmath132 to mean @xmath133 as @xmath134 . then in heston we have @xmath135 and in exponential lvy models with lvy measure @xmath136 satisfying @xmath137 we have @xmath138 ( * ? ? ? * page 21 ) . we therefore see that the small - maturity exponential lvy smile blows up at a much quicker rate than the heston forward smile . d. we have @xmath139 and @xmath140 . higher - order terms are not necessarily continuous at @xmath36 . for example ( when @xmath141 ) we have @xmath142 . e. the at - the - money forward implied volatility ( @xmath36 ) asymptotic is not covered by theorem [ theorem : fwdsmilesmalltime ] and a separate analysis is needed for this case ( see section [ sec : atmfwdsmile ] ) . in particular the proof fails since in this case the key function @xmath143 ( defined through equation ) does not converge to a boundary point , but rather to zero as @xmath12 tends to zero ( see the proof of lemma [ lemma : u^*tausmalltime ] ) . f. it does not make sense to consider the limit of our asymptotic result for fixed @xmath31 as @xmath17 tends to zero since for @xmath24 using the forward time - scale @xmath78 will produce a trivial limiting lmgf and hence none of the results will carry over . the time scale in the spot case is @xmath52 ; see @xcite . our result is only valid in the forward ( not spot ) smile case . g. as seen in the proof , due to the term @xmath144 in the forward - start option asymptotics in theorem [ theorem : fwdstartoptionssmalltime ] , one can only specify the small - maturity forward smile to arbitrary order if @xmath141 . if this is not the case then such an expansion for the forward smile only holds up to order @xmath145 . now the dynamics of the heston volatility @xmath146 is given by @xmath147 with @xmath148 . if we set @xmath141 the volatility becomes gaussian , which corresponds to a specific case of the schbel - zhu stochastic volatility model @xcite . so as the heston volatility dynamics deviate from gaussian volatility dynamics a certain degeneracy occurs such that one can not specify high order forward smile asymptotics in the small - maturity case . interestingly , a similar degenerary occurs when studying the tail probability of the stock price . as proved in @xcite , the square - root behaviour of the variance process induces some singularity and hence a fundamentally different behaviour when @xmath149 . the analysis above excluded the at - the - money case @xmath36 . we show below that this case has a very different behaviour and can be studied with a much simpler machinery . in this section , we shall denote the future implied volatility @xmath150 as the implied volatility corresponding to a european call / put option with strike @xmath151 , maturity @xmath12 , observed at time @xmath17 . we first start with the following model - independent lemma , bridging the gap between the at - the - money future implied volatility @xmath152 and the forward implied volatility @xmath153 . note that a similar result albeit less general was derived in @xcite . we shall denote by @xmath154 the expectation ( under the given risk - neutral probability measure ) with respect to @xmath155 , the filtration at time zero . [ lem : modelindatm ] let @xmath28 . assume that there exists @xmath156 such that the expansion @xmath157 holds and that @xmath158 for @xmath159 . if the at - the - money forward implied volatility satisfies @xmath160 , then @xmath161 for all @xmath162 . in the black - scholes model , we know that for any @xmath163 , @xmath164 , the price at time @xmath17 of a ( re - normalised ) european call option with maturity @xmath165 is @xmath166 $ ] , and its at - the - money expansion as the maturity @xmath12 tends to zero reads ( see ( * ? ? ? * corollary 3.5 ) ) @xmath167 we keep the @xmath168 dependence in the @xmath169 to highlight the fact that , when @xmath168 depends on @xmath12 ( such as @xmath170 ) , one has to be careful not to omit some terms . now , for a given martingale model for the stock price @xmath171 , we shall denote by @xmath172 the price at time @xmath17 of a european call option with payoff @xmath173 at time @xmath165 . the future implied volatility @xmath150 is then the unique solution to @xmath174 . for at - the - money @xmath36 , we obtain the following expansion for short maturity @xmath12 : @xmath175 where we have used here the expansion assumed for @xmath152 . note also that the coefficients @xmath176 are random variables . we follow the probabilistic version of the @xmath177 notation detailed in ( * ? ? ? * section 5 ) , namely the random remainder @xmath178 is @xmath179 as @xmath12 tends to zero if and only if for any @xmath180 there exist a constant @xmath181 and a threshold @xmath182 for which @xmath183 for all @xmath184 . for brevity we abuse the notations slightly here and write @xmath177 instead of @xmath185 . now , the forward - start european call option ( at inception ) in the black - scholes model reads @xmath186 = \mathbb{e}_0\left\{\mathbb{e}\left[\left(\frac{s^{\mathrm{bs}}_{t+\tau}}{s^{\mathrm{bs}}_t}-{\mathrm{e}}^{k}\right)_+\vert { \mathcal{f}}_t\right]\right\ } = { \mathcal{n}}(d_+(\sigma , \tau))-{\mathrm{e}}^{k}{\mathcal{n}}(d_-(\sigma,\tau ) ) = c_0^{\mathrm{bs}}(k , \tau , \sigma),\ ] ] where @xmath187 . for a given model , let us denote by @xmath188 the price of a forward - start european call option . by definition of the forward implied volatility @xmath22 , we have @xmath189 . for at - the - money @xmath36 , it follows that @xmath190 using the assumed expansion for @xmath153 we similarly obtain ( as in ) @xmath191 note that now the coefficients @xmath192 are not random variables , but simple constants . recall now that @xmath193 = \mathbb{e}_0\left\{\mathbb{e}\left[\left(\frac{s_{t+\tau}}{s_t}-{\mathrm{e}}^{k}\right)_+\vert { \mathcal{f}}_t\right]\right\ } = \mathbb{e}_0\left(c_{t}(k,\tau)\right).\ ] ] combining this with and , we find that @xmath194 for @xmath195 . the higher - order terms for the expansion can be proved analogously and the lemma follows . we now apply this to the heston model . recall the definition of the kummer ( confluent hypergeometric ) function @xmath196 : @xmath197 where the pochhammer symbol is defined by @xmath198 for @xmath199 and @xmath200 . for any @xmath201 and @xmath28 we define @xmath202 with @xmath69 defined in . this function is related to the moments of the feller diffusion ( see ( * ? ? ? * theorem 2.4 ) ) : for any @xmath28 , @xmath203=\delta(t , p)$ ] if @xmath201 and is infinite otherwise . note in particular that @xmath204 ( see ( * ? ? ? * page 504 ) ) . the heston forward at - the - money volatility asymptotic is given in the following theorem . [ theorem : fwdatmsmalltime ] the following expansion holds for the forward at - the - money volatility as @xmath12 tends to zero : @xmath205 a. as opposed to the out - of - the - money case , the small - maturity limit here is well defined . b. combining lemma [ lem : modelindatm ] and @xcite , @xmath37 holds for any well - behaved stochastic volatility model @xmath206 . c. the proof does not allow one to conclude any information about higher order terms in heston for the case @xmath207 . a different method would need to be used to compute higher order asymptotics in this case . in heston we recall ( ( * ? ? ? * corollary 4.3 ) or ( * ? ? ? * corollary 3.3 ) ) the asymptotic @xmath208 and so for small @xmath12 we have @xmath209 , with @xmath210 and @xmath211 . lemma [ lem : modelindatm ] and conclude the proof . we first compare the true heston forward smile and the asymptotics developed in the paper . we calculate forward - start option prices using the inverse fourier transform representation in ( * ? ? ? * theorem 5.1 ) and a global adaptive gauss - kronrod quadrature scheme . we then compute the forward smile @xmath42 with a simple root - finding algorithm . the heston model parameters are given by @xmath212 , @xmath213 , @xmath214 and @xmath215 unless otherwise stated in the figures . in figures [ fig : halfmonth ] and [ fig : onemonth ] we compare the true forward smile using fourier inversion and the asymptotic in theorem [ theorem : fwdsmilesmalltime ] . it is clear that the small - maturity asymptotic has very different features relative to `` smoother '' asymptotics derived in @xcite . this is due to the introduction of the forward time - scale and to the fact that the limiting lmgf is not steep . note also from remark [ rem : smmatfwdsmileasymp](iv ) that the asymptotics in theorem [ theorem : fwdsmilesmalltime ] can approach zero or infinity as the strike approaches at - the - money . this appears to be a fundamental feature of non - steep asymptotics ; numerically this implies that the asymptotic may break down for strikes in a region around the at - the - money point . in figure [ fig : smallmatatmfourier ] we compare the true at - the - money forward volatility using fourier inversion and the asymptotic in lemma [ lem : modelindatm ] . results are in line with expectations and the at - the - money asymptotic is more accurate than the out - of - the - money asymptotic . this is because the at - the - money forward volatility ( unlike the out - of - the - money case ) has a well defined limit as @xmath12 tends to zero . in figure [ fig : exploding ] we use these results to gain intuition on how the heston forward smile explodes for small maturities . in @xcite the authors derived a diagonal small - maturity asymptotic expansion for the heston forward smile valid for small forward start - dates and small maturities . in order for the small - maturity asymptotic in this paper to be useful , there needs to be a sufficient amount of variance of variance at the forward - start date . practically this means that the asymptotic performs better as one increases the forward - start date . on the other hand the diagonal - small maturity asymptotic expansion is valid for small forward - start dates . in this sense these asymptotics complement each other . figure [ fig : smallmatvsdiag ] shows the consistency of these two results for small forward - start date and maturity . and @xmath216 . we split the proof of theorems [ theorem : fwdstartoptionssmalltime ] and [ theorem : fwdsmilesmalltime ] into several parts , from section [ sec : hestfwdtime ] to section [ sec : optpricefwdsmile ] below . in section [ sec : hestfwdtime ] we develop the necessary tools to characterise the small - maturity heston forward lmgf domain and derive the heston forward time - scale ( lemma [ lemma : hestfwdtimescale ] ) . in section [ sec : asympmeaschange ] we use the forward time - scale to define a time - dependent asymptotic measure - change and derive expansions for fundamental auxiliary functions needed in the analysis . in section [ sec : charactfuncasymp ] we derive the asymptotics of the characteristic function of a re - scaled version of the forward price process @xmath217 under the asymptotic measure - change defined in section [ sec : asympmeaschange ] . this section also uses fourier transform methods to derive asymptotics of important expectations using this characteristic function expansion . section [ sec : optpricefwdsmile ] finally puts all the pieces together and proves theorems [ theorem : fwdstartoptionssmalltime ] and [ theorem : fwdsmilesmalltime ] . a straightforward application of the tower property for expectations ( see also @xcite ) yields the heston forward logarithmic moment generating function : @xmath218 for all @xmath219 , where @xmath220 @xmath221 the first step is to characterise the forward time - scale in the heston model . in order to achieve this we first need to understand the limiting behaviour of the re - scaled @xmath222 function in that plays a fundamental role in the analysis below . the following lemma shows that using @xmath78 as a time - scale produces the only non - trivial limit for the re - scaled @xmath222 function . we then immediately prove lemma [ lemma : hestfwdtimescale ] which characterises the forward time - scale in the heston model . as @xmath12 tends to zero we have the following asymptotic behaviours for the functions @xmath225 and @xmath226 defined in : @xmath227 where we have set @xmath228 first let @xmath229 . then @xmath230 and so the limit is undefined ( complex infinity ) . next let @xmath231 using we see that @xmath232 where @xmath233 , which is strictly positive for @xmath223 and @xmath234 . it follows that the limit in this case is infinite . next let @xmath235 . here we can write @xmath236 if @xmath237 tends to infinity , so does @xmath238 . when @xmath237 tends to zero then @xmath238 does as well . if @xmath237 converges to a constant @xmath239 , then @xmath238 converges to @xmath240 , and the lemma follows . for any @xmath28 , the random variable @xmath38 in is distributed as @xmath69 times a non - central chi - square random variable with @xmath241 degrees of freedom and non - centrality parameter @xmath242 . it follows that the corresponding moment generating function is given by @xmath243 the re - normalised heston forward logarithmic moment generating function is then computed as @xmath244 = \log\mathbb{e}\left[\mathbb{e}\left({\mathrm{e}}^{u\left(x _ { t+\tau } -x_{t}\right)/h(\tau)}|\mathcal{f } _ { t}\right)\right ] \\ \nonumber & = \log\mathbb{e}\left({\mathrm{e}}^{a\left(u / h(\tau),\tau\right ) + b\left(u / h(\tau),\tau\right)v _ { t}}\right ) = a\left(u / h(\tau),\tau\right ) + \log\lambda_{t}^{v}\left(b\left(u / h(\tau),\tau\right)\right),\nonumber\end{aligned}\ ] ] which agrees with when @xmath245 . this is only valid in some effective domain @xmath246 . the lmgf for @xmath247 is well defined in @xmath248 , and hence @xmath249 , where @xmath250 is the effective domain of the ( spot ) re - normalised heston lmgf . consider first @xmath250 for small @xmath12 . from * proposition 3.1 ) if @xmath251 then the explosion time @xmath252 of the heston mgf is @xmath253 recall the following taylor series expansions , for @xmath254 close to zero : @xmath255 as @xmath12 tends to zero @xmath256 is satisfied since @xmath257 and hence @xmath258 therefore , for @xmath12 small enough , we have @xmath259 for all @xmath77 if @xmath260 tends to zero and @xmath259 for all @xmath261 if @xmath262 , where @xmath263 if @xmath264 tends to infinity , then @xmath265 for all @xmath223 . we are also required to find @xmath266 for small @xmath12 . using lemma [ lemma : blimits ] we see that if @xmath267 then @xmath268 . by the limit of a set we precisely mean the following : @xmath269 if @xmath270 tends to infinity then @xmath271 and if it tends to zero , then @xmath272 . the limiting domains in the lemma follow after taking the appropriate intersections . next we move on to the limits . we only consider the cases where @xmath267 and where @xmath270 tends to zero since these are the only cases for which the forward logarithmic moment generating function is defined . using we see as @xmath12 tends to zero @xmath273 the lemma follows from this and the fact that the function @xmath274 in satisfies @xmath275 . in this section we define the fundamental asymptotic time - dependent measure - change in and derive expansions for critical functions related to this measure - change . in order to proceed with this program we first need to prove some technical lemmas . we use our forward time - scale and define the following rescaled quantities : @xmath276 with @xmath277 , @xmath274 and @xmath222 defined in and respectively . the following lemma gives the asymptotics of the re - scaled quantities @xmath278 , @xmath279 as @xmath12 tends to zero : from the definition of @xmath274 in and the asymptotics in with @xmath78 we obtain @xmath282 substituting the asymptotics for @xmath225 and @xmath226 in we further obtain @xmath283 and therefore using the definition of @xmath222 in we obtain @xmath284 it is still not clear what benefit the forward time - scale has given us since the limiting lmgf is still degenerate . firstly , even though the limiting lmgf is zero on a bounded interval , the re - scaled forward lmgf for fixed @xmath164 is still steep on the domain of definition which implies the existence of a unique solution @xmath285 to the equation @xmath286 further as @xmath12 tends to zero , @xmath285 converges to @xmath287 when @xmath90 and to @xmath288 when @xmath92 ( see lemma [ lemma : u^*tausmalltime ] below ) . the key observation is that the forward time - scale ensures finite boundary points for the effective domain , which in turn implies finite limits for @xmath285 . this is critical to the asymptotic analysis that follows and it will become clear that if any other time - scale were to be used the analysis would break down . the following lemma shows that our definition of @xmath285 is exactly what we need to conduct an asymptotic analysis in this degenerate case . [ lemma : u^*tausmalltime ] for any @xmath289 , @xmath164 , the equation admits a unique solution @xmath285 ; as @xmath12 tends to zero , it converges to @xmath287 ( @xmath288 ) when @xmath90 ( @xmath92 ) , to zero when @xmath36 , and @xmath290 for @xmath12 small enough . we first start by the following claims , which can be proved using the convexity of the forward moment generating function and tedious computations ; we shall not however detail these lengthy computations here for brevity , but figure [ fig : mgfconv ] below provides a visual help . a. for any @xmath164 , the map @xmath291 is strictly increasing and the image of @xmath46 by @xmath292 is @xmath34 ; b. for any @xmath164 , @xmath293 and @xmath294 , i.e. the unique minimum of @xmath295 converges to zero ; c. for each @xmath296 , @xmath297 converges to zero as @xmath12 tends to zero . now , choose @xmath90 ( analogous arguments hold for @xmath92 ) . it is clear from ( i ) that admits a unique solution . since @xmath298 , then there exists @xmath299 such that for any @xmath300 , @xmath301 . note further that ( i ) and ( ii ) imply @xmath302 . from ( iii ) there exists @xmath303 such that the sequence @xmath304 is strictly increasing as @xmath12 goes to zero for @xmath305 . now let @xmath306 and consider @xmath307 . then @xmath308 is bounded above by @xmath287 ( because @xmath301 ) and therefore converges to a limit @xmath309 $ ] . suppose that @xmath310 . since @xmath311 is strictly increasing as @xmath312 tends to zero ( and @xmath313 ) , and @xmath292 is strictly increasing we have @xmath314 ; combining this and ( iii ) yields @xmath315 which contradicts the assumption @xmath90 . therefore @xmath316 and the lemma follows . for sufficiently small @xmath12 we introduce a time - dependent change of measure by @xmath317 by lemma [ lemma : u^*tausmalltime ] , @xmath318 for @xmath12 small enough and so @xmath319 is finite since @xmath320 . also @xmath321 is almost surely strictly positive and by definition @xmath322=1 $ ] . therefore is a valid measure change for all @xmath31 and sufficiently small @xmath12 . equation can be written explicitly as @xmath323 = \left(1 -2 \widehat{b}(u^*_{\tau } ) \beta _ t\right)^2,\ ] ] with @xmath278 and @xmath279 defined in . we now use this to derive an asymptotic expansion for @xmath324 as @xmath12 tends to zero . existence and uniqueness was proved in lemma [ lemma : u^*tausmalltime ] and so we assume the result as an ansatz . consider @xmath90 . from lemma [ lemma : u^*tausmalltime ] it is clear that @xmath330 . the ansatz and lemma [ lemma : abhat ] then imply the following asymptotics as @xmath12 tends to zero ( we drop here the @xmath331-dependence ) : @xmath332 where @xmath333 is defined in . we substitute these asymptotics into and solve at each order . at the @xmath111 order we have two solutions , @xmath334 and we choose the negative root so that @xmath335 for @xmath12 small enough . in a straightforward , yet tedious , manner we continue the procedure and iteratively solve at each order ( the next two equations are linear in @xmath328 and @xmath329 ) to derive the asymptotic expansions in the lemma . an analogous treatment holds in the case @xmath92 . to complete the proof ( and make the ansatz approach above rigorous ) we need to show the existence of this expansion for @xmath308 . fix @xmath31 and set @xmath336 . now let @xmath337 . from lemma [ lemma : u^*tausmalltime ] we know that there exists a solution @xmath338 to the equation @xmath339 and the strict convexity of the forward lmgf implies @xmath340 . further , the two - dimensional map @xmath341 is analytic . it follows by the implicit function theorem ( * ? ? ? * theorem 8.6 , chapter 0 ) that @xmath342 is analytic in some neighbourhood around @xmath343 . since this argument holds for all @xmath337 , this function is also analytic on @xmath344 . also from lemma [ lemma : u^*tausmalltime ] we know that @xmath345 . since we computed the taylor series expansion consistent with this limit and the expansion is unique , it follows that @xmath285 admits this representation . in the forthcoming analysis we will be interested in the asymptotics of @xmath346 as @xmath12 tends to zero . since @xmath347 converges to zero , it is not immediately clear that @xmath348 has a well defined limit . but we can use the asymptotics in to deduce the following lemma : we now define the random variable @xmath355 and the characteristic function @xmath356 of @xmath357 in the @xmath358-measure in as @xmath359 define now the functions @xmath360 by @xmath361 with @xmath362 , @xmath363 defined in , @xmath326 , @xmath350 in , , and @xmath364 , @xmath365 and @xmath366 in . the following lemma provides the asymptotics of @xmath367 : [ rem : weakconsmalltau ] for any @xmath369 , lvy s convergence theorem ( * ? ? ? * page 185 , theorem 18.1 ) implies that @xmath357 converges weakly to a normal random variable with zero mean and variance @xmath370 as @xmath12 tends to zero . from the change of measure and the forward logarithmic moment generating function given in we compute ( we drop the @xmath331-dependence throughout ) for small @xmath12 : @xmath371 \nonumber \\ \nonumber & = { -\frac{1}{\sqrt{\tau}}\lambda_{\tau}^{(t)}\left(u^*_{\tau}\right)-\frac{{\mathtt{i}}uk}{\tau^{1/8}}}+\log\mathbb{e}^{\mathbb{p}}\left[\exp\left({\left(\frac{x_{\tau}^{(t)}}{\sqrt{\tau}}\right)\left({\mathtt{i}}u\tau^{3/8}+u^*_{\tau}\right)}\right)\right ] \\ \label{eq : phiasympsmalltime } & = -\frac{{\mathtt{i}}uk}{\tau^{1/8}}+\frac{1}{\sqrt{\tau}}\left(\lambda_{\tau}^{(t)}\left({\mathtt{i}}u\tau^{3/8 } + u^*_{\tau}\right)-\lambda_{\tau}^{(t)}\left(u^*_{\tau}\right)\right).\end{aligned}\ ] ] using the asymptotics in we have that as @xmath12 tends to zero ( we drop the @xmath331-dependence ) @xmath372 where @xmath333 is defined in . similarly for small @xmath12 , @xmath373 we now use @xmath348 defined in to re - write the term @xmath374 and then use the asymptotics in and lemma [ lemma : etausmalltime ] to find that for small @xmath12 @xmath375 with @xmath105 and @xmath376 defined in and . from the definition of @xmath326 , @xmath350 and @xmath69 we note the simplification @xmath377 similarly we find that as @xmath12 tends to zero @xmath378 again we use @xmath348 defined in to re - write the term @xmath379 = \left(1-\frac{2\beta_t\left(\widehat{b}\left(u^*_{\tau}+{\mathtt{i}}u\tau^{3/8}\right)-\widehat{b}\left(u^*_{\tau}\right)\right)}{e_\tau\tau^{1/4}}\right)^{-2\kappa\theta/\xi^2},\ ] ] and then use the asymptotics in and lemma [ lemma : etausmalltime ] to find that for small @xmath12 @xmath380 using with definition and , property and the asymptotics in , , and we finally calculate the characteristic function for small @xmath12 as @xmath381 with @xmath382 defined in , , and so the lemma follows from the following decomposition @xmath383\tau^{\frac{1}{4 } } + \mathcal{o}(\tau^{\frac{3}{8}})\big\}.\end{aligned}\ ] ] the following technical lemma will be needed in section [ sec : optpricefwdsmile ] where it will be used to give the leading order exponential decay of out - of - the - money forward - start options as @xmath12 tends to zero . [ lemma : u*asymptotics ] the following expansion holds for all @xmath31 as @xmath12 tends to zero : @xmath384 where @xmath385 , @xmath386 and @xmath387 are defined in , @xmath87 is characterised explicitly in lemma [ lemma : u*characterisation ] and @xmath388 is given in . we use the asymptotics in lemma [ lemma : ustartausmalltime ] and the characterisation of @xmath87 in lemma [ lemma : u*characterisation ] to write for small @xmath12 ( we drop the @xmath331-dependence ) @xmath389 using the heston forward lmgf definition in and we can write @xmath390 using the definition of @xmath348 in and the asymptotics in lemma [ lemma : etausmalltime ] we find that for small @xmath12 @xmath391 the the lemma follows after using and , the asymptotics in , and and the simplification @xmath392 . we now use the characteristic function expansion in lemma [ lemma : charactfuncasympsmalltime ] and fourier transform methods to derive the asymptotics for the expectation ( under the measure ) of the modified payoff on the re - scaled forward price process . this lemma will be critical for the analysis in section [ sec : optpricefwdsmile ] . [ lemma : fourierasympsmalltime ] the following expansion holds for all @xmath31 as @xmath12 tends to zero : @xmath393 { 1\hspace{-2.1mm}{1}}_{\{k>0\ } } + \mathbb{e}^{\mathbb{q}_{\tau , k}}\left[{\mathrm{e}}^{-u^*_{\tau}z_{\tau , k}/\tau^{3/8}}\left(1-{\mathrm{e}}^{z_{\tau , k}\tau^{1/8}}\right)^+\right ] { 1\hspace{-2.1mm}{1}}_{\{k<0\ } } \\ & = \frac{\tau^{7/8}\beta_t}{\zeta(k ) \sqrt{2\pi}}\left(1+p_1(k)\tau^{1/4}+o\left(\tau^{1/4}\right)\right),\end{aligned}\ ] ] where @xmath105 is defined in , @xmath394 in and @xmath69 in . we first consider @xmath90 and drop the @xmath331-dependence for the functions below . we denote the fourier transform @xmath395 by @xmath396 for all @xmath397 , @xmath77 . the fourier transform of the payoff @xmath398 is given by @xmath399_{0}^{\infty } -\left[\frac{{\mathrm{e}}^{z\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}\right ) } } { \left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}\right)}\right]_{0}^{\infty } \\ & = \frac{1}{\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}\right)}-\frac{1}{\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}+\tau^{1/8}\right ) } \\ & = \frac{\tau ^{7/8}}{\left(u_{\tau } ^*-{\mathtt{i}}\tau ^{3/8 } u\right ) \left(u_{\tau } ^*-\sqrt{\tau } -{\mathtt{i}}\tau ^{3/8 } u\right)},\end{aligned}\ ] ] if @xmath400 , which holds for @xmath12 small enough since @xmath324 converges to @xmath401 by lemma [ lemma : ustartausmalltime ] . to remark [ rem : weakconsmalltau ] @xmath402 converges weakly to a gaussian random variable and since the gaussian density and the modified payoff are in @xmath403 we can use parseval s theorem ( * ? ? * page 48,theorem 13e ) for small enough @xmath12 to write @xmath404 where we have used that @xmath405 with @xmath406 denoting the complex conjugate for @xmath407 . using the asymptotics of @xmath324 given in lemma [ lemma : ustartausmalltime ] we can taylor expand for small @xmath12 to find that @xmath408 finally combining and the asymptotics of the characteristic function derived in lemma [ lemma : charactfuncasympsmalltime ] with we find that for small @xmath12 @xmath409 where in the last line we have used that @xmath410 since @xmath411 is an odd power of @xmath412 . the result then follows by using simple moment formulae of the normal distribution . fix now @xmath92 . the fourier transform of the payoff @xmath413 is given by @xmath414_{-\infty}^{0}-\left[\frac{{\mathrm{e}}^{z\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}+\tau^{1/8}\right ) } } { \left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}+\tau^{1/8}\right)}\right]_{-\infty}^{0 } \\ & = \frac{1}{\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}\right)}-\frac{1}{\left({\mathtt{i}}u - u^*_{\tau}/\tau^{3/8}+\tau^{1/8}\right ) } \\ & = \frac{\tau ^{7/8}}{\left(u_{\tau } ^*-{\mathtt{i}}\tau ^{3/8 } u\right ) \left(u_{\tau } ^*-\sqrt{\tau } -{\mathtt{i}}\tau ^{3/8 } u\right)},\end{aligned}\ ] ] if @xmath415 , which holds for @xmath12 small enough since @xmath324 converges to @xmath416 by lemma [ lemma : ustartausmalltime ] . the rest of the proof is analogous to @xmath92 above and we omit it for brevity . we have chosen to specify the remainder in the form @xmath417 instead of @xmath418 since it can actually be shown that the term @xmath418 is zero by extending the results in lemma [ lemma : charactfuncasympsmalltime ] and the next non - trivial term is @xmath419 . for brevity we omit this analysis . we use the time - dependent change of measure defined in to write forward - start call option prices for all @xmath90 as @xmath420 & = & { \mathrm{e}}^{\lambda^{(t)}_{\tau}\left(u^*_{\tau}\right)/\sqrt{\tau}}\mathbb{e}^{\mathbb{q}_{k,\tau}}\left[{\mathrm{e}}^{- u^*_{\tau}x_{\tau}^{(t)}/\sqrt{\tau}}\left({\mathrm{e}}^{x^ { ( t)}_{\tau}}-{\mathrm{e}}^k\right)^+\right ] \\ \nonumber & = & { \mathrm{e}}^{-\frac{ku^*_{\tau}-\lambda^{(t)}_{\tau}\left(u^*_{\tau}\right)}{\sqrt{\tau}}}\mathbb{e}^{\mathbb{q}_{k,\tau}}\left[{\mathrm{e}}^{-\frac{u^*_{\tau}}{\sqrt{\tau}}\left(x_{\tau}^{(t)}- k\right)}\left({\mathrm{e}}^{x^ { ( t)}_{\tau}}-{\mathrm{e}}^k\right)^+\right ] \\ \nonumber & = & { \mathrm{e}}^{-\frac{ku^*_{\tau}-\lambda^{(t)}_{\tau}\left(u^*_{\tau}\right)}{\sqrt{\tau}}}{\mathrm{e}}^{k}\mathbb{e}^{\mathbb{q}_{k,\tau}}\left[{\mathrm{e}}^{-\frac{u^*_{\tau}z_{\tau , k}}{\tau^{3/8}}}\left({\mathrm{e}}^{z_{\tau , k}\tau^{1/8}}-1\right)^+\right],\end{aligned}\ ] ] with @xmath357 defined on page . a similar result holds for forward - start put option prices for all @xmath92 . the theorem then follows by applying lemma [ lemma : u*asymptotics ] and lemma [ lemma : fourierasympsmalltime ] and using put - call parity since in the heston model @xmath421 is a true martingale ( * ? ? ? * proposition 2.5 ) . the general machinery to translate option price asymptotics into implied volatility asymptotics has been fully developed by gao and lee @xcite . we simply outline the main steps here . assume the following ansatz for the forward implied volatility as @xmath12 tends to zero : @xmath422 substituting this ansatz into the bsm asymptotics in lemma [ lemma : bsasympsmalltau ] we then obtain @xmath423.\ ] ] equating orders with theorem [ theorem : fwdstartoptionssmalltime ] we solve for @xmath424 and @xmath425 , but we can only solve for higher order terms if @xmath426 or @xmath141 . andersen and a. lipton . asymptotics for exponential lvy processes and their volatility smile : survey and new results . forthcoming in _ international journal of theoretical and applied finance _ , 2013 . deuschel , p.k . friz , a. jacquier and and s. violante . marginal density expansions for diffusions and stochastic volatility , part ii : applications . forthcoming in _ communications on pure and applied mathematics _ , 2013 . j. figueroa - lpez , r. gong and c. houdr . small - time expansions of the distributions , densities , and option prices of stochastic volatility models with lvy jumps . _ stochastic processes and their applications _ , 122 : 1808 - 1839 , 2012 . n. kunitomo and a. takahashi . applications of the asymptotic expansion approach based on malliavin - watanabe calculus in financial problems . _ stochastic processes and applications to mathematical finance _ , world scientific , 195 - 232 , 2004 .

maximum likelihood estimation of a log - concave probability density is formulated as a convex optimization problem and shown to have an equivalent dual formulation as a constrained maximum shannon entropy problem . closely related maximum renyi entropy estimators that impose weaker concavity restrictions on the fitted density are also considered , notably a minimum hellinger discrepancy estimator that constrains the reciprocal of the square - root of the density to be concave . a limiting form of these estimators constrains solutions to the class of quasi - concave densities . and .

1007.4013

our objective is to introduce a general class of shape constraints applicable to the estimation of probability densities , multivariate as well as univariate . elements of the class are represented by restricting certain monotone functions of the density to lie in convex cones . maximum likelihood estimation of log - concave densities constitutes an important special case ; however , the wider class allows us to include a variety of other shapes . a one parameter subclass modeled on the means of order @xmath0 studied by @xcite incorporates all the quasi - concave densities , that is , all densities with convex upper contour sets . estimation methods for these densities , as described below , bring new opportunities for statistical data analysis . log - concave densities play a crucial role in a wide variety of probabilistic models : in reliability theory , search models , social choice and a broad range of other contexts it has proven convenient to assume densities whose logarithm is concave . recognition of the importance of log - concavity was already apparent in the work of schoenberg and karlin on total positivity beginning in the late 1940s . @xcite forged a link between log - concavity and classical statistical properties such as the monotone likelihood ratio property , the theory of sufficient statistics and uniformly most powerful tests . maximum likelihood estimation of densities constrained to be log - concave has recently enjoyed a considerable vogue with important contributions of walther ( @xcite , @xcite , @xcite ) , @xcite , @xcite , @xcite , @xcite , @xcite and @xcite , among others . log - concave densities are constrained to exhibit exponential tail behavior . this restriction motivates a search for weaker forms of the concavity constraint capable of admitting common densities with algebraic tails like the @xmath1 and @xmath2 families . the @xmath0-concave densities introduced in section [ s : mle ] constitute a rich source of candidates . while it would be possible , in principle , to consider maximum likelihood estimation of such densities , duality considerations lead us to consider a more general class of maximum entropy criteria . maximizing shannon entropy in the dual is equivalent to maximum likelihood for the leading log - concave case , but other entropies are also of interest . section [ s : dual ] describes several examples arising in the dual from the class of rnyi entropies , each corresponding to a distinct specification of the concavity constraint , and each corresponding to a distinct fidelity criterion in the primal . the crucial advantage of adapting the fidelity criterion to the form of the concavity constraint is that it assures a convex optimization problem with a tractable computational strategy . a probability density function , @xmath3 , is called _ log - concave _ if @xmath4 is a ( proper ) convex function on the support of @xmath3 . we adhere to the usual conventions of @xcite , which allow convex functions to take infinite values although we will allow only @xmath5 , because all our convex functions will be _ proper_. the _ domain _ of a convex ( concave ) function , @xmath6 , is then the set of @xmath7 such that @xmath8 is finite . we adopt the convention @xmath9 . unimodality of concave functions implies that log - concave densities are unimodal . an interesting connection in the multivariate case was pointed out by @xcite : the number of modes of a kernel density estimate is monotone in the bandwidth when the kernel is log - concave . however , as illustrated by the student @xmath1 family , not every unimodal density is log - concave . laplace densities , with their exponential tail behavior , are ; but heavier , algebraic tails are ruled out . this prohibition motivates a relaxation of the log - concavity requirement . a natural hierarchy of concave functions can be built on the foundation of the weighted means of order @xmath0 studied by @xcite : for any @xmath10 in the unit simplex , @xmath11 , let @xmath12 for @xmath13 ; the limiting case for @xmath14 is @xmath15 the familiar arithmetic , geometric and harmonic means correspond to @xmath0 equal to @xmath16 , @xmath17 and @xmath18 , respectively . following @xcite , a nonnegative , real function @xmath3 , defined on a convex set @xmath19 is called _ @xmath0-concave _ if for any @xmath20 and @xmath21 , @xmath22 in this terminology , @xmath23-concave functions are @xmath17-concave and concave functions are @xmath16-concave . as @xmath24 is monotone , increasing in @xmath0 for @xmath25 and any @xmath21 , it follows that if @xmath3 is @xmath0-concave , then @xmath3 is also @xmath26-concave for any @xmath27 . thus , concave functions are @xmath23-concave , but not vice - versa . in the limit @xmath28 , concave functions satisfy the condition @xmath29 so they are ( and consequently for all @xmath0-concave functions ) _ quasi - concave_. the hierarchy of @xmath0-concave density functions was considered in the economics literature by @xcite in spatial models of voting and imperfect competition ; their results reveal some intriguing connections to tukey s half - space depth in multivariate statistics ; see @xcite . curiously , it appears that the first thorough investigation of the mathematical concept of quasi - concavity was carried out by @xcite . further details and motivation for @xmath0-concave densities can be found in @xcite , @xcite and @xcite . suppose that @xmath30 is a collection of data points in @xmath31 such that the convex hull of @xmath32 , @xmath33 , has a nonempty interior in @xmath31 ; such a configuration occurs with probability @xmath16 if @xmath34 and the @xmath35 behave like a random sample from @xmath36 , a probability density with respect to the lebesgue measure on @xmath31 . viewing the @xmath35 s as a random sample from an unknown , log - concave density @xmath36 , we can find the _ maximum likelihood estimate _ of @xmath36 by solving @xmath37 it is convenient to recast ( [ ! lcmle ] ) in terms of @xmath38 , the estimate becoming @xmath39 , @xmath40 the objective function of ( [ ! lcmlelog ] ) is equal to @xmath5 , given the convention adopted above , unless all @xmath35 are in the domain of @xmath41 . as in @xcite , it proves convenient to move the integral constraint into the objective function , @xmath42 a device that ensures that the solution integrates to one without enforcing this condition explicitly . apart from the multiplier @xmath43 , the crucial difference between ( [ ! lcmlelog ] ) and ( [ ! warden ] ) is that the latter is a convex problem , while the former not . it is well known that nave , unrestricted maximum likelihood estimation is doomed to fail when applied in the general density estimation context : once `` log - concave '' is dropped from the formulation of ( [ ! lcmle ] ) , any sequence of putative maximizers is attracted to the the linear combination of point masses situated at the data points . one escape from this `` dirac catastrophe '' involves regularization by introducing a roughness , or complexity , penalty ; various proposals in this vein can be found in @xcite , @xcite , @xcite and @xcite . another way to obtain a well - posed problem is by imposing shape constraints , a line of development dating back to the celebrated @xcite nonparametric maximum likelihood estimator for monotone densities . while monotonicity regularizes the maximum likelihood estimator , unimodality _ per se_somewhat surprisingly does not . the desired effect is achieved only by enforcing somewhat more stringent shape constraints for instance log - concavity , sometimes also called `` strong unimodality . '' an advantage of shape constraints over regularization based on norm penalties is that it is not encumbered by the need to select additional tuning parameters ; on the other hand , it is limited in scope applicable only when the shape constraint is plausible for the unknown density . expanding the scope of our investigation , we now replace @xmath44 in the integral of the objective function by a generic function @xmath45 and define @xmath46 the following conditions on the form of @xmath47 will be imposed : * @xmath47 is a nonincreasing , proper convex function on @xmath48 . * the domain of @xmath47 is an open interval containing @xmath49 . * the limit , as @xmath50 , of @xmath51 is @xmath5 for every real @xmath52 and any @xmath53 . * @xmath47 is differentiable on the interior of its domain . * @xmath47 is bounded from below by @xmath17 , with @xmath54 when @xmath55 . the most crucial condition is ( a1 ) ensuring the convexity of @xmath56 . condition ( a2 ) assures that @xmath57 is finite for all @xmath58 , while ( a3 ) is required in the proof of the existence of the estimates . the relationship between primal and dual formulations of the estimation problem is facilitated by ( a4 ) , and ( a5 ) rules out possible complications regarding the existence of the integral @xmath59 in ( [ ! probj ] ) , allowing for the convention @xmath60 . in the spirit of the lebesgue integration theory , the integral then _ exists _ , although @xmath45 does not have to be _ summable _ : it is either finite [ which is automatically true for any @xmath41 convex and @xmath61 or @xmath5 . in the latter case , the objective function @xmath62 is considered to be equal to @xmath5 ; @xmath62 is also @xmath5 if @xmath63 for some @xmath35 , which occurs unless all @xmath35 lie in the domain of @xmath41 . on the other hand , any @xmath41 equal to some positive constant on @xmath33 and @xmath5 elsewhere yields @xmath64 . a rigorous treatment without assumption ( a5 ) , that is , for functions @xmath47 not bounded below , would introduce technicalities involving handling of the integrals in the spirit of singular integrals of calculus of variations , a strategy resembling the contrivance of @xcite of subtracting a fixed quantity from the objective function to ensure finiteness of the integral . although we do not believe that such formal complications are unsurmountable , we do not pursue such a development . careful deliberation reveals that replacing @xmath41 by its closure ( lower semicontinuous hull ) does not change the integral term in ( [ ! probj ] ) , and potentially only decreases the first term ; this means that without any restriction of its scope , we may reformulate the estimation problem as @xmath65 where @xmath66 stands for the class of closed ( lower semicontinuous ) convex functions on @xmath31 . unlike in ( [ ! warden ] ) , @xmath45 is not necessarily the estimated density @xmath3 ; the relationship of @xmath41 to @xmath3 will be revealed in section [ s : dual ] , together with the motivation leading to concrete instances of some possible functions @xmath47 . we now establish that the estimates , the solutions of ( [ ! dumb ] ) , admit a finite - dimensional characterization , which is a key to many of their theoretical properties . for every collection @xmath67 of points @xmath68 and @xmath69 , we define a function @xmath70 any function of this type is finitely generated in the sense of @xcite , whose corollary 19.1.2 asserts that it is polyhedral , being the maximum of finitely many affine functions , and therefore convex . the convention @xmath71 used in ( [ ! lch ] ) means that the domain of @xmath72 is equal to @xmath33 . if @xmath73 is a convex function such that @xmath74 , for all @xmath75 , then @xmath76 for all @xmath7 ; the function @xmath72 is thus the maximum of convex functions with this property the lower convex hull of points @xmath77 . for fixed @xmath32 , we will denote the collection of all functions @xmath72 of the form ( [ ! lch ] ) by @xmath78 . the collection @xmath67 determines @xmath72 uniquely , by virtue of its definition ( [ ! lch ] ) . given @xmath32 , we call a vector @xmath79 with components @xmath80 _ discretely convex relative to _ @xmath32 , if there exists a convex function @xmath73 defined on @xmath33 such that @xmath81 . any function @xmath41 from @xmath78 determines a unique discretely convex vector @xmath82 . the converse is also true : there is a one one correspondence between @xmath78 and @xmath83 , the set of all vectors discretely convex relative to @xmath32 . [ t : first ] suppose that assumption holds true . for every convex function @xmath73 on @xmath31 , there is a function @xmath84 such that @xmath85 ; the strict inequality holds whenever @xmath86 and @xmath33 has nonempty interior . the theorem shows that it is sufficient to seek potential solutions of ( [ ! dumb ] ) in @xmath78 ; this means , due to the one one correspondence of the latter to @xmath87 , that the optimization task ( [ ! dumb ] ) is essentially finite dimensional . the theorem also justifies the transition to a more convenient optimization domain in the primal formulation appearing in the next section . the conjugate dual formulation of the primal estimation problem ( [ ! dumb ] ) conveys a maximum entropy interpretation and leads us to several concrete proposals for @xmath47 . to conform to existing mathematical apparatus , we begin by further clarifying the optimization and constraint functional classes of our primal formulation . for definitions and general background on convex analysis , our primary references are rockafellar ( @xcite , @xcite ) and @xcite ; we may also mention @xcite and @xcite . hereafter , @xmath88 will denote the cone of closed ( lower semicontinuous ) convex functions on @xmath33 , the convex hull of @xmath32 . this cone is a subset of @xmath89 , the collection of functions continuous on @xmath33 ; it is important that @xmath89 is a linear topological space , with respect to the topology of uniform convergence . note that @xmath90 . in view of theorem [ t : first ] , any solution of ( [ ! dumb ] ) is also the solution of @xmath91 and conversely ; thus , we will refer to ( [ ! primal ] ) as our _ primal formulation_. the conjugate of @xmath47 is @xmath92 since @xmath47 is nonincreasing , there are no affine functions with positive slope that minorize the graph of @xmath47 , hence @xmath93 for all @xmath94 . if @xmath47 is differentiable on the ( nonempty ) interior of its domain , then @xmath95 can be obtained using differential calculus as the _ legendre transformation _ of @xmath47 ; denoting the derivative @xmath96 by @xmath97 , we have @xmath98 where @xmath99 is any solution , @xmath100 , of the equation @xmath101 . the ( topological ) dual of @xmath102 is @xmath103 , the space of ( signed ) radon measures on @xmath33 ; its distinguished element is @xmath104 , the empirical measure supported by the data points @xmath35 . the _ polar cone _ to @xmath88 is @xmath105 [ t : dual ] suppose that assumptions and hold . the strong ( fenchel ) dual of the primal formulation ( [ ! primal ] ) is @xmath106\\[-8pt ] \eqntext{g \in{\mathcal k}(x)^{-},}\end{aligned}\ ] ] in the sense that the value , @xmath62 , of the primal objective for any @xmath41 satisfying the constraints of ( [ ! primal ] ) , dominates the value , for any @xmath3 satisfying the constraints of ( [ ! dual ] ) , of the objective function in ( [ ! dual ] ) ; the minimal value of ( [ ! primal ] ) and maximal value of ( [ ! dual ] ) coincide . moreover , there exists @xmath3 attaining the maximal value of ( [ ! dual ] ) . any dual feasible function @xmath3 , that is , any @xmath3 satisfying the constraints of ( [ ! dual ] ) and yielding finite objective function of ( [ ! dual ] ) , is a probability density with respect to the lebesgue measure : @xmath107 and @xmath108 . if condition is also satisfied , then the dual and primal optimal solutions satisfy the relationship @xmath109 . it should be emphasized that the expression of absolute continuity in ( [ ! dual ] ) is a requirement on @xmath110 ; the dual objective function is defined as the conjugate to the primal objective function @xmath56 , and is equal to @xmath28 for any radon measure that is _ not _ absolutely continuous with respect to the lebesgue measure . this is how regularization operates here : only those @xmath2 qualify for which @xmath104 gets canceled with the discrete component of @xmath111 . once @xmath2 satisfies this requirement , its density integrates to @xmath16 , as shown in the proof of theorem [ t : dual ] . the nonnegativity for @xmath3 yielding finite dual objective function is the consequence of @xmath112 being infinite for @xmath113 . in practical implementations , it may be prudent to enforce @xmath107 in the dual explicitly as a feasibility constraint . an immediate consequence of theorem [ t : dual ] is that we can reformulate the maximum likelihood problem posed in ( [ ! warden ] ) as an equivalent maximum ( shannon ) entropy problem . [ c : mledual ] maximum likelihood estimation of a log - concave density as posed in ( [ ! warden ] ) has an equivalent dual formulation @xmath114\\[-8pt ] \eqntext{g \in{\mathcal k}(x)^{-},}\end{aligned}\ ] ] whose solution satisfies the relationship @xmath39 , where @xmath41 is the solution of ( [ ! warden ] ) . in particular , the solution of ( [ ! warden ] ) satisfies @xmath115 , therefore problems ( [ ! lcmlelog ] ) and ( [ ! warden ] ) are equivalent . the emergence of the shannon entropy is hardly surprising in view of the well - established connections of maximum likelihood estimation to the kullback leibler divergence and maximum entropy . note that the dual criterion can be also interpreted as choosing the @xmath3 closest in the kullback leibler divergence to the uniform distribution on @xmath33 , from all @xmath3 satisfying the dual constraints . while the outcome of corollary [ c : mledual ] , the equivalence of ( [ ! lcmlelog ] ) and ( [ ! warden ] ) , could be also shown by elementary means , it is important to emphasize that the real value of the dual connection lies in the vista of new possibilities it opens . to explore the link to potential alternatives , we consider the family of entropies originally introduced for @xmath116 by rnyi ( @xcite , @xcite ) , @xmath117 as an extension of the limiting case for @xmath118 , the shannon entropy . for @xmath119 , maximizing ( [ ! rent ] ) over @xmath3 is equivalent to the maximization of @xmath120 the dependence of convexity / concavity properties of @xmath121 necessitates a separate treatment of the cases with @xmath122 , when the conjugate pair is @xmath123 and the cases with @xmath124 , where @xmath125 where @xmath126 and @xmath127 are conjugates in the usual sense that @xmath128 . see figure [ f:1 ] . ( left ) and dual @xmath95 ( right ) for selected @xmath129 from the rnyi family of entropies . ] for @xmath122 , we may write @xmath132 instead of @xmath133 , and then introduce @xmath134 . the resulting primal formulation is @xmath135 by theorem [ t : first ] , this formulation is equivalent to @xmath136 after substituting @xmath137 for @xmath73 , multiplying by @xmath126 , and rewriting in terms of @xmath127 we obtain a new objective function @xmath138 which recalls the `` minimum density power divergence estimators , '' proposed , for @xmath139 , by @xcite in the context of estimation in parametric families . although @xmath141 is a special case of the power divergence family mentioned above , it deserves a special mention . the choice of @xmath142 in the rnyi family leads to the dual formulation @xmath143 the primal formulation can be written , after the application of theorem [ t : first ] , in a particularly simple form @xmath144 which can be interpreted as a variant of the minimum pearson @xmath140 criterion . a similar theme can be found in the dual , which can be interpreted as returning among all densities satisfying its constraints the one with minimal pearson @xmath140 distance to the uniform density on @xmath89 . the relation between primal and dual optimal solutions is @xmath145 ; the convexity constraint on @xmath41 therefore implies that @xmath3 must be concave . replacing @xmath41 in ( [ ! renprim ] ) by @xmath146 and appropriately modifying the cone constraint gives a variant of the `` least - squares estimator , '' studied by @xcite and going back at least to @xcite ; the estimate was defined to estimate a _ convex _ ( and decreasing ) density on @xmath147 , a domain that is apparently still under the scope of theorem [ t : first ] . while the form of the objective function for @xmath141 has some computational advantages , its secondary consequence constraining the density itself to be concave rather than its logarithm is not at all appealing . indeed , all rnyi choices with @xmath122 impose a more restrictive form of concavity than log - concavity . from our perspective , it seems more reasonable to focus attention on weaker forms of concavity , corresponding to @xmath148 . apart from the celebrated log - concave case @xmath118 , a promising alternative would seem to be rnyi entropy with @xmath149 . this choice in the rnyi system leads to the dual @xmath150 and primal , again after the application of theorem [ t : first ] , @xmath151 the estimated density satisfies @xmath152 , which means that the primal constraint , @xmath153 , enforces the convexity of @xmath154 . in the terminology of section [ s : mle ] , the estimated density is now required to be only @xmath155-concave , a significant relaxation of the log - concavity constraint ; in addition to all log - concave densities , all the student @xmath156 densities with @xmath157 satisfy this requirement . the dual problem ( [ ! dhell ] ) can be interpreted as a hellinger fidelity criterion , selecting from the cone of dual feasible densities the one closest in hellinger distance to the uniform distribution on @xmath33 . although the original rnyi system was confined to @xmath116 , a limiting form for @xmath158 can be obtained similarly to the @xmath118 case . it yields the conjugate pair @xmath159 as is apparent from figure [ f:1 ] , this @xmath47 violates our condition ( a5 ) , but may nevertheless deserve a brief consideration . note first that the possible complications with existence of integrals may occur only in the formulation ( [ ! dumb ] ) with unbounded domain not in ( [ ! primal ] ) , where all integrals are of bounded functions over a compact domain . the major technical complications with @xmath47 violating ( a5 ) concern theorems in section [ s : exfish ] , and are briefly discussed there . here we mention only that the resulting dual , adapted directly from ( [ ! dual ] ) , is @xmath160 and the primal becomes @xmath161 in this case @xmath162 , and the estimate is constrained to be @xmath18-concave , a yet still weaker requirement that admits all of the student @xmath156 densities for @xmath163 . if we interpret the dual problem ( [ ! mledual ] ) , for @xmath118 , as choosing a constrained @xmath3 to minimize the kullback leibler divergence of @xmath3 from the uniform distribution on @xmath33 , we can similarly interpret the @xmath158 dual as minimizing the reversed kullback leibler divergence . in parametric estimation , the latter objective is sometimes associated with empirical likelihood , while the former is associated with exponentially tilted empirical likelihood . see , for example , @xcite for related discussion in the context of kernel density estimation , and @xcite . one might try to continue in this fashion marching inexorably toward weaker and weaker concavity requirements . there appears to be no obstacle in considering @xmath164 ; the general form ( [ ! genprim ] ) of the primal is still applicable . the shape constraints corresponding to negative @xmath127 encompass a wider and wider class of quasi - concave densities , eventually arriving at the @xmath28-concave constraint , at which point we would have sanctioned all of the quasi - concave densities . but formal complications , as well as computational difficulties dictate the more prudent strategy of restricting attention to @xmath116 cases . returning to our general setting , existence , uniqueness and fisher consistency are established under mild conditions on the function @xmath47 . theorem [ t : first ] not only justifies the choice of the optimization domain in ( [ ! primal ] ) , but also shows , due to the one one correspondence between @xmath78 and @xmath87 , that the optimization task ( [ ! primal ] ) is essentially finite dimensional , parametrized by the values @xmath165 . this facilitates the proof of the following existence result . [ t : ex ] suppose that assumptions and hold , and that @xmath33 has a nonempty interior . then the formulation ( [ ! dumb ] ) has a solution @xmath166 ; if @xmath47 is strictly convex , then this solution is unique . in our general setting , a comprehensive asymptotic theory for the proposed estimators remains a formidable objective . considerable recent progress has been made on theory for the univariate log - concave ( @xmath118 ) maximum likelihood estimator : @xcite proved consistency in the hellinger metric , @xcite prove consistency in the supremum norm on compact intervals , and @xcite derive asymptotic distributions . for maximum likelihood estimators in @xmath31 , @xcite establish consistency for estimators of a log - concave density , and @xcite for estimators of convex - transformed densities . these results are surely suggestive of the plausibility of analogous results for other @xmath127 and dimensions greater than one . however , the highly technical nature of the proofs , and their strong reliance on special features of the univariate setting indicate that such a development may not be immediate . while anything else in this direction may be viewed as speculative , fisher consistency , a crucial prerequisite for a more detailed asymptotic theory , can be verified in a quite straightforward manner and essentially complete generality . for differentiable @xmath47 , theorem [ t : dual ] gives the relationship between the solution @xmath41 of the optimization task ( [ ! primal ] ) and the density estimate : @xmath167 . using the notation @xmath97 for @xmath96 , and @xmath168 for its inverse , as in section [ s : dual ] , we may write @xmath169 , and subsequently rewrite the formulation ( [ ! dumb ] ) in terms of the estimated density @xmath3 ( omitting , for brevity , the integration variables ) @xmath170\\[-8pt ] \eqntext{\mbox{subject to $ \chi^{-1}(-f ) \in{\mathcal k}$}.}\end{aligned}\ ] ] this yields a new objective function which we nevertheless denote , slightly abusing the notation , also @xmath56 . the population version of this @xmath56 is obtained by replacing @xmath171 by @xmath172 : @xmath173 the fisher consistency for an estimator defined by solving ( [ ! primf ] ) requires that @xmath174 , for every @xmath3 ; however , there may be a formal problem now with the existence of the integral in ( [ ! popsi ] ) , as @xmath175 may take both positive or negative values . a possible way of handling this obstacle is the strategy of @xcite , briefly mentioned in section [ s : mle ] : instead of @xmath56 , we consider a modified objective function @xmath176 which , when minimized over @xmath3 satisfying the constraint of ( [ ! primf ] ) , yields an optimization problem equivalent with ( [ ! primf ] ) , since the difference of @xmath56 and @xmath177 is constant in @xmath3 . however , the population version of @xmath178 @xmath179 is now better suited for the ensuing version of the fisher consistency theorem . [ t : fish ] suppose that @xmath47 satisfies assumptions and . the integrand in ( [ ! pophub ] ) is then nonnegative for any probability density @xmath3 such that @xmath180 , and identically equal to @xmath17 for @xmath181 ; therefore , @xmath182 , where @xmath183 is well defined for every @xmath3 , possibly equal to @xmath5 . in fact , theorem [ t : fish ] can be proved in the same manner for the unmodified @xmath56 , if @xmath184 with @xmath185 . then the inverse of @xmath186 , and hence the range of @xmath168 is bounded from below by @xmath187 . in such a case , @xmath188 , so the first term in ( [ ! popsi ] ) is minorized by an integrable function @xmath189 ; the second term is bounded from below by @xmath17 by assumption ( a5 ) , so the whole integral then exists in the lebesgue sense , being either finite or equal to @xmath5 . if , however , assumption ( a5 ) is not satisfied , then the existence of the integral should be assumed explicitly ; we return to this point briefly at the end of the proof of theorem [ t : fish ] . note that , by comparing ( [ ! legendre ] ) and ( [ ! popsi ] ) , existence of the integral is equivalent to assuming the integrability ( summability ) of @xmath190 that is , the existence and finiteness of the entropy term in the dual ( [ ! dual ] ) . we employed two independent algorithms for solving the convex programming problems posed above : ` mskscopt ` from the mosek software package of @xcite , and the ` pdco ` matlab procedure of @xcite . both algorithms are coded in matlab and employ similar primal - dual , log - barrier methods . further details regarding numerical implementation appear in appendix [ appb ] . the crux of both algorithms is a sequence of newton - type steps that involve solving large , very sparse least squares problems , a task that is very efficiently carried out by modern variants of cholesky decomposition . several other approaches have been explored for computing quasi - concave density estimators that are log - concave . an active set algorithm for univariate log - concave density estimation was described in @xcite and implemented in the r package ` logcondens ` of @xcite . @xcite have recently implemented a promising steepest descent algorithm for multivariate log - concave estimation that may be adaptable to other quasi - concave density estimation problems . to illustrate the application of the foregoing methods , we briefly consider some realistic examples . our first example features data similar to those considered by @xcite , the type of data where shape constraints sometimes arise in a natural manner . the two samples consist of 9092 measurements of radial and 3933 of rotational velocity for the stars from bright star catalog , @xcite . the left and right panels of figure [ figstars ] show the results for the radial and rotational velocity samples , respectively . -concave ) estimates . ] the broken line in the upper panels shows kernel density estimates , each time with default matlab bandwith selection ; the solid lines correspond to one of the norm penalized estimates proposed in @xcite : maximum likelihood penalized by the total variation of the second derivative of the logarithm of the estimated density . this is the @xmath191 version of the silverman s ( @xcite ) estimator penalizing the squared @xmath192 norm of the third derivative . the smoothing parameter for the latter estimate was set quite arbitrarily at @xmath16 ; it seems that this arbitrary choice works quite satisfactorily here , providing somewhat surprisingly , for both samples about the same level of smoothing as the kernel estimator . for the radial velocity sample , the two estimates are essentially the same . for the rotational velocity sample , however , the right upper panel shows that the kernel density estimate differs substantially from the penalized one . both estimators have the unfortunate effect of assigning considerable mass to negative values despite the fact that there are no negative observations . this effect is somewhat more pronounced for the kernel estimate . since the preliminary analyses of the upper panels indicates that the hypothesis of unimodality is plausible for both of the datasets , a natural next step is the application of a shape - constrained estimator a move that , among other things , may relieve us of insecurities related to the arbitrary choice of smoothing parameters . the broken line in the lower panels of figure [ figstars ] shows the log - concave maximum likelihood ( @xmath193 ) , and the solid line the hellinger ( @xmath155-concave ) estimate ( @xmath194 ) . while , as expected , there is almost no difference between the two ( and in fact , among all four ) estimates for the radial velocity dataset , the right lower panel reveals that the log - concave estimate yields for the rotational velocity sample a density that is monotonically decreasing which contradicts the evidence suggested by all other methods . the hellinger estimate , on the other hand , exhibits a subtle , but visible bump at the location of the plausible mode , thus turning out to be visually more informative about the center of the data than the tails . this is somewhat paradoxical given its original heavy - tail motivation , confirming that the real universe of data analysis can be much more subtle than that of the surrounding theoretical constructs . to illustrate our approach in a simple bivariate setting , we reconsidered the well - known @xcite data on the heights and left middle finger lengths of 3000 british criminals . this data was employed by gosset in preliminary simulation work described in @xcite . -concave ) estimate of the density of student s criminals . contours are labeled in units of log - density . ] figure [ fig7 ] illustrates the hellinger ( @xmath155-concave ) fit of this data . contours are labeled in units of log - density . a notable feature of the data is the unusual observation in the middle of the upper edge . this point is highly anomalous , at least for any density with exponential tail behavior . the maximum likelihood estimate of the log - concave model in figure [ fig8 ] has very similar central contours , but the outer contours fall off much more rapidly implying that the log - concave estimate assigns much smaller probability to the region near the unusual point . motivated by a suggestion of one of the referees , we undertook some numerical experiments to explore performance of our shape constrained estimators and evaluate whether consistency appeared to be a plausible conjecture . for the log - concave estimator pal , woodroofe and meyer ( @xcite ) report `` hellinger error '' for a fully crossed design involving five target densities and five sample sizes with 500 replications per cell . in figure [ logc ] , we report results of our attempt to reproduce the pwm experiment expanded somewhat to consider two competing estimators : the adaptive kernel estimator of silverman ( @xcite ) using a gaussian kernel , and the logspline estimator of kooperberg and stone ( @xcite ) as implemented in the ` logspline ` r package of kooperberg . five target densities are considered : ( standard ) normal , laplace , gamma(3 ) , beta(@xmath195 ) and weibull(@xmath196 ) as in pwm . five sample sizes are studied : 50 , 100 , 200 , 500 , 1000 . and two measures of performance are considered : squared hellinger distance as in pwm in the left panel and @xmath197 distance in the right panel . plotted points in these figures represent cell means . both figures support the contention that the rates of convergence are comparable for all three estimators . -concave densities . ] figure [ hellc ] reports results from a similar experiment for the the @xmath155-concave estimator described in section [ sec36 ] . we consider five new target densities : lognormal , @xmath198 , @xmath199 , @xmath200 and pareto(5 ) , all of which fall into the @xmath155-concave class . the same competing estimators and sample sizes are used . in a small fraction of cases for the second group of densities , less than 0.2 percent , there were problems either with the convergence of the logspline or shape - constrained estimator , or with the numerical integration required to evaluate the performance measures , so figure [ hellc ] plots cell medians rather than cell means . again , the figures support the conjecture that the rates of convergence for the shape - constrained estimator are competitive with those of the adaptive kernel and logspline estimators . @ld2.4d2.4d2.4@ & & & + l1 error & -0.417 & -0.366 & -0.393 + & ( 0.018 ) & ( 0.003 ) & ( 0.012 ) + hellinger & -0.875 & -0.498 & -0.698 + & ( 0.032 ) & ( 0.031 ) & ( 0.021 ) + a concise way to summarize results from these experiments is to estimate the simple linear model @xmath201 where @xmath202 denotes a cell average of our two error criteria for one of our three estimators , for target density @xmath75 and sample size @xmath203 . in this rather nave framework , @xmath204 can be interpreted as an empirical rate of convergence for the estimator . in tables [ pwmtab ] and [ helltab ] , we report these estimates suppressing the estimated target density @ld2.4d2.4d2.4@ & & & + l1 error & -0.405 & -0.324 & -0.386 + & ( 0.004 ) & ( 0.008 ) & ( 0.01 ) + hellinger & -0.751 & -0.355 & -0.672 + & ( 0.034 ) & ( 0.023 ) & ( 0.019 ) + specific @xmath205 s . in this comparison too , the shape constrained estimators perform quite well . we have described a rather general approach to qualitatively constrained density estimation . log - concave densities are an important target class , but other , weaker , concavity requirements that permit algebraic tail behavior are also of considerable practical interest . ultimately , the approach accommodates all quasi - concave densities as a limit of the rnyi entropy family . there are many unexplored directions for future research . as we have seen , a consequence of the variational formulation of our concavity constraints is that the estimated densities vanish off the convex hull of the data . various treatments for this malady may be suggested . @xcite have recently suggested applying one of several estimators of the pareto tail index to the smoothed ordinates from the log - concave preliminary density estimator . our inclination would be to prefer solutions that impose further regularization on the initial problem . thus , for example , we can add a new penalty term to the primal problem , penalizing the total variation of the derivative ( gradient ) of @xmath206 , and choosing a suitable value of the associated lagrange multiplier to smooth the tail behavior at the boundary . we have adhered , thus far , to the principle that the entropy choice in the fidelity criterion of the dual problem should dictate the form of the convexity constraint : likelihood thus implies log - concavity , hellinger fidelity implies @xmath207 concavity , etc . one may wish to break this linkage and consider maximum likelihood estimation of general @xmath0-concave densities . this may have some advantages from an inference viewpoint , at the cost of complicating the numerical implementation . embedding shape constrained density estimation of the type considered here into semiparametric methods would seem to be an attractive option in many settings . and , it would obviously be useful to consider inference for the validity of shape constraints in the larger context of penalized density estimation . we hope to pursue some of these issues in future work . proof of theorem [ t : first ] given @xmath73 convex , put @xmath208 and take @xmath209 , the function defined by ( [ ! lch ] ) . the convexity of @xmath73 implies that @xmath210 for every @xmath7 ; since @xmath47 is nonincreasing , we have @xmath211 the definition of @xmath72 implies that @xmath212 ; therefore , the rest of @xmath56 remains unchanged , and ( [ ! intineq ] ) implies that @xmath85 . suppose that @xmath86 . then @xmath213 and the inequality @xmath214 implies that @xmath215 . if @xmath216 , then there is a point @xmath217 from the interior of @xmath218 , because @xmath219 is closed . the continuity of @xmath73 at @xmath220 implies that @xmath221 for all @xmath7 from an open neighborhood of @xmath220 ; this proves that @xmath222 . if @xmath223 , then the polyhedral character of @xmath33 implies through theorem 10.2 of @xcite that @xmath73 is upper semicontinuous relative to @xmath33 at any @xmath224 ; that is , if @xmath225 for some @xmath226 , then the inequality holds for all @xmath7 in an open , relatively to @xmath33 , neighborhood of @xmath220 . such a relative neighborhood has positive lebesgue measure , due to the fact that the interior of @xmath33 is nonempty . hence , we have @xmath222 also in this case and the strict inequality @xmath227 follows . proof of theorem [ t : dual ] we use the conventional notation @xmath228 to denote @xmath229 , the result of the application of a linear functional to @xmath7 . the definition of the conjugate of a convex function @xmath230 in this notation is @xmath231 the resulting function is convex itself , being a sup of affine functions . for any @xmath232 and any radon measure @xmath111 , a linear functional from @xmath233 , we have @xmath234 we start the proof by rewriting ( [ ! primal ] ) as @xmath235 where @xmath56 is the original objective function of ( [ ! primal ] ) and @xmath236 is the indicator function of @xmath237 . the expression for the fenchel dual of this type of problem follows from @xcite ; see also @xcite , section 5 , example 11 : @xmath238 ( note that one of the conjugates , in both cites sources , is in the `` concave '' sense , which explains the negative sign of the argument in the second term , but not in the first ) . the conjugate of the indicator of a convex cone @xmath237 is the indicator of @xmath239 [ @xcite , section 3 , equation ( 3.14 ) ] . the term @xmath240 in the objective can be therefore interpreted as a constraint @xmath241 , that is , @xmath242 . the definition of the conjugate of @xmath56 gives @xmath243\\[-8pt ] & = & \sup_g \biggl ( \langle g - p_n , g \rangle- \int\psi(g(x ) ) \,dx \biggr ) = \psi^{\ast}(g - p_n);\nonumber\end{aligned}\ ] ] the sup is taken over all @xmath41 from @xmath244 where @xmath245 is the functional given by @xmath246 and @xmath247 is its conjugate . the form of the latter is given by @xcite , corollary 4a : if @xmath111 is absolutely continuous with respect to the lebesgue measure , then @xmath248 otherwise @xmath249 . these facts , and expressions ( [ ! one ] ) and ( [ ! two ] ) , yield ( [ ! dual ] ) . rockafellar [ ( @xcite ) , theorem 1 ; see also rockafellar ( @xcite ) , section 8 , example 11@xmath250 gives also a constraint qualification for this type of problem : to prove strong duality , we need to find some @xmath41 where both @xmath56 and @xmath251 are finite and one of them is continuous . such a @xmath41 is provided by a function constant on @xmath33 , say @xmath252 for all @xmath224 . it is convex , thus @xmath253 is finite . so is @xmath62 ; the topology on @xmath102 is that of uniform convergence , and @xmath47 is continuous at @xmath16 , hence there is a neighborhood of @xmath41 containing only functions for which @xmath56 is finite and @xmath56 is continuous at @xmath41 . once the constraint qualification is verified , we know that the primal and dual optimal values coincide ( zero duality gap ) , and that the dual is attained there is an optimal solution to the dual ; see theorem 52.b(3 ) of @xcite . due to the fact that @xmath47 is decreasing , @xmath254 whenever @xmath255 ; thus , if @xmath3 yields a finite dual objective function , then @xmath3 is nonnegative . if @xmath256 , then @xmath257 for every @xmath258 ; consequently , @xmath259 therefore , @xmath260 and for every dual feasible @xmath3 , @xmath261 that is , every dual feasible @xmath3 is a probability density with respect to the lebesgue measure . if a primal solution , @xmath41 , exists the fact that is established by theorem [ t : ex ] , but here we are exploring only the consequences of such a premise it is related to the dual solution , @xmath3 , via extremality ( karush kuhn tucker ) conditions . the form of this relationship asserted by the theorem follows from the second condition of ( 8.24 ) in rockafellar [ ( @xcite ) , section 8 , example 11@xmath250 , together with the form of the subgradient of @xmath245 given by rockafellar [ ( @xcite ) , corollary 4b ] , combined with the fact established above that the estimated density @xmath3 corresponds to @xmath262 . proof of theorem [ t : ex ] by theorem [ t : first ] , any potential solution of ( [ ! dumb ] ) lies within the class @xmath78 of polyhedral functions ; due to the one one correspondence between @xmath78 and @xmath87 , the set of vectors discretely convex relative to @xmath32 , the existence proof can be carried for ( [ ! primal ] ) reparametrized by @xmath263 , the putative values of @xmath264 . in what follows , @xmath32 remains fixed , and @xmath127 , @xmath126 will denote generic coefficients of convex combinations : any real numbers satisfying @xmath265 , @xmath266 . as the correspondence between @xmath78 and @xmath87 is not a linear mapping ( except for @xmath267 ) , the first thing to be shown is that ( [ ! primal ] ) remains a convex problem when reparametrized in terms of a vector @xmath268 , with components @xmath269 . the resulting problem minimizes , over @xmath268 , the objective function @xmath270 note first that @xmath87 is a convex subset of @xmath271 : if @xmath272 , then there exist convex functions @xmath41 , @xmath73 satisfying @xmath82 and @xmath273 ; subsequently , @xmath274 , and as @xmath275 is also a convex function , we obtain that @xmath276 for any convex combination of @xmath79 and @xmath277 . thus , it is sufficient to show that @xmath278 is convex on @xmath279 , which amounts to demonstrating the convexity of the function @xmath280 . let @xmath281 ; the definition ( [ ! lch ] ) gives for their convex combination @xmath282 as @xmath47 is nonincreasing and convex , we obtain @xmath283\\[-8pt ] & & \qquad\leq\int\alpha\psi\bigl(g_{(x , y)}(x)\bigr ) + \beta\psi\bigl(g_{(x , z)}(x)\bigr ) \,dx \nonumber\\ & & \qquad= \alpha\int\psi\bigl(g_{(x , y)}(x)\bigr ) \,dx + \beta\int\psi\bigl(g_{(x , z)}(x)\bigr ) \,dx\nonumber\end{aligned}\ ] ] as was required . note that the integral is also finite whenever @xmath79 has all components in the domain of @xmath47 , due to the polyhedral character of @xmath284 and the fact that @xmath47 is nonincreasing and @xmath33 is bounded . otherwise , it may be equal only to @xmath5 ; hence @xmath285 is a proper convex function . [ l : multi ] suppose that @xmath286 are vectors in @xmath31 such that @xmath287 has constant components , and @xmath277 is arbitrary . for any @xmath288 , @xmath289 note first that by the definition , @xmath290 identically on @xmath33 for constant @xmath79 ; likewise , for every @xmath224 , @xmath291 proving the lemma . choose a real number @xmath52 lying in the domain of @xmath47 , and set @xmath292 . according to lemma [ l : multi ] , @xmath293 for every @xmath224 ; the function constant on @xmath33 is convex , hence @xmath294 . then @xmath295 therefore , @xmath79 lies in the domain of @xmath278 . for arbitrary @xmath296 , not identically @xmath17 , and @xmath288 , we have @xmath297 we know that @xmath278 is a convex function on a finite - dimensional linear space @xmath271 ; to establish the existence of its minimizer , it is sufficient to show that @xmath298 for @xmath50 , which means that we need to verify that the limit of the expression in the parentheses is positive ( possibly @xmath5 ) ; see also @xcite , remark 3.2.8 . let @xmath299 , @xmath300 , @xmath301 be sets in @xmath33 where @xmath302 is , respectively , negative , zero or positive ; we are to examine the limit behavior of @xmath303\\[-8pt ] & & \qquad { } + \int_{e^0 } \frac{\psi(y)}{\tau } \,dx + \int_{e^+ } \frac{\psi(y + \tau g(x))}{\tau } \,dx,\nonumber\end{aligned}\ ] ] when @xmath50 . for the integral over @xmath300 , the limit is obviously zero . if @xmath47 satisfies assumptions ( a1 ) and ( a5 ) , then @xmath47 is nonincreasing and converging to @xmath17 when @xmath50 ; for every @xmath304 then @xmath305 monotonically decreases with increasing @xmath306 , hence the limit of the integral over @xmath301 is zero as well . finally , if @xmath47 satisfies also ( a3 ) , then for every @xmath307 the limit of @xmath305 is @xmath5 ; at the same time , the expression is bounded from below by @xmath17 , due to ( a4 ) . the application of the fatou lemma then gives that the limit of the integral over @xmath299 is @xmath5 , whenever @xmath301 has positive lebesgue measure . the proof of the theorem is then finished by the examination of possible alternatives . if the first term in ( [ ! trint ] ) , the mean of the @xmath308 s , is positive , then the theorem is proved , as all other terms in ( [ ! trint ] ) converge either to @xmath17 or @xmath5 . if the first term of ( [ ! trint ] ) is negative or equal to zero , then there must be some @xmath309 ( the case with all @xmath310 is excluded ) . that means that @xmath311 is negative for some @xmath7 ; this implies that @xmath299 has positive lebesgue measure , and then the limit of the second term in ( [ ! trint ] ) , the integral over @xmath299 , and thus of the whole expression ( [ ! trint ] ) is @xmath5 . this proves the theorem . under the strict convexity of @xmath47 , the strict convexity of @xmath278 follows ( for appropriate @xmath127 , @xmath126 ) from the second inequality in ( [ ! convex ] ) , which becomes sharp this is due to the fact that the sharp inequality holds pointwise for all @xmath7 , and thus for the integrals as well . the strict convexity of @xmath278 then implies the uniqueness of the solution . finally , functions @xmath47 satisfying ( a1)(a3 ) , but not necessarily ( a5 ) require some special considerations . for the integral over @xmath299 , we have to observe that for every @xmath288 we have @xmath312 ; if @xmath313 , then @xmath314 for every @xmath315 , if @xmath316 then @xmath317 for every @xmath288 . hence we have also in this case an integrable constant minorant [ due to the fact that @xmath33 has finite lebesgue measure ] ; this justifies the limit transition via the fatou lemma . finally , for the integral over @xmath301 , we need to assume that the limit of the integrand is @xmath17 for every @xmath7 , and find an integrable minorant ; this may be related to the existence of the integral of the second term in ( [ ! dumb ] ) . proof of theorem [ t : fish ] the proof relies on the application of what is called fenchel s inequality by rockafellar [ ( @xcite ) , page 105 ] or the ( generalized ) young inequality by hardy , littlewood and plya [ ( @xcite ) , section 4.8 ] , or zeidler [ ( @xcite ) , section 51.1 ] . the inequality says says that for arbitrary @xmath318 and convex function @xmath47 @xmath319 applied pointwise to @xmath320 and @xmath321 , the inequality yields @xmath322 which is equivalent to the nonnegativity of the integrand in ( [ ! pophub ] ) . for @xmath181 , the equality ( [ ! legenf0 ] ) implies that @xmath323 which proves the theorem . for functions not satisfying ( a5 ) , integrability of @xmath324 is no longer equivalent to that of @xmath325 however , if we assume the integrability of the latter , then the proof can be carried through in the same way . our computational objective is to provide a unified algorithmic strategy for solving the entire class of problems described above . interior point methods designed for general convex programming and capable of exploiting the inherently sparse structure of the resulting linear algebra offer a powerful , general approach . we have employed two such implementations throughout our development process : the pdco algorithm of @xcite , and the mosek implementation of @xcite . our generic primal problem ( [ ! dumb ] ) involves minimizing an objective function consisting of a linear component , representing likelihood or some generalized notion of fidelity , plus a nonlinear component , representing the integrability constraint . minimization is then subject to a cone constraint imposing convexity . we will first describe our procedure for enforcing convexity , and then turn to the integrability constraint . in dimension one convexity of piecewise linear functions can be imposed easily by enforcing linear inequality constraints on a set of function values , @xmath326 at selected points @xmath327 . for ordered @xmath328 s , the convex cone constraint can be written as @xmath329 for a tridiagonal matrix @xmath330 that does second differencing , adapted to the possible unequal spacing of the @xmath328 s . in dimension two , enforcing convexity becomes more of a challenge . ideally , we would utilize knowledge of the polyhedral character of the optimal @xmath41 , established by theorem [ t : first ] , and implying that the optimal @xmath41 is piecewise linear over _ some _ triangulation of the observations @xmath331 . once we knew the triangulation , it is again straightforward to impose convexity : each interior edge of the triangulation generates one linear inequality on the coefficients @xmath332 . unfortunately , the complexity of traveling over a binary tree of possible triangulations of the observed points makes finding the optimal one difficult . the algorithm implemented in the ` logconcdead ` package of @xcite , for computing log - concave estimates , exploits special features of the log - concave mle problem and thus does not appear to be easily generalizable to our other settings . finite - element methods involving fixed ( delaunay ) triangulation of an expanded set of vertices were also ultimately deemed unsatisfactory . a superior choice , one that circumvents the difficulties of the finite - element , fixed triangulation approach , relies on finite differences . convexity is imposed directly at points on a regular rectangular grid using finite - differences to compute the discrete hessian : @xmath333/4 , \\ h_{21 } ( \xi_1,\xi_2 ) & = & h_{12 } ( \xi_1,\xi_2).\end{aligned}\ ] ] convexity is then enforced by imposing positive semidefiniteness . these constraints convexity at each of the grid points @xmath334produce a semi - definite programming problem . in the bivariate setting the semi - definiteness of each @xmath230 can be reformulated as a rotated quadratic cone constraint ; we need only constrain the signs of the diagonal elements of @xmath230 and its determinant . this simplifies the implementation of the hellinger estimator in mosek . for the relatively fine grid used for figure [ fig7 ] solution requires about 25 seconds , considerably quicker than the log - concave estimate of figure [ fig8 ] computed with the implementation of @xcite . for certain special @xmath47 , one can evaluate the integral term @xmath335 in the objective function of ( [ ! dumb ] ) explicitly as was done for @xmath336 by @xcite . while such a strategy may also be possible for certain other specific @xmath47 , we adopt a more pragmatic approach based on a straightforward riemannian approximation @xmath337 here , @xmath338 are weights derived from the configuration of @xmath328 . of course , with only a modest number of @xmath328 s such an approximation may be poor ; in dimension one we therefore augment the initial collection of the observed data @xmath339 by filling the gaps between their order statistics by further grid points , to ensure that the resulting grid ( not necessarily uniformly spaced ) and consisting of the observed data points as well as the new grid points , provides a sufficiently accurate approximation ( [ ! riemann ] ) . the @xmath338 s are then simply the averages of the adjacent spacings between the ordered @xmath328 s . given the size of problems modern optimization software can successfully handle , it is no problem to add an abundance of new points in dimension one . in dimension two , the approximation ( [ ! riemann ] ) is based on the uniformly spaced grid of the points used in the finite - difference approach described in the previous subsection . as the original data points @xmath35 may no longer lie among the grid points @xmath328 , we have to modify the fidelity component of the objective function : instead of obtaining @xmath264 directly , we obtain it via linear interpolation from the values of @xmath41 at the vertices of the rectangles enclosing @xmath35 . as long as the grid is sufficiently fine , the difference is minimal . we use this approach often also in dimension one , as it provides better numerical stability especially for fine grids and large data sets . adopting the procedures described above , we can write the finite - dimensional version of the primal problem as @xmath340 where @xmath341 denotes now the @xmath342-vector with typical element @xmath343 , @xmath344 is an `` evaluation operator '' which either selects the data elements from @xmath332 , or performs the appropriate linear interpolation from the neighboring ones , so that @xmath345 denotes the @xmath346-vector with typical element , @xmath264 , and @xmath347 is an @xmath346-vector of observation weights , typically @xmath348 . associated with the primal problem ( [ equp ] ) is the dual problem @xmath349 here , @xmath350 is an @xmath342-vector of dual variables and @xmath351 is an @xmath342-vector of function values representing the density evaluated at the @xmath328 s , and @xmath352 . the vector @xmath247 is the convex conjugate of @xmath245 defined coordinate - wise with typical element @xmath353 . problems ( [ equp ] ) and ( [ equd ] ) are strongly dual in the sense of the following result , which may viewed as the discrete counterpart of theorem [ t : dual ] . if @xmath47 is convex and differentiable on the interior @xmath354 of its domain , then the corresponding solutions of ( [ equp ] ) and ( [ equd ] ) satisfy @xmath355 whenever the elements of g are from @xmath354 and the elements of @xmath3 are from the image of @xmath354 under @xmath356 . for @xmath357 with typical element @xmath358 we have @xmath247 with elements @xmath359 , so the dual problem corresponding to maximum likelihood can be interpreted as maximizing the shannon entropy of the estimated density subject to the constraints appearing in ( [ equd ] ) . since @xmath41 was interpreted in ( [ equp ] ) as @xmath206 , this result justifies our interpretation of solutions of ( [ equd ] ) as densities provided that they satisfy our integrability condition . this is easily verified and thus justifies the implicit lagrange multiplier of one on the integrability constraint in ( [ equp ] ) , giving a discrete counterpart of theorem [ t : dual ] . let @xmath360 denote an @xmath342-vector of ones , and suppose in ( [ equp ] ) that @xmath361 and @xmath362 . then solutions @xmath351 of ( [ equd ] ) satisfy @xmath363 and @xmath364 . the crucial element of the proof is that the differencing operator d annihilates the constant vector and therefore the result extends immediately to other norm - type penalties as well as to the other entropy objectives that we have discussed . indeed , since the second difference operator representing our convexity constraint annihilates any affine function it follows by the same argument that the mean of the estimated density also coincides with the sample mean of the observed @xmath35 s . we are grateful to lutz dmbgen , kaspar rufibach , guenther walther and jon wellner for sending us preprints of their work , to the referees for their very constructive comments , and to mu lin for help with the bright star example and fisher consistency proof . dmbgen , l. , hsler , a. and rufibach , k. ( 2007 ) . active set and em algorithms for log - concave densities based on complete and censored data . available at http://arxiv.org/abs/arxiv:0707.4643v2[arxiv:0707.4643v2 ] . huber , p. j. ( 1967 ) . the behavior of maximum likelihood estimates under nonstandard conditions . in _ proc . fifth berkeley sympos . probability _ * 1 * 221233 . univ . california press , berkeley . pal , j. k. , woodroofe , m. and meyer , m. ( 2007 ) . estimating a polya frequency function . in _ complex datasets and inverse problems : tomography , networks and beyond _ ( r. liu , w. strawderman and c .- h . zhang , eds . ) . _ ims lecture notes - monograph series _ * 54 * 239249 . ims , beachwood , oh .

we study numerically multifractal properties of two models of one - dimensional quantum maps , a map with pseudointegrable dynamics and intermediate spectral statistics , and a map with an anderson - like transition recently implemented with cold atoms . using extensive numerical simulations , we compute the multifractal exponents of quantum wave functions and study their properties , with the help of two different numerical methods used for classical multifractal systems ( box - counting method and wavelet method ) . we compare the results of the two methods over a wide range of values . we show that the wave functions of the anderson map display a multifractal behavior similar to eigenfunctions of the three - dimensional anderson transition but of a weaker type . wave functions of the intermediate map share some common properties with eigenfunctions at the anderson transition ( two sets of multifractal exponents , with similar asymptotic behavior ) , but other properties are markedly different ( large linear regime for multifractal exponents even for strong multifractality , different distributions of moments of wave functions , absence of symmetry of the exponents ) . our results thus indicate that the intermediate map presents original properties , different from certain characteristics of the anderson transition derived from the nonlinear sigma model . we also discuss the importance of finite - size effects .

1007.1404

multifractal behavior has been observed in a wide variety of physical systems , from turbulence @xcite to the stock market @xcite and cloud images @xcite . it has been recognized recently that such a behavior can also be visible in quantum wave functions of certain systems . in particular , wave functions in the anderson model of electrons in a disordered potential are multifractal at the metal - insulator transition ( see e.g. @xcite ) . similar behaviors were seen in quantum hall transitions @xcite and in random matrix models such as the power law random banded matrix model ( prbm ) @xcite and ultrametric random matrices @xcite . such properties have also been seen in diffractive systems @xcite and pseudointegrable models , for which there are constants of motion , but where the dynamics takes place in surfaces more complicated than the invariant tori of integrable systems @xcite . in all these models , this behavior of wave functions came with a specific type of spectral statistics , intermediate between the wigner distribution typical of chaotic systems and the poisson distribution characteristic of integrable systems @xcite . recently , a new model of one - dimensional quantum `` intermediate map '' which displays multifractal behavior was proposed @xcite , and a version with random phases was shown semi - rigorously to display intermediate statistics @xcite . this model is especially simple to handle numerically and analytically , and displays different regimes of multifractality depending on a parameter @xcite . in parallel , new experiments allowed recently to observe the anderson transition with cold atoms in an optical potential @xcite using a one - dimensional `` anderson map '' proposed in @xcite . although much progress has been made in the study of these peculiar types of systems , several important questions related to multifractality remain unanswered . in particular , many results were derived or conjectured in the framework of the anderson model , and their applicability to other families of systems is not known . also , the precise link between the multifractal properties of wave functions and the spectral statistics is not elucidated . in order to shed some light on these questions , we systematically investigate several properties of the wave functions of the intermediate quantum map of @xcite , and compare them to results obtained for the anderson map . in these one - dimensional systems , wave functions of very large vector sizes can be obtained and averaged over many realizations . this enables to control the errors and evaluate the reliability of standard methods used in multifractal analysis . this also allows to check and discuss several important conjectures put forth in the context of anderson transitions . our results also enable to study the road to asymptotic behavior in such models , giving hints on which quantities are more prone to finite - size effects or can be visible only with a very large number of random realizations . the manuscript is organized as follows . in section ii , we review the known facts and conjectures about quantum multifractal systems , which were mainly put forth in the context of the anderson transition . in section iii , we present the models that will be studied throughout the paper . section iv discusses the numerical methods used in order to extract multifractal properties of the models . section v presents the results of numerical simulations , allowing first to compare the different numerical methods of section iv , and then to test the conjectures and results developed in the context of the anderson transition to the two families of models at hand . we first recall some basic facts about multifractal analysis . localization properties of wave functions of components @xmath0 , @xmath1 , in a hilbert space of dimension @xmath2 , can be analyzed by means of their moments @xmath3 the asymptotic behavior of the moments for large @xmath2 is governed by a set of multifractal exponents @xmath4 defined by @xmath5 , or by the associated set of multifractal dimensions @xmath6 . equivalently , the singularity spectrum @xmath7 characterizes the fractal dimensions of the set of points @xmath8 where the weights @xmath9 scale as @xmath10 . it is related to the multifractal exponents @xmath4 by a legendre transform @xmath11 compared to classical multifractal analysis , the quantum wave functions in hilbert space of increasing dimensions are considered as the same distribution at smaller and smaller scales . this allows to define properly the multifractality of quantum wave functions , although at a given dimension they correspond to a finite vector . in many physical instances , only a single realization of the wave function is considered . however , as soon as several realizations are considered , as is the case in presence of disorder , the moments are distributed according to a certain probability distribution , and multifractal exponents depend on the way ensemble averages are performed , and in particular on the treatment of the tail of the moments distribution . if the tail decreases exponentially or even algebraically with a large exponent , the two averages should give the same answer . on the other hand , if the moments decrease according to a power - law with a small exponent , the average @xmath12 will be dominated by rare wave functions with much larger moments ( whose magnitude directly depends on the number of eigenvectors considered ) , while the quantity @xmath13 will correspond to the typical value of the moment for the bulk of the wave functions considered . to each of these possible averaging procedures corresponds a set of multifractal exponents , defined by @xmath14 as soon as averages over several realizations are made there can be a discrepancy between @xmath4 and @xmath15 . historically this effect was seen in the context of the anderson transition @xcite and was very recently confirmed by the analytical calculations of @xcite in the same model . more specifically , it was shown that the distribution of the normalized moments @xmath16 is asymptotically independent of @xmath2 and has a power - law tail @xmath17 for large @xmath18 @xcite . the multifractal exponents @xmath4 and @xmath15 only coincide over an interval @xmath19 $ ] where @xmath20 . in the case of heavy tails @xmath21 , the averages @xmath12 and @xmath22 yield different exponents . this phenomenon has a counterpart in the singularity spectra @xmath7 and @xmath23 . while @xmath23 can not take values below 0 and terminates at points @xmath24 such that @xmath25 , the singularity spectrum @xmath7 continues below zero . the two spectra coincide over the interval @xmath26 $ ] . it can be shown that outside the interval @xmath19 $ ] the set of exponents @xmath15 are given by the linear relation @xmath27 for @xmath28 and @xmath29 for @xmath30 @xcite . in @xcite , it was stated that the exponents @xmath31 and @xmath15 can be related through the following relation which depends on the tail exponent of the moments distribution @xmath32 : @xmath33 equation was analytically proven for prbm @xcite for integer values of @xmath32 and also in the limit of large bands for @xmath34 . it remains unclear to which extent eq . is valid for other types of systems . a consequence of the identity for @xmath35 is that @xmath36 . in the regime of weak multifractality where @xmath37 is a linear function , the identity implies that @xmath38 for @xmath39 @xcite(see subsection v.e . for details ) . finally , a further relation that we wish to investigate in the present paper has been predicted based on the nonlinear sigma model and observed in the 3d anderson model at criticality and several other systems . it is a symmetry relation of multifractal exponents @xcite , which can be expressed as @xmath40 with @xmath41 . for the singularity spectrum it gives @xmath42 . the validity of these different relations will be investigated on two particularly simple models of quantum maps that we describe in the next section . the properties of section [ knownfacts ] have been first observed for wave functions in the 3d anderson model at criticality . in the present paper the first model we will consider is a quantum map whose eigenfunctions display similar multifractal properties in momentum representation . it corresponds to a quantization of a classical map , defined on the torus by @xmath43 where @xmath44 is the momentum variable and @xmath45 the angle variable , while @xmath46 and @xmath47 are the same quantities after one iteration of the map . the corresponding quantum evolution can be expressed as @xmath48 in operator notation , or equivalently as a @xmath49 matrix in momentum representation : @xmath50 with @xmath51 @xcite . for generic irrational @xmath52 , the spectral statistics of @xmath53 are expected to follow random matrix theory . when @xmath52 is a rational number , @xmath54 with @xmath55 integers , spectral statistics are of intermediate type and eigenvectors of the map display multifractal properties . in order to study the effect of ensemble averaging on multifractal exponents , we will consider a random version introduced in @xcite , where the phases @xmath56 are replaced by independent random phases @xmath57 . we will also present numerical results obtained for the initial map given by . an important system where multifractal wave functions have been observed corresponds to electrons in a disordered potential in dimension three . indeed , the anderson model @xcite which describes such a situation displays a transition between a localized phase ( exponentially localized wave functions ) and a diffusive phase ( ergodic wave functions ) for a critical strength of disorder . at the transition point , the wave functions display multifractal properties @xcite and the spectral statistics are of the intermediate type @xcite . in order to compare this type of system with the previous one , we have studied a one - dimensional system with incommensurate frequencies , which has been shown to display an anderson - like transition @xcite . in @xcite it was shown that it can be implemented with cold atoms in an optical lattice , which enables to probe experimentally the anderson transition . the system is a generalization of the quantum kicked rotator model , and is described by a unitary operator which evolves the system over one time interval : @xmath58 with @xmath59 ( here time @xmath60 corresponds to number of kicks ) . here @xmath61 and @xmath62 should be two frequencies mutually incommensurate . following @xcite we chose in the simulations @xmath63 , @xmath64 and @xmath65 , where @xmath66 is the real root of the cubic equation @xmath67 . in @xcite it was shown that this system displays an anderson - like transition at the critical value @xmath68 , but multifractality of this system was up to now not verified . the function @xmath69 can be chosen either by taking @xmath70 ( free evolution ) or as in the preceding case by replacing this quantity by independent random phases uniformly distributed in the interval @xmath71 . this is actually what we chose to do in our numerical simulations , in order to increase the stability and accuracy of the numerical results . as a wave packet spreads slowly at the transition , one has to iterate the map for a long time in order to obtain data on a sufficiently large wave function . we found that in order to reach vector sizes of order @xmath72 , it was necessary to iterate the map up to @xmath73 . such values are not realistic for experiments with cold atoms ( limited to a few hundreds kicks ) but allow to obtain more precise results . in both models the evolution of the system during one time step has the form of an operator diagonal in momentum ( kinetic term ) times an operator diagonal in position ( kick term ) . in both cases , it has been common practice in the field to replace the kinetic term by random phases . this allows to obtain a similar dynamics but with a more generic behavior . moreover , it enables averages over random phases to be performed , which makes numerical and analytical studies more precise . in contrast , in experiments with cold atoms it is easier to perform iterations with a true kinetic term rather than with random phases . in order to assess the effect of this modification , we will therefore use both approaches in the study of multifractal properties of wave functions . many works on multifractal wave functions have focused on eigenstates of the hamiltonian ( see e.g. @xcite ) . for the intermediate map , the evolution operator has eigenvectors which can be numerically found and explored . it is also possible to evolve wave packets , for example initially concentrated on one momentum state , and to study the multifractality of the wave packet as time increases . this corresponds more closely to what can be explored in experiments . for the intermediate map , this process can be understood as the dynamics of a superposition of eigenvectors of the evolution operator . however , in the case of the anderson map , the evolution operator is time dependent ( the continuous time problem being not periodic ) , and the connection with eigenvectors is lacking . in the following , we will explore and compare the multifractality of both eigenfunctions and time - evolved wave packets for the intermediate map . as is well - known , the numerical estimation of multifractal dimensions is very sensitive to finite - size effects . in the present work we have analyzed and compared different numerical methods in order to compute accurately the multifractal exponents . in this section we briefly review the methods we used . the most straightforward method is to compute directly the moments of the wave functions through the scaling of the moments given by @xmath74 . if the scaling holds true , then @xmath75 is a linear function of @xmath76 and its slopes yield the exponents @xmath31 . for @xmath77 , coarse - graining over neighboring sites is necessary in order to avoid instabilities due to very small values of @xmath78 . a variation of the moment method is the box - counting method . it consists in taking a vector of fixed size @xmath2 and summing the @xmath9 over boxes of increasing length . if @xmath79 , then we define @xmath80 which corresponds to averaging the measure over boxes of length @xmath81 , @xmath82 . starting from @xmath83 allows to smooth out exceedingly tiny values of @xmath9 which otherwise yield non - accurate estimates of @xmath37 for negative @xmath45 . the two methods above give similar results , as was also carefully checked for the anderson model in @xcite . we will therefore in the following present results using the box - counting method to assess the properties of this type of procedures . recently , alternative procedures to compute the multifractal spectrum based on the wavelet transform were developed @xcite . the wavelets form a basis of functions as does the fourier basis and a function expressed in this new basis gives the wavelet transform ( wt ) . unlike the fourier basis wavelets are localized both in position and in momentum space ( or time and frequency space ) . they are therefore suitable to probe the local variations of a function at different scales . they have become an essential tool for image and sound processing and compression . a wavelet basis is constructed from a single function @xmath84 called analyzing or mother wavelet . the rest of the basis is constructed by translations and compressions ( expansions ) of the analyzing wavelet @xmath84 . the translations define a space variable while the compressions define the scale at which the function is analyzed . we define the wt of a ( real ) function @xmath85 as @xmath86 where @xmath87 is the scale variable and @xmath88 is the space variable . as a consequence , @xmath89 can be interpreted as a measure of how close the function @xmath85 is to the mother wavelet at point @xmath88 and at scale @xmath87 . the @xmath31 can be extracted from the wt in the following way . we define the partition function @xmath90 where @xmath91 are the local maxima at scale size @xmath87 and @xmath45 is real . it can be shown that @xmath31 appears as the exponent in the power law behavior of @xmath92 @xmath93 this is essentially the method known as wavelet transform modulus maxima ( wtmm ) @xcite . this method is developed for continuous wavelet function . if the function @xmath85 is sampled as an @xmath2-dimensional vector with @xmath79 the wavelet transform can be discretized and implemented efficiently by a hierarchical algorithm @xcite resulting in a fast wavelet transform , ( fwt ) . the scale and space parameters take the values @xmath94 respectively . starting from a wave function @xmath95 and using a proper normalization at each scale , we redefine the partition function as @xmath96^q\ ] ] where again @xmath91 are the local maxima at scale @xmath87 . as in the continuous case @xmath92 exhibits the same power law behavior as eq . ( [ powlaw ] ) . in the following , we will present results using this implementation of the wavelet method , using the daubechies 4 mother wavelet @xcite . we note that , although the partition function ( [ zarneomm ] ) is the most standard , recently it was numerically observed @xcite that for a complex multifractal wave function @xmath95 , the exponents @xmath31 can also be obtained from the power law behavior of a modified partition function defined from the complex fwt of @xmath95 and using @xmath97 as exponents in eq . ( [ zarneomm ] ) . and @xmath98 . bottom panel : instance of an iterate of the anderson map with random phases for @xmath99 , @xmath100 and @xmath101 ; @xmath102 with @xmath103 . [ fig : psi2 ] ] as a function of @xmath104 , the logarithm of the box sizes . bottom panels : partition function @xmath105 as a function of the logarithm of the scale @xmath87 . the left panels correspond to eigenvectors of the intermediate map with random phases , @xmath106 , and @xmath98 . here the average is taken over @xmath107 vectors ( resp . @xmath108 vectors for the partition function ) . the right panels correspond to iterates of the anderson map for @xmath99 , @xmath100 , @xmath102 with @xmath103 and @xmath101 , where the average is taken over @xmath109 vectors . the values chosen for @xmath45 are : @xmath110 ( red diamonds ) , @xmath111 ( blue circles ) , and @xmath112 ( green squares ) . the gray shaded regions show the fitting interval . [ fig : logzq ] ] in this subsection , we present numerical results corresponding to the multifractal exponents for the intermediate quantum map model ( [ intermediate_map ] ) and the anderson map ( [ qmap ] ) . examples of wave functions for both models are shown in fig . [ fig : psi2 ] . the two sets of multifractal dimensions @xmath37 and @xmath113 were computed from @xmath114 and @xmath115 respectively for the box - counting method , and from @xmath116 and @xmath117 respectively for the wavelet method . for the anderson map , as less realizations of the random phases could be computed , the same set of random realizations were used for the two methods investigated in order to ensure that comparable quantities were plotted . figure [ fig : logzq ] illustrates the scaling of these quantities for the two models considered . it displays @xmath114 as a function of the logarithm of the box size ( top ) and @xmath118 as a function of the logarithm of the scale parameter @xmath119 ( bottom ) for different values of @xmath45 . we chose to show the scaling of these moments since it is the worst case , the curves for @xmath115 and @xmath117 being always closer to linear functions . nevertheless , fig . [ fig : logzq ] shows that the scaling is indeed linear over several orders of magnitude . the slopes of the linear fits give the multifractal exponents . in the two methods , there is a certain freedom in determining the range of box sizes ( or scales for the wavelet method ) over which the linear fit is made ( the one we chose is indicated by the shaded area in fig . [ fig : logzq ] ) . usually for moderate values of @xmath45 in absolute value , the result does not depend very much on this choice . as can be seen in fig . [ fig : logzq ] , the data are well fitted by a linear function in the range chosen ; however , there is still an uncertainty on the slope , which usually gets larger for large negative @xmath45 . in the next figures of this subsection , the uncertainty of the linear fit for the set of points chosen is plotted together with the mean value , in order to give an estimate of the reliability of the values obtained . for eigenvectors of the intermediate map with random phases , @xmath106 , and @xmath98 . empty / filled red circles correspond to the method of moments / wavelets . the ( light blue , light red ) shaded regions indicate standard error in the least squares fitting . the multifractal analysis was done over @xmath107 vectors with box sizes ranging from @xmath120 to @xmath121 and scales ranging from @xmath122 to @xmath123 . bottom panel : typical multifractal exponents @xmath124 for the same data . in both panels , the gray solid line is the linear approximation @xmath125.[fig : dqb ] ] and @xmath124 for iterates of the anderson map with @xmath99 , @xmath100 . same convention as in fig . [ fig : dqb ] . the multifractal analysis was done over @xmath109 realizations of size @xmath100 with box sizes ranging from @xmath126 to @xmath127 and scales ranging from @xmath128 to @xmath129 . [ fig : dqand ] ] the values of @xmath37 and @xmath113 as a function of @xmath45 are presented in fig . [ fig : dqb ] for the random intermediate map and in fig . [ fig : dqand ] for the anderson map . in both cases , we observe a spectrum typical of multifractal wave functions . in the intermediate case , we observe that the two methods give comparable results with small uncertainty , although it gets larger for large negative @xmath45 . for the anderson map , the uncertainty gets very large for @xmath130 , and besides it begins to depend strongly on the range of box sizes ( box - counting method ) or scales ( wavelet method ) chosen ( data not shown ) . we attribute the larger uncertainty for the wavelet method to the seemingly stronger sensitivity of this method to exceptionally small values . we note that for the anderson map only a vector size of @xmath72 could be reached numerically because of the computational power required to iterate the anderson map for long times . the discrepancy between the two methods is smaller for @xmath113 , which can be understood by the fact that @xmath131 or @xmath132 are more sensitive to rare events than @xmath133 and @xmath117 . our data nevertheless show that the iterates of the anderson map display multifractal behavior . in view of the recent implementation of such maps with cold atoms @xcite , this indicates that in principle one can observe multifractality of this map in cold atom experiments ; we note that in @xcite properties of the wave functions of this experimental system were investigated , but focused on the envelope of the wave packet . in both the intermediate map and the anderson map , @xmath113 goes to a constant @xmath134 for large @xmath45 , which corresponds to the fact already mentioned that @xmath15 is expected to behave as @xmath27 for @xmath28 . this will be studied in more detail in the next section . in @xcite we observed the existence of a linear regime around @xmath135 , with slope @xmath136 for the random intermediate map with parameter @xmath137 . this regime seemed to be valid in quite a large interval around zero . the data displayed in fig . [ fig : dqb ] obtained by two different large - scale computations confirm this result , and show that this regime exists for both types of averages . we will come back to this property in section [ linear_regime ] . and @xmath124 for the anderson map with @xmath99 , @xmath100 , @xmath102 with @xmath103 and @xmath101 . same convention as in fig . [ fig : dqb ] . the multifractal analysis was done over 1302 realizations with box sizes ranging from 4 to 512 . bottom : multifractal exponents @xmath37 and @xmath124 for iterates of the intermediate map with @xmath138 , @xmath139 , @xmath102 with @xmath140 and @xmath101 . the multifractal analysis was done over 2000 realizations with box sizes ranging from 4 to 512 . in both cases we used the box - counting method . [ fig : dqdqtypit],title="fig : " ] + and @xmath124 for the anderson map with @xmath99 , @xmath100 , @xmath102 with @xmath103 and @xmath101 . same convention as in fig . [ fig : dqb ] . the multifractal analysis was done over 1302 realizations with box sizes ranging from 4 to 512 . bottom : multifractal exponents @xmath37 and @xmath124 for iterates of the intermediate map with @xmath138 , @xmath139 , @xmath102 with @xmath140 and @xmath101 . the multifractal analysis was done over 2000 realizations with box sizes ranging from 4 to 512 . in both cases we used the box - counting method . [ fig : dqdqtypit],title="fig : " ] the data presented in fig . [ fig : dqand ] show that the anderson vectors are less fractal than the eigenvectors of the intermediate map for @xmath138 . this can be expected from the fact that anderson vectors correspond to iterates of a quantum map , which in general are less fractal than eigenvectors @xcite . in order to compare similar quantities , we display in fig . [ fig : dqdqtypit ] multifractal dimensions for vectors obtained by iteration of the intermediate map for @xmath138 . one sees that using iterates instead of eigenvectors also in this case reduces the overall multifractality at a given @xmath45 . although a simple relation between multifractal exponents of eigenvectors and iterates is still lacking , this indicates that it might be possible to use experimental results from cold atom experiments to infer properties of the eigenvectors of the 3d anderson transition . for eigenvectors of the intermediate map with ( circles ) and without ( triangles ) random phases . bottom panel : typical multifractal exponents @xmath124 for eigenvectors of the intermediate map with ( squares ) and without ( diamonds ) random phases . in both cases we used the box - counting method . conventions and parameter values are the same as in fig . [ fig : dqb ] except that for non - random phases , a single realization of size @xmath106 was considered . in both panels , the gray solid line is the linear approximation @xmath125.[fig : dqbnonrand ] ] in order to assess the effect of random phases , we present in fig . [ fig : dqbnonrand ] the results of numerical computation of @xmath37 and @xmath113 for the intermediate map without random phases . although spectral statistics for random and non - random vectors are very close @xcite , the obtained fractal dimensions are quite different . moreover it can be seen that there still exists a difference between @xmath37 and @xmath113 , although the map is not random any more . the discrepancy in the two sets of exponents was to our knowledge up to date observed only in disordered systems , where rare events are created by specific realizations of disorder . it is interesting to see that this effect can be observed in a dynamical system without any disorder whatsoever . this discrepancy is due to the fact that the average in eqs . - is performed over several eigenvectors of a single realization of the map , which gives a certain dispersion of the moment distribution . the average over eigenvectors in the intermediate map thus suffices to create the effect even in a deterministic map . therefore one can also observe the separation between the two sets of exponents in quantum systems without disorder . the rare events in this case correspond to rare eigenvectors of the evolution operator having large moments . the numerical results displayed in this subsection indicate that multifractality is indeed present in all models considered . furthermore , our results show that for an appropriate choice of a range of box sizes ( box - counting method ) or scales ( wavelet method ) , both methods are based on curves well - fitted by linear function over a wide interval , with a small uncertainty on the exponent extracted . we believe that the good agreement for @xmath141 between the two methods , and the small uncertainty found for the linear fit , indicates that our numerically extracted multifractal exponents are reliable ( up to finite size effects ) . for @xmath130 , the numerical uncertainty increases for decreasing @xmath45 , and the two methods give increasingly different results . the results presented here show that our data are still reliable , although less precise , for eigenvectors of the intermediate map , even for negative @xmath45 . however , in the case of the anderson map , the uncertainty is too large to give reliable results for @xmath130 . this is certainly due to the fact that both the number of realizations and the vector sizes are smaller in this case , which makes it difficult to find reliable results in the more demanding regime of large negative @xmath45 . in the regime of @xmath141 , our results indicate that the two methods can be used and give similar results . in the more demanding cases ( large negative @xmath45 ) we found the box - counting method more reliable and accurate , and therefore the numerical results presented in the following subsections correspond to this method . in @xcite the first investigations of the multifractal exponents for the random intermediate map showed the presence of a linear regime around @xmath135 over a relatively large range of @xmath45 values . we recall that the map displays multifractal properties of rational values of the parameter @xmath54 . based on semi - heuristic arguments , this linear regime was shown to be described by : @xmath142 the relation ( [ dqlin ] ) enables to link the spectral statistics and the distribution of @xmath37 around @xmath135 in a systematic way since both are controlled by the parameter @xmath143 explicitly . while one expects that some form of linear regime should exist over small intervals for any smooth curve , the extent of it in this particular model indicated a small second derivative near @xmath135 . this feature was seen in the prbm model @xcite but in the regime of weak multifractality where the derivative at @xmath135 of @xmath37 is very close to zero . figure [ fige1 ] displays the extent of the linear regime for the intermediate map for three values of the parameter @xmath137 . the data presented show that the linear regime is present in all three cases , although its extent seems to be larger for large @xmath143 ( weak multifractality ) . this indicates that the linear regime is a robust feature of the random intermediate map . to explore more precisely the dependence of this regime on the value of @xmath143 , the inset of fig . [ fige1 ] shows the separation point between the actual exponent and the linear value , defined by a constant relative difference ( set to @xmath144 ) , for all values of @xmath143 between @xmath145 and @xmath146 . the data presented show that indeed the extent of the linear regime grows with @xmath143 , although the precise law of this growth is difficult to specify . for eigenvectors of the intermediate map with random phases for @xmath98 ( blue solid curve ) , @xmath147 ( green dotted curve ) and @xmath148 ( red dash - dotted curve ) . blue dashed curve shows the same difference for the intermediate map without random phases for @xmath98 . inset : separation points between @xmath37 and @xmath149 ( green dots ) , determined by @xmath150 , for eigenvectors of the intermediate map with random phases for different values of @xmath137 . other parameters as in fig . [ fig : dqbnonrand ] . [ fige1 ] ] these results correspond to the random intermediate map , where the kinetic term is replaced by random phase . it is important to explore also the behavior of the deterministic intermediate map , where the kinetic term is kept as a function of momentum . in this case , [ fige1 ] shows the presence of a much smaller linear regime . this indicates a strong difference between the random model and the deterministic one , and that a large linear regime is a property restricted to a certain kind of models . for iterates of the anderson map , with @xmath149 defined by @xmath151 . same parameters as in fig . [ fig : dqand ] . [ fige3 ] ] the data for the anderson map with random phases , shown in fig . [ fige3 ] , also show a linear regime , comparable with the random intermediate map . again , as the data correspond to iterates of wave packets , not eigenvectors , the multifractality is weaker than for other simulations of the anderson model using eigenvectors @xcite . this might explain why the linear regime that is visible seems larger than for anderson transition eigenstates . ( red solid lower curve ) and @xmath152 ( blue solid upper curve ) as a function of @xmath45 for eigenvectors of the intermediate map with random phases , @xmath106 , and @xmath98 . bottom panel : same but for @xmath148 . in both panels , the black dashed lines are linear fits of @xmath152 at large @xmath153 . the arrows indicate the critical @xmath45 value determined by @xmath154 ( see subsection v.d ) , equal to @xmath155 for @xmath98 and to @xmath156 for @xmath148 . the slopes of the black dashed lines are respectively given by @xmath157 and @xmath158 for @xmath98 and by @xmath159 and @xmath160 for @xmath148 . [ fig : tauq ] ] in disordered systems , the statistical distribution of the moments of the wave function is responsible for a discrepancy between the multifractal exponents @xmath37 and @xmath113 calculated respectively by averaging over the moments themselves or over their logarithm . the two sets of exponents are expected to match only in some region @xmath161 $ ] . outside this range , @xmath15 should follow a linear behavior . figure [ fig : tauq ] displays results for @xmath4 and @xmath15 for the intermediate map with parameter @xmath137 for two representative values of @xmath143 . as expected from the prbm model @xcite , in the case of weak multifractality ( @xmath148 , bottom panel of fig . [ fig : tauq ] ) , the range over which the exponents are equal is wider than for strong multifractality ( @xmath98 , top panel of fig . [ fig : tauq ] ) . beyond that interval the behavior of @xmath15 is linear for both values of @xmath52 . for positive @xmath45 the linear tail appears around @xmath162 for @xmath98 and @xmath163 for @xmath148 . the slopes of the linear tails give @xmath164 and @xmath165 . according to the theory ( see section ii ) , these values of @xmath166 correspond to the terminating point of the singularity spectrum @xmath23 defined in eq . . we obtained comparable results for the anderson map ( data not shown ) . between average and typical exponents for eigenvectors of the intermediate map with random phases for @xmath98 ( red solid curve ) and @xmath148 ( blue dashed curve ) . bottom panel : same figure for iterates of the anderson map . same parameters as in fig . [ fig : dqand ] . [ fig : deltadq ] ] in order to take a closer look at the discrepancy between the two sets of exponents , we plot the difference between the exponents @xmath37 and @xmath113 for both systems . it is clearly seen in fig . [ fig : deltadq ] that in all cases the regime where the exponents coincide is only about @xmath167 . at this scale the separation between @xmath31 and @xmath15 occurs around @xmath168 . in order to obtain more systematically the separation point , we have plotted in fig . [ figdist ] the value of @xmath45 defined by a constant relative value of the difference of the two exponents ( set equal to @xmath169 ) for intermediate maps with different parameters @xmath137 ; this allows to get comparable data independently of the value of the exponents . the results show a clear linear scaling of the separation point with respect to @xmath143 , obeying the formulas @xmath170 and @xmath171 . changing the threshold of relative value from @xmath169 to @xmath172 gives also a linear scaling , with different slope ( data not shown ) . between @xmath37 and @xmath173 ( green dots ) , determined by @xmath174 , for eigenvectors of the intermediate map with random phases for different values of @xmath137 . the blue solid lines are the linear fits @xmath175 and @xmath176 . same parameters as in fig . [ fig : dqand ] . [ figdist ] ] and @xmath23 for eigenvectors of the intermediate map with random phases for @xmath98 . the inset shows the linear behavior of @xmath177 as a function of @xmath143 ( in @xmath137 ) . the values of @xmath178 have been extracted from linear fits of @xmath152 at large @xmath153 . right panel : singularity spectra @xmath7 and @xmath23 for iterates of the anderson map . in both panels , the singularity spectra are given in the range @xmath179 . [ fig : fa ] ] the singularity spectrum @xmath7 defined in eq . is an alternative way to analyze multifractality and the discrepancy between the two sets of multifractal exponents . in fig . [ fig : fa ] we show the singularity spectrum obtained for the intermediate map with random phases ( in @xcite similar curves were obtained using directly the box - counting method to compute @xmath7 ; here we use the legendre transform of the exponents , obtaining similar results ) . as expected @xmath23 terminates at points @xmath134 and @xmath180 given by the large-@xmath45 slopes of @xmath15 , while @xmath7 takes values below zero coming from statistically rare events . in @xcite , it was shown that a linear approximation for @xmath37 yields a parabolic approximation for @xmath7 , giving in turn a behavior @xmath181 . we checked this behavior for all values of @xmath143 between @xmath145 and @xmath146 , thus confirming the validity of this law , even beyond the linear regime ( see inset of fig . [ fig : fa ] ) . figure [ fig : fa ] allows a more direct comparison between multifractality in the intermediate map and the anderson map : the narrower @xmath7 curve for anderson corresponds to a weaker multifractality . the discrepancy between the two sets of multifractal exponents observed in the previous section is due to the fact that the moments @xmath182 defined by have a statistical distribution with a certain width . in particular for multifractal measures the distribution of the normalized moments @xmath16 is expected to have a power - law tail at large @xmath45 as @xmath183 , with an exponent @xmath32 depending on @xmath45 . for @xmath184 of eigenvectors of the random intermediate map for @xmath98 . same parameters as in fig . [ fig : dqb ] . the solid line shows a linear fit ( in log scale ) whose slope yields the tail exponent @xmath185 . [ fig : pmq ] ] in fig . [ fig : pmq ] we show an example of the distribution of the logarithm of the moments @xmath182 for the random intermediate map . the distribution is indeed algebraic with a power law tail depending on @xmath45 . while the linear behavior ( in logarithmic scale ) is clearly observed for small values of @xmath45 ( restricted to the range @xmath186 ) , this is not the case for larger @xmath45 . we calculated the exponent @xmath32 of the tail for a range of values of @xmath45 where this exponent could be extracted . results are displayed in fig . [ fig : xqdirect ] for the intermediate map and fig . [ fig : xqand ] for the anderson map . for eigenvectors of the random intermediate map for @xmath98 with @xmath139 ( blue filled circles ) and @xmath106 ( blue empty circles ) and for @xmath148 with @xmath106 ( red squares ) . same parameters as in fig . [ fig : dqb ] . inset : tail exponent for @xmath148 for larger values of @xmath45 . [ fig : xqdirect ] ] for iterates of the anderson map . same parameters as in fig . [ fig : dqand ] . the inset shows that @xmath187 is well - fitted by a parabola . [ fig : xqand ] ] the value of @xmath45 where @xmath188 should correspond to the value @xmath189 where the two curves of @xmath37 and @xmath113 ( or @xmath31 and @xmath15 ) separate . as one can observe in fig . [ fig : xqdirect ] that value of @xmath45 is rather difficult to estimate numerically with sufficient accuracy as the exponents @xmath32 did not converge to a definite value at the largest vector size available ( @xmath106 ) . however the curves seem to yield an exponent equal to 1 around @xmath190 for @xmath98 and @xmath191 for @xmath148 . these values are indicated with black arrows in fig . [ fig : tauq ] , and at that scale they do seem to coincide with points where @xmath31 and @xmath15 separate . however these points are far beyond the value @xmath192 at which the multifractal dimensions @xmath37 and @xmath113 separate at the scale of fig . [ fig : deltadq ] , and also different from the values obtained by fixing the relative difference of the exponents in fig . [ figdist ] ( equal to @xmath193 for @xmath98 and @xmath194 for @xmath148 ) . the value where the curves separate is dominated by the rare events in the extreme tail of the distribution . the different values obtained indicate that indeed the numerical results are still far from the asymptotic regime . similar conclusions can be drawn from fig . [ fig : xqand ] , which presents the power - law tail exponents @xmath32 obtained for the moment distribution of the anderson vectors : the point where @xmath195 is reached around @xmath196 while fig . [ fig : deltadq ] seems to indicate a separation of the multifractal exponents around @xmath192 . we note that for the finite sizes considered , the value of @xmath32 seems to become infinite as @xmath197 , indicating that in this regime the distribution of moments is not any more fitted by a power law at large moments ( see also fig . [ fig : xq3 ] ) . the behavior of the exponents @xmath32 will be further discussed in the next subsection . as explained in section [ knownfacts ] , it was proposed in @xcite that the exponents @xmath31 and @xmath152 were related to the moment distribution through the relation ( [ xqtq ] ) . this formula was proved only in some very specific cases , such as the prbm model in the regime of weak multifractality , but conjectured to be generically valid . the results of the preceding subsections enable to check numerically whether this formula holds for our models . and @xmath198 for eigenvectors of the intermediate map for @xmath98 . parameter values are the same as in fig . [ fig : dqb ] . blue / green ( black / grey ) circles correspond to values of @xmath45 larger / smaller than @xmath199 ( the values closer to @xmath200 are on the left for both series of points ) . dashed line is the formula ( [ xqtq]).[fig : xqtq ] , width=321 ] and @xmath201 for eigenvectors of the intermediate map for @xmath148 . parameter values are the same as in fig . [ fig : dqb ] . the blue solid line shows a linear fit of the data ( @xmath202 ) . dashed line is the formula ( [ xqtq]).[fig : xqtqb11],width=321 ] in fig . [ fig : xqtq],[fig : xqtqb11 ] we show @xmath203 as a function of @xmath204 for the random intermediate map with parameters @xmath98 and @xmath148 . for @xmath98 ( fig . [ fig : xqtq ] ) , it shows a certain agreement with the conjectured law for small values of @xmath45 . similarly for @xmath148 ( fig . [ fig : xqtqb11 ] ) the agreement with the law is good , but the actual slope seems slightly different . on the other hand , for @xmath98 and larger values of @xmath45 the relation breaks down . we note that there is a certain ambiguity in the formula , since as can be seen in fig . [ fig : xqtq ] one can have two values of @xmath203 for the same value of @xmath201 ( corresponding to two different values of @xmath45 ) . the results indicate that the relation ( [ xqtq ] ) can indeed be seen , even if approximately , in other systems that in anderson transition models . interestingly , the case @xmath148 corresponds to a case of weak multifractality . this might indicate that the relation is better verified in the case of weak multifractality , and only approximate in the general case . but we can not exclude that the regime of weak multifractality leads to weaker finite - size effects and that the results for @xmath98 would eventually converge to the law ( [ xqtq ] ) for larger sizes and many more realizations . an additional problem concerns the different scales of the two figures [ fig : xqtq ] and [ fig : xqtqb11 ] . as the multifractality is weaker in the case @xmath148 , the values of @xmath32 and @xmath31 are larger , leading to a much larger scale for the data in fig . [ fig : xqtqb11 ] . it is possible that the finite - size effects are comparable but show more markedly in fig . [ fig : xqtq ] due to its much smaller scale . in fig . [ fig : xqtqand ] we present the same numerical analysis for the anderson map . the results show that a linear law similar to can be seen . the slope is close to one , but the curve is shifted by a relatively large offset ( @xmath205 ) . note that the formula ( [ xqtq ] ) was predicted for eigenvectors of the anderson model and prbm ; here we are looking at iterates of wave packets , which can show different behavior . and @xmath152 for iterates of the anderson map with @xmath99 , @xmath100 , @xmath102 with @xmath103 and @xmath101 . parameter values are the same as in fig . [ fig : dqand ] . the blue solid line correspond to @xmath206 . dashed line is the formula ( [ xqtq ] ) . [ fig : xqtqand],width=321 ] interesting properties of the exponents @xmath32 can be deduced from the relation . as mentioned in the introduction , if such a relation is verified it implies that for @xmath35 ( or equivalently @xmath21 ) the inverse of the exponents @xmath32 should follow a linear law . indeed , for @xmath35 the exponent @xmath15 is linear and one has @xmath207 , thus @xmath208 is solution of the equation @xmath209 . if this equation has an unique solution , then the quantity @xmath210 is a constant , equal to @xmath189 for @xmath211 , and thus @xmath187 should be linear as a function of @xmath45 for @xmath35 . in order to check whether this holds in our case , we plot in fig . [ fig : xq3 ] the values of @xmath187 as a function of @xmath45 for the intermediate random map with parameters @xmath98 and @xmath148 . in both cases , a linear law agrees well with the data at large @xmath45 , but obeys an equation different from the one predicted , with in particular an extra constant term which depends on @xmath52 . for eigenvectors of the random intermediate map . top panel : @xmath98 ; blue and black dashed lines are linear fits in two different @xmath45 range ; dashed line ( linear fit for large @xmath45 ) is @xmath212 ; inset is a blow - up of the small @xmath45 range . bottom panel : @xmath148 ; blue and black dashed lines are respectively quadratic and linear fits in two different @xmath45 range ; dashed line ( linear fit for large @xmath45 ) is @xmath213 ; inset is a blow - up of the small @xmath45 range , showing in red the best linear fit for the same data . in both cases , the two ends of the error bars correspond to two values of @xmath32 obtained from a fit of the moments probability distribution over two different intervals . parameter values are the same as in fig . [ fig : dqb ] . [ fig : xq3 ] ] a second consequence of relation arises in the case where @xmath37 is given by a linear function . this is in particular the case for the intermediate map in the regime of weak multifractality . in this regime it was observed @xcite that for not too large @xmath45 the multifractal exponents @xmath37 are very closely given by the linear approximation @xmath214 . inserting this relation into , we get that for @xmath215q_-,q_+[$ ] ( where @xmath37 and @xmath113 are equal ) the exponents are given by @xmath216 provided @xmath217 also belongs to @xmath218q_-,q_+[$ ] . we note that this in turn predicts a value @xmath219 for the separation point between the two sets of exponents , contrary to the linear scaling found in the data shown in fig . [ figdist ] . according to these considerations , in the small @xmath45 regime and for weak multifractality , a quadratic behavior of @xmath187 should be observed provided the linear regime extends beyond the point @xmath220 . in the case @xmath148 for the intermediate map , the linear regime is verified quite far away from zero but breaks down before @xmath220 ( see fig . [ fig : deltadq ] ) , thus @xmath221 is not a linear function of its argument . still , fig . [ fig : xq3 ] ( lower panel ) shows that the parabolic behavior of @xmath187 is retained for small values of @xmath45 ; the inset shows that indeed a quadratic fit is much better than a linear fit in this range . in the strong multifractality regime , @xmath37 is not linear any more beyond @xmath168 ; in that case , the inset of the top panel in fig . [ fig : xq3 ] shows that the behavior of @xmath187 is linear rather than quadratic . as described in section [ knownfacts ] , it was predicted analytically and observed in the anderson model at the transition that a symmetry exists between multifractal exponents . indeed , the quantity @xmath41 follows the law @xmath222 ( relation ( [ deltaqrel ] ) ) . this relation was predicted on very general grounds , and expected to hold for all multifractal quantum systems . it was seen in the prbm model @xcite , the anderson model @xcite and ultrametric random matrices @xcite . it was also predicted to occur in simple multifractal cascade models in @xcite . we performed systematic calculation of the quantity @xmath223 for each of the systems considered . ( blue dashed curve ) and @xmath224 ( red solid curve ) for eigenvectors of the intermediate map with random phases , @xmath98 and @xmath106 . parameter values are the same as in fig . [ fig : dqb ] . the black thin line shows the parabola @xmath225 corresponding to the linear regime.[fig : deltaq ] ] figure [ fig : deltaq ] shows the results of this analysis for the random intermediate map . the presence of a large linear regime complicates the picture , since the linear law described above in subsection [ linear_regime ] verifies the symmetry . thus the intermediate map can show deviations from the symmetry only outside the linear regime . it turns out ( comparing figs . [ fige1 ] and [ fig : deltaq ] ) that the symmetry ( [ deltaqrel ] ) is only present in the linear regime and does not extend beyond its validity . this seems to indicate that the symmetry is absent from these models . ( blue dashed curve ) and @xmath224 ( red solid curve ) for eigenvectors of the intermediate map ( [ intermediate_map ] ) with @xmath98 and @xmath106 . the ( light blue , light red ) shaded region indicates standard error in the least squares fitting . the multifractal analysis was done with box sizes ranging from 16 to 1024 . the black thin line shows the parabola @xmath225 corresponding to the linear regime.[figf2 ] ] as the linear regime is much smaller in the case of the intermediate map without random phase , if the symmetry does not hold in this system we should expect a larger discrepancy in the nonrandom case . [ figf2 ] displays @xmath223 and @xmath224 for the nonrandom intermediate case , showing that indeed the symmetry is verified in an even smaller range of @xmath45 values than for the random case , in agreement with the small extent of the linear regime . finally , we have also computed the quantities @xmath223 and @xmath224 for iterates of the random intermediate map and of the random anderson map . as said in subsection v.a , the precision of the exponents degrades for @xmath130 in the case of the anderson map , so verification of the symmetry relation is delicate . nevertheless , our results indicated that while for iterates of the random intermediate map the symmetry ( [ deltaqrel ] ) still does not hold beyond the linear regime , in the case of the anderson map the symmetry remains valid within the numerical error bars ( which are however quite large ) well beyond the linear regime ( data not shown ) . this seems to indicate further that the symmetry is a feature of the anderson model , which is clearly absent from the intermediate map . although we can not prove rigorously that the symmetry relation ( [ deltaqrel ] ) does not hold for intermediate systems , our results strongly indicate that it is violated in these systems as soon as the linear regime breaks down . to further confirm that our numerical method is able to observe the symmetry ( [ deltaqrel ] ) in a system where it is present , we computed @xmath223 and @xmath224 for ultrametric random matrices where the relation is known to hold @xcite . our numerical method was able to confirm unambiguously the presence of the symmetry in this specific case ( data not shown ) . we note that in @xcite the presence of the symmetry ( [ deltaqrel ] ) for the anderson transition was theoretically predicted on the basis of a renormalization group flow whose limit corresponds to a nonlinear sigma model . it would be interesting to see if a different nonlinear sigma model can apply to the intermediate map , or if it is the sign that these models can not describe certain aspects of these systems . in this paper , we have studied the different multifractal exponents one can extract from the wave functions of the intermediate map and the anderson map . both models are one - dimensional , and thus allow much larger system size than the 3d anderson transition , but in contrast to random matrix models such as the prbm they correspond to physical systems with an underlying dynamics . our results enabled to extract the exponents over a large range of @xmath45 values for the intermediate map . we have checked that two methods widely used in other contexts ( classical multifractal systems ) , namely the box - counting and the wavelet method , can be used to obtain reliably the exponents , giving similar results in most cases , although the box - counting method seems more robust for large negative values . our numerical data allow to confirm that the anderson map introduced in @xcite and experimentally implemented with cold atoms @xcite indeed displays multifractal properties at the transition point , although this multifractality is weak . as concerns the intermediate map , our data confirm the existence of a linear regime for the multifractal exponents @xmath37 and @xmath113 which was first seen in @xcite , well beyond the regime of weak multifractality . interestingly enough , the linear regime is much smaller for the nonrandom intermediate map . we checked that the exponents @xmath37 and @xmath113 are different also in the case of the intermediate map , even in the nonrandom case where no disorder is present . our numerical study of the moments of the wave functions and the multifractal exponents show that the generic behavior of @xmath37 and @xmath113 predicted for the anderson transition @xcite are present for the anderson map . our results enable to extract the value of @xmath220 and @xmath226 through the behavior of the moments of the wave functions together with the values of @xmath165 and @xmath180 ; the fact that the value is different from the one obtained by direct computation of the multifractal exponents shows that finite size effects persist in such systems up to very large sizes . note , however , that as our numerical computations correspond to very large system sizes , this might indicate that the asymptotic limit may be difficult to reach even in experimental situations . in addition , our investigations show that the relation ( [ xqtq ] ) between the moments and the exponents conjectured in @xcite is only approximately verified in our systems , even in the anderson map . at last , the exact symmetry relation ( [ deltaqrel ] ) between the multifractal exponents of the anderson transition discovered in @xcite is not present in intermediate systems . our results indicate that intermediate systems , and more generally quantum pseudointegrable systems , represent a type of model with some similarities with the anderson transition model of condensed matter , but with specific properties . in particular , the absence of the symmetry present in the anderson model between the exponents suggests proceeding with care in using the nonlinear sigma model to predict properties of these systems . we thank e. bogomolny , d. delande , k. frahm , c. monthus and a. ossipov for useful discussions . this work was supported in part by the agence nationale de la recherche ( anr ) , project qpprjccq and by a peps - pti from the cnrs . 99 c. meneveau , k. r. sreenivasan , phys . lett . * 59 * , 1424 ( 1987 ) ; j .- f . muzy , e. bacry and a. arneodo , phys . rev . lett . * 67 * , 3515 ( 1991 ) . b. b. mandelbrot , a. j. fisher and l. e. calvet , cowles foundation discussion paper no . 1164 ( 1997 ) . s. lovejoy and d. schertzer , journal of geophysical research * 95 * , 2021 ( 1990 ) . a. d. mirlin , phys . rep . * 326 * , 259 ( 2000 ) . e. cuevas , m. ortuno , v. gasparian and a. perez - garrido , phys . rev . lett . * 88 * , 016401 ( 2001 ) . f. evers and a. d. mirlin , rev . phys . * 80 * , 1355 ( 2008 ) . b. huckestein , rev . mod . phys . * 67 * , 357 ( 1995 ) . a. d. mirlin , y. v. fyodorov , f .- m . dittes , j. quezada , and t. h. seligman , phys . e * 54 * , 3221 ( 1996 ) . v. e. kravtsov and k. a. muttalib , phys . lett . * 79 * , 1913 ( 1997 ) . y. v. fyodorov , a. ossipov and a. rodriguez , j. of stat . * l12001 * ( 2009 ) . a. m. garcia - garcia and j. wang , phys . e * 73 * , 036210 ( 2006 ) . e. b. bogomolny , u. gerland and c. schmit , phys . e * 59 * , r1315 ( 1999 ) . e. b. bogomolny , o. giraud and c. schmit , phys . e * 65 * , 056214 ( 2002 ) . o. giraud , j. marklof and s. okeefe , j. phys . a * 37 * , l303 ( 2004 ) . e. b. bogomolny and c. schmit , phys . 93 * , 254102 ( 2004 ) . j. martin , o. giraud , and b. georgeot , phys . e * 77 * , 035201(r ) ( 2008 ) . j. chab , g. lemari , b. grmaud , d. delande , p. szriftgiser , and j .- c . garreau , phys . * 101 * , 255702 ( 2008 ) ; g. lemari , j. chab , p. szriftgiser , j .- c . garreau , b. grmaud , and d. delande , phys . a * 80 * , 043626 ( 2009 ) . g. lemari , h. lignier , d. delande , p. szriftgiser , and j .- c . garreau ( 2010 ) preprint arxiv:1005.1540 . g. casati , i. guarneri , and d. l. shepelyansky , phys . lett . * 62 * , 345 ( 1989 ) . l. j. vasquez , a. rodriguez and r. a. rmer , phys . rev . b * 78 * , 195106 ( 2008 ) ; a. rodriguez , l. j. vasquez and r. a. rmer , phys . rev . b * 78 * , 195107 ( 2008 ) . a. d. mirlin and f. evers , phys . b * 62 * , 7920 ( 2000 ) . m. s. foster , s. ryu and a. w. w. ludwig , phys . b * 80 * , 075101 ( 2009 ) . c. monthus and t. garel , j. stat . mech . * 2010 * , p06014 ( 2010 ) . f. evers and a. d. mirlin , phys . * 84 * , 3690 ( 2000 ) . a. d. mirlin , y. v. fyodorov , a. mildenberger and f. evers , phys . * 97 * , 046803 ( 2006 ) . b. i. shklovskii , b. shapiro , b. r. sears , p. lambrianides , and h. b. shore , phys . b * 47 * , 11487 ( 1993 ) ; d. braun , g. montambaux , and m. pascaud , phys . rev . lett . * 81 * , 1062 ( 1998 ) . a. arneodo in _ wavelets : theory and applications _ , g. erlebacher , m. y. hussaini and l. m. jameson , eds ( oxford university press , new york , 1996 ) . s. g. mallat , ieee trans . on pattern analysis and machine intelligence , vol . ii , no 7 , 674 ( 1989 ) . i. daubechies , comm . pure appl . math . * 41 * , 909 ( 1988 ) . i. garca - mata , o. giraud and b. georgeot , phys . a * 79 * , 052321 ( 2009 ) . o. giraud and b. georgeot , phys . a * 72 * , 042312 ( 2005 ) .

ewens sampling formula ( esf ) is a one - parameter family of probability distributions with a number of intriguing combinatorial connections . this elegant closed - form formula first arose in biology as the stationary probability distribution of a sample configuration at one locus under the infinite - alleles model of mutation . since its discovery in the early 1970s , the esf has been used in various biological applications , and has sparked several interesting mathematical generalizations . in the population genetics community , extending the underlying random - mating model to include recombination has received much attention in the past , but no general closed - form sampling formula is currently known even for the simplest extension , that is , a model with two loci . in this paper , we show that it is possible to obtain useful closed - form results in the case the population - scaled recombination rate is large but not necessarily infinite . specifically , we consider an asymptotic expansion of the two - locus sampling formula in inverse powers of and obtain closed - form expressions for the first few terms in the expansion . our asymptotic sampling formula applies to arbitrary sample sizes and configurations . and . .

1010.3112

the probability of a sample configuration provides a useful ground for analyzing genetic data . popular applications include obtaining maximum likelihood estimates of model parameters and performing ancestral inference [ see @xcite ] . in principle , model - based full - likelihood analyses , such as that based on the coalescent [ kingman ( @xcite , @xcite ) ] , should be among the most powerful methods since they make full use of the data . however , in most cases , it is intractable to obtain a closed - form formula for the probability of a given data set . a well - known exception to this hurdle is the ewens sampling formula ( esf ) , which describes the stationary probability distribution of a sample configuration under the one - locus infinite - alleles model in the diffusion limit [ @xcite ] . notable biological applications of this closed - form formula include the test of selective neutrality [ see @xcite , slatkin ( @xcite , @xcite ) ] . @xcite provided a plya - like urn model interpretation of the formula , and recently @xcite provided a new combinatorial proof of the esf and extended the framework to obtain new results for the case with a variable population size . we refer the reader to the latter paper for a nice summary of previous works related to the esf . note that the esf also arises in several interesting contexts outside biology , including random partition structures and bayesian statistics ; see @xcite for examples of intricate combinatorial connections . the esf is a special case of the two - parameter sampling formula constructed by pitman ( @xcite , @xcite ) for exchangeable random partitions . @xcite considered generalizing the infinite - alleles model to include recombination and constructed a recursion relation satisfied by the two - locus sampling probability distribution at stationarity in the diffusion limit . @xcite later undertook a more mathematical analysis of the two - locus model and provided several interesting results . however , to date , a general closed - form formula for the two - locus sampling distribution remains unknown . indeed , it is widely recognized that recombination adds a formidably challenging layer of complexity to population genetics analysis . because obtaining exact analytic results in the presence of recombination is difficult , recent research has focused on developing sophisticated and computationally - intensive monte carlo techniques . examples of such techniques applied to the coalescent include monte carlo simulations [ see hudson ( @xcite , @xcite ) ] , importance sampling [ see @xcite , @xcite , @xcite , de iorio and griffiths ( @xcite , @xcite ) , @xcite ] and markov chain monte carlo methods [ see @xcite , @xcite , @xcite ] . being the simplest model with recombination , the two - locus case has been extensively studied in the past [ @xcite , @xcite , @xcite , @xcite , @xcite ] and a renewed wave of interest was recently sparked by @xcite , who proposed a composite likelihood method which uses two - locus sampling probabilities as building blocks . ldhat , a widely - used software package for estimating recombination rates , is based on this composite likelihood approach , and it has been used to produce a fine - scale map of recombination rate variation in the human genome [ @xcite , @xcite ] . ldhat assumes a symmetric diallelic recurrent mutation model at each locus and relies on the importance sampling scheme proposed by @xcite for the coalescent with recombination , to generate exhaustive lookup tables containing two - locus probabilities for all inequivalent sample configurations and a range of relevant parameter values . this process of generating exhaustive lookup tables is very computationally expensive . a fast and accurate method of estimating two - locus probabilities would be of practical value . in this paper , we revisit the tantalizing open question of whether a closed - form sampling formula can be found for the coalescent with recombination . we show that , at least for the two - locus infinite - alleles model with the population - scaled recombination rate @xmath0 large but not necessarily infinite , it is possible to obtain useful closed - form analytic results . note that the aforementioned monte carlo methods generally become less efficient as @xmath0 increases . those methods involve sampling a large collection of genealogical histories consistent with the observed sample configuration , and , when @xmath0 is large , the sampled genealogies tend to be very complicated ; they typically contain many recombination events , and it may take a long time for every locus to reach a most recent common ancestor . however , contrary to this increased complexity in the standard coalescent , we actually expect the evolutionary dynamics to be easier to describe for large @xmath0 , since the loci under consideration would then be less dependent . hence , it seems reasonable to conjecture that there may exist a stochastic process simpler than the standard coalescent with recombination that describes the relevant degrees of freedom in the large @xmath0 limit . we believe that our sampling formula may provide some hints as to what that dual process should be . the work discussed here generalizes previous results [ @xcite , @xcite ] for @xmath1 , in which case the loci become independent and the two - locus sampling distribution is given by a product of one - locus esfs . our main results can be summarized as follows . consider the diffusion limit of the two - locus infinite - alleles model with population - scaled mutation rates @xmath2 and @xmath3 at the two loci . for a sample configuration @xmath4 ( defined later in the text ) , we use @xmath5 to denote the probability of observing @xmath4 given the parameters @xmath6 and @xmath0 . for an arbitrary @xmath4 , our goal is to find an asymptotic expansion of @xmath5 in inverse powers of @xmath0 , that is , for large values of the recombination rate @xmath0 , our goal is to find @xmath7 where @xmath8 , and @xmath9 are independent of @xmath0 . as mentioned before , @xmath10 is given by a product of one - locus esfs . in this paper , we derive a closed - form formula for the first - order term @xmath11 . further , we show that the second - order term @xmath12 can be decomposed into two parts , one for which we obtain a closed - form formula and the other that satisfies a simple strict recursion . the latter can be easily evaluated using dynamic programming . details of these results are described in section [ sec : asy ] . in a similar vein , in section [ sec : joi ] , we obtain a simple asymptotic formula for the joint probability distribution of the number of alleles observed at the two loci . we remark that our work has practical value in genetic analysis . while this paper was under review , we applied the technique developed here to obtain analogous results for an arbitrary finite - alleles recurrent mutation model . see @xcite for details . in that paper , we performed an extensive assessment of the accuracy of our results for a particular finite - alleles model of mutation , and showed that they may be accurate even for moderate values of @xmath0 , including a range that is of biological interest . the accuracy ( not discussed here ) of our results for the infinite - alleles model is very similar to that finite - alleles case . in this section , we review the esf for the one - locus infinite - alleles model , as well as golding s ( @xcite ) recursion relation for the two - locus generalization . our notational convention generally follows that of @xcite . given a positive integer @xmath13 , @xmath14 $ ] denotes the @xmath13-set @xmath15 . for a nonnegative real number @xmath16 and a positive integer @xmath17 , @xmath18 denotes the @xmath17th ascending factorial of @xmath16 . we use @xmath19 to denote either a vector or a matrix of all zeroes ; it will be clear from context which is intended . throughout , we consider the diffusion limit of a neutral haploid exchangeable model of random mating with constant population size @xmath20 . we refer to the haploid individuals in the population as gametes . in the one - locus model , a sample configuration is denoted by a vector of multiplicities @xmath21 , where @xmath22 denotes the number of gametes with allele @xmath23 at the locus and @xmath24 denotes the total number of distinct allelic types observed . we use @xmath17 to denote @xmath25 , the total sample size . under the infinite - alleles model , any two gametes can be compared to determine whether or not they have the same allele , but it is not possible to determine how the alleles are related when they are different . therefore , allelic label is arbitrary . the probability of a mutation event at the locus per gamete per generation is denoted by @xmath26 . in the diffusion limit , @xmath27 and @xmath28 with the population - scaled mutation rate @xmath29 held fixed . each mutation gives rise to a new allele that has never been seen before in the population . for the one - locus model just described , @xcite obtained the following result . at stationarity in the diffusion limit of the one - locus infinite - alleles model with the scaled mutation parameter @xmath30 , the probability of an unordered sample configuration @xmath31 is given by @xmath32 where @xmath33 denotes the number of allele types represented @xmath23 times , that is , @xmath34 . let @xmath35 denote an _ ordered _ configuration of @xmath17 sequentially sampled gametes such that the corresponding unordered configuration is given by @xmath4 . by exchangeability , the probability of @xmath35 is invariant under all permutations of the sampling order . hence , we can write this probability of an ordered sample as @xmath36 without ambiguity . it is given by @xmath37^{-1 } = \biggl [ \prod_{i=1}^k ( n_i - 1 ) ! \biggr]\frac{\theta^k}{(\theta)_{n}},\ ] ] which follows from the fact that there are @xmath38 orderings corresponding to @xmath4 [ @xcite ] . to understand what we mean by ordered and unordered sampling configurations , it is helpful to relate the ewens sampling formula to the theory of random partitions . if the gametes are labeled in order of appearance by @xmath39 , then the resulting sample configuration defines a random partition of @xmath40 $ ] , with gametes belonging to the same block if and only if they have the same allele . the quantity @xmath41 is then the probability of a _ particular _ partition of @xmath40 $ ] whose block sizes are given by the entries in @xmath4 , while the quantity @xmath42 is the probability of observing _ any _ partition of @xmath40 $ ] with these block sizes . for example , if @xmath43 , then there are six partitions of @xmath44 $ ] with these block sizes , and so @xmath45 . it is often more convenient to work with an ordered sample than with an unordered sample . in this paper , we will work with the former ; that is , we will work with @xmath41 rather than @xmath42 . in the coalescent process going backward in time , at each event a lineage is lost either by coalescence or mutation . by consideration of the most recent event back in time , one can show that @xmath41 satisfies @xmath46\\[-8pt ] & & { } + \theta\sum_{i=1}^k \delta_{n_i,1 } q({\mathbf{n}}-{\mathbf { e}}_i\mid\theta),\nonumber\end{aligned}\ ] ] where @xmath47 is the kronecker delta and @xmath48 is a unit vector with the @xmath23th entry equal to one and all other entries equal to zero . the boundary condition is @xmath49 for all @xmath50 $ ] , and @xmath41 is defined to be zero if @xmath4 contains any negative component . it can be easily verified that the formula of @xmath41 shown in ( [ q(a ) ] ) satisfies the recursion ( [ one - locus - recursion ] ) . @xcite also obtained the following result regarding the number of allelic types . let @xmath51 denote the number of distinct allelic types observed in a sample of size @xmath17 . then @xmath52 where @xmath53 are the unsigned stirling numbers of the first kind . note that @xmath54 . it follows from ( [ esf ] ) and ( [ eq : k_n ] ) that @xmath51 is a sufficient statistic for @xmath30 . golding ( @xcite ) first generalized the one - locus recursion ( [ one - locus - recursion ] ) to two loci , and @xcite later undertook a more mathematical study of the model . we denote the two loci by @xmath55 and @xmath56 , and use @xmath2 and @xmath3 to denote the respective population - scaled mutation rates . we use @xmath24 and @xmath57 to denote the number of distinct allelic types observed at locus @xmath55 and locus @xmath56 , respectively . the population - scaled recombination rate is denoted by @xmath58 , where @xmath59 is the probability of a recombination event between the two loci per gamete per generation . a key observation is that to obtain a closed system of equations , the type space must be extended to allow some gametes to be specified only at one of the two loci . the two - locus sample configuration is denoted by @xmath60 , where @xmath61 with @xmath62 being the number of gametes with allele @xmath23 at locus @xmath55 and unspecified alleles at locus @xmath56 , @xmath63 with @xmath64 being the number of gametes with unspecified alleles at locus @xmath55 and allele @xmath65 at locus @xmath56 , and @xmath66 is a @xmath67 matrix with @xmath68 being the multiplicity of gametes with allele @xmath23 at locus @xmath55 and allele @xmath65 at locus @xmath56 . further , we define @xmath69 we use @xmath70 to denote the sampling probability of an ordered sample with configuration @xmath71 . for ease of notation , we do not show the dependence on parameters . for @xmath72 , golding s ( @xcite ) recursion for @xmath70 takes the following form : @xmath73q({\mathbf { a}},{\mathbf{b}},{\mathbf{c } } ) \nonumber\\ & & \qquad= \sum_{i=1}^k a_i(a_i-1 + 2c_{i\cdot})q({\mathbf{a}}-{\mathbf{e } } _ i,{\mathbf{b}},{\mathbf{c}})\nonumber\\ & & \qquad\quad { } + \sum_{j=1}^l b_j(b_j-1 + 2c_{\cdot j})q({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf{c } } ) \nonumber\\ & & \qquad\quad { } + \sum_{i=1}^k\sum_{j=1}^l [ c_{ij}(c_{ij}-1)q({\mathbf{a}},{\mathbf{b}},{\mathbf{c}}-{\mathbf{e}}_{ij } ) \nonumber\\ & & \qquad\quad\hspace*{45.2pt } { } + 2 a_i b_j q({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf{b}}-{\mathbf { e}}_j,{\mathbf { c}}+{\mathbf{e}}_{ij } ) ] \\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \biggl[\sum_{j=1}^l \delta _ { a_i+c_{i\cdot } , 1}\delta_{c_{ij},1 } q({\mathbf{a}},{\mathbf{b}}+{\mathbf { e}}_j,{\mathbf { c}}-{\mathbf{e}}_{ij})\nonumber\\ & & \qquad\quad\hspace*{92.8pt } { } + \delta_{a_i,1}\delta_{c_{i\cdot},0 } q({\mathbf{a}}- { \mathbf{e}}_i,{\mathbf{b}},{\mathbf{c } } ) \biggr ] \nonumber\\ & & \qquad\quad { } + \theta_b\sum_{j=1}^l \biggl[\sum_{i=1}^k \delta_{b_j + c_{\cdot j},1 } \delta_{c_{ij},1 } q({\mathbf{a}}+{\mathbf{e}}_i,{\mathbf { b}},{\mathbf { c}}-{\mathbf{e}}_{ij})\nonumber\\ & & \qquad\quad\hspace*{90.1pt } { } + \delta_{b_j,1}\delta_{c_{\cdot j},0 } q({\mathbf { a}},{\mathbf{b}}- { \mathbf{e}}_j,{\mathbf{c } } ) \biggr ] \nonumber\\ & & \qquad\quad { } + \rho\sum_{i=1}^k\sum_{j=1}^l c_{ij}q({\mathbf{a}}+{\mathbf{e } } _ i,{\mathbf{b}}+{\mathbf{e}}_j,{\mathbf{c}}-{\mathbf { e}}_{ij}).\nonumber\end{aligned}\ ] ] relevant boundary conditions are @xmath74 for all @xmath50 $ ] and @xmath75 $ ] . for notational convenience , we deviate from @xcite and allow each summation to range over all allelic types . to be consistent , we define @xmath76 whenever any entry in @xmath77 , @xmath78 or @xmath79 is negative . for ease of discussion , we define the following terms . the degree of @xmath70 is defined to be @xmath80 . we say that a recursion relation is strictly recursive if it contains only a single term of the highest degree . except in the special case @xmath81 , a closed - form solution for @xmath70 is not known . notice that the terms @xmath82 and @xmath83 on the right - hand side of ( [ golding ] ) have the same degree as @xmath70 on the left - hand side . therefore , ( [ golding ] ) is not strictly recursive . for each degree , we therefore need to solve a system of coupled equations , and this system grows very rapidly with @xmath17 . for example , for a sample with @xmath84 and @xmath85 , computing @xmath86 requires solving a system of more than 20,000 coupled equations [ @xcite ] ; this is around the limit of sample sizes that can be handled in a reasonable time . in the following section , we revisit the problem of obtaining a closed - form formula for @xmath70 and obtain an asymptotic expansion for large @xmath0 . for large @xmath0 , our objective is to find an asymptotic expansion of the form @xmath87 where @xmath88 and @xmath9 are independent of @xmath0 . our closed - form formulas will be expressed using the following notation . for a given multiplicity vector @xmath61 with @xmath89 , we define @xmath90 \frac { \theta_a^k}{(\theta_a)_{a}}.\ ] ] similarly , for a given multiplicity vector @xmath91 with @xmath92 , we define @xmath93\frac { \theta_b^l}{(\theta_b)_{b}}.\ ] ] as discussed in section [ sec : esf ] , @xmath94 ( respectively , @xmath95 ) gives the probability of an ordered sample taken from locus @xmath55 ( respectively , @xmath56 ) . we use @xmath96}$ ] and @xmath97}$ ] to denote the marginal sample configurations of @xmath79 restricted to locus @xmath55 and locus @xmath56 , respectively . the leading - order term @xmath98 is equal to @xmath70 when @xmath99 , in which case the two loci are independent . theorem 2.3 of @xcite states that @xmath100 . more generally , one can obtain the following result for the leading - order contribution . [ prop : leading ] in the asymptotic expansion ( [ expansion ] ) of the two - locus sampling formula , the zeroth - order term @xmath101 is given by @xmath102 although this result is intuitively obvious , in section [ sec : proof_leading ] we provide a detailed new proof , since it well illustrates our general strategy . one of the main results of this paper is a closed - form formula for the next order term @xmath103 . the case with @xmath104 admits a particularly simple solution . [ lemma : first ] in the asymptotic expansion ( [ expansion ] ) of the two - locus sampling formula , the first - order term satisfies @xmath105 for arbitrary @xmath77 and @xmath78 . that @xmath106 vanishes is not expected a priori . below we shall see that @xmath107 in general . for an arbitrary configuration matrix @xmath79 of nonnegative integers , we obtain the following closed - form formula for @xmath103 . [ thm : first ] in the asymptotic expansion ( [ expansion ] ) of the two - locus sampling formula , the first - order term @xmath108 is given by @xmath109\\[-8pt ] & & { } -q^a({\mathbf{a}}+{\mathbf{c}}_a)\sum_{j=1}^l \pmatrix{c_{\cdot j } \cr2 } q^b({\mathbf{b}}+{\mathbf{c}}_b-{\mathbf{e}}_j ) \nonumber \nonumber\\ & & { } + \sum_{i=1}^k \sum_{j=1}^l \pmatrix{c_{ij } \cr2 } q^a({\mathbf { a}}+{\mathbf{c}}_a-{\mathbf{e}}_i)q^b({\mathbf{b}}+{\mathbf{c}}_b -{\mathbf{e}}_j)\nonumber\end{aligned}\ ] ] for arbitrary configurations @xmath110 of nonnegative integers . lemma [ lemma : first ] is used in proving theorem [ thm : first ] . a proof of theorem [ thm : first ] is provided in section [ sec : proof_first ] , while a proof of lemma [ lemma : first ] is given in section [ sec : proof_lemma_first ] . note that the functional form of @xmath101 and @xmath108 in ( [ leading ] ) and ( [ firstorder ] ) has no explicit dependence on mutation ; that is , the dependence on mutation is completely absorbed into the marginal one - locus probabilities . it turns out that ( [ leading ] ) and ( [ firstorder ] ) are _ universal _ in that they also apply to an arbitrary finite - alleles model of mutation , with @xmath94 and @xmath95 replaced with appropriate marginal one - locus probabilities for the assumed mutation model . see @xcite for details . in principle , similar arguments can be used to find the @xmath111th - order term given the @xmath65th , although a general expression does not seem to be easy to obtain . in section [ sec : proof_second ] , we provide a proof of the following result for @xmath112 . [ thm : second ] in the asymptotic expansion ( [ expansion ] ) of the two - locus sampling formula , the second - order term @xmath112 is of the form @xmath113 where @xmath114 is given by the analytic formula shown in the , and @xmath115 satisfies the following strict recursion : @xmath116q_2({\mathbf { a}},{\mathbf { b}},\mathbf{0 } ) \nonumber\\ & & \qquad= \sum_{i=1}^k a_i(a_i-1)q_2({\mathbf{a}}-{\mathbf { e}}_i,{\mathbf { b}},\mathbf{0 } ) + \sum_{j=1}^l b_j(b_j-1)q_2({\mathbf{a}},{\mathbf { b}}-{\mathbf{e}}_j,\mathbf{0 } ) \nonumber\\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \delta_{a_i,1}q_2({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}},\mathbf{0 } ) + \theta_b\sum_{j=1}^l\delta _ { b_j,1}q_2({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,\mathbf{0 } ) \\ & & \qquad\quad { } + 4 \biggl[a\theta_a - ( \theta_a+ a - 1)\sum _ { i=1}^k \delta _ { a_i,1 } \biggr ] \nonumber\\ & & \qquad\quad\hspace*{15.7pt}{}\times \biggl[b\theta_b - ( \theta_b+ b - 1)\sum _ { j=1}^l \delta _ { b_j,1 } \biggr ] q^a({\mathbf{a}})q^b({\mathbf{b}})\nonumber\end{aligned}\ ] ] with boundary conditions @xmath117 for all @xmath50 $ ] and @xmath75 $ ] . in contrast to @xmath106 ( cf . lemma [ lemma : first ] ) , it turns out that @xmath115 does not vanish in general . we do not have an analytic solution for @xmath115 , but note that ( [ q2(a , b,0 ) ] ) is strictly recursive and that it can be easily solved numerically using dynamic programming . numerical study ( not shown ) suggests that the relative contribution of @xmath118 to @xmath70 is in most cases extremely small . [ see @xcite for details . ] deriving an analytic expression for @xmath114 in ( [ q2concise ] ) is a laborious task , as the long equation in the suggests . we have written a computer program to verify numerically that our analytic result is correct . following the same strategy as in the previous section , we can obtain the asymptotic behavior of the joint distribution of the number of alleles observed at the two loci in a sample . to make explicit the dependence of these numbers on the sample size , write the number of alleles at locus @xmath55 as @xmath119 and the number of alleles at locus @xmath56 as @xmath120 . @xcite proved that the probability @xmath121 satisfies the recursion @xmath122p(a , b , c;k , l ) \nonumber\\ & & \qquad = a(a-1 + 2c)p(a-1,b , c;k , l ) + b(b-1 + 2c)p(a , b-1,c;k , l ) \nonumber\\ & & \qquad\quad { } + c(c-1)p(a , b , c-1;k , l ) + 2a b p(a-1,b-1,c+1;k , l ) \nonumber\\ & & \qquad\quad { } + \theta_a [ a p(a-1,b , c;k-1,l ) + c p(a , b+1,c-1;k-1,l ) ] \\ & & \qquad\quad { } + \theta_b [ b p(a , b-1,c;k , l-1 ) + c p(a+1,b , c-1;k , l-1 ) ] \nonumber\\ & & \qquad\quad { } + \rho c p(a+1,b+1,c-1;k , l),\nonumber\end{aligned}\ ] ] where @xmath123 if @xmath124 , @xmath125 , @xmath126 , @xmath127 , @xmath128 , @xmath129 , or @xmath130 . equation ( [ kl ] ) has a unique solution satisfying the initial conditions @xmath131 for @xmath132 . as with golding s recursion , equation ( [ kl ] ) can be solved numerically , but quickly becomes computationally intractable with growing @xmath17 . the only exception is the special case of @xmath1 , for which the distribution is given by the product of ( [ eq : k_n ] ) for each locus . in what follows , we use the following notation in writing an asymptotic series for @xmath133 . for loci @xmath55 and @xmath56 , respectively , we define the analogues of ( [ eq : k_n ] ) as @xmath134 and @xmath135 where @xmath136 and @xmath137 are the stirling numbers of the first kind . we pose the expansion @xmath138 for large @xmath0 . then , in section [ sec : proof_leadingkl ] , we prove the following result for the zeroth - order term . [ prop : leadingkl ] for an asymptotic expansion of the form ( [ expansionkl ] ) satisfying the recursion ( [ kl ] ) , @xmath139 is given by @xmath140 similar to lemma [ lemma : first ] , we obtain the following vanishing result for the first - order term in the case of @xmath141 . [ lemma : firstkl ] for an asymptotic expansion of the form ( [ expansionkl ] ) satisfying the recursion ( [ kl ] ) , we have @xmath142 using this lemma , it is then possible to obtain the following result for an arbitrary @xmath143 . [ prop : firstkl ] for an asymptotic expansion of the form ( [ expansionkl ] ) satisfying the recursion ( [ kl ] ) , @xmath144 is given by @xmath145\nonumber\\[-8pt]\\[-8pt ] & & { } \times [ p^b(b+c;l ) - p^b(b+c-1;l ) ] .\nonumber\end{aligned}\ ] ] proofs of proposition [ prop : firstkl ] and lemma [ lemma : firstkl ] are provided in sections [ sec : proof_firstkl ] and [ sec : proof_lemma_firstkl ] , respectively . in what follows , we provide proofs of the results mentioned in the previous two sections . first , assume @xmath146 . substitute the expansion ( [ expansion ] ) into golding s recursion ( [ golding ] ) , divide by @xmath147 and let @xmath148 . we are then left with @xmath149 now , applying ( [ unwrap0 ] ) repeatedly gives @xmath150\times[l]}c_{ij}!}{c ! } q_0({\mathbf{a}}+{\mathbf{c}}_a,{\mathbf{b}}+{\mathbf{c}}_b , \mathbf{0}),\ ] ] where the summation is over all distinct orderings of the @xmath143 gametes with multiplicity @xmath66 . there are @xmath151 such orderings and since the summand is independent of the ordering , we conclude @xmath152 clearly , ( [ unwrap0b ] ) also holds for @xmath153 . from a coalescent perspective , this equation tells us that any gamete with specified alleles ( i.e. , `` carrying ancestral material '' ) at both loci must undergo recombination instantaneously backward in time . now , by substituting the asymptotic expansion ( [ expansion ] ) with @xmath104 into golding s recursion ( [ golding ] ) and letting @xmath148 , we obtain @xmath154q_0({\mathbf { a}},{\mathbf{b}},\mathbf{0 } ) \nonumber\\ & & \qquad= \sum_{i=1}^k a_i(a_i-1)q_0({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf { b}},\mathbf{0 } ) + \sum_{j=1}^l b_j(b_j-1)q_0({\mathbf{a}},{\mathbf { b}}-{\mathbf{e}}_j,\mathbf{0 } ) \nonumber\\[-8pt]\\[-8pt ] & & \qquad\quad { } + 2\sum_{i=1}^k\sum_{j=1}^l a_i b_j q_0({\mathbf{a}}-{\mathbf{e } } _ i,{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf{e}}_{ij } ) \nonumber\\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \delta_{a_i,1}q_0({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}},\mathbf{0 } ) + \theta_b\sum_{j=1}^l\delta _ { b_j,1}q_0({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,\mathbf{0}).\nonumber\end{aligned}\ ] ] equation ( [ unwrap0b ] ) implies @xmath155 , so with a bit of rearranging we are left with @xmath156q_0({\mathbf { a}},{\mathbf{b}},\mathbf{0 } ) \nonumber\\ & & \qquad= \sum_{i=1}^k a_i(a_i-1)q_0({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf { b}},\mathbf{0 } ) + \sum_{j=1}^l b_j(b_j-1)q_0({\mathbf{a}},{\mathbf { b}}-{\mathbf{e}}_j,\mathbf{0 } ) \\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \delta_{a_i,1}q_0({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}},\mathbf{0 } ) + \theta_b\sum_{j=1}^l\delta _ { b_j,1}q_0({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,\mathbf{0})\nonumber\end{aligned}\ ] ] with boundary conditions @xmath157 for all @xmath50 $ ] and . noting that ( [ goldingc=0 ] ) is the sum of two independent recursions of the form ( [ one - locus - recursion ] ) , one for each locus and each with appropriate boundary condition , we conclude that @xmath158 is given by @xmath159 a product of two ( ordered ) esfs . it is straightforward to verify that ( [ esfx2 ] ) satisfies ( [ goldingc=0 ] ) . finally , using ( [ unwrap0b ] ) and ( [ esfx2 ] ) , we arrive at ( [ leading ] ) . first , assume @xmath160 . substitute the asymptotic expansion ( [ expansion ] ) into golding s recursion ( [ golding ] ) , eliminate terms of order @xmath0 by applying ( [ unwrap0 ] ) , and let @xmath148 . after applying ( [ unwrap0b ] ) to the remaining terms and invoking ( [ goldingc=0 ] ) , with some rearrangement we obtain @xmath161 now , by utilizing ( [ esfx2 ] ) , this can be written in the form @xmath162 where @xmath163\\[-8pt ] & & { } -q^a({\mathbf{a}}+{\mathbf{c}}_a)\sum_{j=1}^l \frac { c_{\cdot j}(c_{\cdot j}-1)}{c}q^b({\mathbf{b}}+{\mathbf{c}}_b -{\mathbf{e}}_j ) \nonumber\\ & & { } + \sum_{i=1}^k\sum_{j=1}^l \frac{c_{ij}(c_{ij}-1)}{c}q^a ( { \mathbf{a}}+{\mathbf{c}}_a-{\mathbf{e}}_i)q^b({\mathbf { b}}+{\mathbf { c}}_b-{\mathbf{e}}_j).\nonumber\end{aligned}\ ] ] above , we assumed @xmath146 . we define @xmath164 if @xmath165 . iterating the recursion ( [ q1+f ] ) , we may write @xmath108 as @xmath166.\end{aligned}\ ] ] similarly , repeatedly iterating ( [ q1+f ] ) yields @xmath167 the key observation is that the right - hand side of ( [ expandf ] ) has a nice probabilistic interpretation which allows us to obtain a closed - form formula . to be more precise , consider the first summation @xmath168 for a fixed sample configuration @xmath79 , this can be interpreted as the sum over all possible ways of throwing away a gamete at random and calculating @xmath169 based on the remaining subsample , which we will denote @xmath170 . equivalently , it is the expected value of @xmath169 with respect to subsampling without replacement @xmath171 of the gametes in @xmath79 . write this as @xmath172,\ ] ] where @xmath173 is the random subsample obtained by sampling without replacement @xmath171 gametes from @xmath79 , and @xmath174 , @xmath175 . note that once the subsample @xmath176 is obtained , then @xmath177 and @xmath178 are fully specified . more generally , consider the @xmath179th sum in ( [ expandf ] ) . a particular term in the summation corresponds to an ordering of @xmath180 gametes in @xmath79 , which , when removed leave a subsample @xmath181 . with respect to this subsample , the summand is @xmath182 and for each such subsample @xmath183 there are @xmath184 distinct orderings of the remaining types in @xmath79 , with each ordering contributing the same amount to the sum . here , @xmath185 denotes the multinomial coefficient : @xmath186 gathering identical terms , the @xmath179th sum in ( [ expandf ] ) can therefore be written over all distinct subsamples of @xmath79 of size @xmath187 : @xmath188,\end{aligned}\ ] ] where , for a fixed @xmath187 , @xmath189 is a multivariate hypergeometric@xmath190 random variable ; that is , @xmath191\times[l ] } \bigl[c_{ij}^{(m ) } = c_{ij}^{(m ) } \bigr ] \biggr ) = \frac{1}{{c\choose m } } \prod_{(i , j)\in[k]\times[l ] } \pmatrix{c_{ij}\cr c_{ij}^{(m)}}.\ ] ] furthermore , marginally we have @xmath192 in summary , ( [ expandf ] ) can be written as @xmath193.\ ] ] according to lemma [ lemma : first ] , the first term @xmath194 vanishes , so we are left with @xmath195.\ ] ] finally , since @xmath196 and @xmath197 , ( [ eq : f ] ) and ( [ subsampling ] ) together imply @xmath198q^a({\mathbf{a}}+{\mathbf { c}}_a-{\mathbf{e}}_i ) \\ & & \qquad\quad\hspace*{21.2pt } { } - q^a({\mathbf{a}}+{\mathbf{c}}_a)\frac{1}{m}\sum_{j=1}^l \mathbb{e}\bigl[c_{\cdot j}^{(m)}\bigl(c_{\cdot j}^{(m)}-1\bigr)\bigr]q^b({\mathbf { b}}+{\mathbf{c}}_b-{\mathbf{e}}_j ) \\ & & \qquad\quad\hspace*{21.2pt } { } + \frac{1}{m}\sum_{i=1}^k \sum_{j=1}^l \mathbb{e } \bigl[c_{ij}^{(m)}\bigl(c_{ij}^{(m)}-1\bigr)\bigr]q^a({\mathbf{a}}+{\mathbf{c}}_a -{\mathbf{e}}_i)q^b({\mathbf{b}}+{\mathbf{c}}_b-{\mathbf{e}}_j)\biggr].\end{aligned}\ ] ] the moments in this equation are easy to compute and one can sum them over @xmath187 to obtain the desired result ( [ firstorder ] ) . first , note that for any sample @xmath199 and any subsample of the form @xmath200 , we have @xmath201 , since every term on right - hand side of ( [ eq : f ] ) has a vanishing coefficient . so , equation ( [ q1concise ] ) implies @xmath202 for any @xmath203\times[l]$ ] . now , substitute the asymptotic expansion ( [ expansion ] ) with @xmath104 into golding s recursion ( [ golding ] ) . note that terms of order @xmath0 are absent since @xmath204 . eliminate terms with coefficients independent of @xmath0 by applying ( [ goldingc=0preliminary ] ) , multiply both sides of the recursion by @xmath0 , and let @xmath148 to obtain the following : @xmath205q_1({\mathbf{a}},{\mathbf { b}},\mathbf{0 } ) \\ & & \qquad= \sum_{i=1}^k a_i(a_i-1)q_1({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf { b}},\mathbf{0 } ) + \sum_{j=1}^l b_j(b_j-1)q_1({\mathbf{a}},{\mathbf { b}}-{\mathbf{e}}_j,\mathbf{0 } ) \\ & & \qquad\quad { } + 2\sum_{i=1}^k \sum_{j=1}^l a_i b_j q_1({\mathbf{a}}-{\mathbf{e } } _ i,{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf{e}}_{ij } ) \\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \delta_{a_i,1}q_1({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}},\mathbf{0 } ) + \theta_b\sum_{j=1}^l\delta _ { b_j,1}q_1({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,\mathbf{0})\end{aligned}\ ] ] with boundary conditions @xmath206 for all @xmath50 $ ] and @xmath75 $ ] . this equation can be made strictly recursive by applying ( [ unwrap1b ] ) to @xmath207 . it therefore follows from the boundary conditions ( for example , by induction ) that @xmath208 . here , we provide only an outline of a proof ; details are similar to the proof of theorem [ thm : first ] . substitute the asymptotic expansion ( [ expansion ] ) into golding s recursion ( [ golding ] ) , eliminate terms with coefficients proportional to @xmath0 or independent of @xmath0 . then , multiply both sides of the recursion by @xmath0 and let @xmath148 to obtain @xmath209 \nonumber\\[-8pt]\\[-8pt ] & & \qquad\quad { } + \theta_a\sum_{i=1}^k \biggl[\sum_{j=1}^l \delta _ { a_i+c_{i\cdot},1}\delta_{c_{ij},1}q_1({\mathbf{a}},{\mathbf { b}}+{\mathbf{e } } _ j,{\mathbf{c}}-{\mathbf{e}}_{ij})\nonumber\\ & & \qquad\quad\hspace*{93.3pt } { } + \delta_{a_i,1}\delta _ { c_{i\cdot } , 0}q_1({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf{b}},{\mathbf { c } } ) \biggr ] \nonumber\\ & & \qquad\quad { } + \theta_b\sum_{j=1}^l \biggl[\sum_{i=1}^k \delta _ { b_j+c_{\cdot j},1}\delta_{c_{ij},1}q_1({\mathbf{a}}+{\mathbf { e}}_i,{\mathbf { b}},{\mathbf{c}}-{\mathbf{e}}_{ij})\nonumber\\ & & \qquad\quad\hspace*{89.3pt } { } + \delta_{b_j,1}\delta _ { c_{\cdot j},0}q_1({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf { c } } ) \biggr ] \nonumber\\ & & \qquad\quad { } - [ n(n-1 ) + \theta_a(a+c ) + \theta _ b(b+c)]q_1({\mathbf { a}},{\mathbf{b}},{\mathbf{c } } ) . \nonumber\end{aligned}\ ] ] by substituting our expression ( [ firstorder ] ) for @xmath108 , the right - hand side can be expressed as a function @xmath210 which is completely known but rather cumbersome to write down . as in the proof of theorem [ thm : first ] , the same `` unwrapping '' maneuver can be applied to rearrange ( [ secondrec ] ) into the form ( [ q2concise ] ) , where @xmath211.\ ] ] this time @xmath212 $ ] is a function of fourth - order moments of the multivariate hypergeometric distribution . the formula shown in the is obtained by evaluating the expectations and summing over @xmath187 . we now show that @xmath115 satisfies the recursion shown in ( [ q2(a , b,0 ) ] ) . we will use the fact that for a sample @xmath213 , we have @xmath214 for an arbitrary @xmath79 , @xmath215 is much more complicated . now , one can adopt the approach used in the proof of lemma [ lemma : first ] to obtain a strict recursion for @xmath118 . first , note that ( [ q2concise ] ) and ( [ g(a1,b1,c1 ) ] ) imply @xmath216 \nonumber\\ & & \qquad= q_2({\mathbf{a}},{\mathbf{b}},\mathbf{0 } ) + g({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf{e}}_{ij } ) \nonumber\\ & & \qquad= q_2({\mathbf{a}},{\mathbf{b}},\mathbf{0 } ) + 2(a-1)(b-1)q^a ( { \mathbf{a}})q^b({\mathbf{b } } ) \\\ & & \qquad\quad { } - 2(b-1)(a_i-1)q^a({\mathbf{a}}-{\mathbf{e}}_i)q^b({\mathbf{b } } ) \nonumber\\ & & \qquad\quad { } - 2(a-1)(b_j-1)q^a({\mathbf{a}})q^b({\mathbf{b}}-{\mathbf{e}}_j ) \nonumber\\ & & \qquad\quad { } + 2(a_i-1)(b_j-1)q^a({\mathbf{a}}-{\mathbf{e}}_i)q^b({\mathbf { b}}-{\mathbf{e}}_j).\nonumber\end{aligned}\ ] ] as before , substitute the asymptotic expansion ( [ expansion ] ) for @xmath217 into golding s recursion ( [ golding ] ) , eliminate terms with coefficients independent of @xmath0 or proportional to @xmath218 , and let @xmath148 to obtain @xmath205q_2({\mathbf{a}},{\mathbf { b}},\mathbf{0 } ) \\ & & \qquad= \sum_{i=1}^k a_i(a_i-1)q_2({\mathbf{a}}-{\mathbf{e}}_i,{\mathbf { b}},\mathbf{0 } ) + \sum_{j=1}^l b_j(b_j-1)q_2({\mathbf{a}},{\mathbf { b}}-{\mathbf{e}}_j,\mathbf{0 } ) \\ & & \qquad\quad { } + 2\sum_{i=1}^k \sum_{j=1}^l a_i b_j q_2({\mathbf{a}}-{\mathbf{e } } _ i,{\mathbf{b}}-{\mathbf{e}}_j,{\mathbf{e}}_{ij } ) \\ & & \qquad\quad { } + \theta_a\sum_{i=1}^k \delta_{a_i,1}q_2({\mathbf { a}}-{\mathbf{e } } _ i,{\mathbf{b}},\mathbf{0 } ) + \theta_b\sum_{j=1}^l\delta _ { b_j,1}q_2({\mathbf{a}},{\mathbf{b}}-{\mathbf{e}}_j,\mathbf{0})\end{aligned}\ ] ] with boundary conditions @xmath117 for all @xmath50 $ ] and @xmath75 $ ] . this equation can be made strictly recursive by applying ( [ unwrap2 ] ) to @xmath219 . after some simplification , this leads to the recursion ( [ q2(a , b,0 ) ] ) . the proof is similar to the proof of proposition [ prop : leading ] , working with the system ( [ kl ] ) rather than golding s recursion ( [ golding ] ) . first , assume @xmath146 . substitute the expansion ( [ expansionkl ] ) into the recursion ( [ kl ] ) , divide by @xmath147 and let @xmath220 . we are left with @xmath221 which implies @xmath222 clearly , ( [ unwrapkl ] ) also holds for @xmath153 . now , by substituting the asymptotic expansion ( [ expansionkl ] ) with @xmath141 into ( [ kl ] ) and letting @xmath148 , we obtain @xmath223p_0(a , b,0;k , l ) \nonumber\\ & & \qquad= a(a-1)p_0(a-1,b,0;k , l ) + b(b-1)p_0(a , b-1,c;k , l ) \nonumber\\ & & \qquad\quad { } + 2ab p_0(a-1,b-1,1;k , l ) \\ & & \qquad\quad { } + \theta_aa p_0(a-1,b,0;k-1,l)\nonumber\\ & & \qquad\quad { } + \theta_bb p_0(a , b-1,0;k , l-1).\nonumber\end{aligned}\ ] ] after invoking ( [ unwrapkl ] ) on @xmath224 and rearranging , we are left with @xmath225p_0(a , b,0;k , l ) \nonumber\\ & & \qquad = a(a-1)p_0(a-1,b,0;k , l ) + b(b-1)p_0(a , b-1,0;k , l ) \nonumber\\[-8pt]\\[-8pt ] & & \qquad\quad{}+ \theta_aa p_0(a-1,b,0;k-1,l)\nonumber\\ & & \qquad\quad { } + \theta_bb p_0(a , b-1,0;k , l-1)\nonumber\end{aligned}\ ] ] with boundary conditions @xmath226 and @xmath227 . equation ( [ klc=0 ] ) can be expressed as a linear sum of two independent recursions : @xmath228 with respective boundary conditions @xmath229 and @xmath230 . these recursions are precisely those considered by ewens [ ( @xcite ) , ( 21 ) ] , with respective solutions ( [ k ] ) and ( [ l ] ) . hence , @xmath231 and @xmath232 , and it is straightforward to verify that @xmath233 satisfies ( [ klc=0 ] ) . substituting this solution into ( [ unwrapkl ] ) , we arrive at ( [ leadingkl ] ) , as required . first , assume @xmath146 . substitute the asymptotic expansion ( [ expansionkl ] ) into the recursion ( [ kl ] ) , eliminate terms with coefficients linear in @xmath0 by applying ( [ unwrapkl ] ) , and let @xmath148 . after applying ( [ unwrapkl ] ) to the remaining terms and invoking ( [ klc=0 ] ) , with some rearrangement we obtain @xmath234\\[-8pt ] & & \qquad\quad\hspace*{33pt}{}- p_0(a+c , b+c-1,0;k , l)\nonumber\\ & & \qquad\quad\hspace*{120.04pt } { } + p_0(a+c-1,b+c-1,0;k , l)].\nonumber\end{aligned}\ ] ] applying the recursion repeatedly , this becomes @xmath235\\[-8pt ] & & \qquad\quad\hspace*{14.05pt } { } - p_0(a+c , b+c-1,0;k , l ) \nonumber\\ & & \qquad\quad\hspace*{100.5pt}{}+ p_0(a+c-1,b+c-1,0;k , l)]\nonumber\\ & & \qquad\quad\hspace*{10.6pt}{}\times\sum_{m=0}^{c-1 } ( c-1-m).\nonumber\end{aligned}\ ] ] according to lemma [ lemma : firstkl ] , the first term @xmath236 vanishes . hence , since @xmath139 is given by ( [ leadingkl ] ) , the right - hand side of ( [ p1+f ] ) is fully known . with some rearrangement , we are left with ( [ firstorderkl ] ) . first , note that ( [ firstreckl ] ) implies @xmath237 now , substitute the asymptotic expansion ( [ expansionkl ] ) with @xmath238 into ( [ kl ] ) , eliminate leading - order terms by applying ( [ klc=0preliminary ] ) , and let @xmath148 . the result is made strictly recursive by invoking ( [ unwrap1kl ] ) , and we obtain @xmath239p_1(a , b,0;k , l ) \\ & & \qquad= a(a-1)p_1(a-1,b,0;k , l)\\ & & \qquad\quad { } + b(b-1)p_1(a , b-1,0;k , l ) \\ & & \qquad\quad{}+ \theta_aa p_1(a-1,b,0;k-1,l)\\ & & \qquad\quad { } + \theta_bb p_1(a , b-1,0;k , l-1)\end{aligned}\ ] ] with boundary conditions @xmath240 , for @xmath241 . it therefore follows ( e.g. , by induction ) that @xmath242 . [ appendixa ] we use @xmath244 to denote @xmath245 , @xmath246 to denote @xmath247 , @xmath248 to denote @xmath249 , and so on . then @xmath250q^aq^b\nonumber\\ & & \qquad\quad { } - \frac{\theta_a(c-1)}{2}\sum_{i=1}^k \delta_{a_i,0}\delta _ { c_{i\cdot } , 1 } q_i^aq^b- \frac{\theta_b(c-1)}{2}\sum_{j=1}^l \delta _ { b_j,0}\delta _ { c_{\cdot j},1 } q^aq_j^b\nonumber\\ & & \qquad\quad { } + \sum_{i=1}^k \biggl[\frac{\theta_a- c(c-3 ) + 2a + 4b - 4}{4 } c_{i\cdot}(c_{i\cdot}-1 ) \nonumber\\ & & \qquad\quad\hspace*{72pt } { } -(2b+c-1 ) c_{i\cdot}(a_i + c_{i\cdot } -1 ) \biggr]q_i^aq^b\nonumber\\ & & \qquad\quad { } + \frac{1}{2}\sum_{i=1}^k \biggl[\frac{\theta_a}{2}\delta _ { c_{i\cdot } , 2 } + \frac{5 - 6a_i-4c_{i\cdot}}{6 } \biggr]c_{i\cdot}(c_{i\cdot } -1)q_{ii}^aq^b\nonumber\\ & & \qquad\quad { } + \sum_{j=1}^l \biggl[\frac{\theta_b- c(c-3 ) + 2b + 4a - 4}{4 } c_{\cdot j}(c_{\cdot j}-1 ) \nonumber\\ & & \qquad\quad\hspace*{71.83pt } { } -(2a+c-1 ) c_{\cdot j}(b_j + c_{\cdot j}-1 ) \biggr]q^aq_j^b\nonumber\\ & & \qquad\quad { } + \frac{1}{2}\sum_{j=1}^l \biggl[\frac{\theta_b}{2}\delta _ { c_{\cdot j},2 } + \frac{5 - 6b_j-4c_{\cdot j}}{6 } \biggr]c_{\cdot j}(c_{\cdot j}-1 ) q^aq_{jj}^b\nonumber\\ & & \qquad\quad { } + \sum_{i , k=1}^k \frac{c_{i\cdot}(c_{i\cdot}-1)c_{k\cdot } ( c_{k\cdot}-1)}{8}q_{ik}^a q^b\\ & & \qquad\quad { } + \sum_{j , l=1}^l \frac{c_{\cdot j}(c_{\cdot j}-1)c_{\cdot l}(c_{\cdot l}-1)}{8 } q^aq_{jl}^b\nonumber\\ & & \qquad\quad { } -\frac{\theta_a+\theta_b - c(c-5)+2a+2b-4}{4}\sum_{i=1}^k\sum_{j=1}^l c_{ij}(c_{ij}-1)q_i^aq_j^b\nonumber\\ & & \qquad\quad { } + \sum_{i=1}^k\sum_{j=1}^l \biggl[\frac{c_{i\cdot}(c_{i\cdot } -1)c_{\cdot j}(c_{\cdot j}-1)}{4}\\ & & \qquad\quad\hspace*{48pt } { } + \frac{c_{ij}(c_{ij}+1 - 2c_{i\cdot } + 2c_{i\cdot}c_{\cdot j}-2c_{\cdot j})}{2 } \nonumber\\ & & \qquad\quad\hspace*{48pt}{}+ c_{ij}b_j(c_{i\cdot } -1 ) + c_{ij}a_i(c_{\cdot j}-1 ) + 2a_i b_j c_{ij } \nonumber\\ & & \qquad\quad\hspace*{48pt}{}+ \frac{\theta _ b}{2}\delta _ { b_j,0}\delta_{c_{\cdot j},1}\delta_{c_{ij},1}(c_{i\cdot } - 1)\\ & & \qquad\quad\hspace*{117pt } { } + \frac { \theta_a}{2}\delta_{a_i,0}\delta_{c_{i\cdot},1}\delta _ { c_{ij},1}(c_{\cdot j } - 1 ) \biggr]q_i^aq_j^b\nonumber\\ & & \qquad\quad { } + \frac{1}{2}\sum_{i=1}^k \biggl[a_i+c_{i\cdot}-1-\frac{\theta_a } { 2}\delta_{c_{i\cdot},2 } \biggr]\sum_{j=1}^l c_{ij}(c_{ij}-1)q_{ii}^a q_j^b\nonumber\\ & & \qquad\quad { } + \frac{1}{2}\sum_{j=1}^l \biggl[b_j+c_{\cdot j}-1-\frac{\theta_b } { 2}\delta_{c_{\cdot j},2 } \biggr]\sum_{i=1}^k c_{ij}(c_{ij}-1)q_i^a q_{jj}^b\nonumber\\ & & \qquad\quad { } - \frac{1}{4}\sum_{i=1}^k\sum_{j=1}^l\sum_{k=1}^k c_{ij}(c_{ij}-1)c_{k\cdot}(c_{k\cdot}-1)q_{ik}^aq_j^b\nonumber\\ & & \qquad\quad { } - \frac{1}{4}\sum_{i=1}^k\sum_{j=1}^l\sum_{l=1}^l c_{ij}(c_{ij}-1)c_{\cdot l}(c_{\cdot l}-1)q_i^aq_{jl}^b\nonumber\\ & & \qquad\quad { } + \frac{1}{8}\sum_{i=1}^k\sum_{j=1}^l\sum_{k=1}^k\sum _ { l=1}^l c_{ij}(c_{ij}-1)c_{kl}(c_{kl}-1)q_{ik}^aq_{jl}^b\nonumber \\ & & \qquad\quad { } - \frac{1}{12}\sum_{i=1}^k\sum_{j=1}^l c_{ij}(c_{ij}-1)(2c_{ij}-1)q_{ii}^aq_{jj}^b.\end{aligned}\ ] ] to check the correctness of the above expression , we also solved the recursion ( [ secondrec ] ) numerically for all sample configurations of sizes @xmath251 , and @xmath252 ( with @xmath253 ) , and confirmed that the above analytic expression agreed in all cases . we also implemented a _ program to solve @xmath254 exactly in the special case @xmath255 . the program can return series expansions in terms of @xmath218 as @xmath148 which are symbolic in @xmath2 and @xmath3 . we could then compare the first three terms against @xmath256 , @xmath257 , and @xmath9 , for various sample configurations @xmath258 . we thank robert c. griffiths , charles h. langley and joshua paul for useful discussions .

we prove that the standard quantum mechanical description of a quantum state change due to measurement , given by lders rules , is a special case of the constrained maximisation of a quantum relative entropy functional . this result is a quantum analogue of the derivation of the bayes laplace rule as a special case of the constrained maximisation of relative entropy . the proof is provided for the umegaki relative entropy of density operators over a hilbert space as well as for the araki relative entropy of normal states over a-algebra . we also introduce a quantum analogue of jeffrey s rule , derive it in the same way as above , and discuss the meaning of these results for quantum bayesianism . * lders and quantum jeffrey s rules as entropic projections * + + ryszard pawe kostecki + + perimeter institute for theoretical physics , 31 caroline street north , n2l 2y5 waterloo , ontario , canada + institute for theoretical physics , department of physics , university of warsaw , hoa 69 , 00 - 681 warszawa , poland + ` [email protected] ` + + august 15 , 2014

1408.3502

an important part of a mathematical setting of probability theory can be considered as a special case of quantum theoretic kinematics . this can be seen most clearly when quantum theoretic kinematics is reformulated in algebraic terms , with quantum states defined as normal positive ( or normalised ) functionals over @xmath0-algebras ( see e.g. @xcite for an overview ) . the borel steinhaus kolmogorov measure theoretic approach to foundations of probability theory is then recovered precisely from commutative @xmath0-algebras and quantum states over them . ) . ] however , both quantum theory and probability theory are equipped with additional structures , which describe possible mappings of probabilities or quantum states . can these prescriptions of information dynamics also be directly related to each other ? in particular , there are various considerations @xcite of von neumann s and lders rules @xcite as noncommutative _ analogues _ of the bayes laplace rule @xcite . it is tempting to ask whether this analogy could be turned to something more definite , both conceptually and mathematically . the main motivation for this paper is a series of results @xcite showing that both bayes laplace and jeffrey s rules ( as well as some other rules @xcite ) can be derived as special cases of the constrained minimisation of various information distances on probabilistic models ( in section [ bayes.section ] we review briefly some of these results ) . this has led us to conjecture @xcite that lders rules may be special cases of constrained maximisation of a quantum relative entropy . in section [ vn.lu.section ] we prove that the weak lders rule of quantum state change due to `` nonselective quantum measurement '' is a special case of quantum entropic projection , provided by the minimisation of the araki distance subject to a specific set of constraints . we also introduce a quantum analogue of jeffrey s rule and derive it as another special case of constrained araki distance minimisation . these are the two main results of this paper . in addition , we show that the strong lders rule of quantum state change due to `` selective quantum measurement '' can be obtained from these results as the limiting case of quantum jeffrey s rule or by regularised araki distance minimisation . in section [ transition.correlation.section ] we show that both weak lders rule and strong von neumann s rule ( which is the same as strong lders rule for pure states ) are also quantum entropic projections for a different quantum distance functional . with an exception of a derivation of a quantum jeffrey s rule , our results hold for arbitrary @xmath0-algebras , so they are applicable in quantum field theoretic and relativistic quantum information problems . these results extend earlier considerations of lders rules as _ analogues _ of the bayes laplace rule with a novel mathematical and conceptual content : all these rules are special cases of the constrained relative entropic inference . in this sense , ` quantum bayesianism ' can be considered a branch of ` quantum relative entropism ' . a discussion of this issue is provided in section [ discussion.section ] . + * history of the problem . * an inference based on minimisation of quantum distance on quantum models was first proposed by herbut @xcite . he derived the weak lders rule from a constrained minimisation of norm distance in the hilbert schmidt operator space . unfortunately , his work has been left unnoticed by all works cited below . several years later marchand and collaborators @xcite used bures distance @xcite ( which is metrical ) , and argued that its constrained minimisation should be considered as a description of the change of state of information _ due to _ `` quantum measurement '' described by a specific form of a coarse graining map @xcite ( conditioned upon a subset of an operator algebra , and predual to a specific form of noncommutative conditional expectation @xcite ) . independently of this body of work , in @xcite it was proposed to use the constrained minimisation of the wgkl distance of probabilities arising from traces of density operators to derive the post - measurement quantum state . the rules of inferential change of quantum states based on a constrained minimisation of other metrical distances on quantum models were later reconsidered by other authors @xcite and some derivations of the strong von neumann rule were obtained ( see section [ transition.correlation.section ] ) . the reinterpretation of von neumann s and lders rules for `` quantum measurement '' as principles of inductive inference conditioned on specific information , and analogous to the bayes laplace rule , was proposed at about the same time by bub @xcite ( however , it can be claimed @xcite , that already von neumann was aware of the possibility of such interpretation ) . these two lines of thought were ( implicitly ) joined in hadjisavvas @xcite postulate that a quantum state change due to acquisition of data ( e.g. in a measurement ) should be provided by means of constrained minimisation of the jmgk distance , as well as in donald s @xcite postulate that a description of `` quantum measurement '' , understood as an inductive inference , should be provided by means of constrained minimisation of the araki distance . at least for injective @xmath0-algebras @xmath1 . ] donald stressed also that this procedure @xcite , where @xmath2 is the constraint set . however , he provided no derivation of any of lders rules ( nor any other `` quantum measurement '' rules ) from this procedure . independently of the above works , warmuth @xcite used constrained minimisation of umegaki s distance to derive a generalisation of the bayes laplace rule to the case of density operators , with conditional probabilities replaced by covariance matrices . this generalisation has not reproduced lders rules . our derivation of lders rule is not only the first such result obtained for umegaki s and araki s distances , but also first result of this type obtained for any nonsymmetric quantum information distance . all results for quantum jeffrey s rule are new . see also a closely related paper @xcite , where the analogous results for weak and strong lders rules are derived using another technique ( based on differentiation , as opposed to generalised pythagorean theorem ) . , where @xmath3 are rank @xmath4 projectors , minimises the umegaki distance @xmath5 . this is a special case of our result for the weak lders rule . the generalisation to our result is stated without proof in @xcite . ] the modern mathematical formulation of the bayes laplace foundations for probability theory @xcite is based on finitely additive boolean algebras @xmath6 and conditional probabilities , defined as maps @xmath7 $ ] . its kinematics is given by the rules @xmath8 from which the @xcite follows , @xmath9 the ` marginal probability ' @xmath10 , called also an ` evidence ' , is a normalising constant calculated from @xmath11 where @xmath12 is a countable set , while the set @xmath13 is exhaustive ( @xmath14 ) and its elements are mutually exclusive ( @xmath15 for @xmath16 ) . in the simple cases this set may consist of two elements : @xmath17 . the dynamics of this approach is given by the @xmath18 the map determines a rule of construction of the new conditional probability assignment associated to @xmath19 under the constraint that certain additional statements ( ` facts ' , ` data ' , ` events ' ) @xmath20 are considered as ( ` appear as ' , ` are known as ' , ` are regarded as ' ) true or false . so , @xmath21 in is interpreted as a ( ` posterior ' ) conditional probability assigned to @xmath19 whenever the truth value of @xmath20 is given ( ` known ' ) . if attribution of a definite truth value to @xmath22 is interpreted as an ` acquisition of data / facts ' , then the bayes laplace rule can be understood as a procedure of statistical inference that transforms ` prior ' information states @xmath23 about _ all _ ` hypotheses ' @xmath24 into ` posterior ' information states @xmath21 , under constraints provided by the acquired data @xmath20 _ and _ the assumed relationship between @xmath20 and @xmath19 which is encoded in the information states @xmath25 . the probability @xmath25 is sometimes called the ` sampling probability ' ( because before the constraint @xmath20 is applied , @xmath25 can represent a probability associated with possible values of constraints for a fixed hypothesis @xmath19 ) or the ` likelihood ' ( because after the constraint @xmath20 is applied , @xmath25 is considered as a probability of @xmath20 as a function over possible hypotheses @xmath19 ) . jeffrey @xcite proposed an alternative dynamical rule , now called , @xmath26 where @xmath27 , @xmath28 is a set of exhaustive and mutually exclusive elements of @xmath6 , and the constraints @xmath29 @xmath30 hold . the defining equality in is equivalent to the condition @xmath31 if @xmath32 with @xmath33 and if @xmath34 , then jeffrey s rule reduces to the bayes laplace rule . the first derivation of the bayes laplace and jeffrey s rules from constrained minimisation of the information distance functional ( more precisely , the wgkl distance ) was provided by williams @xcite . however , as he admitted , his derivation has a restricted validity , because it does not deal with conditional probabilities . this can be solved following van fraassen s suggestion : @xcite . following the results of caticha and giffin @xcite , consider a constrained minimisation of the wgkl distance , @xmath35 for @xmath36 , @xmath37 , with parametrisation @xmath38 allowing to consider a measure space @xmath39 as well as a product measure space @xmath40 , and with constraints given by @xmath41 where @xmath42 and @xmath43 are lagrange multipliers , and @xmath44 is dirac s delta at @xmath45 . the posterior probability selected as a unique solution of this variational problem is given by @xmath46 where @xmath43 is determined via @xmath47 hence , @xmath48 which leads to the bayes laplace rule on @xmath49 , can be interpreted as a density of a conditional measure under some conditions . this is not problematic as long as one interprets the above derivation as a proof of backwards compatibility of entropic projections with the bayes laplace framework ( which is finitary anyway ) , and not as a method of extending this framework from finite to countably additive boolean algebras . ] @xmath50 whenever @xmath51 is such that @xmath52 ( for example , if @xmath53 ) . if the second constraint in is replaced by a more general form , @xmath54 corresponding to a condition @xmath55 with a given probability density @xmath56 , then the entropic projection reproduces jeffrey s rule on @xmath49 , @xmath57 the bayes laplace rule changes information states by means of constraints imposed on the level of propositions , while the entropic projections utilise constraints imposed on probabilities . hence , in order to recover the former rule from the maximum entropy rule one needs a very strong constraint , which forces a unique reference of probability distribution to an underlying space @xmath58 of propositions : @xmath59 . so , while ( for example ) the mean value constraints can be partially dismissed by the new knowledge that is incorporated by the sequential maximum relative entropy updating ( see e.g. @xcite ) , dirac s delta constraints always remain preserved by subsequent updatings . space , while usually the mean value constraints are applied by integration over @xmath58 space , but nevertheless this remark holds in general . ] from this perspective , jeffrey s rule can be understood as arising due to weakening of constraints , which are allowed to carry some additional uncertainty . more generally , diaconis and zabell @xcite have shown that , for a suitable choice of constraints , jeffrey s rule can be derived from a constrained minimisation of any csiszr morimoto @xmath60-distance @xcite for a strictly convex @xmath610,\infty[\,\ra\rr$ ] . in particular , @xcite derived jeffrey s rule from constrained minimisation of @xmath62 . in theorem [ weighted.vnlu.theorem ] of section [ vn.lu.section ] we will derive a quantum analogue of the latter result . in this section we present the mathematical terminology that is used throughout the paper . we also briefly introduce some more advanced notions and results from the theory of operator algebras that are required to set up the mathematical background and notation for the @xmath0-algebraic part of the results presented in sections [ vn.lu.section ] and [ transition.correlation.section ] . see @xcite for a detailed discussion of these structures and their properties . a is a banach space @xmath63 over @xmath64 with unit @xmath65 that is also an algebra over @xmath64 and is equipped with an operation @xmath66 satisfying @xmath67 , @xmath68 , @xmath69 , @xmath70 , and @xmath71 , where @xmath72 is a complex conjugation of @xmath73 . a is defined as such @xmath74-algebra that has a banach predual . if a predual of @xmath74-algebra exists then it is unique . given a @xmath0-algebra @xmath1 , we will denote its predual by @xmath75 . moreover , @xmath76 , @xmath77 , @xmath78 , @xmath79 , @xmath80 , @xmath81 . for @xmath82 , @xmath83 . if @xmath84 is a localisable measure space , then @xmath85 is a commutative @xmath0-algebra , and @xmath86 is its predual . every commutative @xmath0-algebra can be represented in this form . this way the theory of @xmath0-algebras generalises both the localisable measure theory and the theory of bounded operators over hilbert spaces . we define a as a set @xmath87 , where @xmath84 is a localisable measure space . the elements of @xmath88 are radon nikodm quotients of probability measures ( dominated by @xmath89 ) with respect to @xmath89 , and are called . if @xmath1 is a @xmath0-algebra , then we define a as a set @xmath90 . the elements of @xmath91 will be called or ( just ) . a on a @xmath0-algebra @xmath1 is defined as a function @xmath92 $ ] such that @xmath93 , @xmath94 , and @xmath95 , with the convention @xmath96 . a weight is called : iff @xmath97 ; iff @xmath98 ; iff a left ideal in @xmath1 given by @xmath99 is weakly-@xmath100 dense in @xmath1 ; iff @xmath101 ; iff @xmath102 for any uniformly bounded increasing net @xmath103 . a space of all normal semi - finite weights on a @xmath0-algebra @xmath1 is denoted @xmath104 , while the subset of all faithful elements of @xmath104 is denoted @xmath105 . every state is a finite normal weight , and every faithful state is a finite faithful normal state , hence the diagram @xmath106 \ar@{^{(}->}[d ] & \w_0(\n ) \ar@{^{(}->}[d]\\ \n^+_\star \ar@{^{(}->}[r ] & \w(\n ) } \label{wstar.states.weights.comm}\ ] ] commutes . for @xmath107 , @xmath108 for @xmath109 we will write @xmath110 iff @xmath111 . and @xmath112 for @xmath113 , then @xmath114 , so for any @xmath115 with @xmath116 one has @xmath117 iff @xmath118 . ] an element @xmath119 is faithful iff @xmath120 . if @xmath121 is a normal weight on a @xmath0-algebra @xmath1 ( which includes @xmath122 as a special case ) , then the restriction of @xmath121 to a @xmath0-algebra , @xmath123 is a faithful normal weight ( respectively , an element of @xmath124 ) . if @xmath121 is semi - finite , then @xmath125 . hence , given @xmath107 and @xmath126 , @xmath127 iff @xmath128 and @xmath129 . in particular , for @xmath130 and @xmath117 , we have @xmath131 . a of a @xmath74-algebra @xmath63 is defined as a pair @xmath132 of a hilbert space @xmath133 and a @xmath134-homomorphism @xmath135 . a representation @xmath136 is called : iff @xmath137 is dense in @xmath133 ; iff it is continuous with respect to the weak-@xmath100 topologies of @xmath63 and @xmath138 ; iff @xmath139 . an element @xmath140 is called for a @xmath74-algebra @xmath141 iff @xmath142 is norm dense in @xmath138 . a representation @xmath136 of a @xmath74-algebra @xmath63 is called iff there exists @xmath143 that is cyclic for @xmath144 . according to the gelfand namark segal theorem @xcite for every pair @xmath145 of a @xmath74-algebra @xmath63 and @xmath146 there exists a triple @xmath147 of a hilbert space @xmath148 and a cyclic representation @xmath149 with a cyclic vector @xmath150 , and this triple is unique up to unitary equivalence . it is constructed as follows . for a @xmath74-algebra @xmath63 and @xmath146 , one defines the scalar form @xmath151 on @xmath63 , @xmath152 and the @xmath153 which is a left ideal of @xmath63 , closed in the norm topology ( it is also closed in the weak-@xmath100 topology if @xmath154 ) . the form @xmath155 is hermitean on @xmath63 and it becomes a scalar product @xmath155 on @xmath156 . the hilbert space @xmath148 is obtained by the completion of @xmath156 in the topology of norm generated by @xmath155 . consider the morphisms @xmath157_\omega:\c\ni x&\longmapsto [ x]_\omega\in\c/\i_\omega,\\ \pi_\omega(y):[y]_\omega&\longmapsto[xy]_\omega.\end{aligned}\ ] ] the element @xmath146 is uniquely represented in terms of @xmath148 by the vector @xmath158_\omega=:\omega_\omega\in\h_\omega$ ] , which is cyclic for @xmath159 and satisfies @xmath160 . hence @xmath161 an analogue of this theorem for weights follows the similar construction , but lacks cyclicity . if @xmath1 is a @xmath0-algebra , and @xmath162 is a weight on @xmath1 , then there exists the hilbert space @xmath148 , defined as the completion of @xmath163 in the topology of a norm generated by the scalar product @xmath164 , @xmath165 and there exist the maps @xmath157_\omega:\nnn_\omega\ni x&\mapsto [ x]_\omega\in\h_\omega , \label{gns.class.weight}\\ \pi_\omega:\n\ni x&\mapsto([y]_\omega\mapsto[xy]_\omega)\in\bbb(\h_\omega ) , \label{gns.rep.weight}\end{aligned}\ ] ] such that @xmath166_\omega$ ] is linear , @xmath167_\omega)$ ] is dense in @xmath148 , and @xmath168 is a representation of @xmath1 . if @xmath169 then @xmath168 is nondegenerate and normal . it is also faithful if @xmath170 . the of a subalgebra @xmath1 of any algebra @xmath63 is defined as @xmath171 while the of @xmath1 is defined as @xmath172 . a unital @xmath134-subalgebra @xmath1 of an algebra @xmath138 is called the @xcite iff @xmath173 . an image @xmath174 of any representation @xmath132 of a @xmath0-algebra @xmath1 is a von neumann algebra iff @xmath175 is normal and nondegenerate . a subspace @xmath176 of a complex hilbert space @xmath133 is called a iff @xmath177 @xmath178 @xmath179 . a cone @xmath176 is called iff @xmath180 every self - polar cone @xmath176 is pointed ( @xmath181 ) , spans linearly @xmath133 ( @xmath182 ) , and determines a unique conjugation , where @xmath183 , is called a iff it is antilinear , isometric , and involutive ( @xmath184 ) . ] @xmath185 in @xmath133 such that @xmath186 @xcite , as well as a partial order on the set @xmath187 given by , @xmath188 if @xmath1 is a @xmath0-algebra , @xmath133 is a hilbert space , @xmath189 is a self - polar cone , @xmath175 is a nondegenerate faithful normal representation of @xmath1 on @xmath133 , and @xmath185 is conjugation on @xmath133 , then the quadruple @xmath190 is called of @xmath1 and @xmath191 is called of @xmath1 iff the conditions @xcite @xmath192 hold . for any standard representation @xmath193 holds . the map @xmath194 is order preserving . if @xmath195 for some hilbert space @xmath196 , then the standard representation hilbert space is given by the space @xmath197 of hilbert schmidt operators equipped with an inner product @xmath198 , where @xmath199 denotes banach dual space . in this case , the map reads @xmath200 . for a given @xmath0-algebra @xmath1 , @xmath201 , and @xmath170 the map @xmath202_\omega\mapsto[x^*]_\phi\;\;\forall x\in\nnn_\omega\cap\nnn_\phi^ * \label{relative.modular.weights}\ ] ] is a densely defined , closable antilinear operator . its closure admits a unique polar decomposition @xmath203 where @xmath204 is a conjugation operator , called , while @xmath205 is a positive self - adjoint operator on @xmath206 with @xmath207 , called a @xcite . the relative modular operators allow to define a one - parameter family of partial isometries in @xmath208 , called @xcite , @xmath209 where @xmath210 is arbitrary , so it can be set equal to @xmath162 . as shown by araki and masuda @xcite ( see also @xcite ) , the definition of @xmath211 and @xmath212 can be further extended to the case when @xmath213 , by means of a densely defined closable antilinear operator @xmath202_\omega+(\ii-\supp(\overline{[\nnn_\phi]_\omega}))\zeta\mapsto\supp(\omega)[x^*]_\phi\;\;\forall x\in\nnn_\omega\cap\nnn_\phi^*\;\forall\zeta\in\h , \label{relative.modular.for.normal.weights}\ ] ] where @xmath190 is a standard representation of a @xmath0-algebra @xmath1 , and @xmath214 . for @xmath215 this becomes a closable antilinear operator @xcite @xmath216 acting on a dense domain @xmath217 , where @xmath218 denotes a complement of the closure in @xmath133 of the linear span of the action @xmath174 on @xmath219 . in both cases , the relative modular operator is determined by the polar decomposition of the closure @xmath220 of @xmath221 , @xmath222 if or is used instead of , then the formula has to be replaced by @xmath223_\psi}):=\delta^{\ii t}_{\phi,\psi}\delta^{-\ii t}_{\omega,\psi } , \label{connes.for.ns.weights}\ ] ] and @xmath212 is a partial isometry in @xmath224 whenever @xmath225=0 $ ] . given any set @xmath226 , a is defined as a map @xmath227 $ ] such that @xmath228 @xmath229 @xmath230 . a is a map @xmath231 $ ] such that @xmath232 is a distance . a distance is called : iff @xmath233 ; iff @xmath234 ; @xcite iff is it bounded , symmetric and satisfies @xmath235 we will use the symbol @xmath236 instead of @xmath237 to denote metrical distances . a distance on a statistical model will be called a , while a distance on a quantum model will be called a . a term with be used to refer to any of them . let @xmath238 denote a powerset of @xmath226 . if @xmath239 is such that the map @xmath240 is a singleton ( one - element set , @xmath241 ) for all @xmath242 , then we call an . is a unique maximiser of a relative entropy @xmath243 . ] for any @xmath0-algebra @xmath1 , and @xmath244 the following formulas define quantum distances on @xmath245 , @xmath246 @xmath247 @xmath248 hence , @xmath249 all above examples are special cases of a family of quantum @xmath250-distances @xmath251 with @xmath252 $ ] , see @xcite . a special case of @xmath253 is the @xcite @xmath254 if @xmath255 , then takes the form @xcite @xmath256 for a semi - finite @xmath1 , normal faithful semi - finite trace @xmath257 on @xmath1 and @xmath258 and @xmath259 defined by @xmath260 and @xmath261 , the araki distance turns to the @xcite ( cf . also @xcite ) @xmath262 if @xmath117 , and @xmath263 otherwise . a special , but definitely most popular , case of is obtained for @xmath82 and @xmath264 . for a commutative @xmath265 , where @xmath84 is any localisable measure space such that @xmath266 and @xmath267 where @xmath268 @xmath269 and @xmath270 , the araki distance takes a form of the @xcite , @xmath271 under these conditions , and assuming @xmath272 , one has @xmath273 among above distances , only @xmath274 is symmetric , and none of them is metrical . metrical quantum distances will be discussed in section [ transition.correlation.section ] . if @xmath275 , @xmath12 is a countable set , and @xmath276 satisfies @xmath277 and @xmath278 @xmath279 , then the is defined as a map @xcite @xmath280 if @xmath281 , then the is defined as a map @xmath282 with domain restricted by the condition @xmath283 . the , called sometimes a ` partial collapse ' , is defined as a map @xmath284 where @xmath185 is a subset of a countable set @xmath12 corresponding to an orthogonal decomposition @xmath277 , and the domain in is restricted by a condition @xmath285 . if @xmath286 , @xmath12 , and @xmath287 are such as in , @xmath288 , and the condition @xmath289 holds , then can be written as @xmath290 in particular , if @xmath291 , then turns to the direct extension of strong lders rule to @xmath292 , which coincides with on @xmath293 . in the case when @xmath294 , then @xmath295 such that @xmath296 ( i.e. , @xmath286 is a projector onto a closed one - dimensional linear subspace of @xmath133 spanned by @xmath297 ) , and the strong lders rule takes the form of the @xcite , called also a ` state vector reduction ' , @xmath298 [ quantum.jeffrey.rule.definition ] we define as a map @xmath299 where @xmath27 , @xmath300 , @xmath301 , and @xmath302 , @xmath278 , @xmath303 @xmath304 . [ strongluedersfromqjr.remark ] in the case when an orthogonal decomposition of @xmath305 is given by the set @xmath306 , and @xmath307 , then reduces to . the direct analogy between this property and the conditions under which jeffrey s rule reduces to the bayes laplace rule , together with the analogy between and , justify the name we gave to . definition of @xmath308 is equivalent to a condition @xmath309=0\;\;\forall i\in\{1,\ldots , n\ } , \label{equivalent.quantum.jeffrey.condition}\ ] ] with @xmath310=0 $ ] @xmath30 . an implication from to is easy . in the opposite direction , let us consider an arbitrary @xmath311 such that @xmath312=0 $ ] @xmath30 . for an arbitrary countable set @xmath12 the condition @xmath313=0 $ ] @xmath314 is equivalent to @xmath315 and to @xmath316=0 $ ] , where @xmath317 with arbitrary @xmath318 . see @xcite for a clear discussion of these conditions and their equivalence . hence , @xmath319 this gives @xmath320 in what follows we will prove that the weak lders rule is a special case of an entropic projection , determined by constrained minimisation with a @xmath321 distance ( theorems [ vonneumann.lueders.from.entropy.theorem ] and [ gen.vonneumann.lueders.from.entropy.theorem ] ) . next , we will use different constraints to derive quantum jeffrey s rule ( theorem [ weighted.vnlu.theorem ] ) . finally , we will show that the strong lders rule arises as a weakly continuous limit of quantum jeffrey s rule ( remark [ remark.strong.vnlu ] ) , and can be also obtained by a constrained minimisation of a _ regularised _ modification of @xmath321 distance ( theorem [ regularised.mre.vn.theorem ] ) . the necessity of recourse to limit or regularisation indicates that , as opposed to weak lders rule , strong lders rule is not directly derivable from minimisation of @xmath322 distance . however , in section [ transition.correlation.section ] we will show that in some special cases ( which usually turn it to the strong von neumann rule ) it can be derived from minimisation of @xmath274 distance . according to de muynck @xcite ( see also @xcite ) , the strong von neumann and strong lders rules should be viewed as procedures of updating of quantum state , yet not in the _ predictive _ ( inferential ) sense , but in the _ preparative _ ( calibrating ) sense . if one extends this interpretation to the weak lders rule , and assumes that the ` preparative ' use of entropic projection should be implemented by constrained minimisation of information distance in its _ second _ argument , then our result can be interpreted by saying that the weak lders rule of _ quantum state preparation _ can be derived from the constrained minimisation of the araki distance @xmath323 . ( we do not consider this interpretation as necessary . ) in what follows @xmath324 will be an arbitrary set , @xmath325 its arbitrary subset , and @xmath237 will be an arbitrary ( not necessarily bounded ) distance on @xmath324 . if @xmath326 consists of a single element , then we will denote it by @xmath327 . let @xmath328 and @xmath325 . if @xmath329 holds , then we will say that @xmath330 satisfies for @xmath331 at @xmath332 with respect to @xmath237 . let @xmath333 . if @xmath334 then we will say that the pair @xmath335 satisfies with respect to @xmath237 . if @xmath325 satisfies triangle equality for @xmath331 at @xmath332 with respect to @xmath237 , then @xmath336 from @xmath337 and @xmath338 it follows that @xmath339 . from @xmath340 it follows that @xmath326 exists , is unique , and is equal to @xmath286 . [ lemma.subsubprojection ] let @xmath335 satisfy the subprojection property with respect to @xmath237 , and let @xmath341 satisfy triangle equality for @xmath328 at @xmath342 with respect to @xmath237 . then @xmath343 satisfies triangle equality for @xmath328 at @xmath344 with respect to @xmath237 and @xmath345 triangle equalities in this case read @xmath346 now , let @xmath347 . this gives @xmath348 for @xmath349 we have @xmath350 this gives @xmath351 [ lemma.n.triangle ] if @xmath352 satisfies triangle equality for every @xmath353 , and @xmath354 satisfy subprojection property for every @xmath355 , then @xmath356 where @xmath357 , @xmath358 , and @xmath359 satisfies triangle equality at @xmath360 . we will prove this lemma by mathematical induction . let us assume that it holds for some @xmath361 . then @xmath362 and @xmath363 satisfies triangle equality with @xmath364 . let @xmath365 , and consider @xmath366 then @xmath367 for every @xmath368 , and from the subprojection property for @xmath369 it follows that @xmath370 , so @xmath371 . hence , subprojection property holds for @xmath372 and @xmath373 . lemma [ lemma.subsubprojection ] applied to @xmath372 and @xmath373 gives @xmath374 this lemma holds for @xmath375 by subprojection property of @xmath376 and @xmath377 . in what follows , we will assume that @xmath82 and @xmath378 for some hilbert space @xmath133 of arbitrary dimension . the value of @xmath27 will be kept arbitrary but fixed . [ tracelog.triangle.lemma ] if @xmath379 , then the triangle equality at @xmath286 for @xmath331 with respect to @xmath322 is equivalent with @xmath380 if these distance functionals are finite , then @xmath381 [ small.theorem ] let @xmath382=0\}$ ] , and let @xmath383 , where @xmath126 . then @xmath384 the operators @xmath385 are block - diagonal matrices , so any functions of @xmath286 and @xmath121 are also block - diagonal . thus , @xmath386=0 $ ] and @xmath387=0 $ ] . @xmath388 substituting @xmath389 we get @xmath390 similarly , we obtain @xmath391 and it follows that @xmath392 by lemma [ tracelog.triangle.lemma ] , this gives , but it remains to check whether the assumption of this lemma is satisfied . we have @xmath393 now we want to show that @xmath394 . let us denote in matrix form @xmath395 from @xmath396 it follows that @xmath397 @xmath398 @xmath399 . moreover , @xmath400 so , for @xmath401 , the corresponding forms satisfy @xmath402 hence , as operators , @xmath403 hence @xmath404 using @xmath405 together with and , we obtain @xmath406 which follows from @xmath407 [ lemma.pipj ] given @xmath408 , let @xmath409=0\}$ ] for @xmath410 , and @xmath411=0 $ ] . if @xmath412 then @xmath413 from proposition [ small.theorem ] it follows that @xmath414 from @xmath415=0 $ ] we obtain @xmath416= p_i[\psi , p_j]p_i+(\ii - p_i)[\psi , p_j](\ii - p_i).\ ] ] let @xmath417 , @xmath411=0 $ ] @xmath304 , and @xmath418=0\;\forall i\in\{1,\ldots , n\}\}$ ] . then @xmath330 satisfies triangle equality and @xmath419 where @xmath420 for @xmath421 and @xmath358 . follows directly from proposition [ lemma.n.triangle ] , lemma [ lemma.pipj ] and proposition [ small.theorem ] . [ vonneumann.lueders.from.entropy.theorem ] if @xmath417 satisfies @xmath278 @xmath304 , @xmath422 , and @xmath423=0\;\forall i\in\{1,\ldots , n\}\ } , \label{ql.constraint}\ ] ] then @xmath424 by mathematical induction . assume that @xmath425 then @xmath426 what follows from @xmath427 the first step of this induction is satisfied by proposition [ small.theorem ] . [ commutation.as.expectation.lemma ] if @xmath417 satisfies @xmath278 @xmath304 and @xmath422 , then the conditions 1 . @xmath428=0\;\forall i\in\{1,\ldots , n\}\}$ ] 2 . @xmath429)=0\;\forall x\in\n\;\forall f:\{1,\ldots , n\}\ra\cc\}$ ] are equivalent . using the property @xmath430)=\tr([p,\rho]x)$ ] , we have @xmath431)=0\;\forall x\in\n\;\ ; \iff\;\ ; \tr([p,\rho]x)=0\;\forall x\in\n\;\ ; \iff\;\ ; [ p,\rho]=0.\ ] ] since there is finitely many projections , every function in a @xmath0-algebra @xmath1 generated by these projections is their finite sum , and is equivalent to a function on @xmath432 points . as shown by lemma [ commutation.as.expectation.lemma ] , the weak lders rule has no equivalent in the commutative case , because then the condition @xmath433)=0 $ ] is satisfied trivially for arbitrary @xmath434 . hence , one _ can not _ interpret the weak lders rule as a noncommutative _ generalisation _ of the bayes laplace rule . now we will generalise theorem [ vonneumann.lueders.from.entropy.theorem ] to arbitrary @xmath0-algebras @xmath1 and @xmath435 . in order to prove this theorem , we will need also to generalise lemma [ tracelog.triangle.lemma ] , proposition [ small.theorem ] , and lemma [ lemma.pipj ] . this will be provided , respectively , by lemma [ gen.tracelog.triangle.lemma ] , proposition [ gen.small.theorem ] , and lemma [ gen.lemma.pipj ] . our main tool will be the expression for @xmath436 in terms of connes cocycle , introduced by petz in @xcite . [ gen.tracelog.triangle.lemma ] given @xmath437 , consider a gns representation @xmath438 . if @xmath439 , then the triangle equality for @xmath331 at @xmath286 with respect to @xmath322 is equivalent to @xmath440 if @xmath439 , then @xmath441 moreover , given @xmath442 we have @xmath443 hence , @xmath444 so , @xmath445 where the second equation follows from the property @xmath446 @xmath447 for a standard representative @xmath448 of @xmath331 in a standard representation @xmath190 , which in this case is given by the cyclic vector @xmath449 of the gns representation @xmath438 . the triangle equality reads @xmath450 and is equivalent to @xmath451 it remains to calculate @xmath452 [ block.diagonality.of.phi ] if @xmath126 and @xmath453 then @xmath454)=0 $ ] @xmath455 iff @xmath121 is block - diagonal , that is , iff @xmath456 . if @xmath454)=0 $ ] , then @xmath457 , so , for @xmath458 , @xmath459 and similarly @xmath460 . conversely , every @xmath461 has the form @xmath462 hence , @xmath463&=pyp+py(\ii - p)-pyp-(\ii - p)yp = py(\ii - p)yp,\\ \phi([p , y])&=\phi(py(\ii - p)-(\ii - p)yp)=0.\end{aligned}\ ] ] [ gen.small.theorem ] let @xmath464)=0\;\forall x\in\n\}$ ] , and let @xmath465 , where @xmath126 . then @xmath466 we will use block decomposition of @xmath1 into @xmath467 together with the corresponding notation @xmath468 by lemma [ block.diagonality.of.phi ] , the states @xmath385 are block - diagonal , @xmath469 recall from - that for a given standard representation of @xmath1 on a hilbert space @xmath133 , and @xmath140 such that @xmath470 $ ] , the relative modular operator @xmath471 is defined as @xmath472 , where @xcite @xmath473 from @xmath474 we have @xmath475 $ ] , and , analogously , @xmath476 $ ] , so , because @xmath477 , it preserves the dense domain . from @xmath478 we have that @xmath479=0 $ ] on the dense domain . since @xmath480 is bounded , it follows that @xmath471 preserves the decomposition @xmath481 . hence , @xmath471 is block diagonal . the same holds for @xmath482 . thus , @xmath483 is also block diagonal , @xmath484 from @xmath485 , and we obtain @xmath486 hence , by lemma [ gen.tracelog.triangle.lemma ] , triangle equality holds if @xmath439 . it remains to check whether this condition is satisfied . araki and masuda @xcite and donald @xcite prove this relation , but under stronger conditions ( araki and masuda assume that @xmath286 is faithful , while donald assumes also faithfulness of @xmath331 ) , so we need to provide more general proof . let @xmath487 $ ] , and consider operators @xmath488 , @xmath489 and @xmath490 given by @xmath491 then @xmath492 where @xmath493 , so @xmath490 is bounded . from @xmath494 , @xmath495 , and @xmath496 we obtain @xmath497 where @xmath498 is bounded . so @xmath499 araki @xcite proved that @xmath500 and @xmath501 . hence @xmath502 on the other hand , @xmath405 so , for every positive operator @xmath19 on a hilbert space @xmath133 , @xmath503 it follows that @xmath439 . [ gen.lemma.pipj ] given @xmath504 , let @xmath505)=0\;\forall x\in\n\}$ ] for @xmath410 , and @xmath411=0 $ ] . if @xmath412 , then @xmath413 from proposition [ gen.small.theorem ] it follows that @xmath506 denote the right hand side of by @xmath507 . we need to check that @xmath508)=0 $ ] @xmath455 @xmath398 @xmath509)=0 $ ] @xmath455 . but this follows from @xmath510)&=\psi(p_i[p_j , x]p_i+(\ii - p_i)[p_j , x](\ii - p_i))\nonumber\\ & = \psi([p_j , p_ixp_i+(\ii - p_i)x(\ii - p_i)])\nonumber\\ & = 0.\end{aligned}\ ] ] [ gen.vonneumann.lueders.from.entropy.theorem ] if @xmath417 satisfies @xmath278 @xmath304 , @xmath422 , and @xmath511)=0\;\forall x\in\n\;\forall i\in\{1,\ldots , n\}\},\ ] ] then @xmath512 the same as for theorem [ vonneumann.lueders.from.entropy.theorem ] , with the first step of induction satisfied by proposition [ gen.small.theorem ] , and using lemma [ gen.lemma.pipj ] instead of lemma [ lemma.pipj ] . [ weighted.vnlu.theorem ] let @xmath82 and @xmath513 . if @xmath514 , @xmath417 satisfy @xmath278 @xmath304 , @xmath422 , and @xmath515 @xmath30 , if @xmath516 satisfy @xmath517 , and @xmath518 then @xmath519 let @xmath520 , @xmath521 , and @xmath522 , @xmath523 @xmath30 . then , using the fact that each function of block diagonal matrices is block diagonal , we obtain @xmath524 using @xmath525 , we obtain @xmath526 where @xmath527 , @xmath528 , and equality is attained iff @xmath529 . hence , @xmath530 from the condition for equality in , it follows that @xmath286 is the unique minimiser of @xmath531 . [ remark.strong.vnlu ] the strong lders rule can be obtained from minimisation of @xmath322 by two different methods . first method amounts to applying quantum jeffrey s rule and taking the limit @xmath532 , @xmath533 note that is a weakly continuous function of @xmath534 . the limit is also weakly continuous . hence , the strong lders rule can be considered as a weakly continuous limit of an entropic projection @xmath535 . despite the observation carried in remark [ strongluedersfromqjr.remark ] , the direct derivation of along the lines of theorem [ weighted.vnlu.theorem ] with the initial constraints @xmath536 is impossible , because in such case the necessary assumptions for this theorem do not hold . more precisely , the states @xmath162 that satisfy @xmath537 do not exist if @xmath538 . this follows from @xmath539 this situation can be improved by ` regularisation ' of the difference of two distance functionals , using connes cocycle with respect to some well - behaved ` reference ' functional @xmath540 . the natural choice in the case of strong lders rule is @xmath541 . ( donald @xcite also introduces a distance functional that is dependent on the choice of a subset of @xmath74-algebra , but his motivation as well as the resulting mathematical construction differ from ours . ) hence , in order to show that the strong lders rule is an entropic projection without any limits involved , we will use an analogue of a theorem [ gen.vonneumann.lueders.from.entropy.theorem ] for a single projection @xmath480 and a _ regularised _ distance functional @xmath542 , defined as a restriction of @xmath321 to a subspace @xmath543 . @xmath544 is a distance functional on @xmath545 , that is , @xmath546 follows directly from the definition and properties of @xmath547 . [ p.omega.lemma ] if @xmath548 then @xmath549=0 $ ] and @xmath550 . we need to check that @xmath551 , which is equivalent to @xmath552 @xmath455 . but this follows from @xmath553 similarly @xmath554 @xmath455 . [ regularised.mre.vn.theorem ] if @xmath555 , and @xmath514 satisfies @xmath556 , then @xmath557 follows from theorem [ weighted.vnlu.theorem ] , applied to @xmath545 , if we notice that @xmath558 which follows from lemma [ p.omega.lemma ] . in the hilbert space based quantum mechanics a of @xmath559 such that @xmath560 is defined as @xmath561 . \label{trans.prob}\ ] ] different generalisations of the notion of transition probability to the case of quantum states over @xmath0-algebras are possible . the two most important are : the @xcite @xmath562 where @xmath190 is a standard representation of @xmath1 , and the @xcite @xmath563 where @xmath564 is defined by @xmath565 @xmath455 for some representation @xmath132 of @xmath1 , and the supremum varies over all possible representations . for the comparison of with and with some other possibilities , see @xcite . from the geometric perspective it is worth noting that is bijectively related to the distance on @xmath566 defined by the norm of @xmath567 , @xmath568 as well as to the @xmath274 distance on @xmath566 , @xmath569 while is bijectively related to the on @xmath566 @xcite ( cf . also @xcite ) defined by @xmath570 where @xmath571 is defined as above , and @xmath572 varies over the same range as @xmath573 in . in the notation above we have used freely the unitary equivalence between any standard representation @xmath190 of a @xmath0-algebra @xmath1 and its canonical representation @xmath574 , see @xcite . both @xmath575 and @xmath576 are metrical distances . herbut @xcite proved that @xmath577 for @xmath82 , @xmath578 , and @xmath579 given by . this derivation of the weak lders rule was the first result of this type in the literature . minimisation of the same distance function , but under constraints of the type @xmath580 with @xmath311 and @xmath581 , was later considered in @xcite , however with no general results . let @xmath582 be the @xmath0-subalgebras of @xmath138 , let @xmath583 be invertible with @xmath584 , and let @xmath585 . marchand and collaborators @xcite considered a quantum inference problem based on @xmath586 with @xmath587 . the algebra @xmath588 is interpreted as representing operators subjected to a `` partial measurement '' , while @xmath589 is a noncommutative analogue of the radon nikodm quotient that follows from sakai s theorem @xcite and can be thought of as a generalisation of a projection .- algebras , see @xcite . for a generalisation that does not require a subset @xmath588 to be a @xmath0-algebra , see @xcite . ] they derived in @xcite several different `` post - measurement '' states , dependent on the choice of @xmath588 , @xmath589 , and initial correlations in @xmath331 . it was shown by raggio @xcite that the strong lders rule can be directly derived as a special case of constrained _ maximisation _ of the cantoni uhlmann transition probability . let @xmath590 be a convex subset of a real topological vector space @xmath226 . a subset @xmath591 is called a iff @xmath592 let @xmath1 be a @xmath0-algebra , and let @xmath593 be a closed , convex subset of @xmath594 such that @xmath595\;\;\limp\;\;\omega,\phi\in k\right)\;\;\forall\omega,\phi\in\n^+_{\star1}.\ ] ] such set is a face in @xmath594 . for each face @xmath596 there exists a unique @xmath126 such that @xmath597 . if @xmath598 is a face in @xmath594 with a corresponding @xmath126 , @xmath585 , @xmath82 , and @xmath599 , then @xcite @xmath600 this corresponds to domotor s observation @xcite that the faces in @xmath88 form the correct constraints for the bayes laplace rule . on the other hand , gives us @xmath601 from the fact @xcite that @xmath602 whenever @xmath1 is commutative , or @xmath331 is pure , or @xmath603 , or @xmath604 for @xmath605 , we can conclude that in any of these cases @xmath606 thus , under the above conditions , the strong lders rule can be derived as a result of constrained minimisation of @xmath607 . however , these conditions are so restrictive ( eliminating e.g. nonpure density operators ) that it is more proper to say that the equation expresses a derivation of the strong von neumann rule . raggio @xcite showed also that , for a general @xmath331 and @xmath593 the same as in , leads to a different result than . this is also the case for a general @xmath331 and @xmath608 . nevertheless , we can prove the following : for @xmath609 and @xmath579 given by , @xmath610 follows directly from and @xmath611 another result was obtained by hadjisavvas @xcite , who showed that the strong von neumann rule for pure @xmath612 can be derived as @xmath613 where @xmath614 is the metrical @xcite on @xmath615 , @xmath616 it is worth noticing that the original definition @xcite of @xmath614 was provided over the measures on orthonormal orthomodular lattice @xmath617 , where it takes a form @xmath618 while the distance minimised in @xcite can be represented in a form @xmath619 where @xmath6 is a boolean algebra . in this sense , hadjisavvas derivation of the strong von neumann rule from minimisation of @xmath614 is similar to burris derivation of jeffrey s rule as a ( nonunique ) minimiser of ( reported in @xcite . ] ) . in @xcite diu showed that , when applied to nonpure states @xmath331 , @xmath620 does not lead to the strong lders rule ( in @xcite he extended this result to a more general family of metrical distances on quantum states ) . in our opinion , diu s result can not be used as a general argument against using constrained minimisation of @xmath614 , or any other information distance , because strong lders rule is not a uniquely `` correct '' quantum state change rule . however , on the positive side , diu s result and the similar result by raggio on @xmath621 for nonpure @xmath331 exemplify that the choice of an information distance subjected to minimisation preselects the type of possible constraints and the class of admissible results . in this sense , the choice of an information distance ( metrical or not ) amounts to the choice of a specific convention of inference , which in turn determines some range of possible forms of information dynamics and their output states . the information state changes conditioned on certainties ( yes / no truth values corresponding to elements of a boolean algebra or orthonormal lattice of projections ) are definitely not the only way , and also not the most useful way , of defining information dynamics of information states , both statistical and quantum . the possibility of a derivation of the bayes laplace , jeffrey s , lders , and quantum jeffrey s rules from minimisation of different distance functionals subjected to various constraints shows the explanatory ( semantic ) strength of the approach based on entropic projections . while _ in principle _ any quantum state change rule ( such as luders rule , quantum jeffrey s rule , or some quantum channel ) is an _ ad hoc _ postulate , not derived from any other , more fundamental , property of quantum theoretic formalism , _ in practice _ the choice of a particular form of this rule reflects certain assumptions about the relationship between the knowledge about outcomes of experimental procedure and the contents of a quantum theoretical model . for example , weak lders rule assumes the specific type of knowledge about the `` measurement result '' , requiring specification of the choice of the projection operators @xmath622 . this restricts the allowed form of ` experimental evidence ' to nonempty convex closed linear subspaces of a commutative @xmath623 space . a virtue of the approach based on quantum entropic projections is that it allows for a vast generalisation beyond the above restriction , while keeping clear underlying conceptual principles as well as strong mathematical useability . each information distance functional @xmath237 expresses a choice of a specific convention on the preferred / relevant and unpreferred / irrelevant aspects of information states : the relevant aspects are those that more strongly participate in the values of @xmath237 ( one can think of a conventional character of a least squares distance , which expresses certain arbitrary preferences regarding the information content of the data ) . to every choice of an information distance there corresponds a preferred type of constraint ( preferred geometric form of ` experimental evidence ' ) for which this information distance is capable of obtaining a unique minimum . the general way to introduce constraints is then to provide a mapping @xmath624 from the space @xmath625 of ( ` epistemic'/`experimental'/`registration ' ) parameters describing the sets of `` possible outcomes '' into nonempty convex closed subsets of @xmath626 . this way , as opposed to the bayes laplace rule and lders rules , in our framework the conditioning is provided not upon the abstract ` event ' that belongs to a boolean algebra or to an orthomodular lattice , respectively , but upon the value taken in the space @xmath625 of parameters describing the `` possible outcomes '' . this is similar to the semi - spectral ( povm / cp - map ) approach , which uses linear povm - instruments conditioned upon the spaces of `` possible outcomes '' that belong to @xmath627 . this difference ( conditioning upon ` quantitative information ' as opposed to conditioning upon ` event ' ) is the key insight . note that in the semi - spectral approach there is also no bijection between the space of ` abstract events ' and the space of `` possible outcomes '' : a single effect can correspond to various elements @xmath628 @xcite : the ` eigenstate - eigenvalue link ' @xcite breaks down here . however , as opposed to the semi - spectral approach , our approach completely detaches from the reliance on the use of spectral theory in foundations of quantum theory , allowing for more flexible operational specification of the ` experimental evidence ' , and for deriving `` quantum measurement '' rules from a single underlying principle ( entropic projection ) , which is alternative to quantum channels . in particular , the use of quantum distance as an underlying mathematical structure allows for geometric analysis and justification of the choice of a specific convention of quantum inference / dynamics . this is in contrast with the general lack of clear geometric justification for a choice of a specific quantum channel in the semi - spectral approach . an extended development of the approach based on quantum entropic projections as an alternative to povm - instruments , and as a replacement for lders rules , will be carried out in @xcite . for an alternative derivation of lders rules from quantum entropic projections based on @xmath322 , see @xcite . what does it all mean for quantum bayesianism ? quite often ` bayesianism ' is understood as a subjective interpretation of probability equipped with the requirement of using the bayes laplace rule for the purpose of changing probabilities due to learning new information . however , the appearance of other updating rules , such as jeffrey s rule , field s rule @xcite , and constrained maximisation of the wgkl distance @xcite , has undermined the universality of a second component . the usual perspective on the meaning of ` quantum bayesianism ' starts from the semi - spectral approach to mathematical foundations of quantum theory , and aims at recasting ( some suitable class of ) povm - instruments as ( a modified form of ) the bayes laplace rule , while keeping the subjective interpretation of probabilities . such perspective assumes that probability theory and spectral theory should be fundamental constitutents of quantum theory . moreover , it also does not provide justification for using povm - instruments : the mathematical foundations are just taken for granted . the results contained in this paper are intended to serve as a guideline ( via quantum jeffrey s rule ) and as a backwards compatibility proof ( via recovery of lders rules ) for an alternative approach to the mathematical and conceptual meaning of ` quantum bayesianism ' . according to our point of view , quantum states should be used as carriers of intersubjective knowledge on their own mathematical right ( as elements of a noncommutative @xmath629 space ) , without reference to probabilities ( elements of a commutative @xmath629 space ) , while the processes of inductive inference ( information dynamics ) can be fruitfully modelled using quantum entropic projections as a nonlinear alternative to the bayes laplace rule and linear povm - instruments . see @xcite for a detailed account . i would like to thank wojciech kamiski for numerous discussions , comments , and insights that have strongly shaped this work . i thank also carlos guedes , frank hellmann , and stanisaw woronowicz for discussions at the early stage of this work , stanley burris and nicolas hadjisavvas for correspondence , patrick coles for informing about @xcite , and bianca dittrich for a partial support . this research was supported in part by perimeter institute for theoretical physics . research at perimeter institute is supported by the government of canada through industry canada and by the province of ontario through the ministry of research and innovation . this research was also partially financed by the national science center of the republic of poland ( narodowe centrum nauki ) through the grant number n n202 343640 .

by applying a magnetic field whose zeeman energy exceeds the kondo energy by an order of magnitude the ground state of the friedel - anderson impurity is a magnetic state . in recent years the author introduced the friedel artificially inserted resonance ( fair ) method to investigate impurity properties . within this fair approach the magnetic ground state is derived . its full excitation spectrum and the composition of the excitations is calculated and numerically evaluated . from the excitation spectrum the electron density of states is calculated . majority and minority d - resonances are obtained . the width of the resonances is about twice as wide as the mean field theory predicts . this broadening is due to the fact that any change of the occupation of the d - state in one spin band changes the eigenstates in the opposite spin band and causes transitions in both spin bands . this broadening reduces the height of the resonance curve and therefore the density of states by a factor of two . this yields an intuitive understanding for a previous result of the fair approach that the critical value of the coulomb interaction for the formation of a magnetic moment is twice as large as the mean field theory predicts . pacs : 75.20.hr , 71.23.an , 71.27.+a

0902.0033

the properties of magnetic impurities in a metal is one of the most intensively studied problems in solid state physics . the work of friedel @xcite and anderson @xcite laid the foundation to understand why some transition - metal impurities form a local magnetic moment while others do nt . kondo @xcite showed that multiple scattering of conduction electrons by a magnetic impurity yields a divergent contribution to the resistance in perturbation theory . yoshida @xcite introduced the concept that the ( spin 1/2 ) magnetic impurity forms a singlet state with the conduction electrons and is non - magnetic at zero temperature . these new insights stimulated a large body of theoretical and experimental work ( see for example @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) . the majority of experimental and theoretical work has focussed on the singlet kondo ground state . however , the `` magnetic state '' of the impurity is of equal or even greater importance because magnetic impurities are always present , including in micro - chips and nanostructures , and influence the thermodynamic and transport properties of the hosts . since many experiments and almost all technical applications are * not * performed at low temperatures the magnetic impurities are generally far above their kondo temperature @xmath0 and show their full magnetic behavior . the theoretical investigation of the magnetic state has been explored in much less detail than the kondo ground state for spin @xmath1 impurities . in many cases the kondo temperature is very low , in the range of liquid helium temperature . in this case the impurity is in the magnetic state at relatively low temperature . ( the word impurity is in this paper reserved to impurities which possess - at sufficiently high temperature - a magnetic moment ) . when the temperature is several times the kondo temperature one is sufficiently above @xmath0 to destroy the kondo ground state . on the other hand one may expect that the properties of the magnetic state are not yet influenced by the thermal excitations due to the finite temperature . therefore a number of theoretical investigations treat the magnetic state at zero temperature , i.e. as a magnetic ground state . this approach is probably justified but it leaves the work always vulnerable to the criticism that there is no magnetic moment at zero temperature . therefore in this paper i prefer to use the effect of a magnetic field on the kondo state . a magnetic field which is an order of magnitude larger than @xmath2 ( @xmath3=bohr magneton ) destroys the kondo singlet state as well and yields the magnetic state . its side effects are that it changes the energy of the d - states by @xmath4 and shifts the conduction bands by @xmath4 . the latter yields the pauli susceptibility but has otherwise only a negligible effect on the interaction between the impurity and the conduction electrons because the fermi level for spin - up and down electrons readjusts to the same height ( as before ) . friedel @xcite and anderson @xcite derived a criterion for the instability of the paramagnetic state , i.e. the formation of a magnetic moment : take the density of states @xmath5 of the d - resonance at the fermi energy ( in the paramagnetic state ) and multiply it by the coulomb repulsion energy @xmath6 . if the product @xmath7 then a magnetic moment is formed . within mean field theory the d - density of states is given by a lorentz function@xmath8 where @xmath9 is an effective energy of the d - electrons in the spin - up or down state , @xmath10 while @xmath11 is the average occupation of the d - electron with the opposite spin ) and the resonance width @xmath12 is given in mean field theory by@xmath13 here @xmath14 is the s - d - hopping matrix element between a conduction electron and the d - state at the impurity and @xmath15 is the density of states of the conduction electrons . in the mean field theory an occupied @xmath16 electron state can only make transitions into @xmath17-states . ( throughout this paper i express electron states by their creation operators ) . it is well known that the mean field theory has a number of shortcomings . during the last few years the group of the author has developed a new approach to the impurity problem , in particular the friedel - anderson and the kondo impurity . in this * fair * method a * * f**riedel state is * * a**rtificially built from each conduction band and * * i**nserted as a * * r**esonance state into the conduction or s - band of spin - up and spin - down electrons . in the appendix a short review of the fair solution for the friedel impurity is sketched . the fair solution for the magnetic state yields a considerably lower energy for the `` magnetic ground state '' and requires a much larger critical coulomb interaction to form a magnetic state . this is of some practical importance since the mean field approximation is used in a number of numerical spin - density functional theory calculations for the magnetic moment of impurities in an ( s , p ) metal host @xcite , @xcite , @xcite , @xcite , @xcite . in addition to the size of the magnetic moment one would like to know the density of states in the magnetic state . the answer of the mean field theory has been discussed above . but there have been a number of suggestions that the d - resonance is broader than the mean field suggests ( see for example logan @xcite ) . the mean field theory decouples the spin - up d - electron from the spin - down d - electron , but in reality the d - electrons are coupled through the coulomb energy . a transition in the @xmath16 electron state changes the energy and the state of the @xmath18 electron as well . therefore it has been suggested in the past that the d - resonances in the friedel - anderson impurity are larger than the mean field theory predicts . a wider d - resonance in the friedel - anderson impurity together with the condition @xmath7 would require a larger coulomb energy to form a magnetic moment . in this connection the previous result of the author that the fair solution requires a ( two times ) larger coulomb energy to form a magnetic moment would find a simple physical interpretation . it is the goal of this paper to calculate the density of states of the `` magnetic ground state '' in the fair solution and compare it with the mean field density of states . in section ii the theoretical background of the magnetic state of the friedel - anderson impurity is sketched . in section iii electrons and holes are introduced into the magnetic ground state . their interactions and excitation energies are derived . in section iv the results of the numerical calculations are presented . finally in section v and vi the results are discussed together with the conclusion . in the appendix a the basic idea of the fair method is sketched . the simplified hamiltonian for a magnetic impurity is generally described by the friedel - anderson ( fa ) hamiltonian@xmath20\right\ } + un_{d\uparrow } n_{d\downarrow } \label{h_fa}\ ] ] here the operators @xmath21 represent s - electrons , i.e. the conduction band . in the friedel - anderson hamiltonian in equ . ( [ h_fa ] ) the d - state for each spin interacts with every electron in the conduction band . imagine how much easier the task would be if the d - electron would interact only with a single electron state ( in each spin band ) . all other conduction band states would represent just a background or quasi - vacuum . this is the fair approach . during the last few years the author introduced such a solution to the friedel - anderson impurity problem in which only four electron states , the spin - up and spin - down d - states @xmath16 and @xmath18 and two fair states , @xmath22 and @xmath23 interact through the coulomb and s - d - hopping potential . these states @xmath22 and @xmath24 are composed of the spin - up and spin - down conduction band states . they are the friedel artificially inserted resonance states or fair states . the interaction of the remaining conduction electron states with the d - states is insignificant ; they just yield a background . this yields very good ground - state properties . the fair states are composed of the corresponding conduction bands@xmath25{ccc}a_{0\uparrow}={\textstyle\sum_{\nu=0}^{n-1 } } \alpha_{0}^{\nu}c_{\nu\uparrow } & & b_{0\downarrow}={\textstyle\sum_{\nu=0}^{n-1 } } \beta_{0}^{\nu}c_{\nu\downarrow}\end{array}\ ] ] the remaining @xmath26 states in each spin band are constructed orthogonal to the corresponding fair state , orthonormal to each other and sub - diagonal with respect to the band energy hamiltonian @xmath27 this yields new bases for the conduction bands @xmath28 and @xmath29 with @xmath30 . these new bases are uniquely determined by the two fair states . within the new bases the fa - hamiltonian ( [ h_fa ] ) can be expressed as @xmath31 with@xmath32 where @xmath33 \label{h_0'}\]]@xmath34 + \sum_{i=1}^{n-1}v_{i}^{\left ( a\right ) sd}\left [ d_{\uparrow } ^{\dag}a_{i,\uparrow}+a_{i,\uparrow}^{\dag}d_{\uparrow}\right ] \label{h_1'}\ ] ] and the spin - down hamiltonians are obtained by replacing @xmath35 by @xmath36 and the @xmath37-states by @xmath38-states . the hamiltonian @xmath39 is diagonal in the band states @xmath40 and @xmath41 for @xmath42 . the only interaction takes place between the states @xmath43 , @xmath44 , @xmath45 and @xmath46 . i call these states the * nest states*. the ground state of the hamiltonian @xmath39 is straight forward . it consists of the coupled state between the nest states and a partially occupied spin - up and down band . i occupy each spin component with @xmath47 electrons , putting @xmath48 electrons into each conduction band states and one spin - up and one spin - down electron into the nest . this yields the magnetic ground state as described in equ . ( [ psi_ms ] ) . @xmath49 \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle \label{psi_ms}\ ] ] where @xmath50 represents a kind of quasi - vacuum ( @xmath51 ) . the calculation of the coefficients @xmath52 yields a secular hamiltonian @xmath53 which i call the nest - hamiltonian and which has the form @xmath54{cccc}e_{0}^{\left ( a\right ) } + e_{0}^{\left ( b\right ) } & v_{b}^{sd } & v_{a}^{sd } & 0\\ v_{b}^{sd } & e_{0}^{\left ( a\right ) } + e_{d } & 0 & v_{a}^{sd}\\ v_{a}^{sd } & 0 & e_{d}+e_{0}^{\left ( b\right ) } & v_{b}^{sd}\\ 0 & v_{a}^{sd } & v_{b}^{sd } & 2e_{d}+u \end{array } \right)\ ] ] here the abbreviations are used : @xmath55 and @xmath56 ( see equ.([h_0 ] ) ) . the superscript @xmath57 stands for nest and the subscript @xmath58 gives the number of nest electrons in the spin - up and spin - down state . ( the energy of the occupied band states is not included . it yields the same contribution to each component ) . @xmath53 has four eigenvalues and eigenstates . the lowest eigenvalue yields the ground state . the components of the ground state , nest plus the band states , are shown in fig.1 . the first order correction to the energy , i.e. the expectation value of @xmath59 is zero . but in addition the second order perturbation of @xmath59 is extremely small . this is demonstrated in appendix b. it may appear remarkable that the neglect of the interactions between the d - electron and all the band states @xmath60 and @xmath61 yields a realistic ground state . but it is not unheard off that one can obtain an excellent ground state while neglecting a major part of the interaction in the system . the bcs theory is a good example because it only includes the electron - phonon interaction between cooper pairs of time - reversed electrons . the interaction between all the other electrons is neglected although their number is much larger . one major part of the numerical calculation is , of course , the optimization of the two fair states @xmath43 and @xmath62 so that the expectation value of the energy @xmath63 of the hamiltonian @xmath39 has a minimum . the optimization procedure is described at length in previous papers @xcite , @xcite , @xcite and is taken for granted in this paper and will not be described here . ( the fair states are rotated in hilbert space ) . since we do nt count the fair states any more as band states the number of band states is reduced by one and their energy is slightly shifted ( by less than the original energy spacing ) . the band states enter in the energy @xmath64 only through the kinetic ( band ) energy of the occupied band states . in a way they just prepare the nest for the states @xmath65 .$ ] @xmath66{fig1.eps}}\\ & \begin{tabular } [ c]{l}fig.1 : the composition of the magnetic state $ \psi_{ms}$ in the nest is shown.\\ it consists of four slater states . each slater state has a half - full spin - up and\\ down band , two fair states ( circles in within the bands ) and two d - states\\ ( circles on the left and right of the band ) . full black circles represent\\ occupied states and light grey represent empty states . the band\ at the right\\ with the half - filled circles symbolizes the magnetic solution with four\\ slater states . \end{tabular}\end{aligned}\ ] ] in the numerical calculation we will present the results for two examples with the parameters @xmath67 , @xmath68 and @xmath69 and @xmath70 . the smaller value of the s - d - matrix element permits a better fit of the resulting resonance curve with a lorentz curve since the effect of the finite width of the band is smaller . in the construction of the magnetic ground state @xmath71 the hamiltonian @xmath59 has been completely neglected . below we will derive the excitation energies by introducing an additional electron ( hole ) into an empty ( occupied ) states . for this calculation it is important to know whether the empty state is really empty or whether transitions from the ground state into the state due to @xmath59 have partially occupied this state . ( this problem is well known from the calculation of the electron - phonon mass enhancement . in the calculation of the electron - phonon self - energy one injects an electron into an `` empty state '' @xmath72 above the fermi energy . the transitions of this electron via the electron - phonon interaction into other empty states @xmath73 contribute to the self - energy @xmath74 . however , the state @xmath72 was not really empty because transitions from the ground state into @xmath72 already created a finite occupation of @xmath72 . one has to correct the self - energy due to these processes ) . in the appendix i show that transitions from the ground state into empty band states ( due to @xmath59 ) are practically zero . the interference between transitions from the d - state and from the corresponding fair state almost perfectly cancel each other . the total weight in all perturbation states is only of the order of @xmath75 and can be completely neglected . therefore the band states are either completely empty or fully occupied . in fig.2 a spin - up electron is injected into one of the empty states @xmath76 of the @xmath77-band . this yields the slater states ( @xmath78 ) . the slater state ( @xmath79 ) is obtained by injecting an electron into the nest , either into the state @xmath22 or @xmath16 . @xmath80{fig2.eps}}\\ & \begin{tabular } [ c]{l}fig.2 : an electron has been injected into the spin - up band and the spin up nest.\\ this induces electron or hole transitions in the spin - down band . the resulting\\ slater states are shown as ( $ \gamma$ ) and ( $ \delta$ ) . ( each band with half circles consists\\ of two slater states . ) \end{tabular}\end{aligned}\ ] ] the injection into @xmath22 or @xmath16 yields@xmath81{lll}$a_{0\uparrow}^{\dag}\psi_{ms}=\left ( a_{d , b}a_{0}^{\dag}d_{\uparrow}^{\dag } b_{0\downarrow}^{\dag}+a_{d , d}a_{0}^{\dag}d_{\uparrow}^{\dag}d_{\downarrow } ^{\dag}\right ) \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle $ & for & $ a_{0}^{\dag}$\\ $ d_{\uparrow}^{\dag}\psi_{ms}=-\left ( a_{a , b}a_{0\uparrow}^{\dag}d_{\uparrow } ^{\dag}b_{0\downarrow}^{\dag}+a_{a , d}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } d_{\downarrow}^{\dag}\right ) \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & for & $ d_{\uparrow}^{\dag}$\end{tabular } \ \ \ \ \label{a_psi_ms}\ ] ] both final states yield a double occupancy of the spin up nest states . furthermore these states are not eigenstates of the nest . with respect to the ( basis ) states @xmath82 and @xmath83 the nest hamiltonian takes the form@xmath84{cc}\left ( e_{d}+e_{0}^{\left ( a\right ) } \right ) + e_{0}^{\left ( b\right ) } & v_{b}^{sd}\\ v_{b}^{sd } & \left ( e_{d}+e_{0}^{\left ( a\right ) } \right ) + e_{d}+u \end{array } \right ) \label{hnst_3}\ ] ] by diagonalization one obtains the eigenstates @xmath85 and @xmath86 with @xmath87 the states ( @xmath78 ) and ( @xmath79 ) in fig.2 are the initial states which one obtains through injection of a spin - up electron into the ground state . due to the perturbation hamiltonian @xmath59 these states interact with each other and ( @xmath79 ) interacts with the states ( @xmath88 ) and ( @xmath89 ) . in table i the possible states which can be obtained through the injection of a spin - up electron plus linear coupling through @xmath59 are collected . these states are ( @xmath78 ) the two - electron nest ground state plus one electron , ( @xmath79 ) a nest with two spin - up and one spin - down electron , ( @xmath88 ) a full spin - up and empty spin - down nest plus one spin down electron and ( @xmath89 ) a full spin - up and full spin - down nest and one spin down hole . in table i these states , their number and their energies are listed . @xmath90{|l|l|l|l|}\hline $ \mathbf{\psi}_{f}$ & \textbf{number } & \textbf{energy } & \\\hline $ a_{j\uparrow}^{\dag}\psi_{ms}$ & $ n/2 $ & $ e_{j}^{\left ( a\right ) } $ & \\\hline $ \psi_{2/1}^{\left ( \alpha\right ) } $ & $ 2,\alpha=1,2 $ & $ e_{d}+e_{0}^{\left ( a\right ) } + e_{2/1}^{\alpha}-e_{00}$ & \\\hline $ a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}b_{j\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ n/2 $ & $ ed+e_{0}^{\left ( a\right ) } + e_{j}^{\left ( b\right ) } -e_{00}$ & \\\hline $ b_{k\downarrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}b_{0\downarrow}^{\dag } d_{\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ n/2 - 1 $ & $ \begin{array } [ c]{c}2e_{d}+u+e_{0}^{\left ( a\right ) } + e_{0}^{\left ( b\right ) } \\ -e_{k}^{\left ( b\right ) } -e_{00}\end{array } $ & \\\hline \end{tabular } \\ & \begin{tabular } [ c]{l}table i : this table describes the states of fig.2 which are generated\\ by the injection of one spin - up electron into the magnetic ground state\\ and transition from the resulting states through $ h_{1}^{\prime}$. the state $ \psi_{ms}$ is\\ given by equ . ( \ref{psi_ms } ) and $ \psi_{2/1}^{\left ( \alpha\right ) } $ is given by ( \ref{psi2/1 } ) . the\\ energy is measured from the ground - state energy $ e_{00}$. \end{tabular}\end{aligned}\ ] ] fig.2 and table i show all the spin - up electron excitations which interact linearly in @xmath59 . this is a total of @xmath91 states . it is straight forward to construct the secular matrix ( i.e. the excitation hamiltonian @xmath92 ) between the excitations in table i. one may put the two nest states @xmath93 and @xmath86 at the positions one and two , followed by the @xmath94 additional single particle excitations . we denote these @xmath91 states as @xmath95 . the diagonal of the hamiltonian is given by the energies in table i. the off - diagonal elements of @xmath92 are the matrix elements of @xmath59 between the states ( @xmath96 ) in fig.2 . the single particle excitations interact only with the first two nest states through @xmath59 but not among each other . ( in appendix b the corresponding matrix elements are shown in table iii for similar transitions from the ground state . ) this hamiltonian @xmath92 is diagonalized and yields a set of @xmath97 new eigenstates @xmath98 with eigenenergies @xmath99 . the components of the eigenstates @xmath98 in terms of @xmath95 are given as columns @xmath100 . the @xmath101th row of the matrix @xmath100 yields the amplitude of our @xmath101th original state @xmath95 in terms of the new eigenstates @xmath98 . if this @xmath101th original state is , for example , @xmath102 then it can be expressed in the new eigenstates as @xmath103 its density of states is then @xmath104 since electron injection creates only the states ( @xmath78 ) and ( @xmath79 ) one obtains the full ( spin - up ) excitation spectrum by summing over these @xmath105 states . the weight of states @xmath106 is one , however , the weight of @xmath107 is only @xmath108 since @xmath109 \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle \\ & = \left [ a_{d , b}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}b_{0\downarrow } ^{\dag}+a_{d , d}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}d_{\downarrow}^{\dag } \right ] \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle\end{aligned}\ ] ] a similar result is found for the weight of @xmath110 which is @xmath111 . since @xmath107 and @xmath110 are not eigenstatet of @xmath39 they represent a combination of the two eigenstates @xmath112 . in fig.3 is sketched what happens when an electron is injected into either the state @xmath113 or @xmath16 . the electron injection yields a superposition of the two eigenstates @xmath112 . from these states the electron can make a transition into any of the ( @xmath114 ) states via @xmath59 . the two transition amplitudes interfere in this transition . this interference has to be included in the calculation of the spectral weight density of @xmath107 and @xmath110 ( which requires just the scalar products between @xmath107 and @xmath115 ( or @xmath110 and @xmath115 ) . this is discussed in more detail in appendix c. @xmath116{fig3.eps}}\\ & \begin{tabular } [ c]{l}fig.3 : an electron has been injected into the $ a_{0\uparrow}^{\dag}$ state ( or $ d_{\uparrow}^{\dag}$ state,\\ dashed arrow ) . the resulting state is a superposition of two nest states.\\ from these nest states the electron makes ( as one possibility ) a transition\\ into the state $ a_{j\uparrow}^{\dag}$ where the two amplitudes interfere . the energies of the\\ two nest states are shown . the different thickness of the arrows shows\\ different probabilities for the two paths . \end{tabular}\end{aligned}\ ] ] for the full spectrum of excitations one has to include the injection of holes into the occupied states . this is shown in fig.4 . the hole can be injected into the occupied states @xmath76 yielding @xmath117 or into the nest . in the latter case the spin - up part of the nest is emptied . this yields for the secular matrix of the nest in analogy to equ . ( [ hnst_3])@xmath118{cc}e_{0}^{\left ( b\right ) } & v_{b}^{sd}\\ v_{b}^{sd } & e_{d}\end{array } \right ) \label{hnst_1}\]]@xmath119{fig4.eps}}\\ & \begin{tabular } [ c]{l}fig.4 : an hole has been injected into the spin - up band and the spin up nest.\\ this induces electron or hole transitions in the spin - down band . the resulting\\ slater states are shown as ( $ \gamma$ ) and ( $ \delta$ ) . \end{tabular}\end{aligned}\ ] ] @xmath90{|l|l|l|}\hline \textbf{state } & \textbf{number } & \textbf{energy}\\\hline $ a_{j\uparrow}\psi_{ms}$ & $ n/2 - 1 $ & $ -e_{j}^{\left ( a\right ) } = \left\vert e_{j}^{\left ( a\right ) } \right\vert $ \\\hline $ \psi_{\downarrow}^{\left ( \alpha\right ) } = \left ( b_{b}b_{0\downarrow } ^{\dag}+b_{d}d_{\downarrow}^{\dag}\right ) \left\vert \mathbf{0}_{a,\uparrow } \mathbf{0}_{b,\downarrow}\right\rangle $ & $ 2,\alpha=1,2 $ & $ e_{\downarrow } ^{\left ( \alpha\right ) } -e_{00}$\\\hline $ b_{j\downarrow}b_{0\downarrow}^{\dag}d_{\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ n/2 - 1 $ & $ e_{d}+e_{0}^{\left ( b\right ) } -e_{j}^{\left ( b\right ) } -e_{00}$\\\hline $ b_{j\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ n/2 $ & $ e_{j}^{\left ( b\right ) } -e_{00}$\\\hline \end{tabular } \\ & \begin{tabular } [ c]{l}table ii : this table describes the states of fig.4 which are generated\\ by the injection of one spin - up hole into the magnetic ground state\\ and transition from the resulting states through $ h_{1}^{\prime}$. the energy is\\ measured from the ground - state energy $ e_{00}$. \end{tabular}\end{aligned}\ ] ] the construction of the excitation or secular hamiltonian @xmath92 is in complete analogy to the electron injection . this time the number of excitations is @xmath120 . the spectrum is obtained in the same way as before . for the numerical calculation a conduction band with a finite number of states is used . we follow here wilson @xcite by using an s - electron band with constant density of states and the fermi level in the center , which we divide into energy cells @xmath121 . in each energy cell ( which may contain @xmath122 @xmath72-states @xmath123 we rearrange the states ( by an orthogonal transformation ) so that one state @xmath124 accumulates all the interaction with the d - states while the other @xmath125 states have zero interaction with the d - states . wilson normalized the energy in terms of the fermi energy so that his band extended from @xmath126 @xmath127to @xmath128 . wilson s logarithmic scale for the energy cells is not opportune for the present investigation because it is not fine enough at the energy of the d - resonance . therefore i use a linear sub - division of the energy band @xmath129 . for the majority of calculations the energy band is sub - divided into @xmath130 and @xmath131 energy cells . the state @xmath132 represents all the s - electron states in the cell @xmath133 and possesses the average energy @xmath134 ( corresponding to @xmath135 for @xmath136 ) . for @xmath137 energy cells with constant width of @xmath138 the s - d - matrix elements @xmath139 is given by @xmath140 so that @xmath141 . in the following i show the results for the friedel - anderson hamiltonian with the d - level energy @xmath142 the coulomb energy @xmath67 and an s - d - hopping matrix element of @xmath70 . the magnetic moment of this impurity is @xmath143 ( for @xmath144 ) . the calculations are performed with @xmath130 or @xmath131 energy levels of constant spacing . a second set of results is derived for the parameters @xmath68 , @xmath145 and @xmath146 . to assure that the magnetic state @xmath71 is indeed the ground state ( i.e. to prevent the formation of a singlet state ) a magnetic field @xmath147 is applied yielding a magnetic energy @xmath148 . this energy is chosen so that @xmath149 . i estimate the kondo energy from the difference between the energies of the singlet and the triplet state . this energy difference is about @xmath150 for @xmath146 and of the order of @xmath151 for @xmath70 . it turned out that the required magnetic energy is in both cases so small that it yields no noticeable changes . this is partly due to the fact that the absolute smallest band energies which are @xmath152 act as a finite temperature as wilson pointed out @xcite . for @xmath136 this corresponds to a temperature of @xmath153 which is a very large temperature compared with most kondo temperatures . but magnetic field is of academic importance to assure the magnetic state @xmath71 is the appropriate ground state . the use of equidistant energy levels is important to identify the resonance state within the electron bands . but it has the drawback that it does not describe well the behavior of the wave function at low energies . at low energies the logarithmic energy scale which wilson introduced would be more appropriate . but the evaluation of the density of states is much more difficult for a non - linear energy scale . for each spin band one obtains a spectrum with a total weight of @xmath154 , corresponding to @xmath137 s - electron states and one d - electron state . however , the number of energy levels is @xmath155 . ( this is the number of eigenstates of the excitation hamiltonians for electrons and holes together ) . this means that the weight at the individual energies is at least for @xmath156 energies much less than one . in fig.5 the spectral weight at different energies is shown in the energy range from @xmath126 to @xmath157 for the minority band . for negative energies the weight is either very close to one or very small . here one can calculate the density of states by the separation of the levels with weight close to one ( by dividing the weight by the level separation ) . the evaluation is more complicated for positive energies . however , here the sum of neighboring energy levels is close to one . then one can calculated the `` center of weight '' for two neighboring levels which have a total weight close to one and then proceed as before . it turns out that the best approach is to start from the lower and upper ends of the band in the evaluation . @xmath158{fig5.eps}}\\ & \begin{tabular } [ c]{l}fig.5 : the spectral weight for the different excitation energies . \end{tabular}\end{aligned}\ ] ] in fig.6a the density of states of the excitation spectrum for the majority spin is shown . the full circles are obtained with @xmath136 states and the stars use @xmath144 equally spaced wilson states . the full curve represents a lorentz curve with the parameters @xmath159 the resonance energy is is @xmath160 and the resonance ( half ) width is @xmath161 . the corresponding mean field resonance width is @xmath162 . @xmath163{fig6a.eps}}\\ & \begin{tabular } [ c]{l}fig.6a : the density of states for the majority spins . the full circles\\ are calculated with $ n=40 $ states and the stars use $ n=80 $ equally\\ spaced wilson states . the full curve represents a lorentz curve\\ with the resonance energy $ e_{r}=-0.53 $ and the width $ \gamma_{r}=0.08$.\\ this width is twice the mean - field value of $ \gamma_{mf}=0.039$. \end{tabular}\end{aligned}\ ] ] in fig.6b the density of states of the minority spin is drawn . again the full curve represents a lorentz curve with the resonance at @xmath164 and a resonance half - width of @xmath165 . @xmath166{fig6b.eps}}\\ & \begin{tabular } [ c]{l}fig.6b : the density of states for the minority spins . the full circles\\ are calculated with $ n=40 $ states and the stars use $ n=80 $ equally\\ spaced wilson states . the full curve represents a lorentz curve\\ with the resonance energy $ e_{r}=0.52 $ and the width $ \gamma_{r}=0.08$.\\ again this width is twice the mean - field value of $ \gamma_{mf}=0.039$. \end{tabular}\end{aligned}\ ] ] for comparison fig.7 shows the result of a similar calculation and evaluation for a simple friedel resonance , where the coulomb energy @xmath6 is set equal to zero . the evaluation yields a lorentz curve with @xmath167 and @xmath168 the s - d - matrix element is still given by @xmath70 . @xmath169{fig7.eps}}\\ & \begin{tabular } [ c]{l}fig.7 : the density of states for an impurity with $ u=0$. the full circles\\ and the stars are calculated with $ n=40 $ and $ n=80 $ states.\\ the full curve represents a lorentz curve with the resonance energy\\ $ e_{r}=-0.51 $ and the width $ \gamma_{r}=0.044$. \end{tabular}\end{aligned}\ ] ] alternatively we tried to obtain the density of states by broadening the @xmath89-shaped energy spectrum in fig.5 with a gaussian curve @xmath170 $ ] . this method worked quite well . the optimal density curve was obtained when @xmath171 was equal to the level distance @xmath138 . the width of the two resonances was essentially the same as in fig.6a , b and fig.8a , b . only the heights were slightly reduced . however , i prefer to use the other evaluation method in this paper so that there is no doubt that the broadening of the resonance width is a real physical effect and not due to an artificial broadening with a gaussian curve . in a second series the majority and minority density of states have been calculated for the parameters @xmath68 , @xmath145 , and @xmath146 . this impurity possesses a magnetic moment of @xmath172 . in this case the mean - field theory yields a resonance width of @xmath12=0.0@xmath173 . the best fit to the numerical results yields @xmath174 . it is again about twice the value of the mean field . ( the resonance curves are no longer perfectly symmetrical because of the finite width of the conduction band ) . the corresponding friedel density of states ( which is not shown here ) has a resonance with of 0.08 which is quite close to the theoretical value of @xmath175 . @xmath176{fig8a.eps}}\\ & \begin{tabular } [ c]{l}fig.8a : the density of states for the majority spins for $ \left\vert v_{sd}^{0}\right\vert ^{2}=0.05$\end{tabular}\end{aligned}\]]@xmath166{fig8b.eps}}\\ & \begin{tabular } [ c]{l}fig.8b : the density of states for the minority spins for $ \left\vert v_{sd}^{0}\right\vert ^{2}=0.05$\end{tabular}\end{aligned}\]]@xmath19 in the mean field approximation the magnetic state has two d - resonances at the energies @xmath177 . since in the symmetric case one has @xmath178 one finds @xmath179 since @xmath180 . so generally the resonances are closer to fermi energy than @xmath181 in our case for the parameters @xmath142 @xmath145 and @xmath70 the magnetic moment is within 1% @xmath182 ( in units of @xmath183 and one expects the resonance almost at @xmath184 . the resonance width in mean field is given by @xmath185 where @xmath186 is the density of states of the s - electrons at the fermi level . this yields for the above parameters @xmath187 . ( the matrix element @xmath188 and the density of states are normalized to the atomic volume as the sample volume ) . this is the same width that one expects for a friedel resonance with @xmath70 . indeed in fig.7 the friedel resonance has a @xmath189-value of @xmath190 . this agrees within 10% or @xmath191 with the numerical result . the important result is that the resonance width of the minority and majority spins is larger than the friedel resonance width by a factor of two . therefore the obtained resonance width is also twice the mean field resonance width . this suggests that the any calculation which uses mean field yields an incorrect density of states . it will be interesting to check how the spin - density functional theory is affected by this result because the latter uses the mean field approximation . since the coulomb interaction broadens the d - resonance by a factor of two it also reduces the height of the resonance by the same factor of two . therefore it is very plausible that the mean field theory overestimates the tendency to form a magnetic moment . if the criterion for the formation of a magnetic moment , @xmath192 , is accepted then one expects that the critical coulomb energy for the formation of a magnetic moment is increased by a factor of two . this was the previous result by the author @xcite . in the density of states of the majority and minority spins in fig.6ab and fig.8a , b one observes a scattering and a small maximum at zero energy . this is probably due to the fact that i used a constant cell width for the wilson states . this means that i average over all states within an energy cell of the width @xmath193 . this is definitely a poor approximation for the two energy cells @xmath194 and @xmath195 ( which touch the fermi level ) . wilson avoids this problem by using a logarithmic energy scale . however , the present method to evaluate the density of states does not work for an energy - dependent cell width . details of this question will have to be clarified in the future . in this paper the density of states of the friedel - anderson impurity is calculated in the magnetic ground state . the magnetic ground state is enforced by the application of a magnetic field whose zeeman energy is an order of magnitude larger than the kondo energy . ( for the parameters chosen in the numerical calculation the effect of the magnetic field is so small that it can be neglected ) . the fair ground state is the eigenstate of a hamiltonian @xmath39 . to construct this hamiltonian two artificial friedel resonance states @xmath43 and @xmath62 are reverse engineered out of the original spin - up and spin - down conduction bands . there is a numerical procedure to optimize these two fair states . when this is done one has an extremely simple hamiltonian @xmath39 . in @xmath39 only the two s - states @xmath43 and @xmath62 interact with the d - state . these states @xmath196 @xmath62 , @xmath16 and @xmath18 are called the nest states . @xmath39 is diagonal in the ( modified ) conduction band electrons @xmath197 and @xmath61 . a perturbation hamiltonian @xmath198 which is the difference between the original fa - hamiltonian and @xmath39 , has zero energy expectation value in the fair ground state . in addition it is shown in second order self - consistent perturbation theory ( see appendix c ) that the total occupation of all perturbation states is only of the order of @xmath75 , i.e. the fair ground state has still an amplitude of @xmath199 . this is important when an electron or hole is injected into the ground state . the excitation spectrum is obtained by injecting an electron or a hole into the ground state . the resulting excited states interact via the perturbation hamiltonian @xmath59 and yield a spectrum of energy resonances . it turns out that the injection of an electron into the spin - up conduction band creates also transition between the spin - down conduction band and the nest states . the resulting density of states possesses the shape of a resonance curve . however , the resonance width is about twice the value of the mean - field theory . as a consequence the height of the resonance density of states is reduced by a factor of two . since the formation of a magnetic moment depends on the product of the coulomb interaction and the density of d - states , one would expect that the mean field overestimates the tendency towards a magnetic moment . indeed i observed in the first paper about the magnetic ground state that the formation of a magnetic moment requires about twice the coulomb energy that the mean field theory predicts . this consorts well with the present finding of the reduced resonance density of states . the next step in the future investigation is the calculation of the density of states of the kondo resonance within the fair model . for this calculation one has to use a wilson spectrum with a logarithmic energy scale . @xmath19 the basic idea of the fair method can be best explained for a friedel resonance with the hamiltonian @xmath200\ ] ] this is done in the following steps : 1 . from the free ( or s- ) electron basis an artificial friedel resonance ( * fair ) * state is constructed with @xmath201 together with a full orthonormal basis @xmath202 so that the free electron - electron hamiltonian @xmath203 takes the form @xmath204\ ] ] the requirement that the matrix elements between different @xmath205 and @xmath206 @xmath207 vanish has the consequence that a given fair state @xmath208 determines uniquely the full basis @xmath209 . in the new basis the total friedel hamiltonian takes the form@xmath210 where @xmath211\ ] ] the perturbation hamiltonian has the form@xmath212 + \sum_{i=1}^{n-1}v_{i}^{\left ( a\right ) sd}\left [ d^{\dag}a_{i}+a_{i}^{\dag}d\right ] \right\}\ ] ] here the new matrix elements are given as @xmath213 and @xmath214 . a trial state @xmath215 is defined as@xmath216 \left\vert \mathbf{0}_{a}\right\rangle \label{psi_f}\ ] ] where @xmath217 @xmath51 . the right side is abbreviated as@xmath218 where , for example , for @xmath219 one has @xmath220 3 . the energy expectation value @xmath221 of the @xmath39 with respect to the trial state @xmath215 is calculated . the contribution of @xmath59 is not included . this results in a @xmath222 secular matrix @xmath223{cc}{\textstyle\sum_{i=1}^{n-1 } } e_{i}+e_{0}^{\left ( a\right ) } & v_{0}^{\left ( a\right ) sd}\\ v_{0}^{\left ( a\right ) sd } & { \textstyle\sum_{i=1}^{n-1 } } e_{i}+e_{d}\end{array } \right)\ ] ] whose lowest eigenvalue yields @xmath64 and the corresponding eigenvector yields the coefficients @xmath224 . 4 . the fair state is rotated ( by variation ) in the @xmath137-dimensional hilbert space until the lowest eigenvalue of the secular matrix reaches a minimum . it has been shown by the author @xcite , @xcite that this procedure results in the * exact * @xmath225-particle ground state of a friedel hamiltonian ( given by ( [ psi_f ] ) ) . the matrix elements of the perturbation hamiltonian @xmath59 between this ground state and any excited state vanish . it is interesting to look at the result in some more detail . * the states @xmath37 @xmath226 enter the secular matrix only through the total energy of the occupied states @xmath227 and contribute only the the background energy . * the coefficients and the relative weight of the states @xmath208 and @xmath228 are only determined by the energies of the fair and the d - state @xmath229 and their coupling @xmath230 . in a way one can say that the states @xmath37 prepare just the background - a kind of nest - for @xmath208 and @xmath228 . the secular matrix represents an effective hamiltonian for these two states in the nest . in the following i will call the secular matrix without the kinetic energy @xmath227 the * nest hamiltonian*.@xmath231{cc}e_{0}^{\left ( a\right ) } & v_{0}^{\left ( a\right ) sd}\\ v_{0}^{\left ( a\right ) sd } & e_{d}\end{array } \right)\ ] ] the state @xmath208 represents an artificially inserted friedel resonance state . therefore i call @xmath208 a `` * * f**riedel * * a**rtificially * * i**nserted * * r**esonance '' state or * fair*-state . the use of the fair - states is at the heart of my approach to the fa- and kondo impurity problem . therefore i call this approach the * fair * method . the hamiltonian of the friedel - anderson impurity is given in equ . ( [ h_fa ] ) . one obtains the mean - field hamiltonian from equ.([hfa0 ] ) by replacing @xmath232 = @xmath233@xmath234 @xmath235 @xmath236 . after adjusting @xmath237 and @xmath238 self - consistently one obtains two friedel resonance hamiltonians with a spin - dependent energy of the @xmath239-state : @xmath9 @xmath240 . @xmath241\right\}\ ] ] the mean - field wave function is a product of two friedel ground states for spin up and down @xmath242 . now we express each friedel ground state @xmath243 by the fair solution , for example@xmath244 for the two friedel states in the mean - field wave function i use the form of equ . ( [ psi_f ] ) and obtain for the mean - field solution @xmath245 \left [ \left ( a_{b,\downarrow}a_{0-\downarrow}^{\dag } + a_{s,\downarrow}d_{\downarrow}^{\dag}\right ) \prod_{i=1}^{n-1}b_{i\downarrow}^{\dag}\right ] \phi_{0}\label{psi_mf}\\ & = \left [ a_{a , b}a_{0\uparrow}^{\dag}b_{0\downarrow}^{\dag}+a_{a , d}a_{0\uparrow}^{\dag}d_{\downarrow}^{\dag}+a_{d , b}d_{\uparrow}^{\dag } b_{0\downarrow}^{\dag}+a_{d , d}d_{\uparrow}^{\dag}d_{\downarrow}^{\dag } \right ] \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle \nonumber\end{aligned}\ ] ] where @xmath246 and @xmath247 are two ( different ) bases of the @xmath137-dimensional hilbert space . this solution can be rewritten as equation ( [ psi_ms ] ) . in the mean - field solution @xmath248 the coefficients @xmath249 are restricted by two conditions @xmath250 ( @xmath251 ) . therefore this state does not describe well the correlation effects . in contrast the state ( [ psi_ms ] ) opens a wide playing field for improving the solution : ( i ) the fair states @xmath208 and @xmath252 can be individually optimized , each one defining a whole basis @xmath253 and @xmath254 . this yields a much better treatment of the correlation effects . the resulting state is denoted as the ( potentially ) magnetic state @xmath71 . the magnetic state @xmath71 has the same structure as the mean field solution @xmath248 ; the only difference is that its components are optimized for the friedel - anderson hamiltonian . in the construction of the magnetic ground state @xmath71 only the hamiltonian @xmath39 ( equ . [ h_0 ] ) has been used . the expectation value of the `` perturbation '' hamiltonian @xmath59 is zero , @xmath255 where @xmath256 and @xmath257 + \sum_{j=1}^{n-1}v_{j}^{\left ( a\right ) sd}\left [ d_{\sigma}^{\dag } a_{j,\sigma}+a_{j,\sigma}^{\dag}d_{\sigma}\right ] \label{h1p}\ ] ] but @xmath59 yields transitions from the ground state @xmath71 into excited states . from equ . ( [ h1p ] ) one recognizes that @xmath258 only permits transitions between the nest @xmath259 and a band state @xmath260 but no transition among band states . if one considers only final states through this transitions which are linear in @xmath59 then fig.9 shows the possible final states . for the spin - up band this are the transitions from the nest state into an electron excitation @xmath205 , leaving the spin - up part of the nest empty or a transition from an occupied state by creating a hole @xmath261 and filling both state @xmath22 and @xmath16 of the spin - up part of the nest . the corresponding states are generated for the spin - down band . @xmath80{fig9.eps}}\\ & \begin{tabular } [ c]{l}fig.9 : final states which can be obtained by transitions from the\\ ground state $ \psi_{ms}$ through the perturbation hamiltonian $ h_{1}^{\prime } .$\end{tabular}\end{aligned}\ ] ] as an example we consider the transitions in the spin - up band . applying @xmath262 to the ground state @xmath71 yields@xmath263{c}\left [ v_{j}^{\left ( a\right ) fr}a_{a , b}+v_{j}^{\left ( a\right ) sd}a_{d , b}\right ] a_{j,\sigma}^{\dag}b_{0\downarrow}^{\dag}\\ + \left [ v_{j}^{\left ( a\right ) fr}a_{a , d}+v_{j}^{\left ( a\right ) sd}a_{d , d}\right ] a_{j,\sigma}^{\dag}d_{\downarrow}^{\dag}\\ + \left ( -v_{j}^{\left ( a\right ) fr}a_{d , b}+v_{j}^{\left ( a\right ) sd}a_{a , b}\right ) a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } b_{0\downarrow}^{\dag}\\ + \left ( -v_{j}^{\left ( a\right ) fr}a_{d , d}+v_{j}^{\left ( a\right ) sd}a_{a , d}\right ) a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } d_{\downarrow}^{\dag}\end{array } \right\ } \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle\ ] ] the top two lines represent an electron excitation , shown as the left state in fig.9 and the bottom two lines a hole excitation shown as the second state in fig.9 . the final states and the corresponding matrix elements are collected in table iii . @xmath90{|l|l|l|l|}\hline $ \mathbf{\psi}_{i}$ & $ \mathbf{\psi}_{f}$ & $ \left\langle \psi_{i}\left\vert h_{1}^{\prime}\right\vert \psi_{f}\right\rangle $ & \textbf{excitation}\\\hline $ \psi_{ms}$ & $ a_{j\uparrow}^{\dag}b_{0\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ \left ( v_{j}^{\left ( a\right ) fr}a_{a , b}+v_{j}^{\left ( a\right ) sd}a_{d , b}\right ) $ & electron , $ j\geq n$\\\hline $ \psi_{ms}$ & $ a_{j,\uparrow}^{\dag}d_{\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ \left ( v_{j}^{\left ( a\right ) fr}a_{a , d}+v_{j}^{\left ( a\right ) sd}a_{d , d}\right ) $ & electron , $ j\geq n$\\\hline $ \psi_{ms}$ & $ a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } b_{0\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ \left ( -v_{j}^{\left ( a\right ) fr}a_{d , b}+v_{j}^{\left ( a\right ) sd}a_{a , b}\right ) $ & hole , $ 0<j < n$\\\hline $ \psi_{ms}$ & $ a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } d_{\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & $ \left ( -v_{j}^{\left ( a\right ) fr}a_{d , d}+v_{j}^{\left ( a\right ) sd}a_{a , d}\right ) $ & hole , $ 0<j < n$\\\hline \end{tabular } \ \ \ \ \\ & \begin{tabular } [ c]{l}table iii : for the spin - up band the final states $ \psi_{f}$ , and the matrix elements\\ of $ h_{1}^{\prime}$ between the ground state $ \psi_{ms}$ and the final states $ \psi_{f}$ are listed.\\ the final states still have to be expanded into new eigenstates of the nest\\ plus band . for the spin - down band one obtains equivalent transitions . \end{tabular } \ \\end{aligned}\ ] ] there are two important aspects of this result , ( i ) there are always two transitions into each final state , for example @xmath264 and @xmath265 . these two transitions interfere and cancel each other almost completely ( as will be shown below ) . ( ii ) the first two final states in table iii ( as well as the third and forth final state ) are not eigenstates of @xmath39 . one has to expand these final states in terms of the eigenstates of the nest , for example@xmath81{|l|}\hline $ a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}b_{0\downarrow}^{\dag } \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ \\\hline $ a_{j,\uparrow}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag}d_{\downarrow}^{\dag } \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ \\\hline \end{tabular } \ \ \ \ \ \longleftrightarrow\begin{array } [ c]{c}a_{j,\uparrow}\psi_{2/1}^{\left ( 1\right ) } \\ a_{j,\uparrow}\psi_{2/1}^{\left ( 2\right ) } \end{array}\ ] ] this is in complete analogy to the calculation of the excitations in section iii . in fig.10 the logarithm of the effective matrix element is plotted for the first transition in table iii . this represents an electron excitation which is restricted to positive energies . as one recognizes the effective matrix elements are strongly reduced in the energy range in which transitions are possible . the value of @xmath266 lies in the range between @xmath267 and @xmath75 while the original matrix elements for @xmath14 are @xmath268 . this applies for all eight possible excitations . in the energy range where an excitation is permitted the matrix elements are strongly reduced . among the eight possible transitions there is only one transition whose matrix elements exceed @xmath267 this is the second one in table iii where the matrix element reaches values of @xmath269 , still much smaller than the original matrix elements . @xmath270{fig10.eps}}\\ & \begin{tabular } [ c]{l}fig.10 : the logarithm of the effective matrix elements for a transition\\ from the ground state $ \psi_{ms}$ into the excited state $ a_{j\uparrow}^{\dag } b_{0\downarrow}^{\dag}\left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ . this\\ transition is only possible for positive energies , and there the\\ interference between the the d - state $ d_{\uparrow}^{\dag}$ and the fair state $ a_{0\uparrow}^{\dag}$\\ cancels the transition almost completely . \end{tabular}\end{aligned}\ ] ] the strong reduction of the matrix elements is due to the introduction of the fair states which compensate the transitions involving the d - states . it is also the reason why the magnetic ground state is so well represented by @xmath71 . with the eigenenergies of the states in fig.9 and the matrix elements for the transition from the ground state @xmath71 into these states one can now perform a self - consistent perturbation calculation . the number of excited states is @xmath271 for each spin - band . one can build the the full secular matrix for these states which consists of @xmath272 states , where the additional one is the ground state . the result for our standard example ( @xmath142 @xmath145 , @xmath273 , @xmath274 is rather dramatic . the weight of all @xmath275 excitations together is only about @xmath75 . the amplitude of @xmath71 after diagonalization is @xmath276 . this demonstrates the fair solution is an excellent approximation to the ground state . with the secular hamiltonian @xmath92 in section 3.1 one can construct the green s functions ( gf)of the excitations @xmath95@xmath277 or @xmath278 the resulting diagonal elements of @xmath279 are the green s function of the excitations , for example @xmath280 is the green s function of @xmath93 . since the @xmath281 are a function of the energy the above relation is not very practical for a numerical nor analytical calculation of @xmath282 . for a numerical evaluation it is much easier to calculate the eigenvalues @xmath99 and eigenvectors @xmath98 of @xmath92 where @xmath283 . then the green s function @xmath284 of the excitation can be expressed as @xmath285 each energy eigenvalue contributes to the spectrum of the state @xmath286 for the particle density of states we need the green s function ( or the spectrum ) of the @xmath287 states @xmath102 and @xmath107 and @xmath110 . the @xmath47 electron green s functions @xmath288 follow directly from the above calculation . the @xmath289 and @xmath290 for spin up have to be derived from the green s functions of @xmath291 and @xmath86 . in fig.3 a spin - up electron is injected into the nest . this electron can be injected into the @xmath22 or the @xmath16 state . these processes yield complementary amplitudes of @xmath81{lll}$a_{0\uparrow}^{\dag}\psi_{ms}=\left ( a_{d , b}a_{0}^{\dag}d_{\uparrow}^{\dag } b_{0\downarrow}^{\dag}+a_{d , d}a_{0}^{\dag}d_{\uparrow}^{\dag}d_{\downarrow } ^{\dag}\right ) \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow } \right\rangle $ & for & $ a_{0}^{\dag}$\\ $ d_{\uparrow}^{\dag}\psi_{ms}=-\left ( a_{a , b}a_{0\uparrow}^{\dag}d_{\uparrow } ^{\dag}b_{0\downarrow}^{\dag}+a_{a , d}a_{0\uparrow}^{\dag}d_{\uparrow}^{\dag } d_{\downarrow}^{\dag}\right ) \left\vert \mathbf{0}_{a,\uparrow}\mathbf{0}_{b,\downarrow}\right\rangle $ & for & $ d_{\uparrow}^{\dag}$\end{tabular } \ \ \ ] ] each of the two resulting states has to be expanded in the two eigenstates @xmath93 and @xmath86 of the nest . for example , the amplitudes for @xmath107 are obtained through the scalar product between @xmath107 and @xmath112 . then they can make a transition into any of the other states with a single excitation , for example the state @xmath102 . again the resulting amplitudes interfere . so the particle components of the gfs @xmath292 and @xmath290 are composed of the gfs of @xmath112 with a weight which is the square of the scalar product between the amplitude in equ . ( [ a_psi_ms ] ) and @xmath112 . these gfs @xmath289 and @xmath290 have also a hole component . j. friedel , philos . mag . 43 , 153 ( 1952 ) ; adv . phys . 3 , 446 ( 1954 ) ; philos . suppl . 7 , 446 ( 1954 ) ; can . 34 , 1190 ( 1956 ) ; nuovo cimento suppl . 7 , 287 ( 1958 ) ; j. phys . radium 19 , 573 ( 1958 ) , the electronic structure of the transition metals and their alloys and of the heavy metals

analysis of broadband _ hst _ acs and wfpc2 images of the young galactic supernova remnant cassiopeia a reveals a far larger population of outlying , high - velocity knots of ejecta with a broader range of chemical properties than previously suspected . acs filter flux ratios along with follow - up , ground - based spectra are used to investigate some of the kinematic and chemical properties of these outermost ejecta . in this paper , we concentrate on a.5 sq arcmin region located along the eastern limb of the remnant where numerous outer emission knots are optically visible due to an interaction with a local circumstellar cloud , thereby providing a more complete and unbiased look at the remnant s fastest debris fragments . from a study of this region , we identify three main classes of outer ejecta : 1 ) knots dominated by [ n ii ],6583 emission ; 2 ) knots dominated by oxygen emission lines especially [ o ii ],7330 ; and 3 ) knots with emission line strengths similar to the [ s ii ] strong fmk ejecta commonly seen in the main emission shell . we identified a total of 69 [ n ii ] , 40 [ o ii ] , and 120 [ s ii ] knots in the small eastern limb region studied . mean transverse velocities derived from observed proper motion for 63 n - rich , 37 o - rich , and 117 fmk - like knots identified in this region were found to be 8100 , 7900 , and 7600 km s , respectively . the discovery of a significant population of o - rich ejecta situated in between the suspected n - rich outer photospheric layer and s - rich fmk - like ejecta suggests that the cas a progenitor s chemical layers were not completely disrupted by the supernova explosion outside of the remnant s ne and sw high velocity ` jet ' regions . in addition , we find the majority of o - rich outer ejecta at projected locations out beyond ( v = km s ) the remnant s fastest moving fe - rich x - ray emission material ( 6000 km s ) seen in _ chandra _ and _ xmm _ data along the eastern limb . this suggests that penetration of fe - rich material up through the s and si - rich mantle did not extend past the progenitor s n or o - rich outer layers for at least this section of the remnant .

astro-ph0509067

with an estimated explosion date no earlier than 1671@xmath4 ( thorstensen , fesen , & van den bergh 2001 ) and an estimated distance of 3.4@xmath5 kpc @xcite , the bright radio source cassiopeia a ( cas a ) is currently the youngest known galactic remnant of a core - collapse supernova ( sn ) . the remnant consists of an optical , infrared , and x - ray bright @xmath6 diameter ( @xmath04 pc ) emission ring of reverse shock heated sn debris rich in o , si , s , ar , ca and fe @xcite . optically , the remnant s sn debris are seen as condensations and filaments known collectively as fast - moving knots ( fmks ) , with expansion velocities between @xmath7 km s@xmath2 @xcite . estimates for the initial main sequence mass of the cas a progenitor are between 10 and 30 m@xmath8 @xcite . a wolf - rayet star progenitor has often been proposed @xcite and a mass - loss , ring nebula around such a progenitor is consistent with the presence of slow - moving clumps ( ` qsfs ' , v = @xmath9 km s@xmath2 ) of n and he - rich , pre - sn circumstellar mass loss material @xcite . along the outer periphery of the remnant s main ejecta shell , numerous small knots of ejecta have been detected at projected positions near or beyond the forward shock front ( see @xcite and references therein ) . the brightest and best known group of such high - velocity outer ejecta lies in the northeastern ` jet ' , which consists of three or four streams of ejecta knots extending out @xmath10 from the remnant s northeast rim . over a hundred jet knots have been identified from ground - based images and these knots exhibit much higher proper motion derived transverse velocities ( @xmath11 km s@xmath2 ) than the @xmath12 6000 km s@xmath2 fmks seen in the main shell @xcite . optical imaging and spectra of fmks in the jet show strong [ s ii ] @xmath16716,6731 but relatively weak or absent [ o iii ] @xmath14959,5007 emission . a few much fainter outer fmks have been recently found in an apparent southwest ` counterjet ' spatially coincident with faint extended x - ray emission @xcite . in addition to these outlying ne and sw jet fmks , some 50 faint outlying knots with strong [ n ii ] @xmath16548,6583 emission but weak or absent h@xmath13 , [ s ii ] @xmath14 and [ o ii ] @xmath15 line emissions have been detected around much of the remnant s periphery @xcite . radial velocity measurements combined with estimated transverse velocities based on proper motion or knot displacement from the remnant s center of expansion indicate these strong [ n ii ] emission knots have space velocities of @xmath16 km s@xmath2 @xcite . these apparently n - rich knots appear to represent fragments of the progenitor s outer layers and are consistent with the suggestion of a wn wolf - rayet progenitor which experienced substantial pre - sn mass - loss ( the n - rich qsfs ) before exploding as a type ib / c or type ii supernova . in this paper , we report results from a new and much deeper survey of outer ejecta fragments in the cas a snr . the survey reveals a far larger population of high - velocity knots of ejecta and a broader range of chemical properties than previously suspected . a complete catalog of outer ejecta knots is beyond the scope of the present work and will be addressed in a separate paper @xcite . here we concentrate on the remnant s outer ejecta rich eastern limb region and use this area to help define some of the kinematic and chemical properties of the remnant s fastest moving ejecta . the observations are described in @xmath172 , with the results and conclusions given in @xmath173 and @xmath174 , respectively . as part of an imaging survey of the ejecta in the cas a snr , we obtained high resolution images of the remnant between january 2000 and december 2004 using two different cameras on board the _ hubble space telescope _ ( _ hst _ ) . multiband images taken at six pointings covering the entire remnant , including all previously known outlying ejecta knots , were obtained on 2004 march 46 and 2004 december 45 using the wide field channel ( wfc ) of the advanced camera for surveys ( acs ; @xcite ) on board _ hst_. the acs / wfc consists of two @xmath18 ccds providing a field of view @xmath19 with an average pixel size of @xmath20 . four 2-point line dithered images were taken in each of the four acs / wfc sloan digital sky survey ( sdss ) filters , namely f450w , f625w , f775w , and f850lp ( i.e. , sdss g , r , i , and z ) , at each target position to permit cosmic ray removal , coverage of the @xmath21 interchip gap , and to minimize saturation effects of bright stars in the target fields . total integration times in the f450w , f625w , f775w , and f850lp filters were 2000 s , 2400 s , 2000 s , and 2000 s respectively . standard acs pipeline iraf / stsdas data reduction was done including debiasing , flat - fielding , geometric distortion corrections , photometric calibrations , and cosmic ray and hot pixel removal . the stsdas _ drizzle _ task was used to combine exposures in each filter . detected counts in each of the drizzled filter images were converted to flux units by summing the signal in @xmath22 pixel windows , subtracting a local mean background and then multiplying by the mean flux density per unit wavelength generating 1 count s@xmath2 ( i.e. , the ` photflam ' factor ) times the filter effective bandwidth ( ebw ) . for the f625w , f775w , and f850lp filters , the photflam values used in units of 10@xmath23 erg @xmath24 s@xmath2 @xmath2 and the ebw values used were 1.195 , 1.007 , 1.507 , and 415.5 , 434.6 , and 539.4 , respectively @xcite . due to significant reddening toward cas a ( a@xmath25 = @xmath26 mag ; @xcite ) , [ o iii ] @xmath14959,5007 line emission was too weak to be detected for most outlying knots . we have , therefore , not included f450w images in our analysis . table 1 lists the primary emission lines detected by the three other filters , the total system throughput of the telescope + acs / wfc camera + filter , and the resulting relative line flux of a typical main shell cas a fmk ejecta knot . previous _ hst _ images of the remnant taken on 2000 january 23 and 2002 january 16 using the wide field planetary camera 2 ( wfpc2 ) were also analyzed , primarily to measure knot proper motions . while these wfpc2 images mainly targeted cas a s bright main shell @xcite , they also included @xmath27 s f675w filter exposures of portions of the remnant s outer eastern limb . the wfpc2 images have an image scale of @xmath28 pixel@xmath2 which under - sampled _ @xmath29 angular resolution . the f675w filter ( bandpass : @xmath30 ) was sensitive to line emissions of [ o i ] @xmath16300,6364 , [ n ii ] @xmath16548,6583 , [ s ii ] @xmath16716,6731 , [ ar iii ] @xmath317136 , and [ o ii ] @xmath17319,7330 . further descriptions of these data and their reduction can be found in @xcite and @xcite . follow - up , low - dispersion optical spectra of a few outlying emission knots were obtained on three nights in september 2004 using the mdm 2.4 m telescope and the modular spectrograph with a 600 lines mm@xmath2 6000 blaze grating and a @xmath32 slit . typically , we obtained two 1000 s , 1500 s , or 2000 s exposures for each emission knot yielding a spectrum with an effective coverage of 6000 8000 and a spectral resolution of @xmath0 2 . standard iraf data reduction software was used with wavelengths calibrated with hg , ne , and xe lamps and @xcite standard stars . although all three nights were photometric , seeing was @xmath33 leading to considerable slit light losses at times due to variable seeing conditions and guiding errors . these data supplemented previous low - dispersion spectra of ground - based detected outlying emission knots obtained in october 1996 using the mdm 2.4 m telescope with the mark iii spectrograph and a 1024 @xmath34 1024 tektronix ccd detector @xcite . a @xmath35 slit and a 300 lines mm@xmath2 @xmath36 blaze grism were used to obtain sets of two or three @xmath37 s exposures spanning the spectral region @xmath38 with a spectral resolution of @xmath0 8 . in both these and the september 2004 observations , aperture slits were oriented north - south . a comparison of the 2000 and 2002 wfpc2 + f675w filter images of a small portion of the remnant s easternmost limb revealed about 100 high proper motion ejecta knots . most of these knots went undetected in earlier ground - based imaging surveys which had concentrated on detecting strong [ n ii ] @xmath16548,6583 emission knots @xcite . figure 1 shows the dramatic increase in the number of detected outlying ejecta knots using broadband _ hst _ images compared to narrowband ground - based images . the upper left - hand panel of the figure shows a @xmath39 portion of the remnant s eastern limb as seen in a 2.4 m mdm image taken using a 6568 filter ( fwhm = 90 ) sensitive to ejecta knots with strong [ n ii ] emission within a limited radial velocity range ( @xmath40 km s@xmath2 ) . only a few outer knots were detected in this image @xcite . the same region imaged in january 2002 using the wfpc2 + f675w filter is shown in the lower left - hand panel of figure 1 . in contrast to the small handful of knots detected from the ground , over 100 small clumps of ejecta can be seen in the wfpc2 image with knot diameters near the image resolution ( @xmath28 pixel@xmath2 ) . outlying knots visible on the wfpc2 image and confirmed by the detection of proper motions @xmath41 @xmath42 between the 2000 and 2002 wfpc2 images , corresponding to v@xmath43 @xmath41 6000 km s@xmath2 for d = 3.4 kpc , are marked with circles on the 2002 wfpc2 image shown in the upper right - hand panel . the location of these outlying optical ejecta knots with respect to the remnant s outermost x - ray emission , i.e. , associated with the forward shock @xcite , can be seen by comparing the knots positions with the extent of the remnant s x - ray emission shown in the lower right - hand panel image of figure 1 . this image shows a section of the 1 msec chandra acis image ( epoch 2004.3 ; @xcite ) covering the same eastern region as the other panels in the figure . nearly all the outlying optical knots seen here lie ( in projection ) beyond the remnant s bright main emission shell and either close to or eastward ( i.e. , outside ) of the remnant s @xmath0 6000 km s@xmath2 forward shock front as determined by the remnant s easternmost x - ray emission @xcite . the detection of so many new outlying ejecta knots in such a relatively small area suggested that scores of undiscovered fast - moving knots of sn debris might exist elsewhere around the whole remnant . unfortunately , the broad sensitivity of the wfpc2 f675w filter to a variety of elemental emission lines including several oxygen , nitrogen , sulfur , and argon lines precluded an easy discrimination of knot spectral properties . on the other hand , the acs / wfc sdss filters f625w , f775w , and f850lp could be used to distinguish outlying ejecta knots having either strong [ n ii ] or [ o ii ] emissions from the remnant s more commonly seen o + s bright fmk ejecta knots . for example , as shown in table 1 , while the sdss f625w filter , like the wfpc2 f675w filter , is sensitive to emissions from oxygen , nitrogen , and sulfur , the sdss f775w and f850lp filters are mainly sensitive to just oxygen and sulfur emissions , respectively . using the throughputs of the acs / wfc + sdss filter combinations at the wavelengths for several strong emission lines seen in cas a ejecta , we list in table 1 observed ( uncorrected for reddening ) relative fluxes for a typical s , o , ar main shell knot ( ` fmk2 ' ; @xcite ) and the resulting relative fluxes that would be detected using acs / wfc with the sdss filters . the ability of these three sdss filters to distinguish different types of emission knots is shown in table 2 where we list the relative line fluxes for five well - studied main shell ejecta knots ( ` fmk 1- fmk 5 ' ; @xcite ) along with three relatively bright outer ejecta knots . spectra for these three outlying knots covering the @xmath44 wavelength region can be seen in the left - hand panels of figure 2 . as seen from this table , sdss filter flux ratios for cas a s main shell fmks typically lie between @xmath45 . the one exception is fmk 4 whose observed @xmath46 spectrum is dominated , not by [ s iii ] @xmath19069,9531 as is usually the case , but by relatively strong [ o ii ] @xmath17319,7330 line emission resulting in unusually high f775w / f850lp ratio and low f625w / f775w and f625w/(f775w + f850lp ) ratio ( also called f1/(f2 + f3 ) ; see table 3 ) . in contrast to these bright main shell knots , outlying ejecta knots can show significantly different sdss filter flux ratios . for example , knot 15 located well above the remnant s northern limb whose @xmath47 spectrum shows only [ n ii ] @xmath16548,6583 emission ( v@xmath48 = + 4500 km s@xmath2 ; @xcite ) not surprisingly exhibits f625w / f775w and f625w / f850lp values more than an order of magnitude larger than for the five main shell ejecta knots listed . in similar fashion , the outer southeastern knot 17 , which shows an 6000 - 7500 spectrum with only [ o i ] and [ o ii ] line emissions ( see fig . 2 ) , exhibits an o / s sensitive f775w / f850lp ratio nearly three times greater than even the very strong [ o ii ] @xmath17319,7330 emission main shell knot fmk 4 . lastly , the outer western limb knot 19 , which shows a fairly common fmk - like spectrum except for the addition of strong [ n ii ] @xmath16548,6583 line emission which actually dominates the 6000 - 7500 region ( fig . 2 ) , has 24 times higher f625w / f775w and f625w/(f775w + f850lp ) ratios , half the f775w / f850lp ratio , and nearly 23 times the f625w / f850lp ratio of typical main shell ejecta . _ hst _ acs / wfc images covering the whole remnant using the sdss filters were taken in march and december @xmath49 in order to obtain a more complete picture of both the number of outlying debris knots and the diversity of their emission properties . a qualitative look at spectral emission variations in cas a s high - velocity outermost ejecta can be seen in rgb color composite images using the f625w (= red ) , f775w (= green ) , and f850lp (= blue ) image frames . figure 3 shows @xmath50 sections of the individual acs / wfc images covering only a portion of the eastern limb shown above in figure 1 . the three greyscale images were produced by subtracting the march from the december 2004 acs / wfc f625w , f775w , and f850lp filter images . also shown in figure 3 ( bottom right ) is a three color composite image made from the march 2004 filter data . ejecta knots identified through their high proper motions are indicated on each of the frames , with previously ( ground - based ) identified knots ( @xcite ) marked by knot i d numbers . from figure 3 , one sees that many knots are only visible on one or two individual filter images , leading to the strong color differences apparent in the color composite image . similar emission variations were found for ejecta knots located in most other regions around the remnant s periphery . such visual inspections of the sdss filter data set indicated not only a far larger population of high - velocity ejecta knots than previously realized , but also a much broader range of emission line properties over the @xmath51 wavelength range . this , in turn , suggested a greater chemical diversity than just the notion of mostly sulfur - rich fmk - like ejecta in the main shell and jet regions and nitrogen - rich knots elsewhere . published spectra and our new ground - based spectra of a handful of eastern limb knots support this conclusion . however , both their faintness and the small angular separations between many individual outer knots ( often less than @xmath52 ; see fig . 3 ) prevented us from obtaining ground - based spectra of many individual outer ejecta knots . we have therefore used the sdss filter images to classify ejecta knot emission properties into broad categories consistent with available knot spectra . in order to translate observed color composite differences into quantitative spectral emission differences , we measured acs / wfc fluxes from the three sdss filter image sets for all outer knots located along a portion of the remnant s eastern limb using the automated source extraction software package sextractor @xcite . in cases where the sextractor program failed to return a reasonable flux , or failed to return a flux at all , the knot fluxes were calculated by hand . in all cases , the fluxes were calculated using 5 pixel apertures . background estimates were performed , by sextractor , using a @xmath53 pixel rectangular annulus about the isophotal limits of the object . when fluxes were calculated manually , background estimation was performed by calculating the total @xmath54 pixel aperture flux in at least five positions near the object ( avoiding other sources ) and then subtracting the mean computed `` background '' flux from the total object pixel sum . most knots whose fluxes required manual computation were located near a bright background source or very close to another ejecta knot . flux estimates were best for the brightest , well - resolved knots ( above @xmath55 erg @xmath24 s@xmath2 @xmath2 ) , where errors are on the @xmath56 level . moderately bright knots ( @xmath57 erg @xmath24 s@xmath2 @xmath2 ) suffered more from local background variations and uncertainties in knot positions , resulting in flux errors near @xmath58 . error estimates for faint knots ( @xmath59 erg @xmath24 s@xmath2 @xmath2 ) lie approximately at the @xmath60 level . fainter knots , or non - detections in one of the filters , were set to the standard deviation of the background levels of each filter as ( @xmath61 ( f625w ) = @xmath62 erg @xmath24 s@xmath2 @xmath2 , @xmath61 ( f775w ) = @xmath63 erg @xmath24 s@xmath2 @xmath2 , @xmath61 ( f850w ) = @xmath64 erg @xmath24 s@xmath2 @xmath2 ) for the purposes of determining knot type by dominant emission features . for defining the kinematic and chemical properties of the remnant s outlying ejecta , we chose to focus on the remnant s eastern limb . this area exhibits an unusually high number density of knots due to an apparent interaction of the remnant s higher - velocity ejecta with a local circumstellar cloud @xcite . therefore , this region might provide a more complete and unbiased assessment of the remnant s fast sn debris fragments . the region selected lies along cas a s eastern limb region between position angles 78.2 and 140.6 degrees covering @xmath01.5 sq arcmin in size . this avoided the ne jet to the north , which would complicate an analysis of knot emission properties with respect to radial distance , and ends near the southern edge of the faint circumstellar cloud seen in this region . fluxes from the f625w , f775w , and f850lp filter images were measured for a total of 229 outer emission knots identified in this east region through high proper motions measured using the march and december 2004 acs / wfc images . the resulting f625w / f775w and f775w / f850lp flux ratios for these 229 knots are plotted in figure 4 . also shown in this plot are the sdss filter ratios for the five main shell knots ( the purple star symbols ) listed in table 2 , and the three outer ejecta knots ( knots 15 , 17 , and 19 ; circled numbers ) discussed above in @xmath173.1 and also listed in table 2 . we binned the outlying ejecta knots along the remnant s eastern limb into three emission classes , namely , [ n ii ] strong knots , [ o ii ] strong knots , and [ s ii ] strong fmk - like knots . they are color - coded in figure 4 as red , green and blue , respectively , consistent with their colored appearance on our composite images . flux ratio criteria between these three classes were chosen to segregate knots with similar ratios seen for main shell or the outer ejecta knots with existing spectroscopy listed in table 2 . specifically , we chose a flux ratio for f775w/(f625w + f850lp ) @xmath41 1.0 to separate out the [ o ii ] strong fmks from the [ s ii ] strong fmks ; that is , those knots with stronger [ o ii ] @xmath17319,7330 emission detected via the f775w filter than the combined strength of [ n ii ] , [ s iii ] and [ s ii ] emissions detected in the f625w and f850lp filters . similarly , knots with strong [ n ii ] emissions were selected via f625w/(f775w + f850lp ) @xmath41 1.0 , thereby selecting those knots where the combined flux of [ o i ] @xmath16300,6364 , [ s ii ] @xmath16716,6731 , and [ n ii ] @xmath16548,6583 emissions was greater than the sum of f775w flux , due mostly to [ o ii ] , and f850lp flux sensitive to the near - infrared [ s iii ] and [ s ii ] emissions . this latter ratio division is shown as a horizontal line in figure 4 . since the observed [ o i ] flux rarely , if ever , exceeds the [ o ii ] @xmath17319,7330 flux , and the observed [ s ii ] @xmath16716,6731 emission is unlikely to ever exceed the combined flux of [ s iii ] @xmath19069,9531 and [ s ii ] @xmath11028710370 line emissions @xcite , then any knot for which the f625w/(f775w + f850lp ) @xmath41 1.0 requires the presence of significant [ n ii ] @xmath16548,6583 emission . as shown in figure 4 , the chosen ratio criteria successfully grouped the outer fmk - like knots in with typical main shell fmks ( ` fmk 1 - 3 ' and ` fmk 5 ' ; shown as purple stars ) , the strong oxygen emission line ejecta like knot 17 and fmk 4 into an [ o ii ] bright knot class , and ejecta with dominant [ n ii ] emission like knots 15 and 19 into a [ n ii ] knot category . measured f625w , f775w and f850lp fluxes for our selected eastern limb region placed 69 knots in the upper ` [ n ii ] knot ' portion of figure 4 , 40 ` [ o ii ] knots ' in the lower right - hand portion , and 120 ` fmk - like knots ' in the center portion of the figure . an arrow on a knot s plotted position indicates an upper or lower flux limit due to a non - detection for one or more filters ( i.e. , one sigma of the background flux limit , see @xmath173.1.2 ) . for instance , knots with arrows pointing to the left - hand portion of the plot , like some [ n ii ] knots , have no detectable f775w ( [ o ii ] ) emission above background sky variations . a knot marked on figure 4 with a crossbar means it was undetected in one of the filters making up its y - axis value . since the y - axis is a combination of filter fluxes ( y = f1/(f2 + f3 ) ) if there was no detected flux in one filter , say filter f2 , one could not simply mark the knot with a directional arrow on the y - axis . however , the non - detection does lead to an uncertainty in y - axis value for that knot and the crossbar is meant to reflect that uncertainty . the length of the crossbar varies across the plot because : i ) this is a log plot so a constant value has a different length along the axis , and ii ) more importantly , the effect of an uncertainty of say f2 on the y - axis value depends on the specific values of f1 and f3 . crossbars were used instead of arrows because we wished to emphasize that a non - detection in a filter would not move a knot s plotted position into another knot category ( below the line at y = 1 ) no matter what the actual value of flux in the missing filter turned out to be . the line of double arrowed [ n ii ] knots appearing near the top of the [ n ii ] section of figure 4 represent eastern limb knots that have f625w / f775w ratios @xmath41 20 and were undetected in both f775w and f850lp frames . these knots appear to represent highly [ n ii ] emission dominated ejecta , even more so than the outer knot 15 whose spectrum shows no [ s ii ] @xmath16716,6731 or [ o ii ] @xmath17319,7330 emissions ( see fig . 2 ) . knots near the upper part of the [ n ii ] strong knot section of figure 4 have f625w/(f775w + f850lp ) ratios approaching 50 or more . below we briefly discuss three main types of high - velocity , outer emission knots found in this acs / wfc survey . within our selected eastern limb region , dozens of knots were found to show a f625w flux comparable or greater than the total f775w + f850lp fluxes . for these cases , the f625w flux is most probably due to strong [ n ii ] @xmath16548,6583 emission rather than h@xmath13 , [ s ii ] , or [ o i ] emissions . this is because of the 43 strong [ n ii ] emission outer knots with published line strengths [ o i ] and [ s ii ] emissions were not detected and [ n ii ] emission was always much stronger than h@xmath13 , with only three exhibiting h@xmath13 emission at readily detectable levels , i.e. , [ n ii ] @xmath316583/h@xmath13 = 3 4 @xcite . thirty of the 43 knots were bright enough to set meaningful [ n ii ] @xmath316583/h@xmath13 lower limits and had typical values @xmath415 , with 20% above 10 . if the observed f625w emission was due to [ o i ] @xmath16300,6364 and/or [ s ii ] @xmath16716,6731 emission then , based on existing spectra , these knots should have been easily detected through [ o ii ] emission in the f775w images or through strong [ s ii ] and [ s iii ] emission in f850lp images . this was not the case for 69 eastern limb knots indicating these knots have an optical ( @xmath51 ) spectrum with a significant contribution from , if not dominated by , [ n ii ] emission . many of the brighter f625w strong emission knots match known [ n ii ] dominant emission knots previously identified through ground - based images and spectra ( e.g. , knots 4 , 5 , and 7 ) . a @xmath44 spectrum of an eastern limb ejecta clump not previously studied spectroscopically , labeled ` east 1 ' , is shown in the upper right - hand panel of figure 2 ( see table 3 for coordinates and fluxes ) . a spectrum of another eastern limb knot , ` east 3 ' , with f625w emission due in part to the presence of moderately strong [ n ii ] emission is shown in the lower right - hand panel of figure 2 . this knot failed our strong [ n ii ] knot flux criteria test ( i.e. , f625w/(f775w + f850lp ) @xmath41 1.0 ) and was classified as an fmk - like outer knot even though , unlike any main shell fmk , it shows some appreciable [ n ii ] emission . both our color composite images and the plotted knot filter fluxes on the right - hand portion of figure 4 indicated a surprising number of knots with spectra dominated by f775w flux . strong f775w emission indicates a high probability for strong [ o ii ] @xmath17319,7330 line emission since the only other significant emission lines in the f775w filter s passband are the [ ar iii ] @xmath17136,7751 lines . however , even for those rare knots having exceptionally strong argon emission lines ( e.g. , the ` calcium fmk ' ; @xcite ) , [ s ii ] and [ s iii ] emissions still dominate the optical spectra and thus would be plotted on figure 4 in with the more common [ s ii ] strong fmks . therefore , strong f775w emission is a good marker for unusually strong , if not dominate , [ o ii ] line emission . in the eastern limb region , 40 knots were identified as having a spectrum dominated by [ o ii ] emission . examples of these [ o ii ] f775w bright knots can be seen in the color composite image of figure 3 ( lower right - hand panel ) . the group of green knots in the lower right portion of the color composite make up knot 17a , an ejecta clump first identified from ground - based images and exhibiting spectra similar to the more southern knot 17 ( see fig . 2 ; middle panels ) . this knot was one of a small cluster of outer ejecta knots ( knots 16 , 17 , 17a , and 17b ) for which earlier low - dispersion spectra showed strong oxygen emission lines @xcite . their limited spatial distribution along a section of the se limb had initially made them appear to be simply a few odd , high - velocity ejecta clumps , not representatives of a large and well distributed subclass of outer ejecta which we now identify as strong [ o ii ] outer knots . in our color composite images , fmk - like knots appear white or purple due to strong f625w and f850lp fluxes arising principally from [ s ii ] @xmath16716,6731 and [ s iii ] @xmath19069,9531 emissions . these knots typically occupy the middle of the filter flux ratio plot shown in figure 4 , as do the four bright main shell knots previously discussed and shown here as purple star symbols . earlier ground - based surveys had found that nearly all outlying fmk - like knots were located in either the northeast or southwest jet ejecta regions . however , the deeper and higher spatial resolution _ hst _ images reveal a more extensive distribution of such high - velocity fmk - like ejecta . in the eastern limb section for example , we identify 120 [ s ii ] strong fmk - like outer knots . virtually all of these knots are small and faint , and consequently , they fell below the detection limit of the moderately deep and broadband ( fwhm = 250 ) [ s ii ] @xmath16716,6731 imaging survey of the remnant @xcite . histograms for the radial distances for all 229 identified knots in our selected eastern limb region grouped by emission type are shown in figure 5 . radial distances for these knots , measured from the remnant s estimated center of expansion ( @xcite ) , are in the range @xmath65 ( @xmath66 pc ) . as shown in this figure , [ n ii ] bright knots tend to lie at greater distances ( in projection ) than the [ o ii ] bright knots which , in turn , lie coincident with or slightly ahead of the fmk - like knots . mean distances for the [ n ii ] , [ o ii ] , and fmk - like knots are 168@xmath67 , 163@xmath67 , and 158@xmath67 (= 2.8 pc , 2.7 pc , 2.6 pc ) , respectively . the observed range of radial distances , when converted to average transverse velocities ( assuming an age of 320 yr and a distance of 3.4 kpc ) correspond to velocities of @xmath68 km s@xmath2 . for 217 of the 229 eastern limb knots we were able to estimate transverse velocities directly from proper motion measurements . positional displacements were measured using wcs aligned january 2002 wfpc2 f675w and march 2004 acs / wfc f625w , f775w , and f850lp images . measured proper motions relative to observed knot radial distances are plotted in figure 6 . the line shown is a least squares fit to the radial distance vs. proper motion for all knots and has a slope of @xmath69 mas yr@xmath2 arcsec@xmath2 . the [ n ii ] knots exhibit the highest proper motions , up to @xmath70 yr@xmath2 , with somewhat smaller systematic values for the [ o ii ] and fmk - like knots . transverse velocities derived from these proper motions grouped by knot emission type yield mean expansion velocities of 8100 , 7900 , and 7600 km s@xmath2 for 63 n - rich , 37 o - rich , and 117 fmk - like knots , respectively . consistent with their greater average radial distances ( fig . 5 ) , [ n ii ] knots show the highest maximum transverse velocities , up to 9300 km s@xmath2 , followed by the fmk - like knots with velocities up to 9100 km s@xmath2 , and then [ o ii ] knots of up to 9000 km s@xmath2 . these velocities significantly exceed the maximum space velocity of 6000 km s@xmath2 seen for main shell ejecta @xcite . however , knots in all three classes with the smallest radial distances ( @xmath0 @xmath71 ) show a marked decrease in observed proper motions and these tend to lie at a projected positions just ahead of main shell knots . this effect is seen for all three knot emission classes and may indicate stronger deceleration by the reverse shock . conversely , the true slope may actually be steeper than shown with the most distant knots the most decelerated . finally , the location of these outlying optical knots categorized by their measured filter flux ratios relative to the remnant s faint outermost x - ray emission is shown in figure 7 . here the march 2004 red [ n ii ] , green [ o ii ] , and blue fmk optical knot positions are overlaid onto the _ chandra _ 1 ms exposure ( right panel ) , itself color coded by isolating si ( 1.82.0 kev ) , fe k ( 6.57.0 kev ) , and continuum ( 4.26.3 kev ) emissions following @xcite . the three eastern limb knots ( e1e3 ) discussed above for which we obtained spectra are marked in the left - hand panel which shows just the remnant s fe x - ray emission detected by _ chandra_. one sees from this figure that nearly all the outer eastern limb knots lie near or out beyond the remnant s forward shock front . importantly , nearly all the strong [ o ii ] knots which we have identified via filter ratios lie ( in projection ) out ahead of the remnant s bright fe - rich x - ray emitting ejecta . a remnant s outermost debris may carry clues about both the chemistry of the progenitor s outer layers and about the dynamics of the sn explosion . our _ hst _ imaging of the cas a remnant has revealed a surprisingly large population of fast - moving ejecta outside of the ne and sw jet regions . these ejecta knots lie coincident with or out beyond the remnant s current forward shock front position ( as seen in x - rays ; see fig . 1 ) and exhibit a wide range of filter flux ratios indicative of a significant chemical diversity , confirmed by spectral data ( figs . while optically emitting debris constitute only a small fraction of the total ejected mass , the remnant s fastest moving material is perhaps best studied optically . for example , the ne and sw jets can be optically traced about 80 arcsec farther out than in radio or x - rays and only a handful of outer ejecta knots around the rest of the remnant are visible in even the deepest radio or x - ray images . on the other hand , outlying ejecta knots are seen in the optical due to their interaction with surrounding ism / csm which generate @xmath12100 km s@xmath2 internal shocks driven into the knot by the high stagnation pressure behind the knot s bow shock . consequently , what one sees may be a biased view of the distribution of outer ejecta . because the visibility of outlying optical knots depends upon whether they have recently encountered local ism / csm clouds or the forward and reverse shock fronts , combining results from different regions around the remnant might lead to a seemingly greater dispersion of chemical abundance variations with radial distance . this , plus the unlikely smooth and spherical expansion of the progenitor s outer layers , make uncertain apparent correlations of outer ejecta kinematic versus chemical properties when summed across the whole remnant . consequently , we restricted the present study to just a relatively small , limited region along the cas a remnant s eastern boundary where the ejecta appear to be interacting with an extended csm cloud . general spectral properties of outer emission knots were determined by the use of filter flux ratios and we have identified three main classes of outer ejecta : 1 ) knots dominated by [ n ii ] @xmath16548,6583 emission ; 2 ) knots dominated by oxygen emission lines especially [ o ii ] @xmath17319,7330 ; and 3 ) knots with emission line strengths much like those seen in the fmks found in the main shell . the histograms shown in figure 5 make a compelling case that , at least for the eastern limb region studied , the emission - line properties and hence likely chemical properties of cas a s highest velocity ejecta correlate fairly well with expansion velocity . because few spectra are available for the nearly 230 outer knots found within our selected eastern limb region , accurate and unambiguous abundance ranges are not possible for the different knot types from the broadband filter measurements . however , the observed range of flux values , when taken together with representative knot spectra ( see fig . 2 ) , offer qualitative estimates on relative nitrogen , oxygen , and sulfur abundances . using the modified @xcite shock code described in @xcite , several models were run for shock speeds of 50 100 km s@xmath2 assuming knot preshock densities @xmath72 @xmath73 . the log of the abundances of oxygen and nitrogen were varied from 14 to 16 ( h = 12.00 ) with a sulfur abundance of 14 and other elements heavier than ne set at or below solar values . these model results indicate that while [ s iii]/[o ii ] is sensitive to shock velocity , the ratio of [ n ii ] @xmath16548,6583/[o ii ] @xmath17319,7330 is fairly independent of shock speed and the abundances of other elements . for n / o abundance ratios of 0.1 and 10 , we found [ n ii]/[o ii ] values of @xmath0 0.5 and @xmath0 100 , respectively . such values suggest that for strong [ n ii ] emission outer knots , i.e. , knots plotted in the uppermost portion of figure 4 such as east 1 and knot 15 and having spectra with [ n ii]/[o ii ] @xmath74 20 ( cf . fig . 2 ) , the n / o abundance ratio exceeds unity indicating a nitrogen overabundance at least an order of magnitude over solar . the highest - velocity ejecta in the east limb region are clearly those with dominate [ n ii ] emission ( see figs . 5 and 6 ) . this fact , when taken together with the presence of n - rich circumstellar material ( qsfs ) and the observed lack of appreciable hydrogen emission in all but a couple of outermost ejecta @xcite support the notion that the cas a progenitor was probably a wn wolf - rayet star that exploded as a core - collapse sn ib event ( fesen et al.1987 , fesen & becker 1991 , vink 2004 ) . although recent estimates for the progenitor mass based on x - ray abundance analysis have difficulty explaining the lack of c - burning products such as ne and mg , the remnant s estimated oxygen mass of 13 m@xmath75 plus the excellent correlation between si , s , ar , and ca abundances suggest a relatively high main sequence mass @xmath0 1525 m@xmath75 @xcite not unlike that expected for a wr star progenitor . for the strong [ o ii ] emission knots , both our shock models as well as the metal - rich shock models presented in @xcite which assumed somewhat lower preshock densities suggest that a near solar abundance ratio for o / s @xmath12 10 yields an observed [ o ii ] @xmath17319,7330/[s ii ] @xmath16716,6731 ratio of around 1 2 and a f775w / f850lp filter ratio of around one ( after correcting for an a@xmath25 = 5 mag and the acs / wfc throughput ) . because [ o ii ] strong knots show f775w / f850lp ratios above 1.5 ( and can exceed 10 ) and are unlikely to have strong [ s ii ] @xmath16716,6731 emission in light of f625w/(f775w + f850lp ) ratios between 0.07 0.5 , such knots appear to possess o / s abundance ratios many times above the solar value . this qualitative assessment is consistent with spectra taken of of such knots , ( e.g. , knot east 2 and knot 17 ; see fig . 2 ) for which only emission lines of oxygen are detected in the 6000 7500 region . our discovery of a significant population of o - rich ejecta outside the forward shock front both here along the eastern limb region and elsewhere around the remnant @xcite may help in understanding the degree and asymmetry of compositional mixing of the cas a sn debris . on the one hand , the very high - velocity si and s rich ejecta in the ne and sw jets plus the presence of highly fe - rich ejecta produced by explosive si burning appearing outside the remnant s si - rich shell along the nw and se limbs has been taken as evidence for large scale turbulence and mixing due to a non - spherical expansion @xcite . this would appear consistent with some core - collapse models which show significant mixing and overturning of ejecta through rayleigh - taylor instabilities @xcite , and which sn 1987a gave direct evidence for in the form of the apparent transport of freshly synthesized @xmath76ni from the core to the h - rich envelope @xcite . however , the presence of a layer of o - rich ejecta knots situated in between the apparent n - rich outer photospheric layer and the s - rich , fmk - like ejecta layer suggests that , at least for this limited eastern limb region away from the remnant s ne and sw jets , cas a s chemical layers do not appear to have been completely mixed or disrupted all the way out to the surface . the outlying o - rich knots we see in cas a exhibit spectral properties not unlike the o - rich ejecta seen in the lmc / smc remnants n132d and 1e 0102.2 - 7219 where there is no evidence supporting mixing of o - rich ejecta with o - burning products , i.e. , si , s , ar , and ca rich ejecta @xcite . this raises questions about just how well mixed were the o - rich and s - rich layers in the cas a sn on both a local and global scale . the degree and uniformity of elemental mixing in cas a has been a long standing question ( chevalier & kirshner 1978 , 1979 ; johnson & yahil 1984 ; winkler et al . 1991 ; douvion et al . 1999 ; hwang et al . 2000 , willingale et al . strong o and s lines in most main shell ejecta knots indicate that substantial mixing of the o - rich and si+s - rich layers did indeed occur . however , there is also compelling evidence that this mixing was neither microscopic nor homogeneous on large scales . some ejecta clumps have optical spectra with nearly only oxygen emission lines visible @xcite , while others have very weak or nearly absent oxygen emissions but unusually strong lines of ca , ar , and s @xcite . compared to the range of [ s ii]/[o ii ] emission line ratios for optical knots in the main shell , there is a fairly constant [ ar iii ] @xmath317136/[s iii ] @xmath319069 ratio suggesting ar and s formed at about a constant ratio from oxygen burning , in sharp contrast to the observed [ o iii ] @xmath315007/[s iii ] @xmath319069 ratio which varies by as much as a factor of 100 . the observed range of ejecta spectral properties could reflect either different degrees of explosive oxygen burning @xcite or varying levels of mixing between outer o - rich material with inner layers of o - burning products @xcite . ne and ar line emission maps made from mid - infrared observations show an anti - correlation between the presence of neon ( which is most abundant in the outermost o - rich layers ) and silicon - rich ejecta pointing to a mixing process that was not homogeneous on large scales ( douvion et al . 1999 ) . the fact that there is no correlation of main shell ejecta abundances with expansion velocity means that there had to be some inversion of the s - rich inner layers relative to the o - rich outer layers @xcite . large scale rayleigh - taylor instability ` fingers ' or incomplete explosive o - burning across different regions might help explain the degree of mixing observed . in and around r - t fingers , considerable mixing would be expected leading to o - rich and s - rich ejecta regions with a lack of element vs. ejection velocity correlation like that observed . @xcite found for a type ib sn model with 4.2 m@xmath75 that r - t instabilities form at the si / o and o / he compositional interfaces . this lead to fragmentations of the fe - rich core and considerable mixing of the inner 2 m@xmath75 which ends up with an expansion velocity as high as 3500 5500 km s@xmath2 . however , this velocity is less than that seen in cas a and nearly pure fe - rich material out beyond the si - rich layers in cas a s nw and se sections @xcite would seem more consistent with a picture of strong but only regional overturning . @xcite also noted that the pattern of r - t overturning carries information about the geometry of neutrino - driven convection that seeds the r - t instabilities . if that is true , then the observed @xmath77 angular size of the fe - rich areas in cas a may be indicating the rough scale of such neutrino - driven convection seeds . on the other hand , high - velocity ( v = @xmath78 km s@xmath2 ) n - rich and o - rich ejecta at projected locations along the eastern limb out beyond the remnant s fastest moving fe - rich x - ray emission material ( 6000 km s@xmath2 ) show that even though fe - rich core fragments apparently penetrated the s + si - rich mantle ejecta layer , they did not expand past the n or o - rich outer layers as happened in sn 1987a . thus in cas a , at some distance radially above the r - t finger instabilities , the he + n - rich and o - rich layering structure appears to have survived largely intact . a nitrogen - oxygen - iron layering can be seen in figure 7 where the majority of optical n - rich and o - rich knots lie in projection ahead of the fe - rich ejecta detected in the x - rays . while not lying exactly in the plane of the sky , this fe - rich material is seen to extend out to a maximum radius of @xmath79 , in this region , corresponding to an average expansion velocity of around @xmath80 km s@xmath2 . de - projection of its observed radial velocity of @xmath81 to @xmath82 km s@xmath2 @xcite would only move it a few arcseconds farther eastward , insufficient to extend it out beyond either the n - rich or o - rich outer ejecta knots which lie at radial distances of @xmath83 and @xmath84 , respectively ( see fig . @xcite and @xcite comment that any fe - rich debris ejected farther out than seen in the _ chandra _ image would have encountered the reverse shock at an earlier epoch and thus could have cooled down below the x - ray emitting temperature range . however , the coincidence of the inner edge of the o - rich ejecta we have detected with the outermost fe - rich x - ray emitting material seen in _ chandra _ fe k image is striking ( left panel , fig . 7 ) suggesting that we re seeing most of the fastest - moving fe - rich material in this region . this holds true for the cooler fe l emission as well which has about the same spatial distribution and extent as that of fe k @xcite . however , there may be a causal relation between fe - rich ejecta and formation of these outlying ejecta knots . as can been seen from figure 7 , the projected space density of the outer optical ejecta knots may be correlated with the outermost fe - rich x - ray emitting ejecta . for example , the cluster of optical knots around knot e2 lies at the easternmost tip of the detected fe - rich x - ray emission , with the number of knots showing a noticeable decline to both the north and south . without knot radial velocity information , one can not test if such projected correlations are in fact spatially real , yet the coincidences seen here are intriguing . while radially limited layer mixing may explain the abundance pattern seen along the remnant s east limb , this process clearly did not operate in the ne and sw jet regions . thus the global picture of the cas a sn explosion which emerges is one of regional and radially limited overturning of mantle and core material . added to this scenario are the seemingly opposing ne and sw jets of much higher velocity ejecta , rich in explosive o - burning products , whose nature is not well understood . this all created the fairly complex sn remnant ejecta structure exhibiting regionally diverse abundance mixing patterns we observe today . we present an analysis of broadband _ hst _ acs and wfpc2 images of the young galactic supernova remnant cassiopeia a , and concentrate specifically on a @xmath01.5 sq arcmin region located along the easternmost limb of the remnant . this region exhibits numerous outer emission knots that are optically visible due to an apparent interaction with a local circumstellar cloud , and therefore provide a more complete and unbiased look at the remnant s fastest debris fragments . calibrated acs filter flux ratios along with follow - up , ground - based spectra were used to investigate some of the kinematic and chemical properties of these outermost ejecta . these data revealed the following : \1 ) a substantial population of outlying , high - velocity ejecta knots exists around the remnant which exhibit a broader range of chemical properties than previously recognized . based on the outlying knots found in the east limb region , we identify three main classes of outer ejecta : a ) knots dominated by [ n ii ] @xmath16548,6583 emission , b ) knots dominated by oxygen emission lines especially [ o ii ] @xmath17319,7330 , and c ) ejecta with emission line strengths similar to the common [ s ii ] strong fmk ejecta of the main shell . mean transverse velocities derived from observed proper motion for 63 n - rich , 37 o - rich , and 117 fmk - like knots found in this region are 8100 , 7900 , and 7600 km s@xmath2 , respectively . \2 ) the discovery of a significant population of o - rich ejecta situated in between the suspected n - rich outer photospheric layer and s - rich fmk - like ejecta suggests that the cas a progenitor s chemical layers were not completely disrupted by the supernova explosion outside of the remnant s ne and sw high velocity ` jet ' regions . \3 ) we find that most o - rich outer ejecta along the remnant s eastern limb lie at projected locations out beyond ( v = @xmath3 km s@xmath2 ) the remnant s fastest moving fe - rich x - ray emission material ( 6000 km s@xmath2 ) along the eastern limb . this suggests that penetration of fe - rich material up through the s and si - rich mantle did generally not extend past the progenitor s n or o - rich outer layers as had been previously assumed . the chemical and dynamical picture of cas a that emerges is one of large , regional overturning involving both mantle and core material , separated by areas where the initial he + n - rich , o - rich , and si + s - rich layering survived fairly intact . this generated a fairly complex remnant structure with numerous clumps of reverse shocked ejecta showing a wide diversity of chemical abundances . future investigations of the cas a remnant using this _ hst _ image data set will include a full catalog of all optically visible outlying ejecta knots , analysis of the remnant s ejecta expansion asymmetries , and knot emission variability and ablation effects due to high speed passage through the local ism / csm . we thank p. hoeflich for helpful discussions , d. patnaude for help with the chandra x - ray data reduction , and j. thorstensen for assistance with the celestial coordinate transformations . this work was supported by nasa through grants go-8281 , go-9238 , go-9890 , and go-10286 to raf and jm from the space telescope science institute , which is operated by the association of universities for research in astronomy . rac is supported by nsf grant ast-0307366 and clg is supported through uk pparc grant ppa / g / s/2003/00040 . arnett , d. , fryxell , b. , & mueller , e. 1989 , , 341 , l63 bertin , e. , & arnouts , s. 1996 , , 117 , 393 blair , w. p. , morse , j. a. , raymond , j. c. , kirshner , r. p. , hughes , j. p. , dopita , m. a. , sutherland , r. s. , long , k. s. , & winkler , p. f. 2000 , , 537 , 667 chevalier , r. a. , & kirshner , r. p. 1978 , , 219 , 931 chevalier , r. a. , & kirshner , r. p. 1979 , , 233 , 154 delaney , t. , & rudnick , l. 2004 , advances in space research , 33 , 422 douvion , t. , lagage , p. o. , & cesarsky , c. j. 1999 , , 352 , l111 fabian , a. c. , willingale , r. , pye , j. p. , murray , s. s. , & fabbiano , g. 1980 , , 193 , 175 fesen , r. a. 2001 , , 133 , 161 fesen , r. a. , morse , j. a. , chevalier , r. a. , borkowski , k. j. , gerardy , c. l. , lawrence , s. s. , & van den bergh , s. 2001 , , 122 , 2644 fesen , r. a. , & becker , r. h. 1991 , , 371 , 621 fesen , r. a. , becker , r. h. , & blair , w. p. 1987 , , 313 , 378 fesen , r. a. , & gunderson , k. s. 1996 , , 470 , 967 ford , h. c. , et al . 1998 , proc . spie , 3356 , 234 garca - segura , g. , langer , n. , & mac low , m 1996 , , 316 , 133 gotthelf , e. v. , koralesky , b. , rudnick , l. , jones , t. w. , hwang , u. , & petre , r. 2001 , , 552 , l39 hammell , m. c. , & fesen , r. a. 2006 , in preparation hughes , j. p. , rakowski , c. e. , burrows , d. n. , & slane , p. o. 2000 , , 528 , l109 hurford , a. p. , & fesen , r. a. 1996 , , 469 , 246 hwang , u. , holt , s. s. , & petre , r. 2000 , 537 , l119 hwang , u. , & laming , j. m. 2003 , , 597 , 362 hwang , u. , et al . 2004 , , 615 , l117 kamper , k. & van den bergh , s. 1976 , , 32 , 351 kifonidis , k. , plewa , t. , janka , h .- t . , & m " uller , e. 2000 , , 531 , l123 kifonidis , k. , plewa , t. , janka , h .- , & m " uller , e. 2003 , , 408 , 621 kirshner , r. . , & chevalier , r. a. 1977 , , 218 , 142 laming , j. m. , & hwang , u. 2003 , , 597 , 347 laming , j. m. , & hwang , u. 2005 , , 298 , 33 langer , n. , & el eid , m. f. 1986 , , 167 , 265 lawrence , s. s. , macalpine , g. m. , uomoto , a. , woodgate , b. e. , brown , l. w. , oliversen , r. j. , lowenthal , j. d. , & liu , c. 1995 , , 109 , 2635 massey , p. , & gronwald , c. 1990 , , 358 , 344 minkowski , r. 1968 , in `` stars and stellar systems '' , 7 , 623 morse , j. a. , fesen , r. a. , chevalier , r. a. , borkowski , k. j. , gerardy , c. l. , lawrence , s. s. , & van den bergh , s. 2004 , , 614 , 727 pavlovsky , c. , et al . `` acs instrument handbook '' , version 5.0 , ( baltimore : stsci ) peimbert , m. , & van den bergh , s. 1971 , , 167 , 223 raymond , j. c. 1979 , , 39 , 1 reed , j. e. , hester , j. j. , fabian , a. c. , & winkler , p. f. 1995 , , 440 , 706 reynoso , e. m. , & goss , w. m. 2002 , , 575 , 871 sirianni , m. , et al . 2005 , preprint ( astro - ph/0507614 ) thorstensen , j. r. , fesen , r. a. , & van den bergh , s. 2001 , , 122 , 297 van den bergh , s. 1971 , , 165 , 457 van den bergh , s. , & dodd , w. w. 1970 , , 162 , 485 vink , j. , bloemen , h. , kaastra , j. s. , bleeker , j. a. m. 1998 , , 339 , 201 vink , j. 2004 , new astronomy review , 48 , 61 willingale , r. , bleeker , j. a. m. , van der heyden , k. j. , kaastra , j. s. , & vink , j. 2002 , , 381 , 1039 willingale , r. , bleeker , j. a. m. , van der heyden , k. j. , & kaastra , j. s. 2003 , , 398 , 1021 winkler , p. f. , roberts , p. f. , kirshner , r. p. 1991 , in `` supernovae : the tenth santa cruz summer workshop in astronomy and astrophysics '' , ed . woosley ( springer - verlag : new york ) , p 652 woosley , s. e. , langer , n. , & weaver , t. a. 1993 , , 411 , 823 woosley , s. e. , & weaver , t. a. 1995 , , 448 , 315 lcclccc f625w & @xmath85 s & 54507100 & [ o i ] 6300,6364 & 0.42 & 39 & 16 + ( sdss r ) & & & [ s ii ] 6716,6731 & 0.44 & 100 & 44 + & & & [ n ii ] 6548,6583 & 0.43 & 0 & 0 + & & & & & & + f775w & @xmath27 s & 68508600 & [ ar iii ] 7136 & 0.38 & 41 & 15 + ( sdss i ) & & & [ o ii ] 7319,7320 & 0.42 & 171 & 72 + & & & [ ar iii ] 7751 & 0.38 & 14 & 5 + & & & & & + f850lp & @xmath27 s & 850010500 & [ s iii ] 9069 & 0.20 & 210 & 42 + ( sdss z ) & & & [ s iii ] 9531 & 0.14 & 600 & 85 + & & & [ s ii ] 1028710370 & 0.05 & 380 & 19 + lcccccccc [ o i ] 6300,6364 & 55 & 39 & 29 & 152 & 157 & 7 & @xmath865 & 22 + [ n ii ] 6548,6583 & @xmath871 & @xmath871 & @xmath871 & @xmath871 & @xmath871 & 130 & 100 & @xmath868 + [ s ii ] 6716,6731 & 100 & 100 & 100 & 100 & 100 & 62 & @xmath865 & @xmath868 + [ o ii ] 7319,7330 & 275 & 171 & 104 & 1530 & 112 & 33 & @xmath865 & 100 + [ s iii ] 9069,9531 & 689 & 810 & 440 & 1190 & 558 & & & + & & & & & & & & + f625w ( f1 ) & 67 & 60 & 56 & 108 & 55 & 100 & 100 & 100 + f775w ( f2 ) & 134 & 92 & 55 & 673 & 58 & 24 & @xmath868 & 253 + f850lp ( f3 ) & 135 & 146 & 78 & 234 & 103 & 78 & @xmath8610 & 30 + & & & & & & & & + f625w / f775w & 0.50 & 0.65 & 1.00 & 0.16 & 0.95 & 4.17 & @xmath4112.0 & 0.40 + f775w / f850lp & 1.02 & 0.63 & 0.71 & 2.88 & 0.56 & 0.31 & @xmath00.80 & 8.34 + f625w / f850lp & 0.49 & 0.41 & 0.72 & 0.46 & 0.53 & 1.28 & @xmath4110.0 & 3.30 + f1/(f2 + f3 ) & 0.25 & 0.25 & 0.42 & 0.11 & 0.34 & 0.98 & @xmath415.5 & 0.35 +

using resistive magnetohydrodynamics simulation , we investigate circumstellar disk formation in a strongly magnetized cloud . as the initial state , an isolated cloud core embedded in a low - density interstellar medium with a uniform magnetic field is adopted . the cloud evolution is calculated until almost all gas inside the initial cloud falls onto either the circumstellar disk or a protostar , and a part of the gas is ejected into the interstellar medium by the protostellar outflow driven by the circumstellar disk . in the early main accretion phase , the disk size is limited toau because the angular momentum of the circumstellar disk is effectively transferred by both magnetic braking and the protostellar outflow . in the later main accretion phase , however , the circumstellar disk grows rapidly and exceedsau by the end of the main accretion phase . this rapid growth of the circumstellar disk is caused by the depletion of the infalling envelope , while magnetic braking is effective when the infalling envelope is more massive than the circumstellar disk . the infalling envelope can not brake the circumstellar disk when the latter is more massive than the former . in addition , the protostellar outflow weakens and disappears in the later main accretion phase , because the outflow is powered by gas accretion onto the circumstellar disk . although the circumstellar disk formed in a magnetized cloud is considerably smaller than that in an unmagnetized cloud , a circumstellar disk exceedingau can form even in a strongly magnetized cloud .

1009.2140

the magnetic field is an essential ingredient in present - day star formation . observations have shown that molecular cloud cores have magnetic energy comparable to gravitational energy @xcite . the cloud cores also have angular momentum @xcite . stars are born in such cloud cores . as the cloud collapses , the central region of the collapsing cloud rotates rapidly , and this rapid rotation suppresses further collapse and subsequent protostar formation . in the collapsing cloud , however , the magnetic field can effectively transfer the excess angular momentum from the center of the cloud by magnetic braking ( and protostellar outflow ) , and promote further collapse and protostar formation . moreover , after the protostar formation , the magnetic field greatly contributes to the evolution of a protostar and a circumstellar disk . because the angular momentum in the circumstellar disk is transferred by magnetic effects , the gas in the circumstellar disk can fall onto the protostar , increasing the protostellar mass in the main accretion phase . thus , it may be expected that only a small - size ( or no ) disk appears around the protostar if the angular momentum of the disk is effectively transferred by magnetic effects . in various star - forming regions , however , many circumstellar disks with a size of several hundred au can be observed ( e.g. , @xcite ) , indicating that some of the angular momentum survives the magnetic braking catastrophe and contributes to the growth of the circumstellar disk . the formation and evolution of the circumstellar disk are important not only for star formation but also for planet formation . the size and mass of the circumstellar disk determine the mode of formation of a planet ( i.e. , core accretion or gravitational instability mode ; see @xcite ) . the angular velocity of the molecular cloud core that is the host cloud for star formation creates a circumstellar disk in the star formation process . thus , to understand the formation of the circumstellar disk , we need to investigate the evolution of the molecular cloud core from the prestellar stage . numerical simulations can reproduce the disk formation in the molecular cloud core . however , because there is a large difference in both spatial and time scales between the molecular cloud core and the circumstellar disk , we need special numerical techniques such as adaptive mesh refinement or nested grid and smoothed particle hydrodynamics . using these techniques , several studies have investigated the disk formation in the molecular cloud core from the prestellar stage . @xcite and @xcite investigated the formation of the protostar and circumstellar disk in an unmagnetized cloud ; they showed that the first core that forms before protostar formation in the gas - collapsing phase directly evolves into a circumstellar disk after the protostar formation . @xcite and @xcite showed that the circumstellar disk , which originates from the first core , is more massive than or comparable to the protostar in the main accretion phase and has a size of several hundred au when the host cloud is unmagnetized . using numerical simulations , the disk formation in magnetized clouds has also been investigated . these studies reported that the calculated size of the circumstellar disk appearing in a magnetized cloud is much smaller than the observed sizes . using an ideal mhd calculation , @xcite showed that the circumstellar disk formed in a magnetized cloud has a size of @xmath3au at maximum . @xcite also showed that large - size disk formation is suppressed in a magnetized cloud . this is because the angular momentum of the circumstellar disk ( or infalling gas onto the circumstellar disk ) is excessively transferred outward by magnetic braking , and the gas in the disk effectively falls onto the protostar . in reality , however , the magnetic field dissipates in a high - density gas region by ohmic dissipation and ambipolar diffusion , because the degree of ionization is considerably low ( e.g. , @xcite ) . such a high - density gas region , where the neutral gas is decoupled from the magnetic field , corresponds well to the disk - forming region in the early main accretion phase ( @xmath4au , @xcite ) . thus , if the magnetic field can sufficiently dissipate around the protostar , magnetic braking becomes ineffective and a large - size circumstellar disk comparable to observations may appear even in strongly magnetized clouds . in contrast , with non ideal mhd calculations , @xcite and @xcite reported that the disk size is considerably smaller than observations even when the magnetic field dissipates in the high - density gas region , although the disk in non ideal mhd calculations is somewhat larger than that in ideal mhd calculations . these studies , however , overlooked the effect of the finite mass of the host cloud for star and circumstellar disk formation . in the early main accretion phase , a massive gas envelope ( or remnant of the molecular cloud ) remains around the circumstellar disk . the massive envelope rotates much slower than the circumstellar disk , and thus it can brake the circumstellar disk through the magnetic field lines that connect the less massive rapidly rotating disk to the more massive slowly rotating envelope . in other words , the angular momentum of the circumstellar disk is transferred into the infalling envelope by torsional alfv@xmath5n waves . this is the magnetic braking mechanism for the circumstellar disk formation . thus , even with very strong magnetic fields , the circumstellar disk rotation velocity does not decrease when no massive envelope exists around the disk . the mass of the infalling envelope decreases with time in the main accretion phase when an isolated molecular cloud core is assumed . the infalling envelope disappears by the end of the main accretion phase : a part of the envelope is blown away from the cloud by the protostellar outflow @xcite , and the remainder falls onto either the circumstellar disk or the protostar . thus , we can expect that magnetic braking becomes gradually ineffective as the mass of the infalling envelope decreases , and it rarely works when the mass of the circumstellar disk is much larger than that of the infalling envelope . note that , in reality , magnetic braking can work even after the infalling envelope disappears , because the low - density interstellar medium can receive angular momentum from the circumstellar disk . also note that because the efficiency of magnetic braking decreases as the ambient density decreases , the angular momentum of the circumstellar disk is not effectively transferred into the low - density interstellar medium ( see , [ sec : diss ] ) . in addition , in an isolated cloud core , although magnetic braking can redistribute the angular momentum inside the cloud , it can not effectively transfer the angular momentum outside the cloud ( or into the interstellar medium ) . the infalling envelope receives the angular momentum from the circumstellar disk by magnetic braking , while it finally falls onto the circumstellar disk with the angular momentum by the end of the main accretion phase . however , when the infalling envelope is depleted and magnetic braking becomes ineffective , no sufficiently large disk appears if insufficient mass to form the disk remains in the infalling envelope . in summary , to investigate the real size and mass of the circumstellar disk , we have to investigate the disk evolution in the collapsing cloud core , at least till the end of the main accretion phase ( or until the infalling gas disappears ) . in previous studies , the calculation was stopped much before the infalling envelope disappears . thus , it is natural that magnetic braking is effective and no large - size disk appears during this phase ( i.e. , early main accretion phase ) . in this study , however , we calculated the evolution of the circumstellar disk until almost all gas of the infalling envelope disappears and found that a large - size disk comparable to observations appears when the infalling envelope becomes less massive than the circumstellar disk . the structure of this paper is as follows . the framework of our models and the numerical method are given in 2 . the numerical results are presented in 3 . we discuss the magnetic braking timescale and the effect of the infalling envelope on the circumstellar disk formation in 4 . to study the formation and evolution of a circumstellar disk in a magnetized molecular cloud core , we solve the three - dimensional resistive mhd equations including self - gravity : @xmath6 where @xmath7 , @xmath8 , @xmath9 , @xmath10 , @xmath11 , and @xmath12 denote density , velocity , pressure , magnetic flux density , resistivity , and gravitational potential , respectively . to mimic the thermal evolution calculated by @xcite , we adopt the piece wise polytropic equation of state ( see also @xcite ) as @xmath13 , \label{eq : eos}\ ] ] where @xmath14ms@xmath15 and @xmath16 ( @xmath17 ) . with equation ( [ eq : eos ] ) , the gas behaves isothermally for @xmath18 and adiabatically for @xmath19 . for a realistic evolution of the magnetic field in the circumstellar disk , we adopt the resistivity @xmath11 as the fiducial value in @xcite , in which the ohmic dissipation becomes effective for @xmath20 ( for details , see eqs . [ 9 ] and [ 10 ] , and fig . 1 of @xcite ) . as the initial state , we take a spherical cloud with a bonnor ebert ( be ) density profile , which is widely recognized as the density profile of a prestellar cloud @xcite . the initial cloud extends up to @xmath21 , where @xmath22 is the critical be radius . outside the initial cloud ( @xmath23 ) , a uniform density is adopted to mimic the static interstellar medium . we add nonaxisymmetric density perturbation of the m = 2 mode to the initial core ( for details , see @xcite ) . this perturbation ensures that the center of gravity is always located at the origin . we ignore gravity ( self - gravity and gravity of the protostar ) in the region of @xmath23 . in addition , we prohibit the gas inflow at @xmath24 in order to suppress mass input from the interstellar medium into the gravitationally collapsing cloud core . thus , only the gas inside @xmath25 collapses to form the circumstellar disk and protostar . note that although the protostellar outflow driven by the circumstellar disk propagates into the region of @xmath26 and disturbs the interstellar medium over time , we can safely calculate the mass accretion process onto the circumstellar disk because the outflow propagating into the interstellar medium does not affect the inner cloud region ( @xmath25 ) . for the be density profile , we adopt the central density as @xmath27 , @xmath28 ( fiducial model ) , and @xmath29 with an isothermal temperature of @xmath30k . to promote contraction , we increase the density by a factor of @xmath31=1.68 , where @xmath31 is the density enhancement factor that represents the stability of the initial cloud . with @xmath32 , the initial cloud has @xmath33 for @xmath34 and 0.42 for @xmath35 , where @xmath36 and @xmath37 are the thermal and gravitational energy , respectively . although the central density of the initial cloud has values of @xmath38 , @xmath39 and @xmath40 , we describe the initial cloud density without the density enhancement factor @xmath31 as @xmath41 in the following . note that because the critical be radius ( or initial cloud size ) is determined by the central density without @xmath31 , the initial density without @xmath31 is a more suitable parameter for describing the initial cloud . the density and enclosed mass against the initial cloud radius are plotted in figure [ fig:1 ] for the model with @xmath42 and @xmath29 . the density contrast between the center of the cloud ( @xmath43 ) and cloud boundary ( @xmath24 ) is @xmath44 . the cloud with a central density of @xmath45 ( fiducial model ) has a radius of @xmath46au and mass of @xmath47 . on the other hand , the cloud with @xmath48 and @xmath49 have a radius of @xmath50au and @xmath51au , and a mass of @xmath52 and @xmath53 , respectively . as seen in figure [ fig:1 ] , because we adopted the be density profile , the density has a nearly flat distribution around the center of the cloud , while it drops in proportion to @xmath54 in the outer envelope . the cloud rotation generates a rotating disk around the protostar as the cloud collapses . however , it is considered that the magnetic field suppresses the formation of the rotationally supported disk in the collapsing cloud , because the magnetic field effectively transfers the angular momentum outward from the center of the cloud . in this study , to investigate the disk formation in an initially rotating magnetized cloud , we parameterized the magnetic field strength and rotation rate of the initial cloud . we adopt a uniform magnetic field along the @xmath55-axis in the whole computational domain and rigid rotation along the @xmath55-axis in the region of @xmath35 . thus , the rotation axis is parallel to the magnetic field lines at the initial state . to specify the initial cloud , we use two parameters describing the magnetic field strength ( @xmath56 ) and rotation rate ( @xmath57 ) . they are normalized by the central quantities and described as @xmath58 where @xmath59 is the magnetic flux density , and @xmath60 where @xmath61 is the angular velocity at the center of the cloud . model names and their parameters are listed in table [ table:1 ] . we adopt @xmath62 0.1 , 0.3 , and 1 , and @xmath63 , 0.1 and 0.15 , as described in table [ table:1 ] . in this study , we investigate disk formation until almost all the mass of the initial cloud falls onto either the circumstellar disk or the protostar . note that a part of the cloud mass is ejected from the collapsing cloud by protostellar outflow . thus , quantities describing the whole cloud may be useful to specify our models . the parameter @xmath56 ( or the magnetic flux density at the center of the cloud ) can be connected to the mass - to - flux ratio of the whole cloud . the mass - to - flux ratio is described as @xmath64 where @xmath65 is the mass contained within the radius @xmath66 and @xmath67 is the magnetic flux threading the cloud . there exists a critical value of @xmath68 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 @xmath69 where the constant @xmath70 @xcite . the mass - to - flux ratio normalized by the critical value @xmath71 is described as @xmath72 the mass - to - flux ratios normalized by the critical value for different models ( or different @xmath56 ) are plotted against the enclosed mass ( _ left _ ) and the radius ( _ right _ ) of the initial cloud in the upper panels of figure [ fig:2 ] . reflecting the be density profile , the mass - to - flux ratio has a radial distribution inside the initial cloud . for example , with @xmath73 , the mass - to - flux ratio has a peak at @xmath74 ( @xmath75 ) or @xmath76au ( @xmath77au ) for the model with @xmath78 ( @xmath79 ) . in addition , with @xmath73 and @xmath80 , the initial cloud has a mass - to - flux ratio of @xmath81 as the whole . hereafter , for simplicity , we describe the mass - to - flux ratio of the whole cloud ( @xmath35 ) as @xmath71 ( @xmath82)$ ] , and call it the mass - to - flux ratio . as listed in table [ table:1 ] , we adopt @xmath83 , 2 , 3 ( magnetized model ) and @xmath84 ( unmagnetized model ) . 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 . thus , the mass - to - flux ratios adopted in this paper are within a reasonable range . as the parameter for the rotation of the whole cloud , we define the initial ratio of the rotational to the gravitational energy as @xmath87 where @xmath88 is the rotational energy of the initial cloud in the region of @xmath35 . as listed in table [ table:1 ] , models have the parameter in the range of @xmath89 . observations showed that molecular cloud cores have the rotational energy in the range of @xmath90 with a median value of @xmath91 @xcite . thus , the rotational energies in our models are also within reasonable ranges , compared with the observations . in addition , we also listed the centrifugal radius of each model in table [ table:1 ] . the centrifugal radius is defined as @xmath92 where @xmath93 is the specific angular momentum . the centrifugal radii for models with different @xmath57 or different @xmath94 are plotted against the enclosed mass ( _ left _ ) and the radius ( _ right _ ) of the initial cloud in the lower panels of figure [ fig:2 ] . in the figure , the centrifugal radius gradually increases as the enclosed mass increases , because the gas far from the center of the cloud has a large specific angular momentum . as listed in table [ table:1 ] , models have the centrifugal radius in the range of @xmath95 as the whole cloud ( @xmath35 ) . thus , when the angular momentum is well conserved in the collapsing cloud , a disk with a size of @xmath96au can form around the protostar . for calculations on a large spatial scale , the nested grid method is adopted ( for details , see @xcite ) . each level of a rectangular grid has the same number of cells ( @xmath97 ) . the calculation is first performed with five grid levels ( @xmath98 ) . the box size of the coarsest grid @xmath99 is chosen to be @xmath100 , which corresponds to @xmath101au ( @xmath102 ) , @xmath103au ( @xmath104 ) and @xmath105au ( @xmath106 ) . a new finer grid is generated before the jeans condition is violated @xcite . for model with @xmath107 , the maximum level of grids is @xmath108 , which has a box size of @xmath109au and a cell width of @xmath110au . models with different cloud sizes ( or different initial central densities ) have almost the same spatial resolution ( @xmath111au ) of the maximum level grid , in which the maximum grid level is different in models with different initial central density . as described in [ sec : initial ] , we adopt gravity only in the region of @xmath35 to mimic the star - forming cloud core embedded in the interstellar medium , while a large domain ( @xmath112 , interstellar medium ) is left behind from the gravitational contraction . thus , in our setting , the star forming cloud with a size of 5700au ( @xmath113au , 1740au ) exists in the interstellar medium with a box size of @xmath103au ( @xmath101au , @xmath105au ) for a model with @xmath104 ( @xmath114 , @xmath115 ) . the purpose of this study is to investigate the effect of magnetic braking on disk formation . the angular momentum is transferred by magnetic braking with the alfv@xmath5n speed . to suppress artificial reflection of alfv@xmath5n waves at the computational boundary , we adopt a large simulation box to reproduce a realistic interstellar medium around the star forming cloud . the alfv@xmath5n speed in the interstellar medium ( @xmath112 ) for the model with the strongest magnetic field ( @xmath116 g ) and @xmath117 is @xmath118kms@xmath15 . thus , for the alfv@xmath5n wave to reach the computational boundary from the center of the cloud , it takes @xmath119yr ( @xmath120 ) , which corresponds to about 30 times the free - fall timescale ( @xmath121yr ) , where the freefall timescale @xmath122 is normalized by the central density in the initial cloud . in all models , we stopped the calculation within @xmath123 . thus , the alfv@xmath5n wave generated at the center of the cloud ( or the computational boundary ) never reaches the computational boundary ( or the center of the cloud ) . with this setting , we can correctly treat magnetic braking in the collapsing cloud . in the collapsing cloud core , we assume the protostar formation to occur when the number density exceeds @xmath124 at the center of the cloud . to model the protostar , we adopt a sink around the center of the computational domain . in the region of @xmath125au , gas having a number density @xmath124 is removed from the computational domain and added to the protostar as gravity in each timestep ( for details , see @xcite ) . this treatment of the sink makes it possible to calculate the evolution of the collapsing cloud and circumstellar disk for a longer duration . in addition , inside the sink , the magnetic flux is removed by ohmic dissipation , because such a region has a magnetic reynolds @xmath126 number exceeding unity @xmath127 ( for details , see @xcite ) . to understand the process of formation of a circumstellar disk in a magnetized cloud , we prepared nine different initial clouds with various masses , radii , magnetic field strengths , and angular velocities as listed in table [ table:1 ] . these clouds have masses of @xmath128 and radii of @xmath129au . the magnetic field strengths @xmath130 and angular velocities @xmath131 at the center of the initial cloud are in the range of @xmath132 g and @xmath133s@xmath15 , respectively . in the following subsections , after we describe the disk formation for a typical model ( model 1 ) , we compare the disk formation in models with different initial settings . in this subsection , we describe the disk formation for model 1 , which has the parameter of @xmath83 . figure [ fig:3 ] shows the time sequence of the density distribution around the center of the collapsing cloud on the equatorial plane for model 1 . the elapsed time in units of the initial free - fall timescale @xmath122 and year after the calculation begins are described in each panel . the protostellar , circumstellar disk , and envelope masses ( @xmath134 , @xmath135 , and @xmath136 , respectively ) are also described in each panel . the protostellar mass @xmath134 is defined as the mass falling into the sink . to determine the circumstellar disk in the infalling envelope , we estimated the velocity ratio @xmath137 ( @xmath138 , where @xmath139 and @xmath140 are the radial and azimuthal velocity , respectively ) in each cell , and specified the most distant cell having @xmath141 from the center of the cloud . then , we defined the disk radius @xmath142 as the distance of the cell furthest from the origin , and the disk boundary density @xmath143 as the density of the most distant cell . finally , we defined the circumstellar disk as a region that has a density of @xmath144 and estimated the circumstellar mass @xmath135 . @xcite showed that the keplerian rotating disk is well identified with this criterion . in addition , in each timestep , we estimated the total mass @xmath145 that is sum of the total mass inside the gravitational sphere inside @xmath35 and the protostellar mass . then , we subtracted the circumstellar disk mass and protostellar mass from the total mass and derived the mass of the infalling envelope as @xmath146 we confirmed that the sum of the protostellar mass and the total mass is well conserved during the calculation . note that , in the later main accretion phase , because the mass is ejected from the collapsing cloud by the protostellar outflow , mass conservation inside @xmath35 is not realized . figure [ fig:3 ] shows that the circumstellar disk gradually increases its size with time . figure [ fig:3]_a _ indicates that the circumstellar disk has a size of @xmath3au in the early main accretion phase , as described in @xcite . at this epoch , the circumstellar disk has a mass @xmath147 , which is about twice the protostellar mass ( @xmath148 ) . @xcite showed that the circumstellar disk mass is greater than the protostellar mass in the early main accretion phase , because the first adiabatic core , which is formed before the protostar formation with a mass of @xmath149 , directly evolves into the circumstellar disk after the protostar formation . in addition , at the same epoch , the mass of the infalling envelope is much larger than that of the circumstellar disk ( @xmath150 ) . in figure [ fig:4 ] , the radial distribution of the azimuthal velocity normalized by the keplerian velocity ( _ a _ ) , ratio of radial to azimuthal velocity ( _ b _ ) , surface density ( _ c _ ) and plasma beta ( _ d _ ) at each epoch corresponding to each panel in figure [ fig:3 ] are plotted . the azimuthal velocity and ratio of the radial to azimuthal velocity are azimuthally averaged on the equatorial plane . the surface density is estimated as @xmath151 where @xmath152au is adopted , and azimuthally averaged . note that the surface density depends only a little on @xmath153 when @xmath154au ( and/or @xmath155au ) , because the disk mass is concentrated on the equatorial plane . the plasma beta is defined in each cell on the equatorial plane as @xmath156 where @xmath157 and @xmath9 are the magnetic flux density and thermal pressure on each cell , respectively , which are also azimuthally averaged on the equatorial plane . at the same epoch as figure [ fig:3]_a _ ( @xmath158yr , black lines in fig . [ fig:4 ] ) , the disk has a keplerian velocity in the range of @xmath159au ( fig . [ fig:4]_a _ ) , while the disk structure extends up to @xmath160au ( fig . [ fig:4]_c _ ) . in the range of @xmath161 , the disk has @xmath162 of the keplerian velocity . figure [ fig:4]_b _ shows that the azimuthal velocity is greater than the radial velocity in the range of @xmath163au . thus , the disk is mainly supported by the centrifugal force in the range of @xmath164au , while it is also supported by the thermal pressure gradient and lorentz force in addition to the centrifugal force in the range of @xmath165 . the black line in figure [ fig:4]_d _ shows the rapid increase of the plasma beta at @xmath166au . because the thermal pressure increases in the high - density region , the plasma beta tends to increase inside the disk . in addition , the magnetic field dissipates by ohmic dissipation in such a high - density region @xcite . figures [ fig:3 ] and [ fig:4 ] indicate that the circumstellar disk increases its size with time . at @xmath167yr ( fig . [ fig:3]_b _ and red lines in fig . [ fig:4 ] ) , the circumstellar disk extends up to @xmath168au and has a mass of @xmath169 . figure [ fig:4]_a _ shows that the disk rotates with a keplerian velocity in the range of @xmath170au at this epoch . the radial distribution of the surface density indicates that the disk extends up to @xmath171au . in addition , the plasma beta rapidly increases at @xmath172au and reaches @xmath173 near the origin . thus , the whole region of the disk ( @xmath174au ) is mainly supported by the centrifugal force . in other words , the keplerian rotating disk extends up to @xmath168au at this epoch . the circumstellar disk mass becomes comparable to the mass of the infalling envelope at @xmath175yr , which corresponds to @xmath176 ( fig . [ fig:3]_c _ and blue lines in fig . [ fig:4 ] ) . at the initial state , the density at the cloud boundary ( i.e. , @xmath24 ) is 82 times lower than that at the center of the cloud ( @xmath43 ) . thus , the freefall timescale at the cloud boundary @xmath177 is about nine times longer than that at center of the cloud @xmath122 ( i.e. , @xmath178 ) . thus , we can expect that a large fraction of the initial cloud mass has already fallen onto the central region at @xmath179 . note that the cloud rotation and magnetic field slightly slows the cloud collapse and accretion onto the disk ( e.g. , @xcite ) . figure [ fig:3]_c _ shows that the size of the circumstellar disk exceeds 100au and the spiral structure appears in the disk at @xmath180 . the blue lines in figure [ fig:4 ] also indicate that the keplerian rotating disk extends up to @xmath181au at this epoch . at @xmath182 ( fig . [ fig:3]_d _ and green lines in fig . [ fig:4 ] ) , the circumstellar disk mass ( @xmath183 ) is much greater than the mass of the infalling envelope ( @xmath184 ) . thus , as described in [ sec : intro ] , it is expected that magnetic braking is not effective at this epoch ( for details , see [ sec : diss ] ) . the green lines in figure [ fig:4 ] indicates that the circumstellar disk with a keplerian velocity has a size of @xmath185au ( see fig . [ fig:4]_c _ ) at this epoch . thus , the circumstellar disk exceeding @xmath2au can form even in a strongly magnetized cloud core . figure [ fig:5 ] shows the time sequence of the density distribution for model 1 at the same epochs as in figure [ fig:3 ] but for the @xmath186 plane . the outflow is driven near the circumstellar disk , as seen in figure [ fig:5]_a_-_c_. this type of outflow was seen in many previous studies ( e.g. , @xcite ) . in each panel , the keplerian rotation disk is indicated by the black circle , inside which the gas rotates with a ( nearly ) keplerian speed . the figure shows that the high - density keplerian rotating disk ( black line ) is enclosed by a low - density pseudo disk . the pseudo - disk is formed by both the cloud rotation and magnetic field in the gas collapsing phase @xcite . however , the pseudo - disk gradually disappears in the main accretion phase , because its rotation speed does not reach the keplerian velocity . figures [ fig:5]_a_-_c _ show that the driving point of the outflow radially expands with time , because the keplerian rotating disk , which drives the outflow , extends in the radial direction with time . the outflow disappears in @xmath187yr after the cloud collapse begins . in figure [ fig:5]_d _ , there is a sufficiently thin keplerian rotation disk with a size of @xmath185au , while no outflow is driven by the circumstellar disk at this epoch . this is because by this epoch almost all the cloud mass already has fallen onto either the circumstellar disk or the protostar . the outflow is powered by the gas accreting energy onto the circumstellar disk . to investigate the evolution of the infalling envelope , the density distribution on the @xmath186 plane at the same epochs as in figures [ fig:3 ] and [ fig:5 ] but for the initial cloud scale ( @xmath188au ) is plotted in figure [ fig:6 ] . in each panel , the boundary between the gravitational sphere and static interstellar medium ( @xmath24 ) is plotted by a white broken line ; gas only inside the boundary can fall toward the center of the cloud . the figure shows that the cloud density at a larger radius gradually decreases with time , and the gas is concentrated onto the equatorial plane . the gas is also distributed along the @xmath55-axis in figures [ fig:6]_b _ and _ c_. this mass distribution is caused by the mass ejection from the circumstellar disk by the protostellar outflow . figure [ fig:6]_d _ shows that gas in the infalling envelope is depleted and the mass ejection by the protostellar outflow has already stopped . it is natural that the elapsed time at figure [ fig:6]_d _ ( @xmath189 ) is longer than the freefall timescale ( @xmath190 ) at the cloud boundary , and thus the gas located near the cloud boundary at the initial state has already fallen onto the circumstellar disk by this epoch . it is expected that , at this epoch , magnetic braking becomes ineffective because the ( less massive ) infalling envelope can not brake the circumstellar disk . therefore , a large - size keplerian rotating disk can appear at this epoch . note that we may overestimate the mass of the infalling envelope , because we defined it as the mass of the whole cloud except for the keplerian rotating disk . thus , the infalling envelope includes the pseudo disk in our definition . the infalling envelope has a mass of @xmath184 ( i.e. , @xmath191 of the initial cloud mass ) at @xmath192yr , as described in figure [ fig:3]_d_. in reality , however , a very small amount of the infalling gas remains in the region above the circumstellar disk as seen in figure [ fig:6]_d_. the circumstellar disk is connected to the infalling envelope through the magnetic field lines . thus , a slowly rotating massive infalling envelope can brake a rapidly rotating less massive disk through the magnetic field line . thus , the magnetic field lines are strongly twisted when the infalling envelope is more massive than the circumstellar disk or when magnetic braking is effective . the difference between the rotation speeds of the infalling envelope and the circumstellar disk generates a strong toroidal field . instead , when the infalling envelope is much less massive than the circumstellar disk , the infalling envelope can not brake the circumstellar disk , and a strong toroidal field is not generated . in such a case , it is expected that the magnetic field has a nearly force free configuration . figure [ fig:7 ] shows the ratio of the toroidal field to the poloidal field ( @xmath193 ) at the same epochs as in figure [ fig:3 ] . the toroidal field is greater than the poloidal field ( toroidal dominated region ; @xmath194 ) inside the black line . figure [ fig:7]_a _ shows that the toroidal dominated region is distributed along the @xmath55-axis in the early main accretion phase , in which the torsional alfv@xmath5n wave generated by the rotation of the circumstellar disk is propagated along the @xmath55-axis . in figure [ fig:7]_b _ , the toroidal field is much greater than the poloidal field in the region above the circumstellar disk ( i.e. , in the region of @xmath195au , where @xmath196 is the cylindrical radius ) . this indicates that magnetic braking is effective and the angular momentum of the circumstellar disk is transferred into the infalling envelope . at this epoch , strong mass ejection from the circumstellar disk also takes place as seen in figure [ fig:6]_b _ , resulting from the generation of the strong toroidal field . the toroidal field weakens for @xmath197 as seen in figure [ fig:7]_c_. in figure [ fig:7]_d _ , the poloidal field dominates the toroidal field in the whole region of the infalling envelope . by this epoch , a large fraction of the mass in the initial cloud has already fallen onto either the circumstellar disk or the protostar . at this epoch , the mass of the infalling envelope is much less than the circumstellar disk mass ( fig . [ fig:6]_d _ ) . thus , it is expected that , at this epoch , the magnetic field is rarely twisted and magnetic braking is not so effective . figure [ fig:8 ] shows the radial distribution of the angular momentum for model 1 at the same epochs as in figure [ fig:3 ] against the radially integrated mass . for reference , the distribution of the angular momentum for unmagnetized model ( model 5 ) at almost the same epoch as figure [ fig:3]_a _ is also plotted in this figure . at each radius , the angular velocity and mass are integrated in the gravitationally collapsing cloud ( @xmath35 ) as @xmath198 and @xmath199 where the protostellar mass @xmath134 falling into the sink is added to estimate the mass . note that , in figure [ fig:8 ] , the cloud mass does not reach @xmath200 except for unmagnetized model ( dotted line ) because some fraction of the cloud mass ( @xmath3% ) is ejected from the collapsing cloud ( or from the region of @xmath201 ) by the protostellar outflow . the figure indicates that the angular momentum of the whole cloud ( @xmath202 ) decreases gradually with time for model 1 , while it is well conserved in the whole cloud for model 5 . for model 1 , the angular momentum of the cloud can be transferred into the interstellar medium by magnetic braking . however , the rate of reduction of the angular momentum is not very large during the calculation . the initial cloud has an angular momentum of @xmath203 , while the collapsing cloud has @xmath204 as the whole at the end of the calculation . thus , about half of the angular momentum in the initial cloud is transferred into the interstellar medium . this is natural because the timescale of magnetic braking by the interstellar medium is longer than our calculation time ( discussed in [ sec : diss ] ) . in addition , because the protostellar outflow driven by the circumstellar disk propagates into the interstellar medium with a certain amount of angular momentum , the protostellar outflow also transfers the angular momentum into the interstellar medium . figure [ fig:8 ] also indicates that the angular momentum inside the disk ( i.e. , the region on the left side of the diamond symbol ) is considerably smaller than that at the initial state . this difference of the angular momentum between the initial state and collapsing cloud clearly indicates that the angular momentum is effectively transferred outward in the collapsing cloud . in the early main accretion phase , the angular momentum is expected to be transferred by magnetic effects such as magnetic braking and protostellar outflow , since the circumstellar disk has an axisymmetric structure and remains clear axisymmetricity ( figs . [ fig:3]_a _ and _ b _ ) , and hence , gravitational torque seems to be less important in magnetized models . in the later main accretion phase , since non axisymmetric patterns appear in the circumstellar disk ( figs . [ fig:3]_c _ and _ d _ ) , the angular momentum is expected to be transferred also by gravitational torque due to the non axisymmetric perturbation , in addition to the magnetic effects . in [ sec : typical ] , we investigated the disk formation for a typical model ( model 1 ) . however , it is expected that the formation process of the circumstellar disk and its size and mass depend on the initial cloud properties such as the magnetic field strength , rotation rate , and cloud mass ( or size ) . in this subsection , we show the disk formation in clouds with different parameters . to investigate the disk formation in clouds with initially different magnetic field strengths , we plotted the disk radii ( _ a _ ) , disk and protostellar masses ( _ b _ ) , disk - to - envelope mass ratios ( _ c _ ) and the masses of the infalling envelope ( _ d _ ) for models 1 , 3 , 4 and 5 against the elapsed time @xmath205 in figure [ fig:9 ] . the disk is defined as the region of @xmath206 ( see , [ sec : typical ] ) . thus , in our definition , the disk ( or the circumstellar disk ) has a ( nearly ) keplerian rotation . note that the elapsed time @xmath207 , which is defined as the elapsed time after the circumstellar disk formation , differs from @xmath208 , which is defined as the elapsed time after the cloud collapse begins ( or the calculation begins ) . they are related as @xmath209 where @xmath210 is the time from the beginning of the calculation until the disk formation , and is listed in table [ table:2 ] . in this subsection , we use @xmath207 in place of @xmath208 to compare the disk evolution from the time just after the disk formation for different models . for models in figure [ fig:9 ] , model 3 has a three times larger initial mass - to - flux ratio ( @xmath211 ) than model 1 . model 4 has the same mass - to - flux ratio ( @xmath83 ) as model 1 , but the cloud evolution is calculated with the ideal mhd approximation . the initially unmagnetized cloud ( @xmath212 ) is adopted for model 5 . the elapsed time from the beginning of the calculation ( @xmath208 ) , disk radius ( @xmath213 ) , mass of the circumstellar disk ( @xmath135 ) in each model at the end of the calculation are also listed in table [ table:2 ] . figure [ fig:9]_a _ indicates that the disk radius for an unmagnetized cloud ( model 5 ) is much larger than the radii for magnetized clouds ( models 1 , 3 , 4 ) . in other words , in a magnetized cloud , the circumstellar disk is a relatively smaller that that in an unmagnetized cloud . this indicates that the magnetic field suppresses the disk formation , as already described in past studies @xcite . at @xmath214yr , which almost corresponds to the freefall timescale at the cloud boundary , the circumstellar disk for model 5 reaches @xmath215au , which corresponds to about 20% of the centrifugal radius of the whole cloud ( see , table [ table:1 ] ) . for model 5 , the increase and decrease of the disk radius for @xmath216yr is attributed to the development of the spiral structure . as seen in figure [ fig:9]_b _ , because the circumstellar disk mass is greater than the protostellar mass in the main accretion phase , the disk is unstable against its gravity and frequently shows a spiral structure @xcite . the spiral structure effectively transfers the angular momentum outward and promotes a rapid mass accretion from the circumstellar disk onto the protostar . after the rapid mass accretion , the mass and size of circumstellar disk temporally decreases and the spiral structure weakens . for model 5 , the circumstellar disk mass is greater than the mass of the infalling envelope for @xmath217yr ( fig . [ fig:9]_c _ ) . in addition , over 90% of the initial cloud mass already falls either onto the circumstellar disk or protostar at @xmath218yr ( fig . [ fig:9]_d _ ) . thus , it is expected that the main accretion phase almost ends by this epoch , and the circumstellar disk grows only a little by gas accretion after this epoch . on the other hand , at @xmath219yr , the disk radii in magnetized clouds are about @xmath220au , which is considerably smaller than that of the unmagnetized cloud ( @xmath215au ) . the large difference of disk radii between magnetized and unmagnetized models indicates that a large fraction of the angular momentum is transferred by magnetic effects . the difference of disk radii is caused by the different efficiencies of the angular momentum transfer among models . in figure [ fig:9]_a _ , resistive mhd models ( models 1 and 3 ) show a similar trend for the evolution of the disk radius , in which model 3 has a three times larger mass - to - flux ratio than model 1 at the initial state . for these models , the disk radii gradually increases with time for @xmath221yr . reflecting the difference in the initial mass - to - flux ratio , at the same epoch during the early main accretion phase [ @xmath221yr ] , model 3 which has an initially weaker magnetic field , has a slightly larger disk than model 1 . then , for @xmath222yr , the disk radii for these models increase with strong oscillation , which is caused by the spiral structure appearing in the circumstellar disk . figure [ fig:9]_b _ shows that the circumstellar disk is more massive than the protostar by the end of the calculation even in magnetized clouds . in addition , the protostellar mass increases intermittently for @xmath223yr . this is because the spiral structure , which appears in the circumstellar disk for @xmath224yr ( see , fig . [ fig:3]_c _ and _ d _ ) , unsteadily transfers the angular momentum outward and the mass accretion onto the protostar is temporarily amplified . thus , during this phase ( @xmath225yr ) , the angular momentum is transferred not only by the magnetic effects but also by the non axisymmetric perturbation in the circumstellar disk . finally , in resistive models ( models 1 and 3 ) , the disk radius rapidly increases for @xmath226yr and reaches @xmath227au at the end of the calculation . figure [ fig:9]_c _ shows that the circumstellar disk mass is greater than the mass of the infalling envelope for @xmath228yr . note that since the infalling envelope includes the pseudo disk in our definition as described in [ sec : typical ] , we somewhat overestimate the actual infalling envelope in figure [ fig:9]_c_. after this epoch ( @xmath229yr ) , it is expected that magnetic braking becomes ineffective , because the less massive infalling envelope can no longer brake the circumstellar disk . thus , the circumstellar disk can rapidly grow without effective angular momentum transfer by the magnetic effects . as shown in figure [ fig:9]_d _ , for models 1 and 3 , over 80% of the initial cloud mass already falls either onto the circumstellar disk or the protostar by the end of the calculation . note that a part of the initial cloud mass is ejected from the gravitationally collapsing cloud by the protostellar outflow as shown in figure [ fig:6 ] . at the end of the calculation , the circumstellar disk has a size of 290 au for model 1 and 370 au for model 3 . therefore , although the disk size in resistive mhd models is considerably smaller than that in the unmagnetized model , the keplerian rotation disk with a size of @xmath230au can form even in strongly magnetized cloud . figure [ fig:9]_a _ shows that the disk radius in the ideal mhd model ( model 4 ) is somewhat smaller than that in the resistive mhd model ( model 1 ) with the same parameters for the initial cloud . at end of the calculation ( @xmath231yr ) , the circumstellar disk has a size of @xmath232au for model 1 and @xmath233au for model 4 . magnetic braking is more effective in the ideal mhd model than in the resistive mhd model , because the magnetic field largely dissipates inside the circumstellar disk in the resistive mhd model . thus , it is considered that the difference of the disk size between the ideal and resistive mhd models is caused by the difference in the strengths of the magnetic field in the circumstellar disk . figure [ fig:10 ] shows the plasma beta for models 1 ( upper panel ) and 4 ( lower panel ) on the @xmath186 plane at almost the same epoch after the calculation begins . the contour of @xmath234 is shown by a white thick line , outside of which the magnetic energy is greater than the thermal energy . the circumstellar disk exists in the region of @xmath235 ( i.e. , inside the white line ) . because the circumstellar disk has a high density ( or large thermal pressure ) , the thermal energy is greater than the magnetic energy ( i.e. , @xmath236 ) inside the circumstellar disk . figure [ fig:10 ] shows that , for model 1 , the plasma beta is extremely high as @xmath237 in the region of @xmath238au inside the circumstellar disk , because this region has a density of @xmath239 ( see fig . [ fig:5]_c _ ) and the ohmic dissipation is effective . figure [ fig:10 ] lower panel shows that the circumstellar disk for the ideal mhd model is smaller than that for the resistive mhd model . in the circumstellar disk , the plasma beta for the ideal mhd model is about two or three orders of magnitude smaller than that for the resistive mhd model . despite the large difference in the magnetic energy in the circumstellar disk , the difference in the final size of the circumstellar disk is not very large : the disk has a size of @xmath232au for model 1 , and @xmath233au for model 4 . in addition , the evolution of the masses of circumstellar disk , protostar , and infalling envelope for the ideal mhd model is almost the same as that for the resistive mhd model , as shown in figure [ fig:9]_b _ and _ c_. thus , even in the ideal mhd model , magnetic braking becomes ineffective as the mass of the infalling envelope is depleted , and a circumstellar disk with a size of @xmath240au can form . to investigate the effect of the initial cloud rotation on the circumstellar disk formation , the disk radii and circumstellar disk and protostellar masses for models 1 , 6 and 7 are plotted in figure [ fig:11 ] . model 6 has three times the initial angular velocity as model 1 , while model 7 has one third the initial angular velocity as model 1 . figure [ fig:11 ] upper panel shows that at the same epoch after the circumstellar disk formation , the model with rapid initial rotation has a relatively large size disk . in addition , the disk size begins to oscillate from the earlier evolution phase in the cloud with faster rotation . for example , model 6 begins to show an oscillation at @xmath241yr , while model 7 shows it for @xmath242yr . the size oscillation occurs after the spiral structure develops in the circumstellar disk . the spiral structure appears after the circumstellar disk grows sufficiently in size and mass @xcite . figure [ fig:12 ] shows the density distribution for models 6 and 7 on the equatorial plane after the circumstellar disk becomes more massive than the infalling envelope . the figure indicates that the circumstellar disks show a clear spiral structure with a size of @xmath243au . figure [ fig:11 ] lower panel shows that the cloud with faster initial rotation creates a more massive circumstellar disk . thus , in such a cloud , the spiral structure and size oscillation appear in an earlier phase of the disk evolution . in addition , the cloud with faster initial rotation has a less massive protostar before the size oscillation because mass accretion from the circumstellar disk onto the protostar is suppressed by a strong centrifugal force in such a model . after the size oscillation occurs , the protostar unsteadily acquires its mass from the circumstellar disk , as described in [ sec : typical ] . finally , the disk radius in each model increases rapidly for @xmath244yr to reach @xmath227au ( see , table [ table:2 ] ) . this is because , by this epoch , the circumstellar disk mass becomes greater than the protostellar mass , and magnetic braking becomes ineffective . figure [ fig:13 ] plots the disk radius and mass of the infalling envelope normalized by the initial cloud mass for models 1 , 8 , and 9 against the elapsed time after the circumstellar disk formation . each of these models has a different initial cloud mass but the same @xmath71 and @xmath245 . the figure shows that the initially less massive cloud has a larger - size disk at the same epoch after the circumstellar disk formation . the disk radii for models 1 and 8 rapidly increase after the mass of the infalling envelope deceases to @xmath246% of the initial cloud mass . for model 8 , which has an initial cloud mass of @xmath247 , the mass of the infalling envelope normalized by the initial cloud mass decreases to @xmath248 at @xmath249yr , and the circumstellar disk radius rapidly increases for @xmath224yr and reaches @xmath250au by the end of the calculation . on the other hand , for model 9 , which has an initial cloud mass of @xmath52 , the size of circumstellar disk does not exceed @xmath251au by the end of the calculation ( @xmath218yr ) . thus , the disk size for model 9 is less than one - tenth of that for model 1 at @xmath218yr . note that , for model 9 , because the freefall timescale at the cloud boundary is as large as @xmath252yr , we can not calculate the cloud evolution until the infalling envelope is depleted . it is expected that the difference in the disk size for models with different initial cloud mass is caused by the remaining mass of the infalling envelope . as described in [ sec : intro ] , the massive infalling envelope brakes the circumstellar disk through the magnetic field lines . the initial cloud mass and freefall timescale at the center of the cloud or at the cloud boundary are different in models 1 , 8 , and 9 . thus , at the same epoch , a model with an initially massive cloud has a massive infalling envelope that can brake the circumstellar disk and suppresses the growth of the circumstellar disk by magnetic braking . it is difficult to quantitatively relate the disk size and the mass of the infalling envelope among these models , because the initial distribution of the mass - to - flux ratio ( see fig . [ fig:2 ] ) and angular velocities ( see table [ table:1 ] ) are different in these models . however , figure [ fig:1 ] indicates that after the mass of the infalling envelope decreases sufficiently , the effect of magnetic braking weakens and a large - size circumstellar disk can form even in a strongly magnetized cloud . conversely , large - disk formation is suppressed by magnetic braking when a massive infalling envelope remains around the protostar . in this paper , to investigate the formation of the circumstellar disk in a realistic setting , we set a dense molecular cloud core embedded in a low - density interstellar medium in a large simulation box with size 16 times the radius of the molecular cloud core , in which a uniform magnetic field parallel to the rotation axis is adopted . in the gravitationally collapsing cloud , it is expected that the disk formation is suppressed by magnetic braking which effectively transfers the angular momentum from the circumstellar disk , as described in past studies @xcite . in reality , our results indicate that the circumstellar disk formed in a magnetized cloud is much smaller than that in an unmagnetized cloud ( see [ sec : mags ] ) . however , we also showed that the circumstellar disk exceeding @xmath240au can form even in a strongly magnetized cloud ( @xmath253 ) , which contradicts previous studies . the past studies predicted that magnetic braking suppresses the disk formation and limits the size of the circumstellar disk within @xmath3au . this inconsistency is attributed to the initial setting of the host cloud and the calculation time after the circumstellar disk formation . in this study , we calculated the cloud evolution until almost all gas inside the initial cloud falls onto either the circumstellar disk or the protostar . in other words , we execute the calculation until the freefall timescale at the cloud boundary is exceeded . as described in [ sec : intro ] , as the cloud mass is depleted , magnetic braking is expected to be ineffective because the less massive infalling envelope can not brake the circumstellar disk . in [ sec : results ] , we numerically showed that magnetic braking weakens with time and a large - size circumstellar disk can form . in this section , we analytically discuss the timescale of magnetic braking and disk formation . the angular momentum of the cloud or disk is transferred into the external medium through a torsional alfv@xmath5n wave . magnetic braking becomes effective when the angular momentum transferred into the external medium is comparable to that of the rotator ( e.g. , the molecular cloud or disk ) . in other words , magnetic braking becomes effective when the moment of inertia of the gas through which the alfv@xmath5n waves have propagated becomes comparable to the moment of inertia of the cloud or disk . with this condition , the magnetic braking timescale can be described as @xmath254 where @xmath255 , @xmath256 , @xmath257 and @xmath258 are the half thickness of the cloud ( or the disk ) , alfv@xmath5n velocity in the external medium , cloud ( or disk ) density , and density of the external medium , respectively @xcite . thus , magnetic braking becomes effective when a typical timescale of the system ( @xmath259 ) is longer than the magnetic braking timescale @xmath260 . with equation ( [ eq : mb ] ) , the condition for magnetic braking becoming effective is described as @xmath261 the left - hand side of equation ( [ eq : mb2 ] ) corresponds to the line density of the external medium in the range at which the alfv@xmath5n wave can reach in @xmath259 , while the right - hand side corresponds to the line density of the cloud or disk . equation ( [ eq : mb2 ] ) indicates that magnetic braking is effective only when the mass of the external medium is greater than the mass of the rotator . in other words , magnetic braking does not take place when sufficient mass does not exit around the rotator . this is natural because a less massive external medium can not brake a massive object . this condition can be also understood from equation ( [ eq : mb ] ) . when an uniform magnetic field is assumed , equation ( [ eq : mb ] ) indicates that the magnetic braking timescale is proportional to @xmath262 . note that the alfv@xmath5n speed is proportional to @xmath263 . thus , the magnetic braking timescale becomes longer as the external density decreases . for example , when the external density is extremely low , magnetic braking does not take place . we estimate the magnetic braking timescale for model 1 ( initially the strongest magnetic field model ) . first , we estimate the magnetic - braking timescale of the molecular cloud core embedded in the interstellar medium . for model 1 , the density contrast between the center of the cloud core and the external medium is 82 at the initial state , and alfv@xmath5n speed in the external mediums is @xmath264kms@xmath15 . thus , with equation ( [ eq : mb ] ) , the magnetic braking timescale becomes @xmath265yr , which is much longer than the freefall timescale of the cloud ( @xmath266yr ) or our calculation time ( @xmath267yr ) . thus , it is expected that the cloud rarely loses its angular momentum by magnetic braking of the interstellar medium . in [ sec : typical ] , we showed that the total angular momentum of the cloud core is rarely decreased through the calculation . next , we estimate the magnetic - braking timescale of the circumstellar disk . first , we assume that the circumstellar disk exists only in the low - density interstellar medium ( thus , not including the infalling envelope ) to simply estimate the effect of the low - density interstellar medium outside the gravitational sphere . the density contrast between the circumstellar disk ( @xmath268 ) and interstellar medium ( @xmath269 ) is about @xmath270 , and the circumstellar disk has a scale height of @xmath271au . with these parameters , the magnetic braking timescale of the circumstellar disk is @xmath272yr . thus , the interstellar medium can not brake the circumstellar disk in our setting . note that , in this study , we adopted a somewhat higher density of the interstellar medium ( @xmath269 ) than the observed value of the interstellar medium to limit the alfv@xmath5n speed in the interstellar medium , because it is difficult to calculate the cloud evolution over a long time with a large alfv@xmath5n speed . thus , the magnetic braking timescale in the actual interstellar medium is longer than in our setting with a lower external density , because it is proportional to @xmath263 . in summary , magnetic braking from the low - density interstellar medium is completely ignored in the formation and evolution of the circumstellar disk . on the other hand , the infalling envelope in the region of @xmath273 can brake the circumstellar disk in the main accretion phase . the torsional alfv@xmath5n wave generated by the center of the cloud can reach the cloud boundary ( @xmath274au ) in @xmath275yr ( @xmath276 ) , where the alfv@xmath5n speed of the external medium @xmath264kms@xmath15 is adopted . note that , in reality , because the magnetic field is amplified , the torsional alfv@xmath5n wave can reach the cloud boundary in several freefall timescales . thus , when the infalling envelope is more massive than the circumstellar disk , the angular momentum of the circumstellar disk is effectively transferred by magnetic braking . in the main accretion phase , however , the mass of the infalling envelope decreases with time . thus , it is expected that magnetic braking gradually becomes ineffective with time . however , it is difficult to estimate the magnetic braking timescale of the infalling envelope quantitatively , because the infalling envelope drastically changes its density distribution in a several free - fall timescale that corresponds to the timescale for the circumstellar disk formation . thus , the disk formation process and final size of the circumstellar disk can not be investigated using a simple analytical framework . only a less - massive and small - size circumstellar disk may appear in a magnetized cloud when magnetic braking significantly transfers the angular momentum by the end of the main accretion phase . in this study , however , we calculated the cloud evolution by the end of the main accretion phase and showed that the circumstellar disk exceeding @xmath240au can form even in a strongly magnetized cloud . in a magnetized cloud , magnetic braking becomes ineffective and the disk size increases rapidly in the later main accretion phase . it is natural that magnetic braking only redistributes the angular momentum from the circumstellar disk into the infalling envelope . because the infalling envelope finally falls onto the circumstellar disk , the excess angular momentum gives rise to a rotation supporting the disk in the later main accretion phase . instead of magnetic braking , only the protostellar outflow can transfer the angular momentum to the interstellar medium , because the gas with a certain amount of angular momentum is directly ejected into the interstellar medium from the circumstellar disk by the protostellar outflow . because the protostellar outflow is powered by mass accretion onto the circumstellar disk , the outflow disappears in the later main accretion phase , which also corresponds to rapid growth of the circumstellar disk . thus , the circumstellar disk begins to grow after the protostellar outflow sufficiently weakens or disappears . in addition , in the later main accretion phase , the spiral structure appearing in the circumstellar disk can contribute to the redistribution of the angular momentum in the circumstellar disk . thus , it is difficult to specify which mechanism is effective for the angular momentum transfer in a magnetized collapsing cloud . however , the large - size disk comparable to observations can form in a strongly magnetized cloud . in this study , we mainly investigated the evolution of the collapsing cloud with a mass of @xmath277 in the initial state and found that the circumstellar disk with a size of several hundred au appears in the later main accretion phase . on the other hand , we can not calculate the cloud evolution for an initially massive cloud until the infalling envelope disappears . however , a massive cloud has a larger angular momentum ( or larger centrifugal radius ) . thus , a circumstellar disk much larger than @xmath240au may form in an initially massive cloud . finally , we comment on the geometry of the initial magnetic field . in this study , we assumed the magnetic field lines to be parallel to the rotation axis . with ideal mhd calculation , @xcite already reported the formation of the circumstellar disk when the magnetic field lines are inclined from the rotation axis . thus , we need to investigate the inclined field case with non - ideal mhd effects in the future . numerical computations were carried out on nec sx-9 at center for computational astrophysics , cfca , of national astronomical observatory of japan , and nec sx-8 at the yukawa institute computer facility . this work was supported by the grants - in - aid from mext ( 20540238 , 21740136 ) . ) for models with different initial magnetic field strengths @xmath56 or different cloud masses are plotted against the initial cloud mass ( left panel ) and cloud radius ( right panel ) in the upper panels . the centrifugal radii ( @xmath279 ) for different angular velocities @xmath57 or different cloud masses are plotted in the lower panels . , width=529 ] but on the @xmath186 plane . in each panel , the black line corresponds to the circumstellar disk . the boundary between outflowing ( @xmath280 ) and inflowing ( @xmath281 ) gas is described by the white line ; inside this boundary , the gas is outflowing from the circumstellar disk . , width=566 ] on the @xmath186 plane at the same epochs as figs . [ fig:3 ] and [ fig:5 ] except for panel _ a _ ( initial state ) . the white broken line is the boundary between the gravitational sphere and interstellar medium ; gravity ( gas self - gravity and the gravity of the protostar ) is imposed inside this boundary . , width=566 ] on the @xmath186 plane at the same epochs as figures [ fig:3 ] , [ fig:5 ] and [ fig:6 ] . the toroidal component ( @xmath282 ) is greater than the poloidal component ( @xmath283 ) inside the black line . the white line corresponds to the contour of @xmath284 . the arrows indicate the magnetic field direction on the @xmath186 plane . , width=566 ] against the cloud mass . the dashed line is the distribution of the angular momentum for model 5 at @xmath285yr . the initial distribution of the angular momentum is plotted by a thin solid line . the angular momentum and mass are integrated in the region of @xmath286 . the sum of the protostellar mass @xmath134 and circumstellar disk mass @xmath135 of each epoch is plotted by the diamond symbol ( @xmath287 ) on each line . the region on the left - hand side of the diamond symbol corresponds to the circumstellar disk region . , width=566 ] plane for the resistive ( model 1 ; upper panel ) and ideal ( model 4 ; lower panel ) mhd models . the thick white line is the contour of @xmath288 . the arrows indicate the magnetic field direction on the @xmath186 plane . , width=529 ]

the black hole mass function of supermassive black holes describes the evolution of the distribution of black hole mass . it is one of the primary empirical tools available for mapping the growth of supermassive black holes and for constraining theoretical models of their evolution . in this review we discuss methods for estimating the black hole mass function , including their advantages and disadvantages . we also review the results of using these methods for estimating the mass function of both active and inactive black holes . in addition , we review current theoretical models for the growth of supermassive black holes that predict the black hole mass function . we conclude with a discussion of directions for future research which will lead to improvement in both empirical and theoretical determinations of the mass function of supermassive black holes .

1112.1430

understanding how and when supermassive black holes ( smbhs ) grow is currently of central importance in extragalactic astronomy . a significant amount of empirical work has established correlations between smbh mass and host galaxy spheroidal properties , such as luminosity @xcite , stellar velocity dispersion ( the @xmath0@xmath1 relationship , e.g. , @xcite ) , concentration or sersic index @xcite , bulge mass @xcite , and binding energy @xcite . these scaling relationships imply that the evolution of spheroidal galaxies and the growth of smbhs are intricately tied together . the currently favored mechanism for linking the growth of smbhs and their hosts is black hole feedback , whereby black holes grow by accreting gas in so - called `` active '' phases , possibly fueled by a major merger of two gas - rich galaxies , until feedback energy from the smbh expels gas and shuts off the accretion process @xcite . alternatively , it has been suggested that the origin of the scaling relationships does not necessarily require smbh feedback , but emerges from the stochastic nature of the hierarchical assembly of black hole and stellar mass through galaxy mergers . @xcite . feedback - driven ` self - regulated ' growth of black holes has been able to reproduce the local @xmath0@xmath1 relationship in smoothed particle hydrodynamics simulations @xcite . moreover , agn feedback has also been invoked as a means of quenching the growth of the most massive galaxies @xcite . there have been numerous models linking smbh growth , the quasar phase , and galaxy evolution @xcite . while feedback is likely important for regulating the growth of smbhs and galaxies , the fueling mechanisms that contribute to growing the smbh are likely diverse . major - mergers of gas - rich galaxies may fuel quasars at high redshift , and grow the most massive smbhs . however , major - mergers alone do not appear to be sufficient to reproduce the number of x - ray faint agn @xcite , and accretion of ambient gas via internal galactic processes @xcite , may fuel these fainter , lower @xmath0 agn at lower @xmath2 . this is supported by the fact that many agn are observed to live in late - type galaxies out to @xmath3 @xcite , and the x - ray luminosity function of agn hosted by late - type galaxies suggests that fueling by minor interactions or internal instabilities represents a non - negligible contribution to the accretion history of the universe @xcite . the black hole mass function ( bhmf ) provides a complete census of the mass of smbhs and their evolution . because of this , the bhmf is one of the primary empirical tools available for investigating the growth of smbhs , and for constraining theoretical models for the growth of the smbh population . because smbhs and galaxies are thought to be linked in their evolution , the bhmf provides insight into the fueling mechanisms that dominate black hole growth , and therefore into the role of feedback in the evolution of the host galaxy . the bhmf is also an important tool in planning future surveys , as it provides an estimate of the distribution of smbh mass expected for the survey . this in turn is important because mass is a fundamental quantity of the black hole , and therefore is an important observational quantity for empirical studies of black hole accretion physics @xcite . of course , further improvement to our understanding of black hole accretion physics will further improve our modeling and understanding of black hole accretion and feedback , which in turn will improve our understanding of black hole - galaxy coevolution . therefore , the bhmf is an important empirical quantity for smbh studies . in this review we discuss the current status of bhmf estimation and theoretical modeling . in [ s - est ] we discuss the non - trivial task of estimating the bhmf . in [ s - local ] we discuss current estimates of the local smbh bhmf . in [ s - continuity ] we discuss bhmf estimates derived by combining the local bhmf with the agn luminosity function via a continuity equation . in [ s - agn ] we discuss bhmfs estimated for agn only . in [ s - theory ] we review theoretical models for smbh growth that predict the smbh bhmf . finally , in [ s - future ] we discuss directions for future improvements to the empirical and theoretical studies of the bhmf . we note that unlike , say , the luminosity function , the division between ` observational ' and ` theoretical ' studies is not as clear for the bhmf , as some amount of modeling is necessary in order to estimate the bhmf from strictly observational quantities . we have attempted to divide the studies according to whether the bhmf is constrained empirically , as in , say , a formal statistical fitting procedure , or if it is predicted from a theoretical model for smbh growth . in reality , the line between theoretical and empirical studies is blurry , and some procedures which we have considered to be empirical may be thought of as theoretical . the black hole mass function , denoted as @xmath4 , is the number of sources per comoving volume @xmath5 with black hole masses in the range @xmath6 . the black hole mass function is related to the joint probability distribution of @xmath0 and @xmath2 , @xmath7 , as @xmath8 the normalization of the bhmf is @xmath9 , the total number of smbhs in the observable universe , and is given by the integral of @xmath10 over @xmath0 and @xmath5 . similar to luminosity function estimation , the bhmf may be estimated from astronomical surveys . however , while there are many well - established methods for estimating luminosity functions , there are two complications that make bhmf estimation a more difficult problem @xcite . the first is the issue of incompleteness . surveys are typically constructed by finding the set of objects of interest containing a smbh that satisfy a flux criteria , e.g. , all objects brighter than some flux limit . surveys are not constructed by selecting on mass . because there is a distribution of luminosities at a given smbh mass , whether it be the luminosity of the host galaxy or of the agn , some smbhs will scatter above the flux limit and some below . this creates a selection function which is less sensitive to @xmath0 , and it is possible that a survey may be incomplete in all mass bins . the second complication is the large uncertainty in smbh mass among mass estimators . currently it is not possible to obtain reliable mass estimators for large numbers of smbhs through dynamical and modeling of the stellar or gaseous components , and thus scaling relationships are employed . masses may be estimated using scaling relationships between @xmath0 and the properties of the host galaxy bulge or the luminosity and the width of the broad emission lines for agn @xcite . it has also recently been suggested that the x - ray variability properties of agn may also provide another scaling relationship for estimating @xmath0 @xcite , but further work is needed for developing this . while these scaling relationships enable one to estimate @xmath0 for large numbers of smbhs , they also contain a significant intrinsic statistical scatter . gltekin , k. , et al . @xcite find that for early type galaxies there is an intrinsic scatter in @xmath0 of @xmath11 dex and @xmath12 dex at fixed host galaxy bulge dispersion and luminosity , respectively ; the amplitude of the scatter is larger for late type galaxies . for agn with broad emission lines , vestergaard & peterson @xcite estimate the scatter in @xmath0 at fixed luminosity and line width to be @xmath13 dex , depending on which emission line is used . the statistical uncertainty in the mass estimates can have a significant effect on the inferred bhmf . the distribution of the mass estimates is the convolution of the intrinsic bhmf with the error distribution in the mass estimates . in general , it is typically assumed that the error in the mass estimates is independent of the actual value of @xmath0 . this is not the case for @xmath0 estimated through dynamical modeling , however independence between @xmath0 and its error is likely to be a good approximation for @xmath0 estimated using scaling relationships . because scaling relationships are the only feasible manner in which to estimate @xmath0 for a large sample of smbhs , which is necessary for any estimate of the bhmf , we will assume that @xmath0 and its error are independent . under the assumption of independence between the estimated @xmath0 and its error , the bhmf that would be inferred directly from the distribution of the mass estimates is broader than the intrinsic bhmf , and is thus biased . figure [ f - bhmf_illust ] illustrates this effect , where an intrinsic mass function is compared with the distribution of an unbiased mass estimator having a statistical uncertainty of 0.3 , 0.4 , and 0.5 dex , respectively . as can be seen , the distribution of mass estimates is significantly different than the intrinsic mass function . in particular , the distribution of the mass estimates falls off more slowly with increasing @xmath0 , and overpredicts the number of smbhs at the high @xmath0 end of the mass function . the bias is worse when the dispersion in the scatter in the mass estimates becomes larger . end and for large statistical error . ] in order to estimate the smbh mass function in an unbiased manner , it is necessary to match the mass function with the observed distribution of the mass estimates and any additional observational quantities that the selection function depends on . the basic idea is to start with an assumed mass function . then , calculate the distribution of mass estimates implied by this mass function . in addition , calculate the distribution of observational quantities that one s sample is selected on , say , flux , that is implied by the assumed mass function . this step allows one to correct for incompleteness , but requires an additional assumption about how to relate the mass function to the quantity that one s sample is selected on . finally , impose the selection function for the sample , and compare the predicted observed distributions of mass estimates and any other observables ( e.g. , flux ) with the actual distributions . if they are not consistent , then the data rule out the assumed mass function and relationship between @xmath0 and the observable quantities . we can make the above procedure more quantitative by deriving the likelihood function for the smbh mass function . kelly , vestergaard , & fan @xcite derived the likelihood function for the mass function when using masses estimated from agn broad emission lines . they used this likelihood function for developing a bayesian approach to estimating the smbh mass function . although their method was limited to broad line mass estimates , it is straightforward to generalize their formalism for any generic mass estimator . denote the black hole mass estimate as @xmath14 . in addition , denote as @xmath15 the set of observables that one uses to select one s sample . in the majority of cases this will be flux at one or more wavelengths . then , the likelihood function for the bhmf based on a sample of @xmath16 smbhs is @xmath17^{n - n } \prod_{i=1}^n \int_{0}^{\infty } p(\hat{m}_{bh , i}|m_{bh , i } , z_i , x_i ) p(x_i|m_{bh , i},z_i,\psi ) p(m_{bh , i},z_i|\theta)\ dm_{bh , i } , \label{eq - likhood } \end{aligned}\ ] ] where the bhmf is related to @xmath9 and the probability distribution of @xmath0 and @xmath2 via equation ( [ eq - phiconvert ] ) . here , @xmath18 is the binomial coefficient , @xmath19 denotes the parameters for the bhmf , @xmath20 denotes the parameters for the distribution in @xmath15 at fixed @xmath0 and @xmath2 , and @xmath21 is the probability of including a smbh in one s sample as a function @xmath19 and @xmath20 . here , we have assumed that the distribution in the mass estimates at fixed @xmath22 and @xmath23 , is known , although one could include additional free parameters for this as well . the binomial coefficient arises from the fact that the number of objects included in one s survey follows a binomial distribution and @xmath24 , so equation ( [ eq - likhood ] ) converges to the poisson distribution when @xmath25 . see @xcite for further details . ] with @xmath9 trials and probability of success @xmath21 . the probability of including a smbh in one s survey as a function of the bhmf , @xmath21 , is calculated from the survey selection function @xmath26 as @xmath27\ dx . \label{eq - selfunc}\ ] ] it is up to the researcher to choose the particular parameteric form for the smbh mass function , the distribution in the mass estimates at fixed @xmath28 and @xmath15 , and the distribution of the observable that the sample is selected on ( e.g. , flux ) at fixed @xmath0 and @xmath2 . typical choices are log - normal distributions , schechter functions , and mixtures of log - normal distributions . once one has done this , one can use equation ( [ eq - likhood ] ) to compute a maximum - likelihood estimate for the bhmf , or perform bayesian inference . an alternative form of estimating the bhmf can be used when the mass estimates are derived from an observational quantity , @xmath29 , and the intrinsic distribution of @xmath29 is known . this is commonly used to estimate the local mass function using host - galaxy scaling relationships @xcite . in this case , the mass function is @xmath30 where @xmath31 is the comoving number density of smbhs as a function of the quantity @xmath29 . when both @xmath32 and @xmath31 are known then the bhmf follows directly from equation ( [ eq - mfunc_alt ] ) . as an example , if the mass function is derived from the scaling between @xmath0 and host galaxy spheroidal luminosity , @xmath33 , then @xmath34 , @xmath32 is the @xmath0@xmath33 relationship , and @xmath31 is the luminosity function of stellar bulges hosting smbhs . as with bhmfs determined from a mass estimator , improper treatment of the intrinsic scatter in @xmath0 at fixed @xmath29 will lead to a biased estimate of the bhmf . however , when calculating the bhmf from equation ( [ eq - mfunc_alt ] ) , ignoring the intrinsic scatter results in an estimated bhmf that is too narrow , underpredicting the number of smbhs at the high mass end of @xmath35 . this is opposite to the case when one estimates the bhmf directly from mass estimates . the observed scaling between @xmath0 and the properties of the smbh host galaxy bulge have motivated several groups to estimate the local bhmf @xcite , with decreasing statistical uncertainties . these estimates of the local bhmf have formed the basis for many studies which have attempted to map black hole growth by comparing with the agn luminosity function ; this is further discussed in [ s - continuity ] . typically , the local bhmf is estimated using the local @xmath0@xmath1 relationship or the local @xmath0@xmath33 relationship , combined with the local number density of galaxies as a function of stellar velocity dispersion or bulge luminosity . the scaling relationships between @xmath0 and host galaxy properties are only determined for the local universe , and thus most authors have limited their determination of the bhmf based on them to the local bhmf . an exception is tamura , ohta , & ueda @xcite , who estimated the bhmf out to @xmath3 assuming that evolution in the @xmath0@xmath33 relation is driven only by passive evolution in @xmath33 . evolution in the scaling relationships is currently an area of intense study , with most groups finding evidence that the normalization of the scaling relationships increases towards higher @xmath2 @xcite , at least for active smbhs . however , there are still concerns regarding potential biases due to selection effects @xcite , but see treu et al . @xcite and bennert et al . @xcite for procedures aimed at modeling and correcting for selection . there may also be biases due to extrapolating the agn mass estimates derived from the broad emission lines to luminous quasars at high @xmath2 @xcite . as such , the uncertainties on the quantitative form of the evolution in the scaling relationships and their scatter are currently large , limiting their use for determining the bhmf outside of the local universe . when the @xmath0@xmath1 relationship is used to estimate the local bhmf , it is common to use the velocity dispersion distribution derived from the sdss by sheth et al . @xcite , with an additional component representing the brightest cluster galaxies @xcite . sheth et al . @xcite estimate the velocity dispersion distribution for late type galaxies by using the tully - fisher relation to convert the luminosity function of late type galaxies to a circular velocity distribution , and then set @xmath36 . when the @xmath0@xmath33 relation is used it is typical to estimate the distribution of @xmath33 seperately for early and late type galaxies by converting their respective luminosity functions to spheroidal luminosity functions using an assumed ratio of bulge luminosity to total luminosity . from this it has been inferred that the local bhmf is dominated by early type galaxies at @xmath37 @xcite . shankar , weinberg , & miralda - escud @xcite present a compilation of recently determined local bhmfs based on a variety of methods , scaling relations used , and data sets used . in figure [ f - local_bhmf ] we show the range of local bhmfs estimated from the @xmath0@xmath1 , @xmath0@xmath33 , and @xmath0@xmath38 relationships , as presented in shankar et al . @xcite . in general , estimates of the local mass density of smbhs span the range @xmath39@xmath40 for @xmath41 @xcite . @xmath1 , @xmath0@xmath33 , and @xmath0@xmath38 relationships , as compiled by shankar et al . @xcite . based on this estimate the local universe is dominated by smbhs with @xmath42 . ] while the procedure for estimating the local bhmf is , in theory , straight - forward , a number of significant systematics remain . first , there is the observational difficulty that most bhmfs derived from the @xmath0@xmath1 relationship are based on sdss spectra . unfortunately , the sdss velocity dispersions are based on a fixed aperture , and thus the size of the aperture relative to the bulge varies with the apparent size of the galaxy and its inclination . in addition , the spectral resolution of sdss spectra is @xmath43 , making it difficult to reliably measure @xmath1 for smbhs with @xmath42 . another concern is that the local bhmf is derived by assuming that the @xmath0@xmath1 or @xmath0@xmath33 relations are single power - laws with a constant scatter in @xmath0 at fixed @xmath1 or @xmath33 . however , recent work has shown these assumptions to be incorrect . for one , the @xmath0@xmath1 and @xmath0@xmath33 relations diverge at the high @xmath0 end , which lauer et al . @xcite suggest implies that the @xmath0@xmath1 relation is not a single power - law . this divergence creates an inconsistency in the bhmfs derived from these two scaling relationships @xcite . similarly , the @xmath44@xmath45 relationships for the sdss and dynamical @xmath0 smbh samples are inconsistent , suggesting a possible selection bias in the estimated bhmfs @xcite . the scatter in the @xmath0@xmath1 relation is larger for spirals @xcite , and appears to increase at low @xmath0 such that most smbhs lie below the @xmath0@xmath1 relation , ( * ? ? ? several authors have found differences in the slope and scatter of the scaling relations for pseudobulges @xcite ; however , it is unclear that this result is due to differences in the perceived bulge velocity dispersions for bulges as compared to pseudobulges or due to different scaling relationships . kormendy , bender , & cornell @xcite argue that @xmath0 does not correlate with galaxy disks , and only correlates weakly , if at all , with pseudobulges . although there is still much that we do not understand about the @xmath0 and host galaxy scaling relationship , these recent results suggest that the scaling relationships are not a single power - law with constant intrinsic dispersion in @xmath0 , representing a significant source of systematic uncertainty in the estimated local bhmf , especially at the low - mass end . by employing the argument of soltan @xcite , numerous studies have attempted to estimate the bhmf at a variety of redshifts by comparing the accreted mass distribution implied by the quasar luminosity function with the local bhmf @xcite . these methods employ a continuity equation describing the evolution of the number density of smbhs @xcite : @xmath46 here , @xmath47 is the average growth rate of smbhs as a function of @xmath0 and cosmic age , @xmath48 , and @xmath49 is the rate at which the number density of smbhs changes due to mergers of black holes , or ejections of black holes from their host galaxies due to gravitational recoil . technically , @xmath49 can also include a contribution from smbhs which are created , but this has not been thought to occur over the redshift range in which equation ( [ eq - cont ] ) is typically applied , i.e. , @xmath50 . because the merger rate of black holes is currently unknown , many studies that have employed equation ( [ eq - cont ] ) set @xmath51 . under the assumption that smbhs grow during phases of agn activity , agn demographics in combination with the local bhmf may be used to compute @xmath52 . this is because the agn luminosity function maps the accretion history onto smbhs , and the local bhmf acts as a boundary condition on equation ( [ eq - cont ] ) ; it is also possible in principle to include the bhmf for agn , which provides more information . studies that have used equation ( [ eq - cont ] ) to estimate the bhmf generally fall into two categories : those that assume an agn lightcurve , and those that employ the bhmf of agn . we discuss each of these seperately . most authors employing equation ( [ eq - cont ] ) have assumed a parameteric form for @xmath53 . the accretion rate is related to the bolometric luminosity output of the accretion flow onto the smbh as @xmath54 , where @xmath55 is the radiative efficiency of the accretion flow , @xmath56 is the accretion rate of matter onto the smbh , and @xmath57 is the speed of light . the growth rate of the smbh is @xmath58 , due to the fact that a fraction @xmath55 of accreted mass is radiated away as energy . making this substitution , the continuity equation becomes @xmath59 where we have ignored mergers of smbhs . equation ( [ eq - cont2 ] ) shows that it is possible to calculated the bhmf at an time given the local bhmf , an assumed average accretion flow lightcurve as a function of @xmath0 , @xmath60 , and an assumed radiative efficiency . because @xmath61 and @xmath62 imply a luminosity function , the local bhmf and agn luminosity function can be used to place constraints on @xmath55 and @xmath60 this means that , in practice , one also has to assume a bolometeric correction , which itself likely depends on both black hole mass @xcite and @xmath63 @xcite . in addition , an estimate of @xmath60 also enables one to estimate the lifetime and duty cycle of agn activity , modulo some luminosity - dependent definition of an agn ; note that the agn duty cycle defines the fraction of smbhs that are ` active ' at a given @xmath0 and @xmath2 . a variety of lightcurve models have been used when employing equation ( [ eq - cont2 ] ) to reconstruct the evolution of the bhmf . the simplest model is that where smbhs spend a fraction of their time radiating at a constant eddington ratio , and spend the remainder of their time in quiescence . the free parameters in this model are the eddington ratio , agn lifetime or duty cycle , and radiative efficiency . this model has been used by @xcite to study the build - up of the local black hole mass function , although @xcite also considered models where the average accretion rate relative to eddington falls off toward lower @xmath2 and higher @xmath0 . raimundo & fabian @xcite employed a variation on the constant @xmath63 models , assuming three different populations of agn with their own eddington ratio : a population of obscured low @xmath63 agn , a population of obscured agn with higher @xmath63 , and a population of unobscured agn . yu & lu @xcite modeled the quasar lightcurve as radiating at the eddington limit for a period of time , and then transitioning into a power - law decay . cao @xcite also modeled the quasar lightcurve as undergoing a power - law decay . lightcurves undergoing a power - law decay arise from self - regulation models , and describe the evolution of the lightcurve after black hole feedback unbinds the accreting gas , therefore quenching its fuel supply . the power - law decay occurs either as a result of evolution of a blast - wave @xcite or from viscous evolution of the accretion disk @xcite . in figure [ f - lcurve ] we compare the bhmf calculated by shankar et al.@xcite with that calculated by cao @xcite . for shankar @xcite we show their reference model , which assumes a radiative efficiency of @xmath64 , an accretion rate relative to eddington of @xmath65 , and that half of all smbhs are active at @xmath66 . we show the model from cao @xcite which assumes a radiative efficiency of @xmath67 and a quasar lifetime of @xmath68 , as it better matches the shankar et al . @xcite estimates . the two estimates of the bhmf agree fairly well , despite the different quasar lightcurve models . in general , most of the studies that have used equation ( [ eq - cont2 ] ) in combination with an assumed quasar lightcurve have concluded the following : * most smbh growth occurs in periods when the quasar is radiating near the eddington limit . * most , if not all , of the local black hole mass function can be explained as the relic of previous agn activity , implying that mergers of smbhs are not important for building up the local mass function . * smbh growth is anti - hierarchical , with the most massive black holes growing first . this has also been termed ` down - sizing ' of active smbhs . * the lifetime of agn activity is @xmath69 . * most smbhs have non - zero spin , as implied by inferred radiative efficiencies of @xmath70 . however , while equation ( [ eq - cont2 ] ) has proven to be an important tool for studying smbh growth , and estimating the black hole mass function , it must be kept in mind that use of equation ( [ eq - cont2 ] ) often entails some strong assumptions . these methods rely on the assumed form of the quasar lightcurve , distribution of radiative efficiencies , and bolometeric corrections , all of which are subject to considerable uncertainty . moreover , in general these methods also rely on an estimate of the local black hole mass function , which , as discussed in [ s - local ] , is itself subject to considerable uncertainty . indeed , there is a strong degeneracy between the estimated radiative efficiency of accretion and the normalization of the local bhmf , and therefore the uncertainty in @xmath55 is linearly proportional to that in the normalization ( or integral ) of the local bhmf . all of these issues have the potential to introduce systematic error into methods based on equation ( [ eq - cont2 ] ) , and further work is needed in reducing these systematics . an alternative to the methods described in [ s - lcurve ] is to estimate the average value of the accretion rate onto smbhs directly from the observational data . this avoids the issue of assuming a form for the quasar lightcurve , as instead @xmath71 is derived directly from the estimated distribution of @xmath63 . techniques based on this approach require a means of linking the mass function of active smbhs to observational quantities , which is done via scaling relationships . this was the approach of merloni @xcite and merloni & heinz @xcite , who employed the black hole ` fundamental plane ' ( bhfp ) @xcite . the bhfp is a scaling relationship between @xmath0 , radio luminosity , and x - ray luminosity , that exists for low - accretion rate black holes ( i.e. , @xmath72 ) , extending from galactic black holes to supermassive ones . the bhfp likely reflects the connection between @xmath0 and the conversion of the accretion flow into radiative energy and jet power . it , in principle , enables one to connect the radio and x - ray luminosity functions to the mass function of active smbhs . having obtained a distribution of @xmath0 and x - ray luminosity at a given redshift for the active smbh population , merloni @xcite and merloni & heinz @xcite then convert this to a joint distribution of @xmath0 and @xmath56 assuming a conversion from x - ray luminosity to @xmath56 which depends on the eddington ratio . the joint distribution of @xmath0 and @xmath56 at a given redshift for active smbhs therefore enables calculation of the average growth rate @xmath73 , which can then be combined with the continuity equation to calculate the black hole mass function at the next redshift . their estimated bhmf is shown in figure [ f - mh08_bhmf ] , which is a recreation of their figure 5 . similar to methods based on assuming a quasar lightcurve , merloni & heinz @xcite concluded that smbhs grow anti - hierarchically ; however , in contrast to the lightcurve methods , merloni & heinz @xcite concluded that most smbhs have low spin as inferred from their derived radiative efficiency . in addition , merloni & heinz @xcite concluded that the distribution of smbh accretion rates is broad , and that most smbh growth occurs during a radiatively efficient accretion mode . the method of estimating the bhmf from the bhfp developed by merloni @xcite and merloni & heinz @xcite has the advantage that it derives the distribution of accretion rates empirically . however , there are also disadvantages to this approach . the uncertainties regarding the bolometeric correction , estimation of the local bhmf , and radiative efficiency also apply to the bhfp method as well . moreover , as discussed in merloni & heinz @xcite , the bhfp is only defined for low - accretion rate objects , i.e. , objects with @xmath74 . merloni & heinz @xcite extrapolate the bhfp to higher accretion rates , after rescaling the normalization to ensure that the radio luminosity is weak for agn in the radiatively efficient mode ( those objects with @xmath75 and lacking a jet ) . unfortunately , the agn in the radiatively efficient mode make a significant contribution to the x - ray luminosity function , from which @xmath76 is derived . moreover , most studies , including those based on the bhfp have concluded that most smbh growth occurs at @xmath75 , which corresponds to the radiatively efficient mode . because the radiatively efficient mode also corresponds to the regime of largest systematic uncertainty for the bhfp , there is the potential for significant systematic error in estimating the bhmf based on the bhfp , as well as in estimating the primary mode of smbh growth . there is thus a need for further improvement to our understanding of the scaling relationships involving @xmath0 and the agn sed . thus far , we have focused on methods for estimating the mass function of all smbhs . in this section we will describe methods for estimating the bhmf for those smbhs in agn , and the results that have come from the application of these methods . the steady improvement in reverberation mapping of agn @xcite has revealed a correlation between the luminosity of agn and the broad line region radius @xcite . it is therefore possible , in principle , to obtain an estimate of @xmath0 for broad line agn ( blagn ) by combining a luminosity - based estimate of the broad line region size with an estimate of the velocity dispersion of the broad line region gas obtained from the width of the broad emission lines @xcite . these virial mass estimates are then calibrated to the estimates of @xmath0 obtained from reverberation mapping , which themselves are calibrated to be consistent with the local @xmath0@xmath1 relationship @xcite . currently , calibrations exist for h@xmath77 @xcite , h@xmath78 ( * ? ? ? * e.g. ) , mg ii @xcite , and c iv @xcite . the statistical scatter in the virial mass estimates is currently estimated to be @xmath13 dex @xcite , although there are indications that the scatter may be smaller , at least for the most luminous quasars @xcite . moreover , it should be noted that the calibration for mg ii is obtained by enforcing consistency in the mean values of the mg ii mass estimator and the h@xmath78 and c iv ones , and therefore there is currently no direct estimate of the statistical scatter in mg ii - based virial mass estimates . in contrast , the amplitudes of the statistical scatter for h@xmath78 and c iv are estimated by comparing mass estimates derived from these lines with the masses derived from reverberation mapping @xcite . although there is currently very little reverberation mapping data for c iv , the estimate of the dispersion in the c iv - based mass estimates should not be biased so long as the masses based on reverberation mapping are reliable estimates of the true @xmath0 , regardless of which emission line was used in the reverberation mapping campaign . early estimates of the mass function of smbhs in blagn were obtained by binning up the virial mass estimates and applying a @xmath79 correction @xcite , a technique borrowed from luminosity function estimation . greene & ho @xcite estimated the local bhmf for blagn from the sdss dr4 , while vestergaard et al . @xcite estimated the bhmf for blagn over @xmath80 using the uniformly - selected quasar sample from the sdss dr3 @xcite . vestergaard & osmer @xcite estimated the bhmf for the brightest blagn using objects from a variety of surveys , as their sample was designed to complement the uniformly - selected sdss dr3 sample . unfortunately , as discussed in [ s - complications ] , this method of binning up the mass estimates suffers from biases due to the large statistical scatter in the virial mass estimates , and due to the inability of a luminosity - based @xmath81 correction to correct for incompleteness in @xmath0 . subsequent attempts have further improved in their methodology , providing more accurate bhmfs . shen et al . @xcite employed a forward - modeling approach where the mass function and eddington ratio distribution were estimated by matching the observed distribution of mass estimates and luminosity to that implied by the model bhmf and eddington ratio distribution . their method accounts for incompleteness and the statistical scatter in the mass estimates , but lacked statistical rigor in that the matching was done visually . schulze & wisotzki @xcite employed a maximum - likelihood technique for estimating the local bhmf for blagn . their method corrects for incompleteness in @xmath0 but does not correct the bhmf for the broadening caused by the statistical scatter in the virials mass estimates . kelly et al . @xcite developed a bayesian method that corrects for both the statistical scatter in the mass estimates and incompleteness , and used their method to estimate the local bhmf of blagn from the bright quasar survey @xcite . kelly et al . @xcite used the method of @xcite to estimate the bhmf of blagn at @xmath82 from the mass estimates in the sdss dr3 quasar sample @xcite . the blagn bhmfs from a variety of studies are compiled in figure [ f - blqso_bhmf ] , showing the evolution of the bhmf from the local universe out to @xmath83 . more recently , shen & kelly @xcite extended the bayesian method of @xcite to include a possible luminosity - dependent bias in virial mass estimates derived from the emission line @xmath84 , the existence of which was suggested by shen & kelly @xcite . shen & kelly @xcite applied their method to the sdss dr7 uniformly - selected quasar sample , independently estimating the bhmf and eddington ratio distribution in different redshifts bins . bhmfs estimated by vestergaard et al . @xcite ( dashed red line ) and kelly et al . @xcite ( shaded region ) , where the shaded region for the kelly et al . @xcite estimate defines an approximate @xmath85 confidence region . the bhmfs estimated by greene & ho @xcite and vestergaard et al . @xcite are flux - limited bhmfs , as they did not fully correct for incompleteness in @xmath0 . ] similar to the methods based on the continuity equation , investigations of the bhmf for blagn have found evidence for the anti - hierarchical growth of smbhs , i.e. , cosmic ` down - sizing ' of blagn activity . the inferred eddington ratio distributions are wide , and the density of smbhs continues to increase toward eddington ratios which are below the survey completeness limit . in addition , kelly et al . @xcite used the blagn bhmf to estimate the lifetime of broad line quasar activity to be @xmath86 myr among smbhs with @xmath87 , which is similar to quasar lifetimes inferred from the continuity equation . kelly et al . @xcite also used their estimated bhmf to estimate the maximum mass of a smbh to be @xmath88 , which is in agreement with theoretical expectations @xcite . mass functions estimated from scaling relationships for blagn have the advantage that they are derived from estimates of @xmath0 that are obtained for individual sources , providing a more ` direct ' estimate of the mass function than those based on the continuity equation . however , they have the disadvantage that they are only available for a subset of the agn population , which itself is only a subset of the smbh population . this complicates comparison with other smbh mass functions , as the fraction of agn with broad emission lines is poorly constrained , especially as a function of mass . this being said , bhmfs of blagn represent a subset of smbhs that are actively growing at the time that they are observed , and , as the aforementioned studies have demonstrated , their mass function still contains important information on smbh growth . as with all methods of bhmf estimation , the virial mass estimates and the mass functions derived from them still suffer from systematics . first , there is the usual problem of calculating a bolometeric correction , although this only affects the estimated eddington ratio distribution , and not the bhmf . second , there are a few concerns with the virial mass estimates which could introduce systematic error ; some of these have been discussed by greene & ho @xcite . for one , most of the reverberation mapping data is only available for the h@xmath78 line . because of the limited mg ii data , the mg ii scaling relationship is in general not calibrated using objects with black hole mass estimates from reverberation mapping . there may be systematic effects with luminosity or eddington ratio when using the @xmath84-based scaling relationships @xcite , possibly due to a dependence of the broad line region structure on these quantities . systematic effects on broad line region geometry , which can effect the inferred velocity dispersion , are a particular concern for c iv , which is thought to arise in an accretion disk wind @xcite . along these lines , unaccounted for radiation pressure on broad line clouds may also bias the virial masses , especially among those agn radiating near the eddington limit , @xcite ; however , its importance is still debated @xcite . in addition , the reliability of line width measurements can rapidly deteriorate for low @xmath89 data @xcite . and finally , the blagn virial mass estimates are calibrated to the reverberation mapping derived masses , which themselves are calibrated to lie on the local @xmath0@xmath1 relationship . most of the agn that are used to calibrate the reverberation mapping masses to the @xmath0@xmath1 relationship have lower masses and are hosted by late type galaxies , for which there is evidence that the @xmath0@xmath1 relationship begins to break down @xcite . greene et al . @xcite argue that the normalization of the scaling relationships inferred when limiting the calibration to low mass smbhs hosted in late type galaxies may be about a factor of @xmath90 lower than that used for the current broad line mass estimates @xcite . however , dynamical mass estimates exist for two reverberation mapped agn : ngc 3227 @xcite and ngc 4151 @xcite . in both cases the masses derived from dynamical modeling and reverberation mapping agree , so it is unclear if a smaller scaling factor is needed for late - type galaxies . these issues show that there are still many remaining questions regarding virial masses , highlighting the need for further study using high - quality reverberation mapping data . before broad line mass estimates , there were two earlier attempts at estimating the bhmf for agn , which we briefly mention here . siemiginowska & elvis @xcite and hatziminaoglou , siemiginowska , & elvis @xcite used a model for the agn lightcurve arising due to thermal - viscous accretion disk instabilities @xcite to calculate the expected distribution of luminosity at a given black hole mass . based on this calculated distribution , they used the quasar luminosity function to constrain the quasar black hole mass function . siemiginowska & elvis @xcite found evidence for smbh downsizing in agn , consistent with later work . franceschini et al . @xcite found a tight correlation between @xmath0 and the total radio power observed in a sample of local galaxies . they then used their empirical relationship to estimate the local bhmf derived from the local radio luminosity function of galaxies . while many of the objects in their sample are not considered agn in the traditional sense , fraceschini et al . @xcite argue that this correlation is a signature of an advection - dominated accretion flow , thought to dominate at low accretion rates relative to eddington . therefore , while these smbh may not be ` active ' in the quasar sense , the determination of their mass function relies on radio emission from the smbh accretion flow , so this method may still be considered a method for estimating the bhmf for active smbhs . franceschini et al . @xcite compared their bhmf to models of agn activity and found that it was inconsistent with agn activity being continuous and long - lived , but consistent with agn activity being transient and possibly recurrent . there have been numerous theoretical models for the formation and growth of supermassive black holes , and coevolution with their host galaxies . understanding this formation , growth , and coevolution is one of the current most important outstanding issues in extragalactic astrophysics . because the black hole mass function provides a census of the smbh population and its evolution , it is one of the most fundamental observational quantities available for constraining models of smbh formation and growth . as such , many theoretical investigations have predicted a bmhf for comparison with the empirical bhmf . in this section we review some of the models for smbh formation and growth . there have been numerous theoretical models for smbh growth and formation , and it is beyond the scope of this primarily empirically - focused review to review all of them ; instead , we focus on those theoretical models that predict a bhmf . early models for the coevolution of smhbs and galaxies linked the growth of black holes to the properties of host dark matter halos , with periods of smbh growth occuring in quasar phases initiated by mergers . in general , early studies that predicted a bhmf used various perscriptions to relate @xmath0 to the mass of the host halo @xcite . more recent models for the coevolution of smbhs and galaxies have incorporated agn feedback from the smbh . in addition , the availability of empirical bhmfs have enabled modelers to compare their more recent models with observational data . in general , the models are qualitatively in agreement with the empirical results , in that they are able to match the local bhmf fairly well and predict downsizing of smbhs . however , considering the current systematic and statistical uncertainties in the empirical results , it is difficult to place rigorous empirical constraints on the models such that certain models may be ruled out . because of this , we simply summarize some of the different recent models that have been developed which predict the bhmf . granato et al . @xcite developed a model incorporating feedback from agn and supernovae , where the feeding of the smbh is driven by stellar radiation drag on gas . their predicted local bhmf agrees with that estimated by shankar et al . cattaneo et al . @xcite used halo merger trees constructed from @xmath9-body simulations to track the growth of smbhs . in their model the black hole fueling rate was proportional to the star formation rate of the host galaxy burst component and the density of the cold gas in the starburst component . their model predicted smbh downsizing , with the most massive part of the bhmf being built up first , in agreement with the subsequent empirical studies . hopkins et al . @xcite describe a model for the coevolution of smbhs and galaxies whereby all major mergers of gas - rich galaxies trigger a quasar . in this model the final black hole mass is assumed to be on average proportional to the host spheroidal mass , in agreement with the local scaling relationships between smbhs and their host galaxies . hopkins et al . @xcite estimated the merger rate of gas - rich galaxies by combining theoretical constraints of the halo and subhalo mass functions with empirical constraints on halo occupation models . their model also predicts smbh downsizing , and their predicted bhmf matches the local bhmf derived by marconi et al . @xcite . similarly , shen @xcite also assumed that quasars are triggered by major mergers of gas - rich galaxies , with the smbhs growing via accretion in these quasar phases . shen @xcite used a halo merger rate based on theoretical expectations from @xmath9-body simulations , and assumed a universal quasar lightcurve shape having an exponential increase followed by a power - law decay ( see also @xcite ) . the bhmf predicted by shen @xcite broadly agrees with the local one estimated by shankar et al . @xcite and predicts that most smbhs with @xmath91 were in place by @xmath92 , but only @xmath93 of them were assembled by @xmath94 . most recently , fanidakis et al . @xcite extended the model of @xcite , which includes agn feedback , to also follow the spin distribution of smbhs . in their model smbhs are fueled through accretion of cold gas from mergers , disk instabilities , and cooling flows from hot halos . however , the inclusion of smbh spin enabled them to include different radiative efficiencies , which dictates how much accreted material actually grows the black hole , and to provide an improved model for the amount of mechanical feedback imparted through an agn jet , both of which depend on the spin of the black hole . their model predicts that the present - day universe is dominated by smbhs with @xmath95@xmath96 , and that the bhmf at @xmath97 was largely built up at @xmath98 due to an increase in both lower accretion rate ` radio - mode ' growth and mergers of smbhs . almost all models for the cosmological coevolution of smbhs and galaxies that predict a bhmf have been of an analytical or semi - analytical nature . an exception is the study done by di matteo et al . @xcite , who present the results from cosmological hydrodynamic simulations of the @xmath99cdm model that follow the growth of galaxies and smbhs , including their feedback processes , at @xmath100 . direct cosmological simulations such as these should , in principle , provide the most accurate results as to the predicted bhmf , and for identifying the relevant physical processes that are important in shaping the bhmf . however , current cosmological hydrodynamic simulations suffer from the fact that they can not resolve processes on physical scales corresponding to the smbh accretion flow . in fact , di matteo @xcite use a gravitational softening length of @xmath101 kpc . instead , di matteo @xcite employ a subresolution model where the accretion onto the smbh is estimated using a bondi - hoyle - lyttleton parameterization @xcite with a correction factor to account for the fact that the bondi radius is not resolved . they assume a radiative feedback energy efficiency of @xmath102 @xcite , which is the only free parameter in their model and required in order to match the normalization of the observed local @xmath0@xmath1 relationship . their calculated bhmf at @xmath92 matches the local bhmf for @xmath103 . in addition , di matteo @xcite also find downsizing in their model , in agreement with observations , with the high mass end of the bhmf being largely in place by @xmath104 . . some of these authors did not report bhmfs at each redshift shown , so we only show those available at each redshift . ] in figure [ f - theory ] we compile predicted bhmfs from several recent models for smbh formation and growth @xcite . in general , the models tend to agree to within a factor of a few with regards to the bhmf . however , they diverge at @xmath105 , where their predicted smbh number densities can differ by over an order of magnitude . recent work has made improvement to models for the bhmf by focusing on theoretical modeling of the distribution of seed smbhs . the discovery of quasars at @xmath106@xmath107 with @xmath108 @xcite places strong constraints on the formation of smbh seeds due to the very limited amount of time available at that redshift to grow smbhs . lodato & natarajan @xcite derive the bhmf of smbh seeds at @xmath109 that are the result of the collapse of pregalactic disks which have not yet been enriched by metals @xcite . black holes formed through such a mechanism have masses @xmath110 , while black holes which are the remnants of pop iii stars have @xmath111 . a similar seed black hole formation mechanism is through ` quasi - stars ' @xcite , which are also able to produce seed black holes with @xmath112 . volonteri , lodato , & natarajan @xcite describe a model for the growth of smbhs seeded according to the direct collapse model of lodato & natarajan @xcite with varying formation efficiencies . in addition , they also compared the results from this model using smbhs seeded from pop iii remnants . volonteri et al . @xcite grow smbhs through major mergers , and force the black hole mass after the galaxy merger to scale with the circular velocity of the host halo ; additional growth is also provided through black hole mergers . their merger trees are based on a monte carlo algorithm based on the extended press - schechter formalism . they find that most significant differences in the local bhmf with respect to black hole formation efficiency occur at @xmath113 , with the number density of smbhs with @xmath113 increasing with increasing formation efficiency . volonteri & begelman @xcite performed a similar analysis as that of volonteri et al . @xcite but instead used smbh seeds formed via quasi - stars . the bhmfs calculated by volonteri & begelman @xcite match those of merloni & heinz @xcite at the high mass end , at least at @xmath114 . natarajan & volonteri @xcite used a growth and seeding model which is very similar to that employed by volonteri et al . however , they also predict the bhmf for broad line quasars , assuming that @xmath115 of quasars are unobscured . they compare the bhmf derived from their model at @xmath116 to the bhmf for broad line quasars reported by kelly et al . @xcite , and to the bhmf for all smbhs reported by merloni & heinz @xcite . natarajan & volonteri @xcite concluded that seeds from pop iii stars have difficulty reproducing the blagn bhmf , especially at high redshift , while seeds resulting from the direct collapse of pregalactic disks do better at fitting the high mass end of the blagn bhmf at @xmath114 . before concluding this review , we present a discussion of possible future empirical and theoretical work relevant to bhmf studies . these include : * * better characterization of the smbh - host galaxy scaling relationships*. currently , the local bhmf is estimated from the distribution of host galaxy properties assuming that @xmath0 has a constant log - normal scatter about a single - power law scaling relationship . as discussed in [ s - local ] , recent observations have provided reason to doubt this assumption , suggesting that the correlations break down at the highest and lowest masses . this will create biases in the bhmf determined from the scaling relationships , which in turn will also affect the bhmf estimated from the continuity equation . further direct @xmath0 estimates from dynamical and kinematic modeling should be obtained for a variety of galaxy types , especially at the high and low @xmath0 end . the next class of @xmath117 m telescopes should provide a significantly improved picture of the scaling relationships , thus providing us with more accurate estimates of the local bhmf . * * improvements to techniques based on the continuity equation . * most studies that have invoked the continuity equation to link the local bhmf to the agn luminosity function have assumed a single radiative efficiency , which is equivalent to assuming a single black hole spin , and a universal agn lightcurve . neither of these assumptions are likely to be true , and improvements to this type of modeling should include a distribution of smbh spin and agn lightcurves . in addition , we need to better characterize the bolometeric corrections , which remain a significant source of systematic uncertainty . the continuity equation techniques should also be extended to map the evolution of the full joint 3-dimensional distribution of black hole mass , accretion rate , and spin . while this will not necessarily have a direct effect on estimating the bhmf , it will provide insight into the dominant accretion modes experienced by active smbh and into the dominant fueling mechanism for agn activity , as the spin distribution traces the smbh fueling history @xcite . * * better characterization of scaling relationships involving @xmath0 for agn*. the dominant scaling relationship for estimating @xmath0 in agn involves the broad emission lines . however , as discussed in [ s - broad ] a number of systematic uncertainties remain . in order to reduce these systematics , we need to better understand the broad line region geometry for the different emission lines , and how it scales with luminosity , which will provide us with a more accurate conversion from line width to velocity dispersion . moreover , accurate characterization of the broad line region geometry should remove the need to calibrate the scaling relationships to the local @xmath0@xmath1 relationship , which has its own set of systematics . improvements to reverberation mapping campaigns and modeling @xcite , as well as increasing the number of agn monitored for reverberation mapping , will be needed in order to really understand the systematics involved in the broad line mass estimates . + there is also the need to better characterize the black hole fundamental plane . because the bhfp describes how the emission mechanisms responsible for the radio and x - ray flux scale with @xmath0 , the bhfp coefficients depend on these emission mechanisms however , these emission mechanisms depend on the geometry of the accretion state and the existence of a jet , which in turn depends on the accretion rate @xcite , and therefore the bhfp coefficients will be different for different classes of agn @xcite . in particular , the bhfp is currently poorly constrained for ` soft - state ' galactic black holes and radio - quiet agn . therefore , to reduce the systematics involved with the bhfp it will be necessary to characterize the scaling relationships and their scatter for radio - quiet objects . a correlation between the radio and x - ray luminosity has been observed for radio - quiet objects @xcite , implying that a bhfp should also exist for these objects . in order to better characterize the bhfp for radio - quiet objects it will be necessary to obtain radio detections for a well - defined sample of radio - quiet agn with reliable @xmath0 estimates and x - ray detections . + finally , there has recently been the discovery of scaling relationships involving @xmath0 and the optical @xcite and x - ray @xcite variability properties of agn . mass estimators based on these scaling relationships have not been rigorously developed yet , nor have they seen widespread use . however , the existence of these scaling relationships implies that the variability properties may offer another avenue for estimating @xmath0 and bhmfs , which may become increasingly valuable in the era of current and future large time - domain surveys , such as _ pan - starrs _ and _ lsst_. * * understanding the redshift evolution of scaling relations . * from the theoretical point of view , it is clear that high - redshift scaling relations ( or the lack thereof ) between smbh and their hosts provide unique and powerful constraints to models for agn feeding and feedback , which can not be otherwise distinguished ( see e.g. merloni et al.@xcite and references therein ) . + in practical terms , a better understanding of the evolution of scaling relations may also be very advantageous for bhmf studies . as we discussed above , current technique for bh mass estimation at z>0 involve un - obscured , broad emission line qsos . one can argue that , as long as we are restricted to just this class of qsos , we will have to make critical assumptions about the properties of a significant part of the population to draw conclusions about the full bhmf ( if we wanted , for example , to compare with `` continuity - equation - based '' methods ) . on the other hand , large multi - wavelength surveys do and will provide a wealth of information on the host galaxies of obscured agn at high redshift , that represent the numerically dominant part of the growing black holes population @xcite . therefore , if we had an independent way to put constraints on the nature of the bh - host relation for these objects , we could explore the uncharted territory of bhmf for obscured agn ( and for the entire population ) . such an independent information could come , for example , from ir studies of broad emission lines which could act as probes of the bh potential less affected by obscuration . the first exploratory works pursuing this line of research have recently been published , e.g. , @xcite . + from the technical point of view , a lot of work of course is needed to better understand how reliable these estimators are . another big `` technical '' challenge of all studies of the evolution of scaling relations , is the fact that they require a thorough assessment of the many observational biases one encounters in studying high redshift agn and their hosts @xcite . * * accounting for black hole kicks in theoretical models . * most theoretical models for the bhmf do not include recoiling effects caused by the merger of two black holes . however , recent theoretical work on black hole recoils suggests that black holes can spend a significantly large enough amount of time offset from the central region of the host galaxy to alter their growth , thereby increasing the scatter about the scaling relationships and decreasing the final black hole mass @xcite . on the other hand , volonteri , gltekin , & dotti @xcite and volonteri , natarajan , gltekin @xcite find that ejected smbhs are rare at @xmath118 , especially for massive smbhs , suggesting that accounting for ejected black holes will not make a significant difference in the bhmf . further improvement in our understanding of the effects and frequency of black hole recoil will ensure the accurate implementation of black hole recoil into models for smbh growth . * * improvements to our understanding of agn feedback . * most current theoretical models for smbh growth involve agn feedback , and assume a single efficiency for coupling feedback energy to the gas ; this feedback efficiency is usally treated as a free parameter . an improved physical understanding of agn feedback will improve theoretical models for the bhmf , as the feedback efficiency affects the dynamics of the smbh s fuel supply , and therefore the amount that the smbh accretes as a function of redshift . recent high - resolution hydrodynamic simulations in one dimension @xcite and two dimensions @xcite have concluded that agn feedback efficiency increases with the eddington ratio , and that the values are below the value of @xmath119 assumed in many current theoretical models for smbh growth . further improvements to simulations developed for studying agn feedback will lead to a better physical understanding of agn feedback , which will improve theoretical models for smbh growth and the bhmf . + on the observational side , future x - ray observations should provide considerable improvement in our understanding of agn feedback . x - ray spectra are needed in order to determine the toal column density of the gas , and thus its kinetic energy flux , which can be compared to the energy output of the smbh . current x - ray observations from _ chandra _ have found evidence that agn feedback exists in the local universe @xcite . however , x - ray calorimeters on future x - ray satellites will be needed for further improvement as they provide the high throughput and spectral resolution needed to measure column densities and velocities of ionized gas , and consequently the kinetic energy flux , in a large sample of agn across a broad redshift range . * * improvements in resolution and sub - resolution modeling for direct hydrodynamic simulations . * full hydrodynamic cosmological simulations offer the most promising avenue for providing a physically motivated bhmf without free parameters , and for unambiguously identifying the relevant physical processes in building up the bhmf . however , they currently can not resolve scales relevant to the accretion flow onto the smbh . numerical codes based on _ adaptive mesh refinement _ techniques will provide improvement in resolution , but it will likely be a while before hydrodynamic cosmological simulations are able to follow smbh growth in large cosmological volumes while simultaneously resolving the scales relevant for individual black holes . in the meantime , further improvement can be made to the sub - resolution modeling employed by current hydrodynamic simulations . + one way of improving current sub - resolution models may be to implement the results on agn feedback based on the type of work described in the previous bullet - point . another improvement is in modifying the sub - resolution model for the smbh accretion flow . current methods assume the bondi rate combined with a correction factor to account for the fact that the temperature and density of the gas are not resolved at the bondi radius . not surprisingly , the growth of the smbh is sensitive to how this correction factor is modeled @xcite . moreover , sub - resolution models based on the bondi rate neglect the angular momentum of the gas , and thus the bondi rate may not be representative of the actual accretion rate onto the smbh . hopkins & quataert @xcite used high - resolution hydrodynamic simulations to conclude that the bondi rate was a poor estimate of the actual accretion rate onto the smbh , and describe a sub - resolution model which accurately estimated the actual accretion rate in their simulations . in addition , power , nayakshin , & king @xcite suggest an alternative sub - resolution model based on an ` accretion particle ' to provide a more accurate estimate of the black hole accretion rate . the implementation of improved sub - resolution models for accretion rate and feedback into cosmological hydrodynamic simulations , as well as further improvements to the sub - resolution models , will result in more accurate predicted bhmfs , allowing a more insightful comparison with empirical bhmfs . in this paper we have reviewed current estimates of the smbh mass function , as well as theoretical models for the bhmf . as discussed above , each of the methods for estimating the bhmf has their own set of systematics . in figure [ f - compare ] we compare the empirical estimates of the local bhmf ( defined by the shaded region in figure [ f - local_bhmf ] ) with the bhmfs predicted by the theoretical models compiled in figure [ f - theory ] . in addition , in figure [ f - compare ] we also compare the empirical bhmfs at @xmath94 , as estimated using the lightcurve method ( shown in figure [ f - lcurve ] ) and the black hole fundamental plane ( shown in figure [ f - mh08_bhmf ] ) , with bhmfs predicted by the theoretical models . in both figures we also include the bhmfs for broad line agn as estimated by schulze & wisotzki @xcite and kelly et al . @xcite . in general , the theoretical models are consistent to within a factor of a few with the empirical estimates of the bhmf , although there is a large spread in the models and empirical estimates at @xmath94 . moreover , the estimated number densities of broad line agn are significantly lower than those of all smbhs , suggesting that only a small fraction of smbhs are active across a broad range in @xmath0 , except for possibly smbhs at @xmath120 with @xmath105 . despite the differences in the methods for estimating the bhmf , and the theoretical models , they have lead to a number of common conclusions . in particular , the empirical results have presented a picture whereby smbhs grow primarily via accretion in active phases ( eddington ratios @xmath121 ) , that quasar activity is a relatively short - lived phenomenon relative to the lifetime of the smbh and host galaxy ( i.e. , small ` duty cycles ' for agn activity ) , and that smbh growth is anti - hierarchical with the most massive end of the bhmf being built up first . these empirical results are qualitatively in agreement with the steadily improving theoretical models . we would like to thank tommaso treu , priya natarajan , marta volonteri , aneta siemiginowska , xiaohui fan , and marianne vestergaard for helpful comments on an earlier draft of this review . in addition , we would like to thank two anonymous referees whose comments improved our review . bk acknowledges support by nasa through hubble fellowship grant # hf-51243.01 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 .

we have obtained high resolution echelle spectra ( r = 30,000 - 50,000 ) of the na d absorption doublet ( , 5896 ) for six dwarf starburst galaxies and two more luminous starbursts : m82 and ngc 1614 . the absorption features were separated into multiple components and separated into stellar and interstellar parts based on kinematics . we find that three of the dwarfs show outflows , with an average blueshift of 27 km s . this is small compared to the highest velocity components in ngc 1614 and m82 ( blueshifted by 150 km s and 91 km s , respectively ) ; these two brighter galaxies also show more complex absorption profiles than the dwarfs . none of the outflow speeds clearly exceed the escape velocity of the host galaxy . sightlines in ngc 2363 and ngc4214 apparently intersect expanding shells . we compare the shocked gas velocity ( v ) to the ionized gas velocity ( v ) and interpret the velocity difference as either a trapped ionization front ( ngc 4214 ) or a leaky region ( ngc 2363 ) . the dwarfs show n = 10 , while the na d columns in m82 and ngc 1614 are 10 and 10 , respectively . the mass of expelled gas is highly sensitive to outflow geometry , dust depletion , and ionization fraction , but with a simple shell model we estimate neutral outflow gas masses from m to m .

astro-ph0404148

galactic winds are driven by supernovae and stellar winds in starburst galaxies . they drive outflows of metals and dust from the interstellar medium ( ism ) into the galactic halo and even into the intergalactic medium ( igm ) . these outflows may also heat the intracluster medium and enrich it with metals , although the amount of matter permanently escaping the galaxy is still debated . the amount of interstellar gas expelled into the igm by a starburst in a `` blowout '' is thought to depend on the star formation rate within the galaxy , the escape velocity of the galaxy , and the presence of an extended , low - density gaseous halo ( silich & tenorio - tagle 1998 ; legrand et al . 2001 ) . in dwarf starbursts , the mass - loss measurements have focused on the ionized components of this multi - phase flow . in this paper , we explore the kinematics of the cold , neutral clouds in dwarf starburst outflows . due to their low gravitational potential , dwarf starburst galaxies have small binding energies and may be particularly vulnerable to large mass loss via superwinds . previous works have studied x - ray emission from hot gas within the starbursts ( dahlem , weaver , & heckman 1998 ; martin , kobulnicky , & heckman 2002 ) , and optical line emission ( e.g. lehnert & heckman 1996 ) and absorption ( martin & armus 2004 ; rupke , vielleux , & sanders 2002 ; heckman et al . 2000 - hereafter hlsa ) from larger , more luminous galaxies . other studies have observed dwarf galaxies in emission ( marlowe et al . 1995 ; martin 1999 ) , but ours is among the first to study dwarf starbursts in neutral absorption lines at optical wavelengths . we present a small sample of dwarf starburst galaxies and attempt to quantify mass flux in the cold neutral medium ( cnm ) . the advantage of looking at absorption lines is that we can eliminate all ambiguity in the direction of the radial flow of gas . using the starburst itself as a background continuum source , it is known absolutely whether the gas we see in absorption is falling in ( redshifted ) or expanding outward ( blueshifted ) . moreover , unlike emission lines , whose strength scales as density squared , the strength of spectral lines seen in absorption are directly proportional to the column density of the neutral gas along the line of sight . several previous studies ( phillips 1993 ; lequeux et al . 1995 ; gonzalez - delgado et al . 1998 ; hlsa ; martin & armus 2004 ) have detected interstellar absorption lines in starburst galaxies of various sizes and morphologies , finding blueshifted neutral gas and related large - scale outflows of metals . hlsa find an average outflow speed of @xmath8100 km s@xmath1 for luminous infrared galaxies ( lirgs ) and more recently , martin & armus ( 2004 ) and rupke et al . ( 2002 ) find outflow speeds of up to 700 km s@xmath1 for two samples of ultraluminous infrared galaxies ( ulirgs ) . we focus our observations on the absorption doublet ( 5889.95 and 5895.92 , also called the na d doublet ) because it is stronger than any other optical resonance line ( e.g. , ) ; most other cosmically abundant resonance lines are produced in the uv and therefore can not be observed in nearby galaxies with ground - based telescopes . na d is a good tracer of cold neutral gas because its ionization potential ( 5.14 ev ) is less than that of hydrogen , so we can be sure any absorption is occuring in a cold region of the ism . it is also important to note that sodium is not as strongly depleted in diffuse , low - velocity clouds as or ( spitzer 1968 ; savage & sembach 1996 ) . by using high resolution spectroscopy we can resolve individual , kinematically distinct doublet components within the absorption profiles . at lower resolution , deconvolving the absorption components of disparate origins is a more difficult task ( see , e.g. , hlsa ) . it is our goal to observationally determine the kinematics of the cnm in our sample of starburst galaxies . in the second section of this paper , we discuss the observations , data reduction , and basic data analysis . in 3 we detail our method of separating stellar and interstellar sodium , discuss the kinematics and widths of the absorption features , and then use the interstellar sodium to determine a column density . in 4 we present our interpretation of the data , including simple models for outflow scenarios in the individual galaxies . the final section summarizes our major results . we observed eight nearby galaxies using the keck i high resolution echelle spectrometer ( hires ; vogt et al . 1994 ) on 3 - 5 february 2000 . table 1 lists relevant properties of the galaxies . six of the eight galaxies in the sample are classified as dwarfs ; we define dwarf galaxies to have m@xmath12 -18 for our choice of h@xmath13 = 75 km s@xmath1 mpc@xmath1 . ngc 1614 is a lirg ( alonso - herrero et al . m82 is slightly brighter than a dwarf galaxy , but less luminous than a lirg , making it a good object with which to study the transition between the dwarfs and the lirgs ( sofue 1997 ) . the dwarf galaxies were selected mainly for compatibility with martin 1998 , which provides reference h@xmath14 ( @xmath156563 ) narrowband images as well as preselecting galaxies with intense star formation over the range -13.5 @xmath16 m@xmath17 @xmath16 -18.5 . the sample of galaxies observed here are also seen in h@xmath14 to contain expanding supershells of ionized gas . ngc 1614 was not observed in that study , but does have large - scale h@xmath14 structure ( neff et al . the slit in martin s 1998 observations missed the bubble in ngc 5253 ; however , this galaxy was seen to contain an expanding shell in h@xmath14 by marlowe et al . ( 1995 ) . the galaxies were observed with hires using both the red and blue cross dispersers to obtain full coverage of the desired wavelength range ( i.e. from to ) . the red cross disperser covered the spectral range 5400 7700 ; the blue cross - disperser produced spectra in the range 3700 5900 , with echelle orders overlapping in wavelength to give complete spectral coverage to @xmath85000 . one galaxy , ngc 4214 , was observed at two positions , hereafter called ngc 4214 - 1 and ngc 4214 - 2 ( a bright star cluster at the galactic nucleus and a cluster of star - forming knots , respectively see 4.2.4 ) . ngc 4214 - 2 was not observed with the blue cross - disperser . in all galaxies , the slit covered the brightest optical cluster to give the strongest continuum source against which to measure the absorption . decker a12 was used ; the slit size was 10 @xmath18 0.73 for ngc 1614 and ngc 4449 , and 10 @xmath18 1.24 for all other galaxies . the spectra were processed with the echelle package in iraf using standard techniques . the ccd chip overscan region was used to bias - subtract the images , and a quartz lamp was used to flat - field the images in ccdproc . cosmic rays were removed from the ccd images by averaging multiple images of the same galaxy taken in succession on the same night ; if multiple frames were unavailable , the iraf task cosmicray was used to remove singular , bright pixels from the image . we did not flux - calibrate the spectra , as we are concerned primarily with measuring the equivalent width of absorption lines , which are purely a function of the normalized continuum intensity ( spitzer 1978 ) . a thorium - argon lamp was observed for spectral wavelength calibration , resulting in rms residuals in each echelle order of less than 0.5 km s@xmath1 . the resolution of the spectrum is 11 km s@xmath1 ( 6 km s@xmath1 ) for the 1.24 ( 0.73 ) slit , as measured from night sky lines . after the orders were extracted and the sky background was subtracted using the task doecslit , the spectrum for each order was divided by the continuum to provide a normalized spectrum . the absorption line profiles were fit with gaussian components using the iraf task specfit ( kriss 1994 ) . this allowed us to fit multiple pairs of sodium lines to each profile and accurately measure the equivalent width of the lines . the na d doublet was detected in six of the eight galaxies : ngc 1569 , ngc 1614 , ngc 4214 - 1 and -2 , ngc 4449 , ngc 5253 , and m82 . no absorption was seen in ngc 2363 or i zw 18 . we believe the na d in ngc 5253 , ngc 4214 - 1 , and specific components of ngc 1569 and ngc 4449 are likely stellar in nature , via arguments discussed in section 3.1 . the spectra of the five galaxies in which interstellar sodium absorption was detected are shown in figure 1 ; there are a total of ten interstellar na d absorption components . it is interesting to compare the absorption line kinematics to those of emission - line gas . kinematics were previously measured across large regions in the dwarfs in this paper and m82 to map out superbubbles by martin 1998 . our observations directly measured the h@xmath14 also albeit over a much smaller region . nonetheless , we can use the h@xmath14 along these particular sightlines to tie na d kinematics into these larger structures . the detailed results of these spectra will be discussed individually in 4.2 in order to make a statement about the properties of the interstellar sodium in these galaxies , we need to first determine that the sodium we see is actually interstellar . the na d resonance doublet is prominent in the spectra of cooler stars . the spectrum of a dwarf will be dominated by k - type giants and supergiants if the starburst is older than @xmath810 myr ( leitherer et al . 1999 ) ; these stars show strong photospheric sodium absorption . therefore , we must find a way of deciphering the nature of the neutral sodium observed . we have made use of the available stellar spectral lines in our data to help distinguish between stellar and interstellar sodium . in particular , the mg b band triplet ( 5167.32 , 5172.68 , and 5183.60 ) is a good indicator of the presence of certain stellar populations . the mg b triplet is present in f through m stars , with maximum absorption in the range k0 to m3 , and is not present in the cnm because it is a highly excited line . therefore , we can be certain that any mg b absorption we see is from a stellar atmosphere . the equivalent width of this absorption feature is also well correlated with that of the stellar na d absorption ; we can use the mg b to disjoin the stellar and interstellar sodium by fitting the na d absorption profiles with multiple components , one of which corresponds to a purely stellar feature . using the stellar atlas of jacoby , hunter , & christian ( 1984 ) , we have measured the equivalent widths of the mg b and na d absorption lines in a sample of stars and fit the data with a least - squares fit . we find @xmath19 this is shown with the data from the stellar atlas in figure 2 . this is consistent with the calculations of rupke et al . ( 2002 ) that w(nad ) @xmath8 0.5w(mgb ) . we can use this relationship to predict the strength , width and velocity of the stellar component of the na d absorption . there are two galaxies in our sample , ngc 1569 and ngc 4449 , which have both a stellar and interstellar component . ngc 5253 has what appears to be a purely stellar component , and therefore is not included in the sample of galaxies showing interstellar neutral sodium absorption . the equivalent widths of the stellar and interstellar na d and mg b for these three galaxies are shown in figure 3 , and the spectra are presented in figure 4 . the central component of the m82 absorption profile is near the systemic velocity ( it is redshifted by only 4 km s@xmath1 ) ; however , no mg b absorption is seen to a level of 0.24 . since the equivalent width of the na d absorption in this central component is 1.44 , we can safely say that there is no significant population of cool giants and supergiants contaminating the spectrum at this position , and we will assume this component is purely interstellar . hlsa classify their `` interstellar dominated '' sample galaxies as having a larger line width than their `` strong stellar contamination '' galaxies . since our spectra are well resolved and we are able to distingush between stellar and interstellar lines , it is interesting to note that we can use this method , rather than the overall width of the absorption profile , to categorize the na d. there are only two galaxies ( ngc 1569 and ngc 4449 ) where we have both stellar and interstellar lines , and both of these absorption line systems have a larger interstellar width than stellar . similarly , hlsa found that systems with largers widths are interstellar dominated . to discuss the kinematics of the cnm , it is imperative to know the systemic velocity ( v@xmath20 ) of the galaxy , i.e. the heliocentric velocity of the starburst . however , one may wish to compare data with previous observations in which the velocity of the galactic center of mass was used . usually the center - of - mass velocity and the stellar velocity are similar , unless the galaxy is inclined and the starburst region is far from the axis of rotation ; this could create a velocity offset between the starburst velocity and that of the galactic nucleus . in this case , it may be better to use the velocity of the stars themselves . we were able to get stellar velocities for three galaxies ( ngc 1569 , ngc 4449 , and ngc 5253 ) by using stellar absorption lines ( see 3.1 ) . in the other five galaxies , we set the systemic velocity using ( in order of preference ) co maps ( ngc 1614 , ngc 4214 , and m82 ) , optical lines ( ngc 2363 ) , and stellar velocities ( i zw 18 ) . table 1 gives the systemic velocities used in this analysis and their sources . a heliocentric velocity scale is used throughout the paper . the hlsa sample of 32 lirgs showed significant blueshifts in the na d doublet for their `` interstellar dominated '' outflows , indicating nuclear outflows with typical velocities of @xmath8100 km s@xmath1 . our data for the dwarf galaxies also show the na d absorption to be blueshifted , although with overall slower outflow speeds ( @xmath830 km s@xmath1 ) . ngc 1614 and m82 show both redshifted and blueshifted components ; these are further discussed in 4.2 . the velocities for individual interstellar absorption components are given in table 2 . the line width ( fwhm ) of the interstellar components for the dwarfs and m82 ranges from @xmath830 to @xmath880 km s@xmath1 , with an average of 43 km s@xmath1 . the existence of such broad absorption lines in neutral sodium leads us to conclude that there are multiple clouds of cold gas producing each component of the absorption ; the thermal width of a cloud of cold gas at @xmath8100 k would be only @xmath80.20 km s@xmath1 . moreover , the line widths of cold clouds in the galactic halo are found to be just a few km s@xmath1 ( spitzer & fitzpatrick 1995 ) . the individual line widths in ngc 1614 are significantly larger ( @xmath8100 km s@xmath1 or more ) than those in the dwarfs or m82 ( @xmath845 km s@xmath1 ) . the total width of all components in m82 is relatively large ( 236 km s@xmath1 ) . among the dwarfs in the sample , we do not see a correlation between the fwhm of the lines and the rest - frame equivalent width . however , when the data for m82 and ngc 1614 are added , there is a trend wherein equivalent width increases in proportion to fwhm , for these n points . given the lack of correlation seen by hlsa using a much larger data set , we conclude only that the largest values of fwhm ( and similarly the largest equivalent widths ) are much smaller in dwarfs than in brighter galaxies . the doublet lines are fit with gaussian profiles in pairs with specfit to measure the velocity , fwhm , and equivalent width w of each line ( see table 2 ) . if the absorbing gas is optically thin , then the blue ( @xmath155890 ) line will be twice as strong as the red ( @xmath155896 ) line , and the `` doublet ratio , '' w@xmath21/w@xmath22 = 2 . if the gas is optically thick , w@xmath21/w@xmath22 approaches unity as the optical depth @xmath23 approaches infinity . in our data , just one component is optically thin , the rest are optically thick . we can easily measure how much of the continuum source is being obscured by absorbing gas clouds by studying the residual normalized intensity i@xmath24 of a line at wavelength @xmath15 . if i@xmath24 = 0 for the blue line of a doublet ( i.e. the stronger line ) , then no continuum radiation is `` leaking '' into the absorption line . in other words , the absorbing cloud completely covers the continuum source and the covering factor c@xmath25 = 1 . if a cloud does not cover the whole source , the continuum radiation will `` leak '' into the absorption profile and increase the minimum flux , i@xmath24 , and c@xmath25 @xmath16 1 . specifically , following the derivation of barlow & sargent ( 1997 ) , we can find the covering factor : @xmath26 in our data , the covering factor ranges from 0.23 to 1.0 ( see table 2 ) . now that we know the covering factor of the absorbing gas , we can determine the na d column density . the curve of growth relates the optical depth at line center , @xmath27 , to the doublet ratio via a function f(@xmath27 ) , determined numerically ( spitzer 1978 ) . the column density is then @xmath28 ( spitzer 1968 ) , where @xmath29 is the oscillator strength ( morton 1991 ) , and w@xmath24 and @xmath15 are given in angstroms . using this method , we find that n(nad ) in the dwarfs ranges from 6.2 @xmath18 @xmath30 @xmath5 in ngc 1569 to 5.1 @xmath18 @xmath31 @xmath5 in ngc 4449 . we find the total neutral sodium column densities in ngc 1614 and m82 to be 1.0 @xmath18 @xmath32 @xmath5 and 2.4 @xmath18 @xmath31 @xmath5 , respectively . the measured equivalent widths , and the calculated covering factors and na d column densities , are presented in table 2 . from the column density of sodium , we can then use the metallicity of the galaxy ( given in table 2 ; ( na / h)@xmath10 = 2.04@xmath33 martin & zalubas 1981 ) to find the total column probed . a source of systematic uncertainty is the contingence of neutral sodium column density on the ionization parameter , @xmath34 , as well as the fractional depletion of sodium onto grains , @xmath35 . since dwarfs typically have a low metallicity , we expect lower dust content and depletion level than in larger galaxies . it is the product of these , @xmath36 , which is the important factor in our calculations , as is seen in wakker & mathis ( 2000 ) . they measure the abundance of neutral sodium in galactic high- and intermediate - velocity clouds . using their results and a typical column of 10@xmath37 @xmath5 , we calculate @xmath36 = @xmath8300 . we therefore parameterize the column densities by a simple factor of ( @xmath36/100 ) to reflect the typical correction , and calculate the total column using the equation @xmath38 we find that the corresponding n ( ) in the dwarfs could be as large as 10@xmath37(@xmath39 ) @xmath5 ( ngc 1569 ) to 10@xmath40(@xmath39 ) @xmath5 ( ngc 4449 ) . in m82 and ngc 1614 , this conversion gives a total column of 10@xmath41(@xmath39 ) @xmath5 and 10@xmath40(@xmath39 ) @xmath5 , respectively . in addition to na d , the resonance absorption line ( @xmath42 ) is also present in two of our spectra . and have similar ionization potentials ( 4.34 ev and 5.14 ev , respectively ) and similar depletion levels ( savage & sembach 1991 , wakker & mathis 2000 ) , so it is expected that the species will be spatially coincident in the cnm . it is unfortunate that in nearby galaxies , is found in spectral regions that are strongly contaminated by the fraunhofer a - band absorption from atmospheric o@xmath43 ( near 7600 ) . in only two galaxies were we able to discern the absorption : ngc 1569 and ngc 1614 . ngc 1569 is luckily blueshifted slightly away from the sky absorption , while ngc 1614 is redshifted out of the contaminated region entirely . other galaxies are either completely contaminated by the fraunhofer absorption , the doublet is too weak to measure , or the lines lie between echelle orders . the profile is arguably a better constraint on the cold gas kinematics than the na d profile because the doublet spacing is wider , and the oscillator strength is lower so the lines are less saturated . we fit the profile in ngc 1614 with specfit and found components at -149 km s@xmath1 and + 70 km s@xmath1 . this result helped us pin down the best component velocities for the na d fits , which were previously poorly constrained . due to the low signal - to - noise of the continuum , the doublet ratio and covering factor are uncertain . the results of the fitting are presented in table 3 . without the addition of , the na d profile for ngc 1569 is blended with galactic absorption and atmospheric emission . the stellar profile from the starburst is constrained by absorption , leaving a range of possibilities for the central velocity of the interstellar na d. we find absorption at -24 km s@xmath1 , thereby defining the velocity at which the interstellar na d lies . due to the fraunhofer absorption , the normalization of the continuum in the ngc 1569 spectrum is likely imperfect and the column densities in table 3 are highly uncertain . the spectra do not provide a useful column density constraint . although we sample just one sightline , we can estimate how much of the cnm is brought out into the outflow , if any . only five of the sample galaxies show interstellar sodium absorption , just three of which are dwarfs . due to the intricacies in geometry and modeling of the individual galaxies , as well as the small number of sample galaxies , it is preferable to look at each galaxy individually . below we discuss the galaxies independently , detailing the information gleaned from their spectra , and presenting a model of the starburst kinematics and any outflow / infall structure . ideally , we would like to be able to calculate the amount of cold , neutral gas flowing into the igm , when possible . this requires knowledge not only of the kinematics and column densities , but also the geometry of the starbursting region itself . one important application of measuring the column density of a superbubble or shell region around a starburst is to measure the mass of gas being expelled from the starburst . we can do this , so long as we make a few assumptions . we use the wind / bubble model introduced by weaver et al . ( 1977 ) , wherein a starburst is surrounded by concentric shells of ( in order of increasing radius ) a free - flowing wind , a shocked wind , and a shocked shell of neutral ism . the strmgren sphere of the stars may be trapped in the shell , or may completely photoionize the region . we postulate that the h@xmath14 emission is found in the warm , dense ionized gas region , while the na d absorption is found in the colder , shocked neutral gas region . this model is applicable in light of the expanding shells of warm gas seen in h@xmath14 images by martin 1998 . we assume the shells are perfectly spherical , making the mass a function of the column density within the shell and the area of the shell . given a shell of radius r , the mass of the shell in neutral hydrogen is @xmath44 all of the following models use parameters ( r , z / z@xmath10 ) as are detailed in table 4 and column densities as given in table 3 and transformed into n ( ) using equation 4 . the column densities assume ionization fractions and depletion parameters as discussed in 3.3 , namely that @xmath45 . the sightline to ngc1569 intersects the midplane of the disk . we observe the shell 1569-c in h@xmath14 at a radius of @xmath83 ( 32 pc ) from the continuum source . this is most likely the same region in which mhle et al . ( 2001 ) parameterized an `` hole , '' associated with the starburst , 1 kpc in diameter expanding at 7 km s@xmath1 . the neutral gas outflow we observe in the nearby dwarf galaxy ngc 1569 is moving more slowly than the bipolar outflow seen in h@xmath14 by martin ( 1998 ) . if we assume the neutral sodium is spatially coincident with the edge of this hole , i.e. the neutral sodium is swept up in a slowly expanding shell around the starbursting area , then we find the total mass of cold , neutral gas swept up to be @xmath46 m@xmath47 . the mass of this galaxy is found from radio observations to be @xmath48 m@xmath10 . therefore , the neutral gas in the shell modeled is a small fraction ( @xmath4910% ) of the total neutral gas mass of the galaxy . martin et al . ( 2002 ) imaged ngc 1569 in x - rays with _ chandra _ and argued that the observed outflowing wind is strong enough to blow through the galactic halo . in fact , they found that nearly all of the metals ejected by the starburst are carried into the igm by the galactic superwind , and are primarily from the stellar ejecta itself , not the entrained ism . therefore , it is not surprising that this galaxy shows the smallest amount of outflowing neutral gas of our entire sample . the outflowing mass is measured to be @xmath50 m@xmath10 in hot gas and @xmath51 m@xmath10 in warm gas ( martin 1998 ) . in contrast , we find @xmath52 m@xmath10 in for a 500 pc ( radius ) shell . from mg b band absorption we know there must be a stellar component of the na d profile at the systemic velocity . by fitting a corresponding stellar component to the na d profile , and by fitting the atmospheric absorption and galactic emission , we find there is one interstellar absorption component required for a good fit ( see figure 5 ) . this gas , blueshifted 24 km s@xmath1 from the systemic , corresponds to the absorption seen at the same velocity . we use a value of v@xmath20 = -40 km s@xmath53 0.6 km s@xmath1 , obtained from the mg b band absorption in the stellar atmospheres within the starburst . it is important to note that our value of the systemic velocity is somewhat lower than the typical values for this galaxy : meier et al . ( 2001 ) used @xmath54co to find a systemic velocity of -68 km s@xmath1 , while stil & israel ( 2002 ) used the galaxy s rotation curve to find a mean systemic velocity of -82 km s@xmath1 . in fact , when comparing the heliocentric velocity of the na d with the above values of the systemic velocity , it is possible that while the na d is blueshifted in relation to the , it may be at the systemic velocity or even redshifted . additional uncertainty in the analysis of the cold interstellar gas arises from possible absorption by neutral gas in intervening galactic intermediate- or high - velocity clouds ( kuntz & danly 1996 ) . moreover , in the galactic thin disk is seen towards ngc 1569 ( hartmann & burton 1997 ) , and could overlap in velocity with the lines from ngc 1569 . we observed ngc 1614 in order to extend our observations beyond the scale of dwarf galaxies and into a higher luminosity class . this relatively close lirg ( d = 64 mpc for h@xmath55 = 75 km s@xmath1 mpc@xmath1 ) has a violent starburst in the central region which is thought to have been triggered by interaction with another massive galaxy , as evidenced by outer structures including plumes , tails , and a possible secondary nucleus as seen by alonso - herrero et al . we observe the primary / nuclear starburst . the disk of this galaxy is inclined at @xmath56 to the line of sight ( alonso - herrero et al . 2001 ) , so our sightline intersects the outflow . our observations of this galaxy successfully resolve the absorption profile into two separate components a weaker outflowing component at -149 km s@xmath1 , and a stronger infalling component at + 70 km s@xmath1 . the spectra of , @xmath57 , and h@xmath14 are shown , along with the individual components of and , in figure 6 . alonso - herrero et al . ( 2001 ) use the integrated far - infrared luminosity of the galaxy to find a star formation rate of 52 m@xmath10 yr@xmath1 . combined with their value of the total gas mass fueling the starburst ( @xmath58 m@xmath10 ) , we find the starburst age to be 5.8 myr . using this age and the velocity of the outflowing neutral sodium absorption component , we can determine the radius of the shell assuming a constant velocity expansion . we find in this model that the outflowing component supports a shell 1 kpc in radius , with a total outflow mass of @xmath59 m@xmath60 . assuming the redshifted / infalling component is a spherical shell , we parameterize it in terms of a bubble with radius 1 kpc . this case gives a total infalling mass of @xmath61 m@xmath62 . the total mass of this galaxy is given by bushouse ( 1987 ) to be @xmath63 m@xmath10 , so the simple shell geometry can not completely model the cold ism of this galaxy . this complex system is difficult to sort out . the completely uncoupled kinematics of the h@xmath14 and na d spectra suggests that the shock fronts and ionization fronts are moving independently . it is possible that we are seeing redshifted absorption due to neutral gas in one of the filaments of gas or spiral arms of the galaxy the violent interaction of ngc 1614 colliding with another galaxy has created a complex system of filaments and hot spots which may cause the infall of cold , neutral gas . alternatively , there may be separate continuum sources , one redshifted and one blueshifted , and at least one of which is offset from the galactic rotation axis . ngc 2363 is an extremely luminous giant region connected to ngc 2366 . our observations of the nuclear starburst show a very straightforward double - peaked emission line structure with components at an average velocity of 83 km s@xmath1 and 107 km s@xmath1 . this is in agreement with the models of roy et al . ( 1991 , and references therein ) , who observed this galaxy in @xmath64o iii@xmath65 $ ] ( @xmath155007 ) and also found very clear line splitting . we see no na d absorption , and place an upper limit on the column density of n(na i ) @xmath66 @xmath5 . with our emission line data both spectra and echellegrams we can support the claim of roy et al . ( 1991 ) that there is a relatively simple system with one superbubble ( presumably expanding ) around this giant region . our @xmath64o iii@xmath65 $ ] line splitting of 29.4 km s@xmath1 ( i.e. an expansion velocity of 14.7 km s@xmath1 ) , compared with the data of roy et al . ( 1991 ) , leads us to believe that we are looking at a region about 7 arcsec or a physical size of 122 pc at a given distance of 3.6 mpc from the center of the superbubble . from this distance and the blending of the emission lines in our data , we can deduce that we are seeing the edge of the bubble . the difference in the size of the bubble given by our measurements and those of roy et al . ( 1991 ) can be attributed to the difference in assumed distance to the galaxy . our upper limit on the total mass of cold neutral gas is @xmath67 m@xmath68 . the ionized mass of ngc 2363 was found by luridiana , peimbert , & leitherer ( 1999 ) to be @xmath69 m@xmath10 , which indicates that the outflowing cold gas is a small fraction of the galactic mass . the absence of na d absorption tells us that the ionization front created by the starburst may have `` caught up '' to the shock front , so that the ionization front is no longer trapped within the expanding shell . hence , we find a lack of neutral gas , even though there is very obviously a shell present both in previous observations and in our h@xmath14 observations . this results in a `` leaky , '' expanding region , with ionizing radiation escaping beyond the edge of the shell . this nearby dwarf galaxy went through a starburst phase which started roughly 10@xmath70 years ago ( huchra et al . we observed two different positions in the galaxy , as discussed in 2 . ngc 4214 - 1 ( north - west complex , j2000 r.a.=12153948 , decl.=36193100 ) corresponds to a bright super star cluster ( ssc ) and likely the galactic center , while ngc 4214 - 2 ( south - east complex , r.a.=12153799 , decl.=36194420 ) is made of several smaller star - forming knots . we will discuss these regions separately due to their disparate geometry and kinematics . note that our positions ngc 4214 - 1 and -2 correspond to knots 1 and 2 of sargent & fillipenko ( 1991 ) and regions i and ii in mackenty et al . ( 2000 ) . ngc 4214 - 1 is the brightest starburst knot in this galaxy . the central knot in this region is a massive ssc containing several hundred o stars ( leitherer et al . 1996 ; sargent & fillipenko 1991 ) . observations of ionized and neutral hydrogen around this massive star - forming site have found large cavities in the ism , suggesting material was blown out of these areas by stellar winds and supernovae . the starburst can be characterized , using the stellar population , as an instantaneous starburst aged around 3.5 myr ( cervio & mas - hesse 1994 ) . no absorption is seen in this spectra as there is low signal - to - noise near the mg b triplet . additionally , the na d absorption seen is at the local systemic velocity , so we can not make a conclusive statement about the nature of the absorption . however , accounting for the upper limit on both the equivalent width of na d ( 0.04 ) and mg b ( 0.03 ) , we can say that upon comparison with equation ( 1 ) and figure 3 , it is highly likely that the na d absorption seen is stellar . if this is not the case and the absorption is interstellar , then this may be an extension of the observations of leitherer et al . ( 1996 ) , who observe interstellar uv lines at the systemic velocity . assuming the na d absorption is completely interstellar ( i.e. an upper limit ) , then the covering factor is 0.52 and the column density of is @xmath71 @xmath5 . assuming a bubble radius of 700 pc ( an educated guess based on the h@xmath14 morphology seen in martin 1998 ) , we find an upper limit on the mass of the stationary shell of @xmath72 m@xmath73 . this region has a much different morphology than ngc 4214 - 1 , with ionization by three small ob associations ( @xmath84,000 roughly coeval ob stars observed as h@xmath14 knots ) . our h@xmath14 profile is double - peaked , showing a bright emission line blueshifted from v@xmath20 and a weaker redshifted component , suggesting an expanding shell in h@xmath14 . indeed , in looking at the h@xmath14 echellegram , one can easily see the `` hole '' which indicates the center of the shell ( see figure 7 ) . the na d and h@xmath14 spectra are plotted against velocity for ngc 4214 - 2 in figure 8 . since we have no blue spectrum for this position , we can not use to define the nature of the gas . however , the na d profile is blueshifted from the local systemic velocity as given by co ( becker et al . 1995 ) ; this leads us to believe the absorption is interstellar . the neutral sodium is observed to be expanding at 23 km s@xmath1 , with a width of 40 km s@xmath1 . the h@xmath14 and na d profiles have the same doppler width . the kinematics indicate that there is an expanding neutral shell and an ionized shell . the ionized shell is presumably interior to a concentric shell of neutral gas , and the h@xmath14 shell seems to be expanding a few km s@xmath1 faster than the na d shell ( i.e. is more blueshifted ) . assuming a single , expanding shell , the total mass of neutral gas flowing out of ngc 4214 - 2 is @xmath74 m@xmath75 . ngc 4214 has a total mass of @xmath76 m@xmath10 ( kobulnicky & skillman 1996 ; mcintyre 1998 ) . the combined mass of the maximum possible outflow in ngc 4214 - 1 and the outflow in ngc 4214 - 2 is @xmath77 m@xmath10 . this indicates that while the outflow is significant , it is but a fraction of the total neutral ism ; its gas structure as a whole should be considerably affected by the starbursting region . this nearby galaxy has an unusually large b - band luminosity brighter than any other dwarf in this sample ( see table 1 ) . _ chandra _ x - ray imaging shows an expanding superbubble within a cavity of h@xmath14 emission in the region we observe ( summers et al . 2003 ) , and observations show an unusually massive galactic halo ( bajaja , huchtmeier , & klein 1994 ) . imaging in h@xmath14 ( martin 1998 ) shows a system of large shells breaking out along a nuclear bar structure , with smaller bubbles growing along the edge of the galactic disk . it is possible that these smaller bubbles merge into a larger bubble . the hires spectra show that the cold , neutral gas in ngc 4449 is kinematically uncoupled from the warm , ionized gas as seen in h@xmath14 ( see figure 9 ) . the region we observe is the center of the h@xmath14 emission ( hunter et al . 1998 ) , and the neutral sodium we see expanding at 34 km s@xmath1 is most likely a product of several superbubbles expanding out of the galactic plane along the line of sight . indeed , the column density , n@xmath2 = @xmath78 @xmath79 @xmath5 , is larger than in any other dwarf . the unusually large sodium column may be due to additional neutral gas in the halo , explaining the existence of as much as 300 times more neutral matter in this galaxy as compared to ngc 1569 or ngc 4214 ( bajaja et al . if the cnm being expelled from the nuclear region is being pushed by the conglomerate effect of the h@xmath14 bubbles , we can approximate the geometry of the neutral gas by a single bubble with a radius @xmath81 kpc . this radius is on the order of the radius of the largest complex seen in h@xmath14 imaging , which martin ( 1998 ) calls ngc 4449-a ( radius = 950 pc ) . given the high column and radius , we find that this shell contains a total neutral gas mass of @xmath80 m@xmath81 . from radio observations , bajaja et al . ( 1994 ) find a total galactic mass of @xmath82 m@xmath10 , which suggests that we may be seeing the galaxy s massive halo along our line of sight . we detect na d absorption at a 3@xmath83 level in ngc 5253 , but as discussed in 3.1 , this absorption is believed to be stellar because it is at the same velocity as the absorption and it falls directly on the relation for stars ( see figure 3 ) . marlowe et al . ( 1995 ) observe two kpc - scale superbubbles in h@xmath14 flanking the galactic center along its minor axis . we see these in emission only : a brighter line blueshifted by 39 km s@xmath1 and a weaker line blueshifted by 73 km s@xmath1 . our sightline to the nuclear starburst ( 5253 - 1 ; calzetti et al . 1997 ) could intersect either bubble , or both ; the geometry is unclear . the upper limit on interstellar sodium absorption we detect is n(nad ) = @xmath84 @xmath5 , which gives a mass of @xmath85 m@xmath10 ( r/870 pc)@xmath86 using the bubble radius seen in h@xmath14 by martin ( 1998 ) . this is still a large amount of mass , so this is by no means a definitive statement that there is no expanding shell ; our sightline simply misses the shells as shown in martin 1998 ( figure 2@xmath87 in that paper ) . the mass of ngc 5253 is roughly @xmath88 m@xmath10 ( kobulnicky & skillman 1995 ) , indicating that the mass of cold outflowing gas detected is small compared to the total neutral gas mass of the galaxy . our observations of na d in m82 show an extremely complex absorption system , best fit with five separate absorption doublets . the two blueshifted components are the strongest ; there is a single component within 4 km s@xmath1 of the systemic velocity ( v@xmath20 = 214 km s@xmath1 see table 1 ) , and there are two fairly strong redshifted doublet lines . all of the components are highly saturated with the exception of the redmost component , which is optically thin ( i.e. the doublet ratio = 2 ) . m82 is viewed edge - on , and our sightline intersects the nuclear region of the galactic disk . a bright , bipolar outflow is observed out to distances of @xmath8911 kpc along the minor axis of the galaxy ( lehnert , heckman , & weaver 1999 ) . co observations by wei et al . ( 1999 ) , combined with the schematic diagram presented by ohyama et al . ( 2002 , see their figure 7 ) , leads us to believe we are looking through several filaments separated by regions of warmer , nuclear emission . the co images are centered on the most powerful supernova remnant in the galaxy , snr 41.9 + 58 , roughly 17 from our sightline ( r.a.=0955531 , decl.=6940472 ) . using our value of the systemic velocity ( 214 km s@xmath1 , lo et al . 1987 ) , the most redshifted component in our na d spectrum is at approximately the same velocity as the co. this suggests that the redshifted atomic gas is absorption in the galactic disk close to the molecular gas , near the starburst . h@xmath14 images ( martin 1998 ) and images ( wills , pedlar , & muxlow 2002 ) show that a large , expanding superbubble is seen along the minor axis of the galaxy , and is likely made up of several smaller bubbles forming an outflow as they merge . we can make an educated guess on the distance from the continuum source to the absorbing gas by using the ages derived from h@xmath14 shell expansion in martin ( 1998 ) , and find the dynamical age of the shells to be @xmath88 myr . this gives shell sizes on the order of a few hundred pc . assuming solar metallicity , we can find the mass expelled in the simple case of independently expanding spheres . the shells range in neutral gas mass from @xmath90 m@xmath91 to @xmath92 m@xmath93 . these are relatively small compared to the other similarly luminous galaxies where absorption is detected ; it should be noted that the radii given , and therefore the masses , are a lower limit . i zw 18 is important to study due to its extremely low metallicity - o / h @xmath94 0.02(o / h)@xmath10 - and high column , making it a prime candidate for a young galaxy undergoing its first phase of star formation ( lequeux et al . 1979 ) . moreover , the galaxy has a relatively low escape velocity , and feedback from star formation could drastically affect the galaxy ( martin 1996 ) . we observed i zw 18 hoping to see absorption from cold gas in the largest region in the galaxy . however , we see no na d absorption and our spectra show a relatively featureless continuum , with a prominent hydrogen balmer series and weak emission from [ ] ( @xmath04363 , 4959 , 5007 ) , and only the stronger line from the [ ] doublet at @xmath156716 . we can place an upper limit on the column density of na d of @xmath810@xmath54 @xmath5 , using a s / n @xmath84 from the continuum and an assumed line width of 30 km s@xmath1 on par with the line widths in other dwarfs . this translates to an upper limit of @xmath95 m@xmath96 on the mass of neutral gas in i zw 18 , if the cold , neutral gas is entrained in the 970 pc superbubble parameterized in martin 1998 . it would be somewhat surprising to find even this much neutral sodium in this bubble , considering the incredibly low metallicity of this galaxy . using a metallicity of 2% solar , we find an upper limit on the column of @xmath97 @xmath5 . this is calculated given an abundance [ na / h ] = -1.70 , which is consistent with the value of [ fe / h ] = -1.76 derived from _ fuse _ observations by aloisi et al . this roughly result agrees with previous measurements of the column of a few @xmath18 @xmath98 @xmath5 ( aloisi et al . 2003 ; lecavelier des etangs et al . 2004 ; van zee et al . 1998 ) , and is consistent with a @xmath82% solar abundance of sodium . we have obtained high - resolution echelle spectra of six dwarf starburst galaxies , ngc 1614 , and m82 in order to study the cold interstellar gas in dwarf starbursts . interstellar neutral sodium column densities were obtained by measuring the equivalent width of the na d absorption doublet . we find that out of the eight galaxies , ngc 1569 , ngc 1614 , ngc 4214 - 2 , ngc 4449 , and m82 unambiguously show interstellar sodium absorption , while ngc 2363 , ngc 4214 - 1 , ngc 5253 , and i zw 18 do not . the dwarf galaxies ngc 1569 , ngc 4214 , and ngc 4449 exhibit single - component outflows of neutral gas . ngc 1614 and m82 have multiple interstellar components , some outflowing and some infalling , and both galaxies show more absorption from outflowing gas than any dwarf galaxy . the dwarf galaxies show trends similar to brighter , larger galaxies , but on a smaller scale . while samples of lirgs and ulirgs are shown to have average outflow speeds of @xmath8100 km s@xmath1 ( hlsa ) and @xmath8700 km s@xmath1 ( rupke et al . 2002 ; martin & armus 2004 ) , respectively , the three dwarfs show an average outflow speed of only @xmath830 km s@xmath1 . additionally , most ulirgs ( e.g. 73% in rupke et al . 2002 ) show an outflow , whereas only half of the galaxies in our sample have an outflow region . spectral lines in dwarf galaxies also have smaller velocity spreads . it is particularly interesting to see how complex the absorption line spectra become at high resolution . m82 is an excellent example of an absorption line system previously believed to be a single pair of lines , but within each member of the doublet we find a wealth of structure . we resolve at least five line pairs , showing the incredibly complicated nature of the cnm in this galaxy . this is a very intriguing result and prompts us to wonder how many of the galaxies in previous samples could be resolved into complex systems of multiple components . ngc 1614 shows complex absorption as well , though it has a far smoother profile in na d ( and ) than m82 . the three dwarfs with outflows do not seem to exhibit this behavior , showing far simpler spectra . for the first time we are able to combine measurements of the kinematics of the warm , ionized gas in emission with absorption spectra of the cold , neutral gas . for sightlines intersecting a single shell , we can determine whether the ionization front is trapped inside the shock front . this works extremely well for ngc 2363 , which is a simple expanding bubble with an ionization front that has expanded beyond the shock front . ngc 4214 - 2 also presents a relatively simple shell , and this galaxy also shows na d absorption . from the kinematics we postulate that the ionization front is expanding faster than the shock front , though it has clearly not `` caught up '' yet , as there is still a concentric , outer expanding bubble of cold , neutral gas . in other words , the ionization front is still trapped in the shell . the kinematics of na d and h@xmath14 are very different in other galaxies , and can not be easily explained by this straightforward picture . combining our measurements of column density with previous observations of superbubbles and supershells in the sample galaxies , we have parameterized the mass of neutral gas ( or a limiting case thereof ) flowing out of the starbursting region . using a simple spherical shell model , we find the total neutral gas outflow masses to be roughly 10@xmath9 to 10@xmath11 m@xmath10 . ngc 4449 shows far more outflowing cold gas than any other dwarf galaxy . compared to the measurements of outflowing warm and hot gas from these galaxies ( martin 1998 ) , it is likely that the bulk of the energy in the outflow is carried by the warm and hot gas , rather than cold gas . financial support was provided by the david and lucille packard foundation and the alfred p. sloan foundation . 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 . this research has made use of the nasa astrophysics data system abstract service . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . we appreciate comments and clarifications from an anonymous referee . aloisi , a. , savaglio , s. , heckman , t. m. , hoopes , c. g. , leitherer , c. , & sembach , k.r . 2003 , apj , in press alonso - herrero , a. , engelbracht , c. w. , rieke , m. j. , rieke , g. h. , & quillen , a. c. 2001 , apj , 546 , 952 bajaja , e. , huchtmeier , w. k. , & klein , u. 1994 , a&a , 285 , 385 barlow , t. a. , & sargent , w. l. w. 1997 , aj , 113 , 136 becker , r. , henkel , c. , bomans , d. j. , & wilson , t. l. 1995 , a&a , 295 , 302 bica , e. , pastoriza , m. , da silva , l. , dottori , h. , & maia , m. 1991 , aj , 102 , 1702 bushouse , h. a. 1987 , apj , 320 , 49 calzetti , d. , meurer , g. r. , bohlin , r. c. , garnett , d. r. , kinney , a. l. , leitherer , c. , & storchi - bergmann , t. 1997 , aj , 114 , 1834 cameron , a. g. w. 1973 , ssrv , 15 , 121 casoli , f. , dupraz , c. , combes , f. , & kasez , i. , 1991 , a&a , 251 , 1 cervio , m. , & mas - hesse , j. m. 1994 , a&a 284 , 789 dahlem , m. , weaver , k. a. , & heckman , t. m. 1998 , apjs , 118 , 401 gonzalez - delgado , r. , leitherer , c. , heckman , t. , lowenthal , j. , ferguson , h. , & robert , c. 1998 , apj , 495 , 698 hartmann , d. & burton , w. b. 1997 , atlas of galactic neutral hydrogen ( new york : cambridge university press ) heckman , t. , lehnert , m. , strickland , d. , & armus , l. 2000 , apjs , 129 , 493 ( hlsa ) huchra , j.p . , geller , m.j . , gallagher , j. , hunter , d. , hartmann , l. , fabbiano , g. , & aaronson , m. 1983 , apj , 274 , 125 hunter , d. a. , wilcots , e. m. , van woerden , h. , gallagher , j. s. , & kohle , s. 1998 , apj , 495 , l47 jacoby , g. , hunter , d. , & christian , c. 1984 , apjs , 56 , 257 karachentsev , i. d. , makarov , d. i. , & huchtmeier , w. k. 1999 , a&as , 139,97 kobulnicky , h. a. , & skillman , e. d. 1995 , apj , 454 , l121 . 1996 , apj , 471 , 211 kriss , g. 1994 , asp conf . 61 , in astronomical data analysis software and systems iii , ed . d. r. crabtree , r. j. hanisch , & j. barnes ( san francisco : asp ) , 437 kuntz , k. d. & danly , l. 1996 , apj , 457 , 703 lecavelier des etangs , a. , et al . 2004 , a&a , 413 , 131 legrand , f. , tenorio - tagle , g. , silich , s. , knuth , d. , & cervio , m. 2001 , apj , 560 , 630 lehnert , m. d. , & heckman , t. m. 1996 , apj , 472 , 546 lehnert , m. d. , heckman , t. m. , & weaver , k. a. 1999 , apj , 523 , 575 leitherer , c. , vacca , w. d. , conti , p. s. , filippenko , a. v. , carmelle , r. , & sargent , w. l. w. 1996 , apj , 465 , 717 leitherer , c. et al . 1999 , apjs , 123 , 3 lequeux , j. , peimbert , m. , rayo , j. f. , serrano , a. , & torres - peimbert , s. 1979 , a&a , 80 , 155 lequeux , j. , knuth , d. , mas - hesse , j. , & sargent , w. 1995 , a&a , 301 , 18 lo , k .- y . , cheung , k. , masson , c. , phillips , t. , scott , s. , & woody , d. 1987 , apj , 312 , 574 luridiana , v. , peimbert , m. , & leitherer , c. 1999 , apj , 527 , 110 mackenty , j. w. , maz - apellniz , j. , pickens , c. e. , norman , c. a. , & walborn , n. r. 2000 , aj , 120 , 3007 marlowe , a. t. , heckman , t. m. , wyse , r. f. g. , & schommer , r. 1995 , apj , 438 , 563 martin , c. l. 1996 , apj , 465 , 680 . 1998 , apj , 506 , 222 . 1999 , apj , 513 , 156 martin , c. l. & armus , l. 2004 , in preparation martin , w. c. & zalubas , r. 1981 , j. phys . chem . ref . data , 10 , 153 martin , c. l. , kobulnicky , h. a. , & heckman , t. m. 2002 , , 574 , 663 mcintyre , v. j. 1998 , publications of the astronomical society of australia , 15 , 157 meier , d. s. , turner , j. l. , crosthwaite , l. p. , & beck , s. c. 2001 , , 121 , 740 morton , d. c. 1991 , apjs , 77 , 119 mhle , s. , httemeister , s. , klein , u. , & wilcots , e. m. 2001 , astronomische gesellschaft meeting abstracts , 18 , 546 neff , s. g. , hutchins , j. b. , standord , s. a. , & unger , s. w. 1990 , aj , 99 , 1088 ohyama , y. , et al . 2002 , in asp conf . 289 , the proceedings of the iau 8th asian - pacific regional meeting , ed . s. ikeuchi , j. hearnshaw , & t. hanawa ( san francisco : asp ) , 285 papaderos , p. , loose , h .- h . thuan , t. x. , & fricke , k. j.,1996 , a&as , 120 , 207 phillips , a. 1993 , aj , 105 , 486 roy , j .- r . , boulesteix , j. , joncas , g. , & grundseth , b. 1991 , apj , 367 , 141 rupke , d. s. , veilleux , s. , & sanders , d. b. 2002 , apj , 570 , 588 sargent , w. l. w. , & fillipenko , a. v. 1991 , aj , 102 , 107 savage , b. d. , & sembach , k. r. 1991 , apj , 379 , 245 . 1996 , ara&a , 34 , 279 silich , s. a. , & tenorio - tagle , g. 1998 , mnras , 299 , 249 skillman , e. d. , kennicutt , r. c. , & hodge , p. w. 1989 , apj , 347 , 875 sofue , y. 1997 , pasj , 49 , 17 spitzer , l. 1968 , diffuse matter in space ( new york : interscience publication ) . 1978 , physical processes in the interstellar medium ( new york : wiley - interscience ) spitzer , l. & fitzpatrick , e. l. 1995 , apj , 445 , 196 stil , j. m. , & israel , f. p. 2002 , , 392 , 473 storchi - bergmann , t. , calzetti , d. , kinney , a. l. 1994 , apj , 429 , 572 summers , l. k. , stevens , i. r. , strickland , d. k. , & heckman , t. m. 2003 , mnras , 342 , 690 umeda , h. , nomoto , k. , tsuru , t. g. , & matsumoto , h. 2002 , apj , 578 , 855 van zee , l. , westpfahl , d. , haynes , m. p. , & salzer , j. j. 1998 , aj , 115 , 1000 vogt , s. s. , et al . 1994 , proc . spie , 2198 , 362 wakker , b. p. & mathis , j. s. 2000 , apj , 544 , l107 wei , a. , walter , f. , neininger , n. , & klein , u. 1999 , a&a , 345 , l23 wills , k. a. , pedlar , a. , & muxlow , t. w. b. 2002 , , 331 , 313 lcccccccc ngc 1569 & ibm & -40 & 28 & 2.2 & -17.26 & 5.9 & 0.92 & 1 , 2 + ngc 1614 & sb(s)c pec & 4730 & 210 & 64.0 & -20.84 & 170 & 11.25 & 3 , 4 + ngc 2363 & ib(s)m & 70 & 53 & 3.6 & -16.75 & 79 & ... & 5 , 2 + ngc 4214 & iab(s)m & 304 & 35 & 3.6 & -17.65 & 110 & 1.05 & 6 , 2 + ngc 4449 & ibm & 192 & 65 & 3.6 & -17.86 & 135 & 0.34 & 1 , 2 + ngc 5253 & im(pec ? ) & 389 & @xmath1615 & 4.1 & -17.62 & 20 & 0.73 & 1 , 2 + m82 & i0 & 214 & 135 & 3.6 & -18.95 & 88 & 7.98 & 7 , 8 + i zw 18 & ... & 761 & 50 & 10.0 & -13.84 & 2.6 & ... & 9 , 10 + ngc 1569 & -24 & 0.20 & 1.38 & 39.7 & 0.23 & 0.25 & 0.62 @xmath99 0.22 + ngc 1614 & -149 & 4.85 & 1.15 & 294.7 & 0.80 & 0.70 & 42.52 @xmath99 0.20 + & + 70 & 1.13 & 1.08 & 102.7 & 0.52 & ... & 58.93 @xmath99 0.20 + ngc 4214 - 2 & -23 & 0.50 & 1.39 & 39.9 & 0.71 & 0.25 & 2.2 @xmath99 0.28 + ngc 4449 & -34 & 0.17 & 1.07 & 30.9 & 0.27 & 0.25 & 50.9 @xmath99 0.28 + m82 & -91 & 0.62 & 1.36 & 49.1 & 0.69 & 1.0 & 3.0 @xmath99 0.17 + & -35 & 1.03 & 1.11 & 47.2 & 1.00 & ... & 1.7 @xmath99 0.17 + & + 4 & 0.77 & 1.13 & 36.7 & 1.00 & ... & 7.7 @xmath99 0.17 + & + 45 & 1.52 & 1.19 & 78.5 & 1.00 & ... & 9.4 @xmath99 0.17 + & + 86 & 0.46 & 1.98 & 24.6 & 1.00 & ... & 2.4 @xmath99 0.17 + + ngc 2363 & ... & 0.11 & ... & ... & ... & 0.25 & @xmath160.57 + ngc 4214 - 1 & ... & 0.04 & ... & ... & ... & 0.25 & @xmath160.37 + ngc 5253 & ... & 0.07 & ... & ... & ... & 0.37 & @xmath160.20 + i zw 18 & ... & 0.25 & ... & ... & ... & 0.02 & @xmath161.27 + lcccc ngc 1569 & -24 & 500 & 36@xmath991 & @xmath100 + ngc 1614 & -149 & 1000 & @xmath10110 & @xmath102 + & + 70 & 1000 & @xmath10310 & @xmath104 + ngc 2363 & ... & 122 & @xmath161.0 & @xmath105 + ngc 4214 - 1 & ... & 700 & @xmath1642 & @xmath106 + ngc 4214 - 2 & -23 & 500 & 127@xmath9916 & @xmath107 + ngc 4449 & -34 & 1000 & @xmath10865 & 1.4@xmath109 + ngc 5253 & ... & 870 & @xmath1634 & @xmath110 + m82 & -91 & 743 & 381@xmath9921 & @xmath111 + & -35 & 286 & [email protected] & @xmath112 + & + 4 & 500 & 450@xmath9910 & @xmath113 + & + 45 & 500 & 540@xmath9910 & @xmath114 + & + 86 & 500 & 140@xmath9910 & @xmath115 + i zw 18 & ... & 970 & @xmath16275 & @xmath116 +

we describe the evolution of interstellar gas in a family of low luminosity elliptical galaxies all having but with different degrees of flattening ( e0 , e2 , and e6 ) and two current supernova rates , snu = 0.01 and 0.04 . the galaxies are composed of 90 percent dark matter , are rotationally flattened and have isotropic stellar velocity dispersions . the soft x - ray luminosity of the hot interstellar gas after evolving for 15 gyrs decreases dramatically with increasing galactic rotation . as the rotating hot interstellar gas loses energy in the galactic potential , it cools onto a large disk . the outer radius of the disk can be much reduced by increasing the supernova rate which drives a gentle galactic wind transporting high angular momentum gas out of the galaxy . the total mass of cooled disk gas is less sensitive to the supernova rate . although the hot interstellar gas may be difficult to observe in rotating low - luminosity ellipticals , the cooled disk gas can be observed ( i ) in optical line emission since part of the cooled disk gas is photoionized by stellar uv and ( ii ) in the optical continuum , assuming the colder disk gas forms into luminous stars . the mass of hii gas ( ) may be much greater than previously realized since rotationally supported , low density hii contributes little to the global optical line emission . we interpret the stellar disks that are common ( or ubiquitous ) in low luminosity ellipticals as stars that have formed in the cold disk gas . the total mass of cold disk gas available for star formation is similar to the masses of stellar disks observed . the high stellar h photometric index observed in disky ellipticals can be understood by combining the light of young disk stellar populations with that of the old bulge population . _ subject headings _ : galaxies : evolution galaxies : disks galaxies : interstellar gas # 1#2#3#4 # 1#2 .2 in

astro-ph9706276

in a previous paper we discussed the evolution and x - ray appearance of hot interstellar gas in a family of massive , slowly rotating , isolated elliptical galaxies having different ellipticities and rotation rates ( brighenti & mathews 1996 , a.k.a . paper 1 ) . when large , non - rotating ellipticals are flattened by anisotropic stellar velocities we found that the x - ray images are rather insensitive to variations in the galactic ellipticity however , when a small rotation is introduced the x - ray images become significantly flatter than the optical image when viewed perpendicular to the axis of rotation . as the hot gas flows inward in the galactic `` cooling flow , '' it forms a large disk comparable in radius to the effective radius and spins up to the local disk - plane circular velocity which can be quite large for massive ellipticals , @xmath4 km s@xmath5 , a velocity that should be resolved with axaf . after evolving from 1 gyr , when early galactic winds are assumed to have subsided , to 15 gyrs , about ten percent of the total baryonic mass of the galaxy cooled into this large disk . the final disposition of this cooled gas is uncertain ; low - mass star formation is one possibility but a significant amount of gas may remain as low density hii gas ionized by the stellar uv . now we wish to examine the evolution and x - ray appearance of interstellar gas in a second type of elliptical galaxy having a lower luminosity and larger rotation . in recent years it has become apparent that the stellar properties of elliptical galaxies divide into two types : ellipticals of high luminosity having ( modestly ) triaxial shapes , boxy isophotes , more pronounced central cores , less rotation and ( limited ) flattening by anisotropic stellar dispersion ; by contrast , ellipticals of low luminosity generally have disky isophotes , dense central stellar cusps , rotationally induced flattening , approximately axisymmetric geometry and isotropic stellar velocities ( davies et al . 1983 ; nieto et al . 1991 ; kormendy & bender 1996 ; tremblay & merritt 1996 ; faber et al . 1997 ) the changeover in elliptical properties occurs over a range of luminosities , @xmath6 to @xmath7 where both types coexist . both high and low luminosity ellipticals coexist on the fundamental plane ( faber et al . 1997 ) . our interest in the evolution of hot interstellar gas in low luminosity ellipticals has been motivated in part by the prospect of better x - ray observations that will be provided by axaf in the near future . in low luminosity ellipticals the x - ray emission from the hot gas is likely to be masked by the collective emission from ( low - mass ) x - ray binary stars ( e.g. kim , fabbiano & trinchieri 1992 ) . however , the stellar x - ray component is expected to have the same surface brightness distribution as the optical image so it should be possible in principle to subtract the stellar contribution from high quality axaf images , leaving only the emission of the hot gas . although such a subtraction would be difficult because of the low x - ray luminosity of the gas , it would be helped by the different surface brightness distributions expected for stars and gas ( see below ) , by consistent x - ray surface brightness fluctuations anticipated from the stellar component , and by the dissimilar x - ray spectra of stars and gas . the characteristic stellar velocity dispersion ( stellar temperature ) in ellipticals decreases with total stellar mass , @xmath8 ( faber et al . 1997 ) so the temperature of thermalized interstellar gas in virial equilibrium should also decrease with galactic mass and luminosity . however , the galactic binding energy is also proportional to @xmath9 so any additional source of energy in the interstellar gas , such as supernova explosions , may cause the interstellar gas in low luminosity ellipticals to become unbound and flow out of the galaxy as a wind . little is known at present about the x - ray emission from rapidly rotating , coreless , disky , low luminosity ellipticals ; ngc 4697 may be the only such elliptical with measured x - ray fluxes ( fabbiano , kim , & trinchieri 1992 ) . while the evidence for hot gas is currently weak or nonexistent , more may be learned in the near future . the gas - dynamical history of low luminosity , disky ellipticals is also of interest because the stellar disks may result from star formation in the disk of cold gas that is a natural product of rotating cooling flows . the sense of rotation of the stellar disks is identical to that of the bulge component , supporting an internal origin for the stellar disks rather than formation in a merging event . recently de jong & davies ( 1997 ) have reported that disky ellipticals have higher h@xmath3 photometric indices , indicating the presence of youthful stars . they suggest that stars in the disk component , if sufficiently young , could account for the apparent youthful age of the entire galaxy as determined by the population studies of worthey ( 1994 ) and others . it is of some interest therefore to explore the possibility that star formation has occurred in the disks of cold gas that are expected in rapidly rotating ellipticals . finally , the large rotationally supported disks of cooled gas are exposed to ultraviolet ionizing radiation present in galactic starlight . this raises the interesting possibility that the ionized component of these disks can be ( or has been ) observed in faint optical line emission . we find that surprisingly large masses ( @xmath10 @xmath2 ) of warm ( @xmath11 k ) , low density hii gas can reside in the outer parts of these gaseous disks where is it supported by rotation at the local galactic circular velocity and in pressure equilibrium with the ambient hot interstellar gas . the density of most of this hii gas is very low and radiates line emission extremely weakly . the presence of this warm component , which has not previously been discussed , eases somewhat the mystery regarding the ultimate disposition of galactic cooling flow gas after it cools . in the following we discuss the evolution of hot interstellar gas in a family of six low luminosity , ellipticals having different ellipticities , angular momenta and supernova heating rates . as in paper 1 we consider the rotating galaxies to be isolated ; gas inflow or ram pressure effects expected in a cluster environment are not considered here . we find that the radial extent of the disks , the mass of gas entering the disks , and the current mass of hot interstellar gas throughout the galaxies are all very sensitive to the assumed type ia supernova rate , assuming that the supernova energy is shared throughout the interstellar gas . in the most favorable circumstances for disk formation in rapidly rotating galaxies , the disks are very extended . a variety of double power law density models has been proposed for spherical elliptical galaxies ( jaffe 1983 ; hernquist 1990 ; dehnen 1993 ; tremaine et al . 1994 ) . some of these provide excellent global approximations to de vaucouleurs profiles while others ( faber et al . 1997 ) are designed to match central density profiles observed with hst . however , apart from their success in fitting the projected surface brightness distribution , the lack of dark matter and infinite extent of these double power law models make them inappropriate for determining stellar velocities for our models . on the other hand we find that the projected surface brightness of king profiles are adequate approximations to de vaucouleurs profiles , certainly for our purposes here , and also have much mathematical simplicity particularly when extended to include oblate galaxies . for these reasons we have decided to retain the models used in our previous paper ( brighenti & mathews 1996 , paper 1 ) on more luminous ellipticals : simplified king profiles for the stellar density and pseudo - isothermal structures for the dark matter distribution , i.e. @xmath12 the two density distributions depend on a single parameter @xmath13 and are both truncated at some large elliptical surface defined by @xmath14 and @xmath15 where @xmath16 is the eccentricity of both the stellar and dark matter . we generate a family of equal - mass elliptical en galaxies all having the same @xmath17 and @xmath18 by scaling the @xmath19core radius according to @xmath20 where @xmath21 is the core radius of the spherical galaxy . the detailed procedure used in constructing the models and the total galactic potential ( both within and beyond the galactic mass distribution ) are described in paper 1 . as in paper 1 we solve the two dimensional jeans equations to find the stellar temperature @xmath22 at every position in the galaxy . here @xmath23 is the stellar velocity dispersion in the meridional plane and @xmath24 is the random stellar dispersion in the azimuthal direction . in paper 1 we employed the satoh decomposition of the mean square azimuthal speed into a random dispersion and a systematic rotation : @xmath25 where @xmath26 ( satoh 1980 ) . in general @xmath27 is an unknown function of both spatial coordinates , but is usually regarded as constant throughout the galaxy . in this current paper we consider only galaxies that are rotationally flattened , i.e. @xmath28 and which therefore must have have isotropic velocity ellipsoids . the evaluation of the potential and stellar velocity dispersion simplifies considerably in the spherical limit , see paper 1 for further details . we use an extended hubble nomenclature to describe our models : ena;k@xmath29 where en is the normal hubble designation with @xmath30 $ ] , @xmath31 is the satoh parameter and a = h or l is an additional optional distinction that characterizes the `` high '' or `` low '' supernova rate assumed ( as explained below ) . we consider here a family of galactic models all having the same luminosity , central densities , and total mass but with different degrees of rotationally induced flattening . in table 1 we list the parameters of the spherical member of the family , e0;0 . the basic parameters that define the stellar distribution , @xmath32 , @xmath33 and @xmath34 are chosen so that the galaxy lies on the fundamental plane ( tsai & mathews 1995 ; paper 1 ) . these parameters are also nicely consistent with recent fits to hst observations ( faber et al . 1997 ) with one exception : our stellar core radius @xmath35 pc is about ten times larger than upper limits for possible cores in `` power - law '' galaxies of the same absolute magnitude . if we decrease the core radius to be consistent with the hst upper limits while also maintaining the tsai - mathews fundamental plane relationships , we find that the overall luminosity of the galaxy becomes too faint to be of much interest ( @xmath36 ) . nevertheless , the galaxy described in table 1 is entirely satisfactory for our models of the global large - scale evolution of interstellar gas throughout the galaxy in which the stellar core region has no influence and is poorly resolved by the hydrodynamic grid . finally we note that the central density and core radius of the dark matter component have been chosen so that the total dark mass is 9 times that in the stellar component and its distribution is such that dark matter has little influence on stellar velocities within @xmath37 effective radius . figure 1 shows the distribution of stellar temperature @xmath38 and azimuthal velocity @xmath39 in a quadrant of the e2;1 member of the galactic family corresponding to the e0;0 galaxy in table 1 . at the outer edge of the galaxy the stellar velocity dispersions in the meridional plane must vanish since no stars can move beyond this surface ; however , when @xmath40 the dispersion is isotropic and the azimuthal dispersion @xmath24 ( and @xmath41 ) must also vanish at the outer galactic boundary . consequently , the contours of @xmath38 are much more elliptical than those of the ( slowly rotating ) e2;0.25 galaxy shown in paper 1 . the azimuthal stellar velocity @xmath42 and circular velocity on the equatorial plane are illustrated in figure 2 . the rather strong maximum in @xmath42 at a few hundred parsecs from the galactic center , particularly for the e6;1 galaxy , may differ from stellar rotation curves observed in some low luminosity ellipticals [ e.g. ngc 4697 observed by binney , davies , and illingworth ( 1990 ) ] , but stellar rotation curves are unavailable for most of the low luminosity ellipticals known to have power - law density profiles . in any case , the global interstellar gas dynamics are not strongly influenced by this maximum in @xmath42 . as the hot interstellar gas cools and flows inward , it is expected to spin up to the local circular velocity @xmath43 and become rotationally supported . the gas dynamical equations are identical to those used in paper 1 . interstellar gas is created by normal mass loss from an evolving population of old stars at a rate @xmath44 where @xmath45 gyr represents the present time and @xmath46 s@xmath5 ( mathews 1989 ) . the contribution of gas from type ia supernovae to the interstellar mass is negligible ( @xmath47 ) but is more energetic and iron - rich . the energy generated at early times by type ii supernovae is expected to expel most of the interstellar gas created prior to some time which we adopt as 1 gyr . between 1 and 15 gyrs about one tenth of the stellar mass remaining at @xmath48 gyr enters the interstellar medium . for simplicity , we do not vary the galactic potential during this time interval to account for this small stellar mass loss . the temperature of the interstellar gas is similar to the local stellar temperature since both components are in hydrostatic equilibrium in the galactic potential , although the gas can also be heated by supernova explosions . the characteristic source - term temperature for the gas is @xmath49 where @xmath47 is assumed . heating by supernova explosions is given by @xmath50 ( see mathews 1996 ) where @xmath51 is the reduced hubble constant and we adopt @xmath52 ergs as the typical hydrodynamic energy released in a supernova event . the temperature @xmath53 ( @xmath54 proton mass ) depends on the average mass lost per supernova @xmath55 . the total rate that supernovae supply gas to a galaxy of mass @xmath56 is @xmath57 where @xmath58 is the supernova rate in sec@xmath5 ; to convert @xmath58 to snu units ( number of supernovae per 100 years from stars of luminosity @xmath59 ) we use the mass to light ratio determined by van der marel ( 1991 ) , @xmath60 where @xmath33 is in @xmath61 . the supernova rate is assumed to decrease with time as a power law : @xmath62 although there is at present no compelling observational reason to adopt this form . we consider two values for the current supernova rate , @xmath63 and @xmath64 . these rates are comparable with the most recent observed value @xmath65 ( turatto , cappellaro & benetti 1994 ) . we do not expect exact agreement between the observed value @xmath66 and the factor @xmath67 in the expression above for @xmath68 since snu@xmath69 is only one of several uncertain parameters that influence @xmath68 in our gas dynamical models . with the higher value of @xmath70 we find that the iron abundance in the interstellar gas exceeds those observed by the asca satellite ; possible origins of this discrepancy have been discussed by arimoto et al . ( 1997 ) . for simplicity we distribute the supernova energy ( and iron ! ) smoothly throughout the interstellar gas ; the accuracy of this often - used assumption is difficult to access since the initial deposition of supernova energy is localized in hot interstellar bubbles ( mathews 1990 ) . since @xmath71 decreases with time more rapidly than snu@xmath69 , the influence of supernova heating ( and the likelihood of galactic winds ) increases with time in our models . as in paper 1 we assume that the galaxy is completely isolated , allowing hot gas to flow freely beyond the stellar edge . the gas flow is regarded as invicid , i.e. the magnetic field is sufficiently large to drastically reduce the plasma mean free path yet the field is sufficiently disordered on small scales ( @xmath72 ) not to communicate stresses over large distances in the interstellar gas . for numerical solutions describing the evolution of the interstellar gas we used the ncsa eulerian hydrocode zeus2d , appropriately modified as described in paper 1 . the influence of galactic rotation on the nature and evolution of interstellar gas in ellipticals can be determined by comparison with otherwise identical spherical , non - rotating models . properties of the interstellar gas in the non - rotating e0;0 galaxy at time @xmath45 gyr are shown in figure 3 . the interstellar gas flows in a axisymmetric cylindrical computational grid of 120 @xmath73 120 zones with gradually increasing zone size ; the smallest central zone is half the stellar core radius . solutions are initiated at time @xmath48 gyr , when it is assumed that snii explosions and early galactic winds have subsided , and continued to 15 gyrs with two values of the current supernova rate , snu@xmath74 ( e0l;0 ) and snu@xmath75 ( e0h;0 ) . the linear excursions of the gas density and temperature at small radii in figures 3a and 3b are artifacts of the limited spatial resolution in the innermost grid points . the temperature of the gas is slightly larger than the virial temperature for the solution with the higher supernova rate ( e0h;0 ) so some gas flows out of the galaxy . as a result the gas density and x - ray surface brightness in the e0h;0 galaxy are lower than for the e0l;0 galaxy while the gas temperature in the e0h;0 galaxy is higher , particularly in the outer parts of the galaxy . the x - ray surface brightness is calculated for the 0.1 - 2.4 kev energy band to approximate rosat observations . the effect of changing the supernova rate is also apparent in table 2 which lists several global properties of these models . the mass of hot gas within the galactic boundary @xmath76 and ( especially ) the total soft x - ray luminosity @xmath77 ( 0.1 - 2.4 kev ) are both reduced when the supernova rate is increased . the total amount of cold gas that accumulates in the galactic core @xmath78 is also significantly lower as a result of gas outflow . the gas velocity in these spherical galaxies is very subsonic everywhere . as gas loses energy by radiation it sinks subsonically in the galactic potential and is maintained at approximately constant temperature by @xmath79 compression in the galactic potential . superimposed on this global flow are low - amplitude , highly subsonic random motions ( @xmath80 km s@xmath5 ) . we believe these random velocities are numerical in origin , but similar small velocities could be generated in real galaxies by small physical irregularities . these small velocities are nevertheless sufficient to prevent the appearance of a `` galactic drip '' or global thermal instability ( mathews 1997 ) that appears when these same calculations are performed with a 1d spherically symmetric grid . the implications of these results for the reality of galactic drips are unclear at present . as expected from angular momentum conservation , in rotating galaxies the gas cools into a large , rotationally supported disks . gas that cools below @xmath81 k is held at this temperature throughout the remainder of the calculation . the mass of this cold gas in the disk generally increases with time . the total vertical column density in cooled gas is shown at three times in figures 4a , b , and c for the disks of the e2l;1 , e2h;1 , and e6l;1 galaxies respectively . the e2h;1 elliptical with the higher supernova rate ( fig . 4b ) has a much less extended disk owing to the outflow of ( high angular momentum ) gas from the galaxy in a slow wind . at the last time plotted in figure 4b the e2h;1 disk has actually become smaller probably as a result of local heating of gas by supernovae . however , the large disks that form in the e2l;1 and e6l;1 galaxies are very similar . in both disks the outer edge increases with time , similar to the growth pattern of disks in more slowly rotating ellipticals ( paper 1 ) and the surface density increases slowly within the body of the disk , indicating that gas enters the disk at all radii , not just at the outer edge . the similarity of figures 4a and 4c indicates that the column density beyond several stellar core radii is almost independent of the degree of galactic rotational flattening . the e2l;1 and e6l;1 galaxies both have power - law , not exponential , disks with surface densities given by @xmath82 . even in these maximally rotating galaxies with @xmath28 , gas ejected from stars moves a considerable distance from its point of origin toward the rotation axis before settling into the cold disk . as gas moves along a streamline toward the disk , it mixes with locally produced gas having different angular momentum . in spite of this complication , it is possible to estimate the radius @xmath83 in the disk where mass lost from stars at @xmath84 will eventually cool ( see paper 1 ) . in figure 5 we plot contours of equal @xmath83 for the e2:1 galaxy . these contours are ( implicit ) solutions of @xmath85 , the condition that an element of gas ejected by stars at @xmath84 strictly preserves its specific angular momentum until it reaches the disk at radius @xmath86 . for example , figure 5 indicates that gas ejected from stars at @xmath87 , @xmath88 kpc has the same angular momentum as gas in the disk at @xmath89 kpc . the flow in galaxy e2l;1 shown in figure 6 exhibits subsonic kinetic activity and corresponding irregularities in the density contours , particularly in the outer galaxy . within about 10 kpc , however , the flow toward the central disk is regular and the density contours have an ellipticity similar to that of the local stars . however , notice that a relatively dense and cold region has appeared in the flow at time @xmath45 gyr near @xmath90 , @xmath91 kpc . this cold , transient lump of gas is falling toward the disk plane where it will eventually reside . such intermittent contributions to the far outer edge of the cold disk are also evident from figure 4 and may be a generic feature in rapidly rotating cooling flows . similar inhomogeneities have also appeared in the calculations of dercole & ciotti ( 1997 ) who are studying the evolution of hot gas in s0 galaxies . at time @xmath92 gyrs the most rapid rotation above the disk plane , @xmath93 km s@xmath5 , is near @xmath94 kpc ; the full width of emission line profiles in the x - ray spectrum of this galaxy is @xmath95 km s@xmath5 , within the resolution capability of axaf . the irregular velocity field in figure 6 , unlike the laminar flow in the slowly rotating galaxies described in paper 1 , may result from shear instabilities that first develop in the low density outer galaxy . the azimuthal velocity of gas just within the galaxy is driven somewhat by ejecta from azimuthally streaming stars . such an interaction may enhance shear instabilities at the galactic boundary , but we have not investigated in detail the physical origin of these very subsonic velocities . when the supernova rate is increased by a factor of 4 , a subsonic wind develops throughout most of the galaxy at time @xmath45 gyr as shown in figure 7 . the gas density is lower and its spatial variation is more regular in this gentle galactic wind than in the flow in figure 6 . since we assume that mass enters the flow at a rate @xmath96 that decreases faster than the supernova rate , snu@xmath97 , the thermal energy delivered to each gram of interstellar gas slowly increases with time and the overall solution tends toward a galactic wind as the calculation proceeds . finally , in figure 8 we show the density and velocity fields at @xmath98 gyr in the rapidly rotating e6l;1 galaxy . in such flat galaxies incipient winds first appear near the outer edge of the equatorial plane near @xmath99 kpc . winds occur first in the equatorial region because ( i ) the ( more spherical ) gravitational potential is less here and ( ii ) the presence of rotation reduces the effective potential . notice that there is another `` lump '' of cooling gas descending toward the equatorial plane at @xmath100 kpc similar to the cool lump discussed in figure 6 . the appearance of the hot interstellar gas when viewed in soft x - rays ( 0.1 - 2.4 kev ) is shown for three galaxies in figure 9 . the galactic breeze driven by supernova heating in the e2h;1 galaxy has a dramatic effect in circularizing the x - ray image compared to that of the e2l;1 galaxy of the same age . however , it is unclear if the emission from the hot gas illustrated in figure 9 can be observed because it is masked by the collective emission from low mass x - ray binaries in the stellar component . ideally , with high signal to noise axaf observations it would be possible to subtract the soft x - ray contribution from stars since the stellar surface brightness distribution is known from optical observations . although the possibility of such a decomposition to find the x - ray emission from the gas alone was one of the initial motivations for our series of calculations , the x - ray luminosities of the interstellar gas in the rotating galaxies in table 2 may be too small to be detectable . the stellar @xmath77 from ellipticals is poorly known , but ciotti et al . ( 1991 ) suggest @xmath101 ergs s@xmath5 which for our galaxy is @xmath102 so the @xmath77 from the interstellar gas will be a disky component only a few percent of the total a challenging observation even for axaf . the x - ray emission from non - rotating galaxies in table 2 could be observed quite easily , but it is unlikely that many low luminosity galaxies are either spherical or non - rotating ( tremblay & merritt 1996 ) . the final physical configuration of cooled gas in galactic and cluster cooling flows remains one of the most perplexing unsolved problems in galactic evolution . the usual solution is to assume that stars of very low mass and luminosity form in these high - pressure environments ; there is currently no strong observational evidence against this hypothesis . in the following discussion , however , we examine two new aspects of this problem that arise when galactic rotation is considered . first , regardless of the final endstate of the cooled gas , there will always be some disk gas that can not form into stars because it is photoionized by the galactic and intergalactic ionizing radiation fields . we discuss below the mass and possible observability of this ionized gas . second , we consider the hypothesis that relatively young luminous stars have formed in the cooled gaseous disks and that these disk stars are responsible for the systematically larger h@xmath3 photometric indices observed in disky ellipticals of low luminosity . some of the gas that cools onto the rotating disk is kept ionized by ambient ionizing radiation and remains at @xmath103 k , the same temperature we have adopted for all the cooled gas . in the inner parts of our disks the maximum column depth of hii gas that can be ionized by the available radiation , @xmath104 , is much less than the total column of cooled disk gas @xmath105 . such a layer of ionized gas should be stable toward forming gravitational irregularities . however , regions deeper in the disk than @xmath104 are in reality expected to be colder than @xmath106 k and are probably unstable to cloud ( and star ? ) formation . if dense condensations should form in this colder gas , the hii gas should still be present as a distributed component having a column density @xmath107 . in the inner disk therefore we expect a column @xmath104 of hii gas to always be present even if most of the colder disk gas is highly clumped or stellar . in addition , our solutions indicate that a considerable mass of hii gas may be present in the outer disk provided the supernova rate is low . figure 10 shows a plot of the variation of the total column of cold gas @xmath105 and the maximum amount that can be ionized @xmath104 . beyond about @xmath108 kpc in the e2l;1 galaxy all the disk is fully ionized . we describe below how we have estimated the column @xmath104 and the total mass of hii gas . we also discuss the likelihood of detectable optical line emission from the hii disk . the ionizing radiation has both galactic and extra - galactic components . stellar uv flux has been measured in seven ellipticals from 1800 @xmath109 to the lyman edge ( ferguson & davidsen 1993 ; brown , ferguson , & davidsen 1995 ) and binette et al . ( 1994 ) have estimated the entire uv spectral energy distribution and the total galactic ionizing photon luminosity @xmath110 from the collective radiation of post - agb stars in an old stellar population . such estimates are very uncertain since the evolutionary tracks of highly evolved stars and their atmospheres are poorly understood . at time @xmath111 gyr following a single burst of star formation , binette et al . find @xmath112 s@xmath5 @xmath113 ; for our galaxy this is @xmath114 s@xmath5 . we have attempted to reconcile this estimate with the uv observations by averaging values of @xmath115 at 1100 @xmath109 and 1500 @xmath109 for the six galaxies observed by brown , ferguson , & davidsen ( 1995 ) . our values of @xmath33 are based on @xmath116 and the galactic properties listed by faber et al . individual values of @xmath115 at both wavelengths range over a factor of about four , but the ratio of the mean values at these two wavelengths agrees almost exactly with the slope of the uv spectral energy distribution in this part of the spectrum given in arbitrary units by binette et al . ( 1994 ) . using the uv observations at these two wavelengths to calibrate the ionizing continuum of binette et al . ( 1994 ) we find @xmath117 s@xmath5 which for our galaxy ( e.g. table 1 ) is @xmath118 s@xmath5 in good agreement with the estimate of of binnette et al . we shall therefore adopt here a value @xmath119 s@xmath5 at @xmath92 gyr which is slightly less than the value from binnette et al . at @xmath111 gyr since @xmath110 slowly decreases with time . some of this lyman continuum radiation is consumed ionizing dense circumstellar gas in newly formed planetary nebulae , but we expect this fraction to be small since the nebular gas is swept back along the stellar orbit by ram pressure on time scales short compared to the total uv lifetime of the central stars ( mathews 1990 ) . if the total stellar ionizing photon emissivity follows the starlight , @xmath120^{-3/2}$ ] , we determine the coefficient @xmath121 @xmath122 s@xmath5 and the ionizing photon density at any radius in the galaxy by using the expressions for the radiation density from tsai and mathews ( 1995 ) . for simplicity we use the spherical relations of tsai and mathews to estimate the stellar ionizing photon density and ionized column in the disk of the moderately flattened e2l;1 galaxy . for the intergalactic contribution to the ionizing photon density we use the value @xmath123 @xmath122 ( dove & shull 1994 ) which is constant throughout the galaxy . the ionized column along the disk plane @xmath124 is shown in figure 10 for the e2l;1 galaxy at time @xmath92 gyrs . beyond the galactic core radius in the e2l;1 galaxy the ionizing photon density varies as @xmath125 @xmath122 . this stellar ionizing uv dominates the intergalactic component throughout most of the galactic volume within @xmath126 kpc which includes all of the disk region . we assume that the hii gas in the disk is in pressure equilibrium with the local hot interstellar gas and has temperature @xmath103 k everywhere . since the temperature of the cooled gas in our computations is constrained to @xmath106 , it already has the appropriate density for the hii gas although the density in our cold isothermal disks decreases somewhat away from the disk midplane . beyond the galactic core radius @xmath127 along the midplane of the disk the density of hii gas decreases sharply with radius : @xmath128 @xmath122 . estimates of the total mass of hii gas in ellipticals based on a single assumed density are therefore misleading and incorrect . in the dense inner parts of the disk only a small fraction of the cold disk gas can be ionized but beyond about 9 kpc all the disk gas is ionized . the total mass of ionized gas is very large , @xmath129 @xmath2 , more than ten percent of the total amount of gas that cooled in the disk since 1 gyr ( table 2 ) . the mass of hii gas within radius r along the disk @xmath130 @xmath2 ( valid from @xmath131 to 12.6 ) increases dramatically toward the outer parts of the disk where most of the hii mass resides . the total column density of all the gas that has cooled after @xmath92 gyr is large , @xmath132 @xmath133 ( valid from @xmath134 to 10 ) . this gas can not be smoothly distributed in the disk ( i.e. not in form of clouds or stars ) since it would be easily observed by x - ray absorption out to 2.5 kpc where @xmath135 @xmath133 . however , when self - gravity of the hii gas is included we find that it is gravitationally stable against collapse at every radius in the disk so the hii gas will not form into clouds . the total h@xmath3 luminosity from the disk in our e2l;1 galaxy , @xmath136 ergs s@xmath5 , corresponds to a flux @xmath137 ergs @xmath133 s@xmath5 when viewed at a distance of 20 mpc . this luminosity and flux are comparable to those observed in ellipticals of similar optical brightness ( e.g. trinchieri & di serego alighieri 1991 ) so we can expect that ionized disks contribute a substantial fraction of the total line emission from early type galaxies . because of the strong variation of hii gas density and ionizing photon density with galactic radius , only about 4 percent of the total mass of hii in the dense central regions , @xmath138 kpc , contributes half of the total h@xmath3 luminosity while half of the total hii mass lies far out in the galaxy beyond @xmath139 kpc where all the disk gas is ionized . this low density hii gas at @xmath140 kpc emits only @xmath141 percent of the total @xmath142 over a vast surface area and may be undetectable with present technology . however , the total balmer line luminosity @xmath143 ergs s@xmath5 far exceeds the minimum luminosity expected in galactic cooling flows due to the recombination of gas cooling for the first time ; for the e2l;1 galaxy this latter value is @xmath144 ergs s@xmath5 at time @xmath45 gyr . finally we note that @xmath142 is about 45 times larger than the soft x - ray luminosity @xmath77 from the e2l;1 galaxy listed in table 2 and therefore represents a significantly larger fraction of the overall galactic radiation budget . we have examined in a similar manner the hii mass and line luminosity in the other galaxies listed in table 2 . the radial size of the disk in the e2h;1 galaxy , @xmath145 kpc , is greatly reduced by the higher supernova heating rate assumed . the total mass of hii gas expected from the available ionizing photons is much smaller , @xmath146 @xmath2 , but @xmath147 erg s@xmath5 is only six times less than that of the e2l;1 galaxy since most of the h@xmath3 emission is from dense gas in the inner disk . the corresponding flux at 20 mpc is @xmath148 ergs @xmath133 s@xmath5 . the relative insensitivity of the h@xmath3 luminosity to the physical size of the disk is interesting because it can not explain the enormous range in @xmath149 observed ; for example trinchieri & di serego alighieri ( 1991 ) find that @xmath150 varies over a factor of @xmath151 among ellipticals having @xmath152 @xmath61 . the explanation for this large spread in balmer luminosities could lie in varying magnetic field strengths in the galactic centers ; a field of @xmath153 gauss in the dense hii gas near the center of our e2l;1 galaxy could sharply reduce the gas density and the h@xmath3 emissivity . the e6l;1 galaxy has an extended disk with the following properties : @xmath154 @xmath2 , @xmath155 erg s@xmath5 , and @xmath156 ergs @xmath133 s@xmath5 . what evidence exists for the presence of hii disks in ellipticals ? smooth , extended images of ellipticals in optical emission lines ( e. g. trinchieri & di serego alighieri 1991 ; macchetto et al 1996 ) do not necessarily require emission from ionized disks and indeed some hii emission from young planetary nebulae and other ionized stellar ejecta is expected throughout the galactic volume ( mathews 1990 ) . unfortunately it is difficult to estimate the precise amount of hii gas that has been recently ejected from stars but has not yet entered the hot phase since the rate of this transition depends on the influence of the magnetic field configuration on thermal conductivity . buson et al . ( 1994 ) present optical line images of a sample of ellipticals known to be unusually bright in optical emission lines . these ellipticals also have liner spectra perhaps suggesting an additional non - stellar ionization source although the line emission symmetry does not have the biconical pattern often observed in seyfert galaxies . buson et al . typically find smooth h@xmath157 + [ nii ] images within @xmath158 that are extended along the major axis ( ngc 1395 , ngc 1453 , ngc 2974 , ic 1459 ) . the kinematics of the hii in these ellipticals indicates regular , organized rotation around the galactic center ( zeilinger et al . 1996 ) ; this rotation is consistent with disk emission but some ( smaller ) rotation would also be present in line emission from recent stellar ejecta distributed throughout the galactic cooling flow . in general the emission line surface density in the hii `` disks '' observed by buson et al . is confined to the central regions and drops faster with radius than the stellar light . in a few ellipticals buson find that the major axis of the hii gas is misaligned with the optical axis of the stellar image ( ngc 3962 ) or is counter - rotating ( ngc 6868 , ngc 7097 ) ; in these galaxies the hii disks are evidently disturbed or created possibly as a result of a recent merger event . several straightforward observations would help clarify the contributions of disks to the hii emission from ellipticals . the disk component is expected to have a higher ellipticity than the stellar image and this difference should be most pronounced in ellipticals having high diskyness or @xmath159 that are as a class viewed more nearly perpendicular to the axis of rotation . in low luminosity ellipticals of high @xmath159 the hii rotation velocity should be representative of the circular velocity on the disk plane which is always greater than the rotation of the bulk of the galactic stars . the h@xmath3 index is one of the many observable characteristics that distinguish elliptical galaxies of high and low luminosity . this photometric index measures the strength of balmer line absorption in the galactic spectrum and is the principal observational means of breaking the `` age - metallicity degeneracy '' to reveal the presence of a subpopulation of youthful stars ( gonzalez 1993 ; worthey 1994 ; faber et al 1995 ; worthey et al . recently de jong & davies ( 1997 ) have shown that h@xmath3 is systematically larger in ( low - luminosity ) disky ellipticals ( kormendy & bender 1996 ) ; they also conjectured that the young stellar subpopulation may be the stars in the disk . using guy worthey s stellar population models , de jong & davies combined the h@xmath3 indices from a young ( 2 gyr ) stellar disk containing about 10 percent of the total galactic stellar mass with a much older component ( 12 gyr ) representing the bulk of the old stellar population in the galaxy . they found that the combined h@xmath3 index of the disk and bulge is similar to the higher h@xmath3 indices observed within @xmath160 in disky ellipticals . we have described here how cold gas gradually collects in a rotationally supported disk during the slow evolution of the galactic interstellar medium . this disk provides an ideal environment to form a younger population of disk stars . often in theoretical studies of cooling flows it is assumed that only stars of very low mass form in the cooled gas expelled from the older generation of galactic stars . we shall assume instead that luminous , moderately massive stars form in the cold disk gas but the stellar masses in the young disk population must not extend above about 8 @xmath2 since no type ii supernova have been observed in ellipticals . to test this possibility more quantitatively , we have computed stellar h@xmath3 indices assuming that the disk consists of a number of discrete population bursts having mass @xmath161 and age @xmath162 . these quantized bursts are listed in table 3 for the e2l;1 and e2h;1 galactic models . the burst masses and that of the main galactic bulge are all evaluated within @xmath160 for comparison with the observed values of de jong & davies ( 1997 ) . for simplicity we assume all stellar populations have solar abundance . after creating a variety of stellar population models using guy worthey s website ( http://www.astro.lsa.umich,edu/users/worthey/ ) , it became clear that the h@xmath3 index is rather insensitive to the slope or precise mass cutoffs of the imf but does depend strongly on the ages of the youngest stars considered . stars younger than the youngest stellar population available at the worthey website ( 1 gyr ) can contribute significantly to the overall h@xmath3 . in view of this , guy worthey has generously provided us with a version of his program based on padova evolution tracks that allows stellar populations as young as 0.4 gyr . for the same stellar population , h@xmath3 indices based on padova evolution tracks are @xmath163 lower than the evolutionary tracks used at the website . the imf we use here has a salpeter slope and is terminated at mass limits @xmath164 and @xmath165 @xmath2 . for the population ages we consider ( table 3 ) , stars near the high mass limit do not contribute to the h@xmath3 index , they are only relevant to estimate the remnant mass . our remnant masses are based on evolution tracks currently available at the worthey website program . in table 4 we list the h@xmath3 index of the main galactic bulge , the disk component and the two populations combined all evaluated within @xmath160 . the h@xmath3 indices are slightly overestimated since we have only considered the stars within a physical radius of @xmath160 , not what would be observed in projection at this radius through the entire galaxy . on the other hand the indices are underestimated since we ( i ) have not considered the contribution from stars younger than 0.4 gyr and ( ii ) assume that there is no time delay between the appearance of cold gas in the disk and star formation . in any case , the total h@xmath3 indices listed in table 4 for galaxy plus disk in the e2l;1 and e2h;1 models nicely span the range of observed values given by de jong & davies , h@xmath166 . this greatly supports the hypothesis that the disk stars are simply a natural outcome of cooling flow evolution . there is also some observational evidence that the disks have higher h@xmath3 indices than the rest of the elliptical ( scorza & bender 1996 ) , but this needs further confirmation . our disks are expected to have power - law , not exponential space density and surface brightness profiles . the surface brightness distribution observed in stellar disks of very flattened disky ellipticals ( e4 - e6 ) deviates from exponential toward power law distributions ( scorza & bender 1995 ) . in disky ellipticals the angular momenta of bulge and disk are always parallel ( scorza & bender 1995 ; 1996 ) ; this is consistent with star formation in cooling flow disks and rules out a disk origin from random mergers . finally , we note that our disks are formed in ellipticals that are completely supported by rotation with fully isotropic stellar velocities ( @xmath167 . if some anisotropy were included ( @xmath168 ) , the resulting disks would be less extended with more steeply varying surface brightness in the resulting stars . hot interstellar gas in low - luminosity ellipticals is more vulnerable to outflow than that in brighter ellipticals because the specific gravitational binding energy decreases with galactic luminosity along the fundamental plane . low - luminosity ellipticals are also thought to be rotationally flattened which causes the x - ray luminosity of interstellar gas to decrease sharply with increasing ellipticity . in our family of galaxies all with luminosity @xmath169 @xmath170 we find that @xmath77 is also very sensitive to the assumed supernova rate . in ellipticals having a `` low '' current supernova rate , snu@xmath74 , @xmath171 varies along the sequence e0 ; e2 ; e6 in the ratios 1.0 ; 0.0091 ; 0.0027 . for a higher supernova rate , snu@xmath75 , the corresponding ratios are : 1.0 ; 0.025 ; 0.0033 . while the x - ray emission from hot interstellar gas in the e0 galaxies would be easily observed with axaf ( by subtracting the stellar component if that were possible ) , the prospects for observing the interstellar contribution to @xmath77 for more realistic rotating galaxies of this @xmath33 are less optimistic . furthermore , low luminosity e0 galaxies are rare or non existent ( tremblay & merritt 1996 ) . the influence of past and current supernova rates on @xmath77 and other important observational parameters is not entirely clear . the higher of the two current supernova rates we consider , snu@xmath75 , generates an iron abundance within the half - light radius that is 1.5 to 3 times the stellar iron abundance . of course any observation of the x - ray iron features in rotating , low - luminosity ellipticals is likely to be representative of stellar x - ray sources , not the gas . but in more massive ellipticals the iron abundance in the interstellar gas is often much _ less _ than that in the stars ( arimoto et al . 1997 ) . this could imply that the supernova iron and possibly also its energy is not distributed throughout the interstellar medium as we have assumed here ; indeed the supernova energy is concentrated in hot bubbles that may rise in the cooling flow atmosphere . because of this as yet unsolved problem , we can not be sure that we have treated the supernova energy correctly in our models , particularly for the high supernova rate solutions . if our treatment of supernova energy is correct , however , the disks of cool gas are severely truncated when snu@xmath172 due to an enhanced outflow of high angular momentum gas from the outer parts of the galaxy . the mass of the disk remaining at time @xmath45 gyr @xmath78 is reduced by @xmath173 as snu@xmath174 increases from 0.01 to 0.04 . more luminous ellipticals such as those we discussed in paper 1 require much higher snu@xmath174 to alter the cold disks . of particular interest is the possibility of observing disks of cold gas that are expected in rotating ellipticals either directly as optical line emission or indirectly in the form of a younger stellar population . the column of unionized disk gas in our disks would produce profound and observable absorption of soft x - rays , but this can be avoided since gravitational clumping is expected in this neutral or molecular gas . we have estimated the density of ionizing photons along the disk plane and computed the column density of warm hii disk gas . the presence of hii disk gas should be independent of the likelihood of star formation in the colder neutral gas . for our low value of snu@xmath174 , we find that large masses of hii gas can be rotationally supported beyond @xmath175 kpc amounting to about one tenth of all the gas ejected from galactic stars but the density of this gas is too low to contribute to the observed optical line flux . in previous estimates of the total hii mass in ellipticals a constant hii density is usually assumed ; these masses may seriously underestimate the true mass by several orders of magnitude . in the future we intend to investigate the survivability of gas in the outer disk if the galaxy is moving through an external ( inter - cluster ) medium . most of the optical line emission in the disk comes from the central parts of the disk which are less sensitive to rotation and the supernova rate . the total h@xmath3 luminosities expected from disk hii are similar to those observed in ellipticals , implying that a significant component of optical line emission from these galaxies may originate in cooling flow disks . if the disk hii emission is not masked by other sources of warm gas , the optical line - emitting gas should be ( i ) systematically flatter than the stellar image and ( ii ) rotating at the circular velocity which exceeds the local mean line of sight velocity of galactic stars . we have found considerable support for the notion that stellar disks in low luminosity ellipticals are a natural result of the evolution of the interstellar gas . the inevitability of cold disk formation in galactic cooling flows is consistent with the finding that all low luminosity ellipticals may contain stellar disks since many are hidden due to low inclinations ( rix & white 1990 ) . the stellar disk and bulge are observed to share the same sense of rotation ; this argues against random merging events and supports star formation in cold disks as we propose here . the total mass of gas that enters the cold disk in 1 - 15 gyrs , about 10 percent of the total stellar mass @xmath56 , is very similar to the typical stellar mass observed in elliptical disks . within the approximation of our galactic models ( e.g. king and isothermal density profiles ; constant satoh @xmath176 factor , etc . ) and that of current observations of stellar disks ( most of which have been studied only in very flat e5 or e6 ellipticals ) , we believe that our non - exponential disks are similar to those observed . only three of the nine stellar disks studied by scorza & bender ( 1995 ) could be modeled with exponential disks and exponential fits can always be made over a limited range in radius . finally , we have demonstrated that the range of observed h@xmath3 photometric indices observed in disky ellipticals is just spanned by our models if luminous stars form in the cold gaseous disks . we are implicitly assuming here that the type ia supernova rate may vary among ellipticals to generate a range of disk masses ; at present there is no compelling reason to believe otherwise . a similar variation of disk masses could be obtained by beginning our calculation shortly before or after 1 gyr . if our ideas about star formation in disky ellipticals are correct , these galaxies could become ideal laboratories for studying successive generations of star formation . in principle abundances in the parent population ( bulge ) stars , in the interstellar gas that they have expelled , and in subsequent stellar ( disk ) generations can all be directly observed . while luminous star formation in disks provides a satisfactory explanation for the ultimate fate of cooled gas in low luminosity ellipticals , it is at present unclear why more luminous , slowly rotating ellipticals do not also have visible stellar disks since cold disks are expected there too ( paper 1 ) . within the many uncertainties regarding the star formation process , the stellar properties of all elliptical disks can be understood if stars of lower mass are favored in high pressure environments . maximum interstellar pressures in the most massive ellipticals are about 1000 times that in the interstellar medium of our galaxy so even the most massive stars formed may not be optically luminous ; in low - luminosity ellipticals such as we consider here the pressures are lower so stars at the upper imf cutoff may be optically luminous but insufficiently massive ( @xmath177 @xmath2 ) to produce snii . alternatively , cold and dynamically fragile stellar disks may have formed in luminous ellipticals and subsequently been destroyed or altered perhaps in the same merging events responsible for the overall boxy shapes of these galaxies . finally , hot ambient ( cluster ) gas of low angular momentum may move into luminous ellipticals and greatly reduce the size of cooling flow disks . we hope to address these problems in the near future . we are greatly indebted to guy worthey for preparing for us a version of his dial - a - model program that can accommodate younger stars . the referee is thanked for helpful comments and corrections . our work on the evolution of hot gas in ellipticals is supported by grant nag 5 - 3060 of nasa to whom we are very grateful . in addition wgm is partially supported by a ucsc faculty research grant and fb is supported in part by grant asi-95-rs-152 from the agenzia spaziale italiana . .3 in + arimoto , n. , matsushita , k. , ishimaru , y. , ohashi , t. , & renzini , a. 1997 , ( preprint ) + binette , l. , magris , c. g. , stasinska , g. , & bruzual , a. g. 1994 , a&a , 292 , 13 + binney , j. , davies , r. l. , & illingworth , g.d . 1990 , apj , 361 , 78 + brighenti , f. & mathews , w. g. 1996 , apj , 470 , 747 ( paper 1 ) + brown , t. m. , ferguson , h. c. , & davidsen , a. f. 1994 , apj , 454 , l15 + buson , l. m. , sadler , e. m. , zeilinger , w. w. , bertin , g. bertola , f. , danziger , j. , dejonghe , h. , saglia , r. p. , & de zeeuw , p. t. 1993 , aasup , 280 , 409 + ciotti , l. , dercole , a. , pellegrini , s. , & renzini , a. , 1991 , apj , 376 , 380 + dercole , a. & ciotti , l. 1997 , private communication + davies , r. l. , efstathiou , g. , fall , s. m. , illingworth , g. , & schechter , p. l. 1983 , apj , 266 , 41 + dehnen , w. 1993 , mnras , 265 , 250 + de jong , r. s. & davies , r. l. 1996,mnras ( in print ) + dove , j. b. & shull , j. m. , 1994 , apj , 423 , 196 + fabbiano , g. , d .- w . kim , & trinchieri , g. 1992 , apjs , 80 , 531 + faber , s. m. , tremaine , s. , ajhar , e. a. , byun y - i . , bressler , a. , gebhardt , k. , grillmair , c. , kormendy , j. , lauer , t. r. , & richstone , d. 1997 , apj , ( in press ) + faber , s. m. , trager , s. c. , gonzalez , j. j. , & worthey , g. , 1995 , in _ stellar populations _ gilmore , g. , van der kruit , p. c. , proc . 164 , ( kluwer : dordreht ) , 249 + gonzalez , j. j. , 1993 , ph.d . thesis , univ . of california at santa cruz + hernquist , l. 1990 , apj , 356 , 359 + jaffe , w. 1983 , mnras , 202 , 995 + kim , d .- w . , fabbiano , g. , & trinchieri , g. 1992 , apj , 393 , 134 + kormendy , j. & bender , r. 1996 , apj , 464 , l119 + macchetto , f. , pastoriza , m. , caon , n. , sparks , w. b. , giavalisco , m. , bender , r. & capaccioli , m. 1996 , a&asup , 120,463 + magris , g. , & bruzual , g. 1993 , apj , 417 , 102 + mathews , w. g. 1989 , aj , 97 , 42 + mathews , w. g. , 1990 , apj , 354 , 468 + mathews , w. g. , 1997 , aj , 113 , 755 + nieto , j .- l . , bender , r. , arnaud , j. , & surma , p. 1991 , a&a , 244 , l25 + phillips , m. m. , jenkins , c. r. , dopita , m. a. , sadler , e. m. , & binette , l. 1986 , aj , 91 , 1062 + rix , h .- w . , & white , s. 1990 , apj , 362 , 52 + satoh , c. 1980 , publ . astr . soc . japan , 32 , 41 + scorza , c. & bender , r. 1995 , a&a , 293 , 20 + scorza , c. & bender , r. 1996 , in _ new light on galaxy formation _ , eds . r. bender & r. l. davies ( kluwer : dordrecht ) , 55 + turatto , m. , cappellaro , e. , & benetti , s. 1994,aj , 108 , 202 + tremblay , b. & merritt , d . 1996 , aj , 111 , 2243 + tremaine , s. , richstone , d. o. , byun , y .- i . , dressler , a , faber , s. m. , grillmair , c. , kormendy , j. , & lauer , t. r. 1994 , aj , 107 , 634 + trinchieri , g. , & di serego alighieri , sd . 1991 , aj , 101 , 1647 + tsai , j. c. & mathews , w. g. 1995 , apj , 448 , 84 + van der marel , r. p. 1991 , mnras , 253 , 710 + worthey g. , 1994 , apjs , 95 , 107 + worthey , g. , trager , s. c. , & faber , s. m. , 1995 , in _ fresh views of elliptical galaxies _ , eds . buzzoni , a. , renzini , a. , & serrano , a. , asp conf . series 86 , 203 + zeilinger , w. w. , pizzella , a. , amico , p. , bertin , g. , bertola , f. , buson , l. m. , danziger , i. j. , dejonghe , h. , sadler , e. m. , saglia , r. p. , & de zeeuw , p. t. 1996 aasup , 120 , 257 + @xmath127 & 63.65 pc@xmath32 & 1.725 kpc@xmath14 & 63.65 kpc@xmath178 & @xmath179 gm @xmath122@xmath56 & @xmath180 @xmath2@xmath181 & 31.623@xmath182 & @xmath183@xmath184 & 9 @xmath56@xmath33 & @xmath185@xmath186 & -19.89@xmath187 & -20.83@xmath188 & 5.71@xmath34 & 291 km s@xmath5 e0l;0 & 4 & 0.74 & 3.98 & 121.8 & ... & 10 & 0.59 & 6.41 & 34.6 & ... & 15 & 0.49 & 7.38 & 19.8 & ... & & & & & e0h;0 & 4 & 0.76 & 2.77 & 72.5 & ... & 10 & 0.49 & 4.17 & 14.1 & ... & 15 & 0.32 & 4.57 & 5.10 & ... & & & & & e2l;1 & 4 & 0.65 & 4.03 & 3.30 & @xmath1897.5 & 10 & 0.40 & 6.47 & 0.50 & @xmath18913.5 & 15 & 0.14 & 7.55 & 0.18 & @xmath18931.5 & & & & & e2h;1 & 4 & 0.74 & 2.78 & 7.48 & @xmath1890.75 & 10 & 0.44 & 4.14 & 2.37 & @xmath1890.75 & 15 & 0.29 & 4.40 & 0.13 & @xmath1890.65 & & & & & e6l;1 & 4 & 0.53 & 4.11 & 1.10 & @xmath18927 & 10 & 0.16 & 6.55 & 0.16 & @xmath18942.5 & 15 & 0.08 & 7.63 & 0.053 & @xmath18952 & & & & & e6h;1 & 4 & 0.67 & 2.72 & 2.34 & @xmath1892.3 & 10 & 0.37 & 4.13 & 0.20 & @xmath1894.5 & 15 & 0.08 & 4.40 & 0.017 & @xmath1893.8 + * figure 1 : * contours of the stellar temperature @xmath190 ( light lines ) and azimuthal stellar velocity @xmath191 ( heavy lines ) for the e2;1 galaxy . values of @xmath190 and @xmath191 for two contours are labeled ( units : k and k s@xmath5 ) ; the contours are equally spaced in @xmath192 and @xmath193 . * figure 2 : * variation of the circular velocity @xmath43 ( solid lines ) and stellar azimuthal velocity @xmath194 on the equatorial plane in the e2;1 ( bold lines ) and e6;1 ( light lines ) galaxies . all velocities in km s@xmath5 . * figure 3 : * properties of the e0;0 galaxy ( dashed lines ) and its interstellar gas after 15 gyr for two supernova rates : e0l;0 ( heavy solid lines ) and e0h;0 ( light solid lines ) . a ) variation of stellar density @xmath195 and gas density @xmath196 ( gm @xmath122 ) ; b ) variation of stellar temperature @xmath197 and gas temperature @xmath198 c ) distribution of 0.5 - 4.5 kev x - ray surface brightness @xmath199 ( solid lines ) and stellar surface brightness @xmath200 ( dashed line ) with projected radius ; the vertical normalization is arbitrary . * figure 4 : * plots of the surface density in the disk of cold gas @xmath201 ( gm @xmath133 ) as a function of radius @xmath202 at three times : @xmath203 gyr ( dot - dashed line ) , 10 gyr ( dashed line ) and 15 gyr ( light solid line ) for the a ) e2l;1 galaxy , b ) e2h;1 galaxy , and c ) e6l;1 galaxy . * figure 5 : * equally spaced contours of the logarithm of the ultimate disk radius in parsecs , @xmath204 , for gas ejected from stars throughout the e2l;1 galaxy . two contours are labeled with values of @xmath204 in pc . * figure 6 : * four panels describe the interstellar flow in galaxy e2l;1 at time @xmath45 gyr ; all contours are equally spaced . the physical limit of the galaxy is shown with a dashed line . _ upper left _ : contours of @xmath205 with two contours labeled . velocity arrow at upper left is 50 km s@xmath5 ; _ upper right _ : contours of @xmath205 in the central galaxy with two contours labeled . velocity arrow at upper left is 50 km s@xmath5 ; _ lower left _ : contours of @xmath206 with two labeled ; _ lower right _ : contours of azimuthal velocity @xmath207 with two contours labeled in km s@xmath5 . * figure 7 : * flow and thermal conditions in the e2h;1 galaxy at @xmath45 gyr ; all contours are equally spaced . the physical limit of the galaxy is shown with a dashed line . _ upper panel _ : contours of @xmath205 with two labeled . velocity arrow at upper left is 50 km s@xmath5 ; _ lower panel _ : contours of @xmath206 with two labeled . * figure 8 : * gas flow in the e6l;1 galaxy at @xmath45 gyr ; all contours are equally spaced . the physical limit of the galaxy is shown with a dashed line . two contours of @xmath205 are labeled . velocity arrow at upper left is 50 km s@xmath5 . * figure 9 : * x - ray surface brightness @xmath208 ( 0.5 - 4.5 kev ) distribution in three galaxies after @xmath92 gyr viewed perpendicular to the axis of rotation . two contours are labeled with values of @xmath209 in erg s@xmath5 @xmath133 ; all contours are equally spaced . _ top _ : @xmath208 for galaxy e2l;1 ; _ middle _ : @xmath208 for galaxy e2h;1 ; _ bottom _ : @xmath208 for galaxy e6l;1 . * figure 10 : * column densities along the disk plane of the e2l;1 galaxy at time @xmath45 gyr . _ dashed line : _ the total vertical column of cooled gas @xmath105 ; _ solid line : _ the maximum column @xmath104 that can be photoionized hii gas . the photoionized column at any radius is @xmath124 .

a pedagogical selection of topics in probability and statistics is presented . choice and emphasis are driven by the author s personal experience , predominantly in the context of physics analyses using experimental data from high - energy physics detectors .

1405.3402

these notes are based on a series of three lectures on probability and statistics , given at the aepshep 2012 physics school . while most of the attendance was composed of phd students , it also included master students and young post - docs ; in consequence , a variable level of familiarity with the topics discussed was implicit . for consistency , the scope of the lectures spanned from very general concepts up to more advanced , recent developments . the first lecture reviewed basic concepts in probability and statistics ; the second lecture focussed on maximum likelihood and multivariate techniques for statistical analysis of experimental data ; the third and last lecture covered topics on hypothesis - testing and interval estimation . whenever possible , the notation aligns with common usage in experimental high - energy physics ( hep ) , and the discussion is illustrated with examples related to recent physics results , mostly from the @xmath0-factories and the lhc experiments . mathematical probability is an abstract axiomatic concept , and probability theory is the conceptual framework to assess the knowledge of random processes . a detailed discussion of its development and formalism lies outside the scope of these notes . other than standard classic books , like @xcite , there are excellent references available , often written by ( high - energy ) physicists , and well - suited for the needs of physicists . a non - comprehensive list includes @xcite , and can guide the reader into more advanced topics . the sections on statistics and probability in the pdg @xcite are also a useful reference ; often also , the large experimental collaborations have ( internal ) forums and working groups , with many useful links and references . for a process to be called random , two main conditions are required : its outcome can not be predicted with complete certainty , and if the process is repeated under the very same conditions , the new resulting outcomes can be different each time . in the context of experimental particle physics , such an outcome could be `` a collision '' , or `` a decay '' . in practice , the sources of uncertainty leading to random processes can be * due to reducible measurement errors , i.e. practical limitations that can in principle be overcome by means of higher - performance instruments or improved control of experimental conditions ; * due to quasi - irreducible random measurement errors , i.e. thermal effects ; * fundamental , if the underlying physics is intrinsically uncertain , i.e. quantum mechanics is not a deterministic theory . obviously in particle physics , all three kinds of uncertainties are at play . a key feature of collider physics is that events resulting from particle collisions are independent of each other , and provide a quasi - perfect laboratory of quantum - mechanical probability processes . similarly , unstable particles produced in hep experiments obey quantum - mechanical decay probabilities . let @xmath1 be the total universe of possible outcomes of a random process , and let @xmath2 be elements ( or realizations ) of @xmath1 ; a set of such realizations is called a sample . a probability function @xmath3 is defined as a map onto the real numbers : @xmath4 \ , \nonumber \\ x & \rightarrow & { \cal p}(x ) \ . \end{aligned}\ ] ] this mapping must satisfy the following axioms : @xmath5 from which various useful properties can be easily derived , i.e. @xmath6 ( where @xmath7 is the complement of @xmath8 ) . two elements @xmath8 and @xmath9 are said to the independent ( that is , their realizations are not linked in any way ) if @xmath10 conditional probability @xmath11 is defined as the probability of @xmath8 , given @xmath9 . the simplest example of conditional probability is for independent outcomes : from the definition of independence in eq . ( [ eq : independence ] ) , it follows that if @xmath8 and @xmath9 are actually independent , the condition @xmath12 is satisfied . the general case is given by bayes theorem : in view of the relation @xmath13 , it follows that @xmath14 an useful corollary follows as consequence of bayes theorem : if @xmath1 can be divided into a number of disjoint subsets @xmath15 ( this division process is called a partition ) , then @xmath16 in the context of these lectures , the relevant scenario is when the outcome of a random process can be stated in numerical form ( i.e. it corresponds to a measurement ) : then to each element @xmath8 ( which for hep - oriented notation purposes , it is preferable to design as an event ) corresponds a variable @xmath17 ( that can be real or integer ) . for continuous @xmath17 , its probability density function ( pdf ) @xmath18 is defined as @xmath19 ) \ = \ p(x)dx \ , \end{aligned}\ ] ] where @xmath18 is positive - defined for all values of @xmath17 , and satisfies the normalization condition @xmath20 for a discrete @xmath21 , the above definition can be adapted in a straightworward way : @xmath22 finite probabilities are obtained by integration over a non - infinitesimal range . it is sometimes convenient to refer to the cumulative density function ( cdf ) : @xmath23 so that finite probabilities can be obtained by evaluating the cdf on the boundaries of the range of interest : @xmath24 other than the conditions of normalization eq . ( [ eq : normalization ] ) and positive - defined ( or more precisely , to have a compact support , which implies that the pdf must become vanishingly small outside of some finite boundary ) the pdfs can be arbitrary otherwise , and exhibit one or several local maxima or local minima . in contrast , the cdf is a monotonically increasing function of @xmath17 , as shown on figure [ fig : pdfcanvas ] , where a generic pdf and its corresponding cdf are represented . ; the pdf is assumed to have negligible values outside of the plotted range . right : the corresponding cumulative density function ( cdf ) , plotted in the same range.,title="fig:",width=283 ] ; the pdf is assumed to have negligible values outside of the plotted range . right : the corresponding cumulative density function ( cdf ) , plotted in the same range.,title="fig:",width=283 ] when more than on random number is produced as outcome in a same event , it is convenient to introduce a @xmath25-dimensional set of random elements @xmath26 , together with its corresponding set of random variables @xmath27 and their multidimensional pdf : @xmath28 lower - dimensional pdfs can be derived from eq . ( [ eq : multidimensionalpdf ] ) ; for instance , when one specific variable @xmath29 ( for fixed @xmath30 , with @xmath31 ) is of particlar relevance , its one - dimensional marginal probability density @xmath32 is extracted by integrating @xmath33 over the remaining @xmath34 dimensions ( excluding the @xmath30-th ) : @xmath35 without loss of generality , the discussion can be restricted to the two - dimensional case , with random elements @xmath8 and @xmath9 and random variables @xmath36 . the finite probability in a rectangular two - dimensional range is @xmath37 for a fixed value of @xmath9 , the conditional density function for @xmath8 is given by @xmath38 as already mentioned , the relation @xmath39 holds only if @xmath8 and @xmath9 are independent ; for instance , the two - dimensional density function in figure [ fig : pdf2dcanvas ] is an example of non - independent variables , for which @xmath40 . and @xmath41 : the pattern implies that in general , the probability densities are larger when @xmath17 and @xmath41 are both large or both small ( i.e. they are positively correlated ) , and thus the variables are not independent . the pdf is assumed to have negligible values outside of the plotted two - dimensional range.,width=359 ] the description of a random process via density functions is called a model . loosely speaking , a parametric model assumes that its pdfs can be completely described using a finite number of parameters . a straightforward implementation of a parametric pdf is when its parameters are analytical arguments of the density function ; the notation @xmath42 indicates the functional dependence of the pdf ( also called its shape ) in terms of variables @xmath43 and parameters @xmath44 . consider a random variable @xmath8 with pdf @xmath18 . for a generic function @xmath45 , its expectation value @xmath46 $ ] is the pdf - weighted average over the @xmath17 range : @xmath47 \ = \ \int dx p(x ) f(x ) \ .\end{aligned}\ ] ] being often used , some common expectation values have their own names . for one - dimensional pdfs , the mean value and variance are defined as @xmath48 \ = \ \int dx p(x ) x \ , \\ { \rm variance } & : \ \sigma^2 \ = & \ v[x ] \ = \ e[x^2 ] - \mu^2 \ = \ e[(x-\mu)^2 ] \ ; \end{aligned}\ ] ] for multidimensional pdfs , the covariance @xmath49 and the dimensionless linear correlation coefficient @xmath50 are defined as : @xmath51 -\mu_i\mu_j \ = \ e[(x_i-\mu_i)(x_j-\mu_j ) ] \ , \\ { \rm linear } \ { \rm correlation } & : \ \rho_{ij } \ = & \ \frac{c_{ij}}{\sigma_i\sigma_j } \ .\end{aligned}\ ] ] by construction , linear correlation coefficients have values in the @xmath52 interval . the sign of the @xmath53 coefficient indicates the dominant trend in the @xmath54 pattern : for positive correlation , the probability density is larger when @xmath21 and @xmath55 are both small or large , while a negative correlation indicates that large values of @xmath21 are preferred when small values of @xmath55 are realized ( and viceversa ) . when the random variables @xmath15 and @xmath56 are independent , that is @xmath57 , one has @xmath58 \ = \ \int\int dx_idx_jp(x_i , x_j ) x_ix_j \ = \ \mu_i\mu_j \ , \end{aligned}\ ] ] and thus @xmath59 : independent variables have a zero linear correlation coefficient . note that the converse needs not be true : non - linear correlation patterns among non - independent variables may `` conspire '' and yield null values of the linear correlation coefficient , for instance if negative and positive correlation patterns in different regions of the @xmath60 plane cancel out . figure [ fig : corrcanvas ] shows examples of two - dimensional samples , illustrating a few representative correlation patterns among their variables . ) ; no correlation , with a zero linear correlation coefficient ; a slightly milder , negative correlation ( with @xmath61 ) ; and a more complex correlation pattern , with variables very strongly correlated , but in such a way that the linear correlation coefficient is zero.,title="fig:",width=141 ] ) ; no correlation , with a zero linear correlation coefficient ; a slightly milder , negative correlation ( with @xmath61 ) ; and a more complex correlation pattern , with variables very strongly correlated , but in such a way that the linear correlation coefficient is zero.,title="fig:",width=141 ] ) ; no correlation , with a zero linear correlation coefficient ; a slightly milder , negative correlation ( with @xmath61 ) ; and a more complex correlation pattern , with variables very strongly correlated , but in such a way that the linear correlation coefficient is zero.,title="fig:",width=141 ] ) ; no correlation , with a zero linear correlation coefficient ; a slightly milder , negative correlation ( with @xmath61 ) ; and a more complex correlation pattern , with variables very strongly correlated , but in such a way that the linear correlation coefficient is zero.,title="fig:",width=141 ] in practice , the true probability density function may not be known , and the accessible information can only be extracted from a finite - size sample ( say consisting of @xmath62 events ) , which is assumed to have been originated from an unknown pdf . assuming that this underlying pdf is parametric , a procedure to estimate its functional dependence and the values of its parameters is called a characterisation of its shape . now , only a finite number of expectation values can be estimated from a finite - size sample . therefore , when choosing the set of parameters to be estimated , each should provide information as useful and complementary as possible ; such procedure , despite being intrinsically incomplete , can nevertheless prove quite powerful . the procedure of shape characterisation is first illustrated with the one - dimensional case of a single random variable @xmath17 . consider the empirical average @xmath63 ( also called sample mean ) @xmath64 as shown later in sec . [ sec : musigma ] , @xmath63 is a good estimator of the mean value @xmath65 of the underlying distribution @xmath18 . in the same spirit , the quadratic sample mean ( or root - mean - square rms ) , @xmath66 is a reasonable estimator of its variance @xmath67 ( an improved estimator of variance can be easily derived from the rms , as discussed later in sec . [ sec : musigma ] ) . intuitively speaking , these two estimators together provide complementary information on the `` location '' and `` spread '' of the region with highest event density in @xmath17 . the previous approach can be discussed in a more systematic way , by means of the characteristic function , which is a transformation from the @xmath17-dependence of the pdf @xmath18 onto a @xmath68-dependence of @xmath69 $ ] , defined as @xmath70 \ = \ e\left[e^{ik\frac{x-\mu}{\sigma}}\right ] \ = \ \sum_j \frac{(ik)^j}{j!}\mu_j \ .\end{aligned}\ ] ] the coefficients @xmath65 of the expansion are called reduced moments ; by construction , the first moments are @xmath71 and @xmath72 ; these values indicate that in terms of the rescaled variable @xmath73 , the pdf has been shifted to have zero mean and scaled to have unity variance . in principle , the larger the number of momenta @xmath74 that are estimated , the more detailed is the characterisation of the pdf shape . among higher - order moments , the third and fourth have specific names , and their values can be interpreted , in terms of shape , in a relatively straightforward manner . the third moment is called skewness : a symmetric distribution has zero skewness , and a negative ( positive ) value indicates a larger spread to the left ( right ) of its median . the fourth moment is called kurtosis ; it is a positive - defined quantity , that ( roughly speaking ) can be related to the `` peakedness '' of a distribution : a large value indicates a sharp peak and long - range tails ( such a distribution is sometimes called leptokurtic ) , while a smaller value reflects a broad central peak with short - range tails ( a so - called platykurtic distribution ) . figure [ fig : kurtosisetal ] shows a few distributions , chosen to illustrate the relation of momenta and shapes . the characterization of the shape of a pdf through a sequential estimation of shape parameters discussed in sec . [ sec : shape ] , aimed at a qualitative introduction to the concept of parameter estimation ( also called point estimation in the litterature ) . a more general approach is now discussed in this paragraph . consider a @xmath25-dimensional , @xmath68-parametric pdf , @xmath75 for which the values @xmath76 are to be estimated on a finite - sized sample , by means of a set of estimators denoted @xmath77 . the estimators themselves are random variables , with their own mean values and variances : their values differ when estimated on different samples . the estimators should satisfy two key properties : to be consistent and unbaised . consistency ensures that , in the infinite - sized sample limit , the estimator converges to the true parameter value ; absence of bias ensures that the expectation value of the estimator is , for all sample sizes , the true parameter value . a biased , but consistent estimator ( also called asymptotically unbiased ) is such that the bias decreases with increasing sample size . additional criteria can be used to characterize the quality and performance of estimators ; two often mentioned are * efficiency : an estimator with small variance is said to be more efficient than one with larger variance ; * robustness : this criterion characterizes the sensitivity of an estimator to uncertainties in the underlying pdf . for example , the mean value is robust against uncertainties on even - order moments , but is less robust with respect to changes on odd - order ones . note that these criteria may sometimes be mutually conflicting ; for practical reasons , it may be preferable to use an efficient , biased estimator to an unbiased , but poorly convergent one . the convergence and bias requirements can be suitably illustrated with two classical , useful examples , often encountered an many practical situations : the mean value and the variance . the empirical average @xmath63 is a convergent , unbiased estimation of the mean value @xmath65 of its underlying distribution : @xmath78 . this statement can easily be demostrated , by evaluating the expectation value and variance of @xmath63 : @xmath79 & = & \frac{1}{n } \sum_{i=1}^n e\left [ x \right ] \ = \ \mu \ , \label{eq : samplemean } \\ v\left [ \overline{x } \right ] & = & e\left [ ( \overline{x}-\mu)^2 \right ] \ = \ \frac{\sigma^2}{n } \ .\end{aligned}\ ] ] in contrast , the empirical rms of a sample is a biased , asymptotically unbiased estimator of the variance @xmath67 : this can be demonstrated by first rewriting its square ( also called sample variance ) in terms of the mean value : @xmath80 and so its expectation value is @xmath81 \ = \ \sigma^2 - v\left[\overline{x}\right ] \ = \ \frac{n-1}{n}\sigma^2 \ , \end{aligned}\ ] ] which , while properly converging to the true variance @xmath67 in the @xmath82 limit , systematically underestimates its value for a finite - sized sample . one can instead define a modified estimator @xmath83 that ensures , for finite - sized samples , an unbiased estimation of the variance . in summary , consistent and unbiased estimators for the mean value @xmath65 and variance @xmath67 of an unknown underlying pdf can be extracted from a finite - sized sample realized out of this pdf : @xmath84 as a side note , the @xmath85 and @xmath86 factors for @xmath87 in eq . ( [ eq : muhat ] ) and for @xmath88 in eq . ( [ eq : sigmahat ] ) can be intuitively understood as follows : while the empirical average be estimated even on the smallest sample consisting of a single event , at least two events are needed to estimate their empirical dispersion . these two classical examples discussed in [ sec : musigma ] refer to a single random variable . in presence of several random variables , expressed as a random @xmath25-dimensional vector @xmath89 , the discussion leads to the definition of the empirical covariance matrix , whose elements @xmath90 can be estimated on a sample of @xmath62 events as @xmath91 ( the @xmath92 indices run over random variables , @xmath93 ) . assuming the covariance is known ( i.e. by means of the estimator above , or from first principles ) , the variance of an arbitrary function of these random variables @xmath94 can be evaluated from a taylor - expansion around the mean values @xmath95 as @xmath96 which leads to @xmath97 \ \simeq \ f(\hat{\vec{\mu}})$ ] ; similarly , @xmath98 \ f^2(\hat{\vec{\mu } } ) + \sum_{a , b=1}^n\left.\frac{df}{dx_a}\frac{df}{dx_b}\right|_{\vec{x}=\hat{\vec{\mu } } } \hat{c}_{ab } \ , \end{aligned}\ ] ] and thus the variance of @xmath99 can be estimated as @xmath100 this expression in eq . ( [ eq : errorpropagation ] ) , called the error propagation formula , allows to estimate the variance of a generic function @xmath94 from the estimators of mean values and covariances . a few particular examples of error propagation deserve being mentioning explicitly : * if all random variables @xmath101 are uncorrelated , the covariance matrix is diagonal , @xmath102 and the covariance of @xmath94 reduces to @xmath103 * for the sum of two random variables @xmath104 , the variance is @xmath105 and the corresponding generalization to more than two variables is straightforward : @xmath106 in absence of correlations , one says that absolute errors are added in quadrature : hence the expression in eq . ( [ eq : sumquadrature ] ) is often written as @xmath107 . * for the product of two random variables @xmath108 , the variance is @xmath109 with its generalization to more than two variables : @xmath110 in absence of correlations , one says that relative errors are added in quadrature , and eq . ( [ eq : prodquadrature ] ) @xmath111 . * for a generic power law function , @xmath112 , if all variables are uncorrelated , the variance is @xmath113 in this section , a brief description of distributions , often encountered in practical applications , is presented . the rationale leading to this choice of pdfs is driven either by their specific mathematical properties , and/or in view of their common usage in the modellingi of important physical processes ; such features are correspondingly emphasized in the discussion . consider a scenario with two possible outcomes : `` success '' or `` failure '' , with a fixed probability @xmath114 of `` success '' being realized ( this is also called a bernouilli trial ) . if @xmath25 trials are performed , @xmath115 may actually result in `` success '' ; it is assumed that the sequence of trials is irrelevant , and only the number of `` success '' @xmath68 is considered of interest . the integer number @xmath68 follows the so - called binomial distribution @xmath116 : @xmath117 where @xmath68 is the random variable , while @xmath25 and @xmath114 are parameters . the mean value and variance are @xmath118 & = & \sum_{k=1}^n kp\left(k;n , p\right ) \ = \ np \ , \nonumber \\ v\left[k\right ] & = & np(1-p ) \ .\end{aligned}\ ] ] , @xmath119 and @xmath120 , respectively.,title="fig:",width=283 ] , @xmath119 and @xmath120 , respectively.,title="fig:",width=283 ] in the @xmath121 , @xmath122 limit ( with @xmath123 finite and non - zero ) for the binomial distribution , the random variable @xmath68 follows the poisson distribution @xmath124 , @xmath125 for which @xmath126 is the unique parameter . for poisson , the mean value and variance are the same : @xmath118 \ = \ v\left[k\right ] \ = \ \lambda \ .\end{aligned}\ ] ] the poisson distribution , sometimes called law of rare events ( in view of the @xmath122 limit ) , is a useful model for describing event - counting rates . examples of a binomial and a poisson distribution , for @xmath127 , @xmath119 and for @xmath120 , respectively , are shown on figure [ fig : binomialpoisson ] . the first two continuous random variables discussed in this paragraph are the uniform and the exponential distributions , illustrated in figure [ fig : uniformexponential ] . , @xmath128 ; for the exponential , two values @xmath129 ( solid line ) and @xmath130 ( dashed line ) are used.,title="fig:",width=283 ] , @xmath128 ; for the exponential , two values @xmath129 ( solid line ) and @xmath130 ( dashed line ) are used.,title="fig:",width=283 ] consider a continuous random variable @xmath17 , with a probability density @xmath131 that is non - zero only inside a finite interval @xmath132 $ ] : @xmath133 for this uniform distribution , the mean value and variance are @xmath134 & = & \frac{a+b}{2 } \ , \\ v\left[x\right ] & = & \frac{(b - a)^2}{12 } \ .\end{aligned}\ ] ] while not being the most efficient one , a straightforward monte - carlo generation approach would be based on a uniform distribution in the @xmath135 $ ] range , and use randomly generated values of @xmath17 in the @xmath136 $ ] range as implementation of an accept - reject algorithm with success probability @xmath114 . consider a continuous variable @xmath17 , with a probability density @xmath137 given by @xmath138 whose mean value and variance are @xmath134 & = & \xi \ , \\ v\left[x\right ] & = & \xi^2 \ .\end{aligned}\ ] ] a common application of this exponential distribution is the description of phenomena occuring independently at a constant rate , such as decay lengths and lifetimes . in view of the self - similar feature of the exponential function : @xmath139 the exponential distribution is sometimes said to be memoryless . and @xmath140.,width=359 ] now turn to the normal ( or gaussian ) distribution . consider a random variable @xmath17 , with probability density @xmath141 and with mean value and variance given by @xmath134 \ & = \ \mu \ , \\ v\left[x\right ] \ & = \ \sigma \ .\end{aligned}\ ] ] the pdf corresponding to the special @xmath142 , @xmath143 case is usually called `` reduced normal '' . on purpose , the same symbols @xmath65 and @xmath144 have been used both for the parameters of the gaussian pdf and for the mean value and variance . this is an important feature : the gaussian distribution is uniquely characterized by its first and second moments . for all gaussians , the @xmath145 $ ] covers @xmath146 of pdf integral , and is customary called a `` one - sigma interval '' ; similarly for the two - sigma interval and its @xmath147 . the dispersion of a peaked distribution is sometimes characterised in terms of its fwhm ( full width at half - maximum ) ; for gaussian distributions , this quantity is uniquely related to its variance , as @xmath148 ; figure [ fig : gaussian ] provides a graphical illustration of the gaussian pdf and its parameters . in terms of conceptual relevance and practical applications , the gaussian certainly outnumbers all other common distributions ; this feature is largely due to the central limit theorem , which asserts that gaussian distributions are the limit of processes arising from multiple random fluctuations . consider @xmath25 independent random variables @xmath149 , each with mean and variances @xmath150 and @xmath151 ; the variable @xmath152 , built as the sum of reduced variables @xmath153 can be shown to have a distribution that , in the large-@xmath25 limit , converges to a reduced normal distribution , as illustrated in figure [ fig : theoremcanvas ] for the sum of up to five uniform distributions . not surprisingly , any measurement subject to multiple sources of fluctuations is likely to follow a distribution that can be approximated with a gaussian distribution to a good approximation , regardless of the specific details of the processes at play . , with a normal pdf ( solid line ) of same mean and variance ; similarly , the plots on the top right , bottom left and bottom right plot the corresponding histograms for the sums @xmath154 , @xmath155 and @xmath156 of two , three and five reduced uniform variables @xmath157 , respectively . the sequence of variables follow distributions that quickly converge to a reduced gaussian.,width=548 ] the gaussian is also encountered as the limiting distribution for the binomial and and poisson ones , in the large @xmath25 and large @xmath126 limits , respectively : @xmath158 note that , when using a gaussian as approximation , an appropriate continuity correction needs to be taken into account : the range of the gaussian extends to negative values , while binomial and poisson are only defined in the positive range . now the @xmath159 ( or chi - squared ) distributions are considered . the following pdf @xmath160 with a single parameter @xmath25 , and where @xmath161 denotes the gamma function , has mean value and variance given by @xmath134 \ n \ , \nonumber \\ v\left[x\right ] \ & = \ 2n \ .\end{aligned}\ ] ] the shape of the @xmath159 distribution depends thus on @xmath25 , as shown in figure [ fig : chi2canvas ] where the @xmath159 pdf , for the first six integer values of the @xmath25 parameter , are shown . probability density functions , for integer numbers of degrees of freedom @xmath25 . the width of solid lines increases monotonically with @xmath25 in the @xmath162 range.,width=359 ] it can be shown that the @xmath159 distribution can be written as the sum of squares of @xmath25 normal - reduced variables @xmath21 , each with mean @xmath150 and variance @xmath151 : @xmath163 in view of this important feature of the @xmath159 distribution , the quantity @xmath25 is called `` number of degrees of freedom '' ; this name refers to the expected behaviour of a least - square fit , where @xmath164 data points are used to estimate @xmath165 parameters ; the corresponding number of degrees of freedom is @xmath166 . for a well - behaved fit , the @xmath159 value should follow a @xmath159 distribution . as discussed in [ sec : chi2test ] , the comparison of an observed @xmath159 value with its expectation , is an example of goodness - of - fit test . the survey closes with the discussion of two physics - motivated pdfs . the first is the breit - wigner function , which is defined as @xmath167 whose parameters are the most probable value @xmath168 ( which specifies the peak of the distribution ) , and the fwhm @xmath169 . the breit - wigner is also called lorentzian by physicists , and in matematics it is often referred to as the cauchy distribution . it has a peculiar feature , as a consequence of its long - range tails : the empirical average and empirical rms are ill - defined ( their variance increase with the size of the samples ) , and can not be used as estimators of the breit - wigner parameters . the truncated mean and interquartile range , which are obtained by removing events in the low and high ends of the sample , are safer estimators of the breit - wigner parameters . in the context of relativistic kinematics , the breit - wigner function provides a good description of a resonant process ( for example the invariant mass of decay products from a resonant intermediate state ) ; for a resonance , the parameters @xmath168 and @xmath169 are referred to as its mass and its natural width , respectively . finally , the voigtian function is the convolution of a breit - wigner with a gaussian , @xmath170 and is thus a three - parameter distribution : mass @xmath168 , natural width @xmath169 and resolution @xmath144 . while there is no straightforward analytical form for the voigtian , efficient numerical implementations are available , i.e. the tmath::voigt member function in the root @xcite data analysis framework , and the roovoigtian class in the roofit @xcite toolkit for data modeling . for values of @xmath169 and @xmath144 sufficiently similar , the fwhm of a voigtian can be approximated as a combination of direct sums and sums in quadrature of the @xmath169 and @xmath144 parameters . a simple , crude approximation yields : @xmath171 + \gamma/2 \ .\end{aligned}\ ] ] when the instrumental resolutions are sufficiently well described by a gaussian , a voigtian distribution is a good model for observed experimental distributions of a resonant process . figure [ fig : bwvoigt ] represents an analytical breit - wigner pdf ( evaluated using the @xmath172 boson mass and width as parameters ) , and a dimuon invariant mass spectrum around the @xmath172 boson mass peak , as measured by the atlas experiment using 8 tev data @xcite . the width of the observed peak is interpreted in terms of experimental resolution effects , as indicated by the good data / simulation agreement . of course , the atlas dimuon mass resolution is more complicated than a simple gaussian ( as hinted by the slightly asymmetric shape of the distribution ) , therefore a simple voigtian function would not have reached the level of accuracy provided by the complete simulation of the detector resolution . boson mass and width as parameters . right : the invariant dimuon mass spectrum around the @xmath172 boson mass peak ; the figure is from an atlas measurement using 8 tev data @xcite.,title="fig:",width=257 ] boson mass and width as parameters . right : the invariant dimuon mass spectrum around the @xmath172 boson mass peak ; the figure is from an atlas measurement using 8 tev data @xcite.,title="fig:",width=309 ] the discussion in section [ sec : parameterestimation ] was aimed at describing a few intuitive examples of parameter estimation and their properties . obviously , such case - by - case approaches are not general enough . the maximum likelihood estimation ( mle ) is an important general method for parameter estimation , and is based on properties following the maximum likelihood ( ml ) theorem . consider a sample made of @xmath62 independent outcomes of random variables @xmath173 , arising from a @xmath25-parametric pdf @xmath174 , and whose analytical dependence on variables and parameters is known , but for which the value(s ) of at least one of its parameters @xmath175 is unknown . with the mle , these values are extracted from an analytical expression , called the likelihood function , that has a functional dependence derived from the pdf , and is designed to maximise the probability of realizing the observed outcomes . the likelihood function @xmath176 can be written as @xmath177 this notation shows implictly the functional dependence of the likelihood on the parameters @xmath175 , and on the @xmath62 realizations @xmath173 of the sample . the ml theorem states that the @xmath178 values that maximize @xmath176 , @xmath179 are estimators of the unknown parameters @xmath175 , with variances @xmath180 that are extracted from the covariance of @xmath176 around its maximum . in a few cases , the mle can be solved analytically . a classical example is the estimation of the mean value and variance of an arbitrary sample , that can be analytically derived under the assumption that the underlying pdf is a gaussian . most often though , a numerical study of the likelihood around the @xmath175 parameter space is needed to localize the @xmath181 point that minimizes @xmath182 ( the `` negative log - likelihood '' , or nll ) ; this procedure is called a ml fit . formally speaking , several conditions are required for the ml theorem to hold . for instance , @xmath176 has to be at least twice derivable with respect to all its @xmath175 parameters ; constraints on ( asymptotical ) unbiased behaviour and efficiency must be satisfied ; and the shape of @xmath176 around its maximum must be normally distributed . this last condition is particularly relevant , as it ensures the accurate extraction of errors . when it holds , the likelihood is said ( in a slightly abusive manner ) to have a `` parabolic '' shape ( more in reference to @xmath183 than to the likelihood itself ) , and its expansion around @xmath181 can be written as @xmath184 in eq . ( [ eq : parabolic ] ) the covariance matrix @xmath185 has been introduced , and its elements are given by : @xmath186 \ .\end{aligned}\ ] ] when moving away from its maximum , the likelihood value decreases by amounts that depend on the covariance matrix elements : @xmath187 \ = \ \sum_{i , j}^ { } \left ( \theta_i - \hat{\theta}_i \right ) \mathbf{\sigma}_{ij}^{-1 } \left ( \theta_i - \hat{\theta}_j \right ) \ .\end{aligned}\ ] ] this result , together with the result on error propagation in eq . ( [ eq : errorpropagation ] ) , indicates that the covariance matrix defines contour maps around the ml point , corresponding to confidence intervals . in the case of a single - parameter likelihood @xmath188 , the interval contained in a @xmath189 contour around @xmath181 defines a @xmath190 confidence interval , corresponding to a @xmath191 range around the ml point ; in consequence the result of this mle can be quoted as @xmath192 . in a typical scenario , the outcome of a random process may arise from multiple sources . to be specific , consider that events in the data sample are a composition of two `` species '' , called generically `` signal '' and `` background '' ( the generalization to scenarios with more than two species is straighforward ) . each species is supposed to be realized from its own probability densities , yielding similar ( but not identical ) signatures in terms of the random variables in the data sample ; it is precisely these residual differences in pdf shapes that are used by the ml fit for a statistical separation of the sample into species . in the two - species example , the underlying total pdf is a combination of both signal and background pdfs , and the corresponding likelihood function is given by @xmath193 \ , \end{aligned}\ ] ] where @xmath194 and @xmath195 are the pdfs for the signal and background species , respectively , and the signal fraction @xmath196 is the parameter quantifying the signal purity in the sample : @xmath197 . note that , since both @xmath194 and @xmath195 satisfy the pdf normalization condition from eq . ( [ eq : normalization ] ) , the total pdf used in eq . ( [ eq : mlfit1 ] ) is also normalized . it is worth mentioning that some of the parameters @xmath175 can be common to both signal and background pdfs , and others may be specific to a single species . then , depending on the process and the study under consideration , the signal fraction can either have a known value , or belong to the set of unknown parameters @xmath175 to be estimated in a ml fit . in event - counting experiments , the actual number of observed events of a given species is a quantity of interest ; it is then convenient to treat the number of events as an additional parameter @xmath126 of the likelihood function . in the case of a single species , this amounts to `` extending '' the likelihod , @xmath198 where an additional multiplicative term , corresponding to the poisson distribution ( c.f . section [ sec : binomialpoisson ] ) , has been introduced . ( the @xmath199 term in the denominator can be safely dropped ; a global factor has no impact on the shape of the likelihood function nor on the ml fit results ) . it is straightforward to verify that the poisson likelihod in eq . ( [ eq : emlfit1 ] ) is maximal when @xmath200 , as intended ; now , if some of the pdfs also depend on @xmath126 , the value of @xmath201 that maximises @xmath176 may differ from @xmath62 . the generalization to more than one species is straightforward as well ; for each species , a multiplicative poisson term is included in the extended likelihood , and the pdfs of each species are weighted by their corresponding relative event fractions ; in presence of two species , the extended version of the likelihood in eq . [ eq : mlfit1 ] becomes : @xmath202 \ .\end{aligned}\ ] ] as discussed in section [ sec : contours ] , the condition in eq . ( [ eq : parabolic ] ) about the asymptotically normal distribution of the likelihood around its maximum is crucial to ensure a proper interpretation of @xmath203 contours in terms of confidence intervals . in this paragraph , two scenarios in which this condition can break down are discussed . plane , obtained from a global electroweak fit ( ew ) performed by the gfitter collaboration @xcite . center : the change in the ew likelihood as a function of @xmath204 , expressed in terms of @xmath205 . both plots illustrate the non - parabolic shape of the ew likelihood . , title="fig:",width=283 ] plane , obtained from a global electroweak fit ( ew ) performed by the gfitter collaboration @xcite . center : the change in the ew likelihood as a function of @xmath204 , expressed in terms of @xmath205 . both plots illustrate the non - parabolic shape of the ew likelihood . , title="fig:",width=283 ] ; the definition of cl is such that @xmath206 at the solutions of the ml fit . an eight - fold constraint on @xmath207 is extracted from measurements in @xmath208 decays by the experiment @xcite . the two pairs of solutions around @xmath209 are very close in values , and barely distinguishable in the figure . , width=359 ] a first scenario concerns a likelihood that is not totally symmetric around its maximum . such a feature may occur , when studying low - statistics data samples , in view of the binomial or poissonian behaviour of event - counting related quantities ( c.f . figure [ fig : binomialpoisson ] ) . but it can also occur on larger - sized data samples , indicating that the model has a limited sensitivity to the parameter @xmath175 being estimated , or as a consequence of strong non - linear relations in the likelihood function . as illustration , examples of two- and one - dimensional likelihood contours , with clear non - parabolic shapes , are shown in figure [ fig : gfitter ] . also , the likelihood function may possess one or more local maxima , or even completely degenerate maxima . there are various possible sources for such degeneracies . for example in models with various species , if some pdf shapes are not different enough to provide a robust statistical discrimination among their corresponding species . for example , if swapping the pdfs for a pair of species yields a sufficiently good ml fit , a local maximum may emerge ; on different sample realizations , the roles of local and global maxima may alternate among the correct and swapped combinations . the degeneracies could also arise as a reflexion of explicit , physical symmetries in the model : for example , time - dependent asymmetries in @xmath210 decays are sensitive to the ckm angle @xmath207 , but the physical observable is a function of @xmath211 , with an addtional phase @xmath212 at play ; in consequence , the model brings up to eight indistinguishible solutions for the ckm angle @xmath207 , as illustrated in figure [ fig : alphababar ] . in all such cases , the ( possibly disjoint ) @xmath189 interval(s ) around the @xmath181 central value(s ) can not be simply reported with a symmetric uncertainty @xmath180 only . in presence of a single , asymmetric solution , the measurement can be quoted with asymmetric errors , i.e. @xmath213 , or better yet , by providing the detailed shape of @xmath203 as a function of the estimated parameter @xmath175 . for multiple solutions , more information needs to be provided : for example , amplitude analyses ( `` dalitz - plot '' analyses ) produced by the b - factories and belle , often reported the complete covariance matrices around each local solution ( see e.g. @xcite as examples ) . consider a process with two possible outcomes : `` yes '' and `` no '' . the intuitive estimator of the efficiency @xmath214 is a simple ratio , expressed in terms of the number of outcomes @xmath215 and @xmath216 of each kind : @xmath217 for which the variance is given by @xmath218 \ = \ \frac{\hat{\varepsilon}(1-\hat{\varepsilon})}{n } \ , \end{aligned}\ ] ] where @xmath219 is the total number of outcomes realized . this estimator @xmath220 clearly breaks down for low @xmath25 , and in the very low or very high efficiency regimes . the mle technique offers a robust approach to estimate efficiencies : consider a pdf @xmath221 to model the sample , and include in it an additional discrete , bivariate random variable @xmath222 , so that the pdf becomes @xmath223 \ .\end{aligned}\ ] ] in this way , the efficiency is no longer a single number , but a function of @xmath17 ( plus some parameters @xmath175 that may be needed to characterize its shape ) . with this function , the efficiency can be extracted in different @xmath17 domains , as illustated in figure [ fig : efficiencycanvas ] , or can be used to produce a multidimensional efficiency map . . the sample contains two categories of events : `` accepted '' and `` rejected '' . the red dots indicate the bins of the histogram for those `` accepted '' only , and the black dots for the two categories cumulated . right : the efficiency as a parametric function of @xmath17 , with parameter values extracted from a ml fit to the data sample . figure extracted from the roofit user s guide @xcite.,width=434 ] as discussed in section [ sec : contours ] , in mle the covariance of @xmath176 is the estimator of statistical uncertainties . other potential sources of ( systematical ) uncertainties are usually at play , and need to be quantified . for this discussion , a slight change in the notation with respect to eq.([eq : likelihood ] ) is useful ; in this notation , the likelihood function is now written as @xmath224 where the parameters are explicitly partitioned into a set @xmath225 , called parameters of interest ( poi ) , that correspond to the actual quantities that are to be estimated ; and a set @xmath76 , called nuisance parameters ( np ) , that represent potential sources of systematic biases : if inaccurate or wrong values are assigned to some nps , the shapes of the pdfs can be distorted , the estimators of pois can become biased . the systematic uncertainties due to nps are usually classified in two categories . the first category refers to `` type - i '' errors , for which the sample ( or other control data samples ) can ( in principle ) provide information on the nps under consideration , and are ( in principle ) supposed to decrease with sample size . the second category refers to `` type - ii '' errors , which arise from incorrect assumptions in the model ( i.e. a choice of inadequate functional dependences in the pdfs ) , or uncontrolled features in data that can not be described by the model , like the presence of unaccounted species . clearly , for type - ii errors the task of assigning systematic uncertainties to them may not be well - defined , and may spoil the statistical interpretation of errors in termos of confidence levels . to deal with type - i nuisance parameters , a popular approach is to use the so - called profile - likelihood method . this approach consists of assigning a specific likelihood to the nuisance parameters , so that the original likelihood is modified to have two different components : @xmath226 then , for a fixed value of @xmath65 , the likelihood is maximized with respect to the nuisance @xmath175 ; the sequencial outcome of this procedure , called profile likelihood , is a function that depends only on @xmath65 : it is then said that the nuisance parameter has been profiled out of the likelihood . as an example , consider the measurement of the cross - section of a generic process , @xmath227 . if only a fraction of the processes is actually detected , the efficiency @xmath214 of reconstructing the final state is needed to convert the observed event rate @xmath228 into a measurement @xmath229 . this efficiency is clearly a nuisance : a wrong value of @xmath214 directly affects the value of @xmath229 , regardless of how accurately @xmath228 may have been measured . by estimating @xmath220 on a quality control sample ( for example , high - statistics simulation , or a high - purity control data sample ) , the impact of this nuisance can be attenuated . for example , an elegant analysis would produce a simultaneous fit to the data and control samples , so that the values and uncertainties of nps are estimated in the ml fit , and are correctly propagated to the values and variances of the pois . as another example , consider the search for a resonance ( `` a bump '' ) over a uniform background . if the signal fraction is very small , the width @xmath169 of the bump can not be directly estimated on data , and the value used in the signal pdf has to be inferred from external sources . this width is clearly a nuisance : using an overestimated value would translate into an underestimation of the signal - to - background ratio , and thus an increase in the variance of the signal pois , and possibly biases in their central values as well , i.e. the signal rate would tend to be overestimated . similar considerations can be applied in case of underestimation of the width . if an estimation @xmath230 of the width is available , this information can be implemented as in eq . ( [ eq : profilelikelihood ] ) , by using a gaussian pdf , with mean value @xmath231 and width @xmath231 , in the @xmath232 component of the likelihood . this term acts as a penalty in the ml fit , and thus constraints the impact of the nuisance @xmath169 on the pois . often , there are large regions in sample space where backgrounds are overwhelming , and/or signals are absent . by restricting the data sample to `` signal - enriched '' subsets of the complete space , the loss of information may be minimal , and other advantages may compensate the potential losses : in particular for multi - dimensional samples , it can be difficult to characterize the shapes in regions away from the core , where the event densities are low ; also reducing the sample size can relieve speed and memory consumption in numerical computations . the simplest method of sample reduction is by requiring a set of variables to be restricted into finite intervals . in practice , such `` cut - based '' selections appear at many levels in the definition of sample space : thresholds on online trigger decisions , filters at various levels of data acquisition , removal of data failing quality criteria ... but at more advanced stages of a data analysis , such `` accept - reject '' sharp selections may have to be replaced by more sophisticated procedures , generically called multivariate techniques . a multi - dimensional ml fit is an example of a multivariate technique . for a mle to be considered , a key requirement is to ensure a robust knowledge of all pdfs over the space of random variables . consider a set of @xmath25 random variables @xmath233 . if all variables are shown to be uncorrelated , their @xmath25-dimensional pdf is completely determined by the product of their @xmath25 one - dimensional pdfs ; now , if variables are correlated , but their correlation patterns are completely linear , one can instead use variables @xmath234 , linear combinations of @xmath173 obtained by diagonalizing the inverse covariance . for some non - linear correlation patterns , it may be possible to find analytical descriptions ; for instance , the ( mildly ) non - linear correlation pattern represented in figure [ fig : pdf2dcanvas ] , was produced with the roofit package , by applying the conditional option in rooprodpdf to build a product of pdfs . in practice , this elegant approach can not be easily extended to more than two dimensions , and is not guaranteed to reproduce complex , non - linear patterns . in such scenarios , the approach of dimensional reduction can potentially bring more effective results . a typical scenario for dimensional reduction is when several variables carry common information ( and thus exhibit strong correlations ) , together with some diluted ( but relevant ) pieces of independent information . an example is the characterization of showers in calorimeters ; for detectors with good transverse and/or longitudinal segmentation , the signals deposited in nearby calorimetric channels can be used to reconstruct details from the shower development ; for example , a function that combines informations from longitudinal and transverse shower shapes , can be used to discriminate among electromagnetic and hadronic showers . the simplest algorithm for dimensional reduction is the fisher discriminant : it is a linear function of variables , with coefficients adjusted to match an optimal criterion , called separation among two species , which is is the ratio of the variance between the species to the variance within the species , and can be expressed in a close , simple analytical form . in presence of more complex , non - linear correlation patterns , a large variety of techniques and tools are available . the tmva @xcite package is a popular implementation of dimensional - reduction algorithms ; other than linear and likelihood - based discriminants , it provides easy training and testing methods for artificial neural networks and ( boosted ) decision trees , which are among those most often encountered in hep analyses . as a general rule , a multivariate analyzer uses a collection of variables , realized on two different samples ( corresponding to `` signal '' and `` background '' species ) , to perform a method - dependent training , guided by some optimization criteria ; then performances of the trained analyzer are evaluated on independent realizations of the species ( this distinction between the training and testing stages is crucial to avoid `` over - training '' effects ) . figure [ fig : tmva ] shows a graphical representation of a figure - of - merit comparison of various analyzers implemented in tmva . sections [ sec : paramestimation ] and [ sec : mltheorem ] mainly discussed procedures for extracting numerical information from data samples ; that is , to perform measurements and report them in terms of values and uncertanties . the next step in data analysis is to extract qualitative information from data : this is the domain of statistical hypothesis testing . the analytical tool to assess the agreement of an hypothesis with observations is called a test statistic , ( or a statistic in short ) ; the outcome of a test is given in terms of a @xmath114-value , or probability of a `` worse '' agreement than the one actually observed . for a set of @xmath25 independent measurements @xmath21 , their deviation with respect to predicted values @xmath150 , expressed in units of their variances @xmath235 is called the @xmath159 test , and is defined as @xmath236 an ensemble of @xmath159 tests is a random variable that , as mentioned in section [ sec : examples ] , follows the @xmath237 distribution ( c.f . . [ eq : chi2 ] ) for @xmath25 degrees of freedom . its expectation value is @xmath25 and its variance @xmath238 ; therefore one does expect the @xmath159 value observed in a given test not to deviate much from the number of degrees of freedom , and so this value can be used to probe the agreement of prediction and observation . more precisely , one expects @xmath190 of tests to be contained within a @xmath239 interval , and the @xmath114-value , or probability of having a test with values larger than a given @xmath159 value is @xmath240 roughly speaking , one would tend to be suspicious of small observed @xmath114-values , as they may indicate a trouble , either with the prediction or the data quality . the interpretation of the observed @xmath114-value ( i.e. to decide whether it is too small or large enough ) is an important topic , and is discussed below in a more general approach . consider two mutually excluding hypotheses @xmath241 and @xmath242 , that may describe some data sample ; the hypothesis testing procedure states how robust is @xmath241 to describe the observed data , and how incompatible is @xmath242 with the observed data . the hypothesis @xmath241 being tested is called the `` null '' hypothesis , while @xmath242 is the `` alternative '' hypothesis , note that in the context of the search for a ( yet unknown ) signal , the null @xmath241 corresponds to a `` background - only '' scenario , and the alternative @xmath242 to `` signal - plus - background '' ; while in the context of excluding a ( supposedly inexistent ) signal , the roles of the null and alternative hypotheses are reversed : the null @xmath241 is `` signal - plus - background '' and the alternative @xmath242 is `` background - only '' . the sequence of a generic test , aiming at accepting ( or rejecting ) the null hypothesis @xmath241 by confronting it to a data sample , can be sketched as follows : * build a test statistic @xmath243 , that is , a function that reduces a data sample to a single numerical value ; * define a confidence interval @xmath244 $ ] ; * measure @xmath245 ; * if @xmath245 is contained in @xmath246 , declare the null hypothesis accepted ; otherwise , declare it rejected . to characterize the outcome of this sequence , two criteria are defined : a `` type - i error '' is incurred in , if @xmath241 is rejected despite being true ; while a `` type - ii error '' is incurred in , if @xmath241 is accepted despite being false ( and thus @xmath242 being true ) . the rates of type - i and type - ii errors are called @xmath207 and @xmath247 respectively , and are determined by integrating the @xmath241 and @xmath242 probability densities over the confidence interal @xmath246 , @xmath248 the rate @xmath207 is also called `` size of the test '' ( or size , in short ) , as fixing @xmath207 determines the size of the confidence interval @xmath246 . similarly , @xmath249 is also called `` power '' . together , size and power characterize the performance of a test statistic ; the neyman - pearson lemma states that the optimal statistic is the likelihood ratio @xmath250 , @xmath251 the significance of the test is given by the @xmath114-value , @xmath252 which is often quoted in terms of `` sigmas '' , @xmath253 so that for example a @xmath254 outcome can be reported as a `` two - sigma '' effect . alternatively , it is common practice to quote the complement of the @xmath114-value as a confidence level ( c.l . ) . the definition of a @xmath114-value as in eq . ( [ eq : pvalue ] ) ( or similarly in eq . ( [ eq : chi2test ] ) for the example for a @xmath159 test ) is clear and unambiguous . but interpretation of @xmath114-values is partly subjective : the convenience of a numerical `` threshold of tolerance '' may depend on the kind of hypothesis being tested , or on common practice . in hep usage , three different traditional benchmarks are conventionally employed : * in exclusion logic , a @xmath255 c.l . threshold on a signal - plus - background test to claim exclusion ; * in discovery logic , a three - sigma threshold ( @xmath256 ) on a background - only test to claim `` evidence '' ; * and a five - sigma threshold ( @xmath257 ) on the background - only test is required to reach the `` observation '' benchmark . in experimental hep , there is a tradition of reaching consensual agreement on the choices of test statistics . the goal is to ensure that , in the combination of results from different samples and instruments , the detector - related components ( specific to each experiment ) factor out from the physics - related observables ( which are supposed to be universal ) . for example in the context of searches for the standard model ( sm ) higgs boson , the four lep experiments agreed on analyzing their data using the following template for their likelihoods : @xmath258 where @xmath259 is the number of higgs decay channels studied , @xmath260 is the observed number of event candidates in channel @xmath261 , @xmath262 and @xmath263 ( @xmath264 and @xmath265 ) are the pdf and event yield for the signal ( background ) species in that channel . also , the test statistic @xmath126 , derived from a likelihood ratio , is @xmath266 so that roughly speaking , positive values of @xmath126 favour a `` background - like '' scenario , and negative ones are more in tune with a `` signal - plus - background '' scenario ; values close to zero indicate poor sensitivity to distinguish among the two scenarios . the values use to test these two hypotheses are : * under the background - only hypothesis , @xmath267 is the probability of having a @xmath268 value larger than the observed one ; * under the signal+plus+background hypothesis , @xmath269 is the probability of having a @xmath268 value larger than the observed one . figure [ fig : lep1d ] shows , for three different higgs mass hypotheses , the @xmath268 values from the combination of the four lep experiments in their searches for the sm higgs boson , overlaid with the expected @xmath269 and @xmath270 distributions . figure [ fig : lep2d ] shows the evolution of @xmath268 values as a function of the hypothetized higgs boson mass ; ( as stated in the captions , the color conventions in the one- and two - dimensional plots are different ) note that for masses below @xmath271 gev , a positive value of the @xmath268 test statistic would have provided evidence for a signal ; and sensitivity is quickly lost above that mass . gev hypotheses for the higgs mass are used . the dashed blue and dashed red lines correspond to the pdfs for the background - only and signal - plus - background hypotheses , respectively . the observed values of the test statistic @xmath268 are marked by black vertical lines . the yellow areas indicate the @xmath270 values for the background - only hypothesis , and the green areas the @xmath269 value for signal - plus - background . the three figures are taken from @xcite . , title="fig:",width=185 ] gev hypotheses for the higgs mass are used . the dashed blue and dashed red lines correspond to the pdfs for the background - only and signal - plus - background hypotheses , respectively . the observed values of the test statistic @xmath268 are marked by black vertical lines . the yellow areas indicate the @xmath270 values for the background - only hypothesis , and the green areas the @xmath269 value for signal - plus - background . the three figures are taken from @xcite . , title="fig:",width=185 ] gev hypotheses for the higgs mass are used . the dashed blue and dashed red lines correspond to the pdfs for the background - only and signal - plus - background hypotheses , respectively . the observed values of the test statistic @xmath268 are marked by black vertical lines . the yellow areas indicate the @xmath270 values for the background - only hypothesis , and the green areas the @xmath269 value for signal - plus - background . the three figures are taken from @xcite . , title="fig:",width=185 ] as a function of the higgs boson mass @xmath272 , obtained by combining the data of the four lep experiments . the dashed line is the mean value of the background - only distribution at each mass @xmath272 , and the green and yellow areas represent @xmath190 and @xmath255 contours for the background - only distribution at each mass @xmath272 . the black line follows the @xmath268 value observed in data as a function of @xmath272 . figure taken from @xcite . , width=359 ] the choice of @xmath269 to test the signal - plus - background hypothesis , while suitably defined as a @xmath114-value . may drag some subjective concern : in case of a fortuitous simultaneous downward fluctuation in both signal and background , the standard @xmath255 benchmark may lead to an exclusion of the signal , even if the sensitivity is poor . a modification of the exclusion benchmark called @xmath273 , has been introduced in this spirit@xcite , and is defined as @xmath274 this test , while not corresponding to a @xmath114-value ( a ratio of probabilities is not a probability ) , has the desired property of protecting against downwards fluctuations of the background , and is commonly used in exclusion results , including the searches for the sm higgs boson from the tevatron and lhc experiments . following the recommendations from the lhc higgs combination group @xcite , the atlas and cms experiments have agreed on using a common test statistic , called profiled likelihood ratio and defined as @xmath275 where the pio @xmath276 is the `` signal strength modifier '' , or higgs signal rate expressed in units of the sm predicted rate , @xmath277 are the fitted values of the nps at fixed values of the signal strength , and @xmath87 and @xmath181 are the fitted values when both @xmath65 and nps are all free to vary in the ml fit . the lower constraint on @xmath278 ensures that the signal rate is positive , and the upper constraint imposes that an upward fluctuation would not disfavor the sm signal hypothesis . for an observed statistic value @xmath279 , the @xmath114-values for testing the signal - plus - background and background - only hypotheses , @xmath280 and @xmath281 , are @xmath282 and the results on searches are reported both in terms of the exclusion significance using the @xmath273 observed and expected values , and the observation significance expressed in terms of the `` local '' @xmath281 expected and observed values . to conclude the discussion on hypothesis testing , there is certainly no better illustration than figure [ fig : fourthjuly ] , taken from the results announced by the atlas and cms experiments on july 4th , 2012 : in the context of the search for the sm higgs boson , both collaborations established the observation of a new particle with a mass around 125 gev . these lectures aimed at providing a pedagogical overview of probability and statistics . the choice of topics was driven by practical considerations , based on the tools and concepts actually encountered in experimental high - energy physics . a bias in the choices , induced by the author s perspective and personal experience is not to be excluded . the discussion was completed with examples from recent results , mostly ( although not exclusively ) stemming from the @xmath0-factories and the lhc experiments . i would like to express my gratitude to all who helped make the aepshep 2012 school a success : the organizers , the students , my fellow lecturers and discussion leaders . the highly stimulating atmosphere brought fruitful interactions during lectures , discussion sessions and conviviality activities . a. stuart , j.k . ord , and s. arnold , _ kendall s advanced theory of statistics _ 2a : _ classical inference and the linear model _ 6th ed . , oxford univ . press ( 1999 ) , and earlier editions by kendall and stuart .

neutron stars that are cold enough should have two or more superfluids / supercondutors in their inner crusts and cores . the implication of superfluidity / superconductivity for equilibrium and dynamical neutron star states is that each individual particle species that forms a condensate must have its own , independent number density current and equation of motion that determines that current . an important consequence of the quasiparticle nature of each condensate is the so - called entrainment effect , i.e. the momentum of a condensate is a linear combination of its own current and those of the other condensates . we present here the first fully relativistic modelling of slowly rotating superfluid neutron stars with entrainment that is accurate to the second - order in the rotation rates . the stars consist of superfluid neutrons , superconducting protons , and a highly degenerate , relativistic gas of electrons . we use a relativistic - mean field model for the equation of state of the matter and the entrainment . we determine the effect of a relative rotation between the neutrons and protons on a star s total mass , shape , and kepler , mass - shedding limit . 00a_0 ^ 0 00b_0 ^ 0 00c_0 ^ 0 00d_0 ^ 0 0.m_*|_o

gr-qc0402015

there is now an extended body of evidence in support of superfluidity in dense , nucleonic matter , such as that which is believed to exist in neutron stars . using conservative estimates one can argue for a fermi temperature on the order of @xmath2 for neutrons in media with supranuclear densities , and that the transition temperature to a superfluid state is about @xmath3 . this is a signficiant fact since it is generally accepted that nascent neutron stars formed from supernovae should cool fairly quickly , and consequently their internal temperatures should pass quickly through the transition value . observational support for such transitions is supplied by the well established glitch phenomenon in pulsars @xcite . these are rapid decreases in the rotational periods followed by a slow recovery @xcite , much too slow to be explained by ordinary fluid viscosity @xcite . the best description is based on superfluid quantized vortices , and how they pin , unpin and then repin as the pulsar s rotation rate evolves @xcite . we present here a fully relativistic formalism to model the rotational properties of superfluid neutron stars within a slow rotation approximation , which is an extension to second order in the rotation rates the previous work of comer and joynt @xcite . the bulk of the studies of neutron star superfluidity have been in the newtonian regime , where the intent is not to be quantitatively descriptive but rather to gain qualitative understandings . however , it is becoming increasingly apparent that general relativity is required to obtain even qualitative understandings . and even if the qualitative does not vary between the two regimes , the number of examples is growing where general relativity can yield factors of two difference from newtonian calculations ( as opposed to `` merely '' @xmath4 to @xmath5 corrections ) . for example , recent modelling of supernovae has revealed that when general relativity is included the range of model parameters that exhibit multiple bounces is significantly smaller than the range found for the newtonian case @xcite . the use of general relativistic hydrodynamics has also led to predictions of the shock radius ( during the shock reheating phase ) being reduced by a factor of two and a corresponding increase by a factor of two for the inflow speed of the material behind the shock @xcite . certainly the need to use general relativity must only be enhanced as supernova remnants become more compact . weber @xcite provides an excellent overview of the many suggestions for the matter content of these remnants , the neutron stars . they range from the traditional neutron / proton / electron models to more exotic configurations with kaon or muon condensates in the core , or hyperons , or even strange stars with absolutely stable u , d , s quark matter . in our study we will be somewhat conservative by considering typical traditional neutron star models , composed of superfluid neutrons , superconducting protons ( with proton fractions on the order of @xmath6 ) , and a highly degenerate relativistic gas of electrons . each species extends from the center of the star all the way to the surface , although in principle one could consider a more realistic scenario wherein the protons and electrons extend out further than the neutrons @xcite and in that way mimic features of a crust . this same technique could be applied in neutron star cores , allowing for the possibility of alternating regions of ordinary fluid and superfluid . the local thermodynamic equilibrium of the matter will be modelled using a relativistic @xmath0 - @xmath1 mean field approach of the type that is attributed to walecka @xcite and discussed in detail for neutron stars by glendenning @xcite . we consider a relativistic approach to be important on two different levels . on the macroscopic level there is the need for general relativity that was discussed above . but on a microscopic level , recall that any fluid approximation for matter has built into it the notion of local , fluid elements . they are small enough that they can be considered to be points with respect to the rest of the star , and yet large enough to contain , say , an avogadro s number of particles . at the densities expected for neutron stars , the local , fermi levels for the nucleons can become high enough that the effective velocities of the nucleons with respect to their fluid elements are relativistic . as for the fluid elements themselves , they can also , in principle , approach speeds near that of light , although in practice ( e.g. for quasinormal mode @xcite or slow rotation calculations @xcite ) they will typically have speeds a few percent of that of light . because of the superfluidity of the neutrons , and superconductivity of the protons , the fluid formalism to be used differs fundamentally from the standard perfect fluid approach in that the neutrons can flow independently of the protons and electrons . there are thus two fluid degrees of freedom in the system , requiring two sets of fluid elements , one set for the neutrons and another for the charged constituents since the electromagnetic interaction very effectively `` ties '' the electrons to the protons . the matter description , therefore , must take into account two fermi levels for the nucleons , and a displacement in momentum space between their respective fermi spheres that will result when one fluid flows with respect to the other . the specification of a local thermodynamic equilibrium for the two fluids requires that the local neutron and `` proton '' ( i.e. a conglomeration of the protons and electrons ) number densities be known as well as the local , relative velocity of the proton fluid elements , say , with respect to those of the neutrons . when the fermi spheres for the nucleons are displaced with respect to each other there results an important effect for neutron star dynamics known as entrainment @xcite . sauls @xcite describes the entrainment effect using quasiparticle language , i.e. just because the neutrons are superfluid and protons superconducting does not mean they no longer feel the strong force . on the contrary , an individual neutron should be understood as being surrounded by a polarization cloud of other neutrons and protons . when this neutron moves , it will be accompanied by this cloud of nucleons . the net effect at the level of the fluid elements is that the momentum of a neutron fluid element is a linear combination of the neutron _ and _ proton number density currents . parameters that are important for entrainment in neutron stars have been calculated in the newtonian regime using a fermi - liquid approach @xcite and in the relativistic regime via a @xmath0 - @xmath1 mean field model @xcite . comer and joynt @xcite find that one of the key parameters that has been much used in superfluid neutron star modelling extends over a much larger range of values than what the newtonian analysis of borumand et al would imply @xcite . comer and joynt also used their formalism to obtain first order rotational corrections to a superfluid neutron star s equilibrium state , which for general relativity means determining the frame - dragging , and the angular momentum . we will extend those calculations to the second order in the rotation rates , and thereby determine rotational corrections to the metric , distribution of particles , the total mass , shape , and the kepler , mass - shedding limit . we construct sequences of equilibrium configurations where members of the sequence have the same relative rotation between the neutrons and protons but are distinguished by their central neutron number densities . we also construct sequences where the central neutron number density is fixed and the relative rotation is allowed to vary ( cf . @xcite for discussions on when to expect a relative rotation ) . one other straightforward , but important , extension here of the work of comer and joynt is a proof that the matter coefficients obtained by them are sufficient for our second order calculations . in order to have a reasonably self - contained document , and to define all the variables , we review in sec . [ review ] the general relativistic superfluid formalism and its application to slowly rotating neutron stars . it is in this section that we prove the matter coefficients obtained by comer and joynt are all that is needed for the extension to second order . in sec . [ meanfield ] we discuss the highlights of the relativistic @xmath0 - @xmath1 model and its mean field limit . we also determine the model s slow rotation limit . in sec . [ results ] we join the slow rotation formalism with the mean field model and produce numerical solutions . after reviewing the main results , the final , concluding section discusses applications beyond those considered here and points out where the formalism should be improved . for convenience we have restated in the appendix results of comer and joynt for the various matter coefficients that are required input for the field equations . we use `` mtw '' @xcite conventions throughout and geometrical units . we will use the formalism developed by carter , langlois , and their various collaborators @xcite . the fundamental fluid variables consist of the conserved neutron @xmath7 and proton @xmath8 number density currents , from which are formed the three scalars @xmath9 , @xmath10 , and @xmath11 . given a master function @xmath12 ( i.e. the superfluid analog of the equation of state ) , then the stress - energy tensor is t^_= ^_+ ^_+ ^ _ , where = - n^_- p^_[press ] is the generalized pressure and _ = b n_+ a p _ , _ = a n_+ c p _ , are the chemical potential covectors which also function as the respective momenta for the fluid elements . the @xmath13 , @xmath14 , and @xmath15 coefficients are obtained from the master function via the partial derivatives = - , = - 2 , = - 2 . the fact that the neutron momentum @xmath16 , say , is not simply proportional to its number density current @xmath17 is a result of entrainment between the neutrons and protons , which we see vanishes if the @xmath13 coefficient is zero . finally the equations for the neutrons and protons consist of the two conservation equations _ ^ = 0 , _ ^ = 0 , and the two euler equations ^ _ [ _ ] = 0 , ^ _ [ _ ] = 0 , where the square braces mean antisymmetrization of the enclosed indices . comer @xcite and prix et al @xcite discuss in some detail why the assumption of separate conservation for the two fluids should be reasonable for slow rotation and quasinormal mode calculations . andersson and comer @xcite have adapted to the superfluid case the ordinary fluid slow rotation scheme originally developed by hartle @xcite . the configurations are assumed to be axisymmetric , asymptotically flat , and stationary , with the metric taking the form @xmath18 ^ 2\right){\rm d}t^2 + v { \rm d } r^2 - 2 { \rm sin}^2\theta k n^{\phi } { \rm d}t~ { \rm d } \phi + \cr & & \cr & & k \left({\rm d}\theta^2 + { \rm sin}^2\theta { \rm d } \phi^2 \right ) \ . \label{finmetric}\end{aligned}\ ] ] the neutrons and protons are assumed to be rigidly rotating about the symmetry axis , with rates @xmath19 and @xmath20 , respectively , and unit four - velocities written as u_^ = , u_^ = , where @xmath21 is the killing vector associated with the stationarity , and @xmath22 with the axisymmetry . the slow - rotation approximation assumes the rotation rates @xmath23 and @xmath24 are small in the sense that they should respect the inequalities ( cf . @xcite ) ^2 _ or ^2 _ or _ _ < < ( ) ^2 , where the speed of light @xmath25 and newton s constant @xmath26 have been restored , and @xmath27 and @xmath28 are the mass and radius , respectively , of the non - rotating configuration . because @xmath29 , the inequalities naturally imply @xmath30 and @xmath31 . the slow - rotation scheme has been shown , for instance in prix et al @xcite , to be a very good approximation for the fastest known pulsar , and starts to fail by about @xmath32 to @xmath4 for stars rotating at their kepler limit . in the same manner as hartle , the metric is expanded like @xmath33 where @xmath1 is understood to be of @xmath34 and @xmath35 , @xmath36 , and @xmath37 of @xmath38 . for later convenience we will also introduce _ = - _ _ = - _ . the expansion for the neutron and proton number densities @xmath39 and @xmath40 , respectively , are written as = _ o(r ) ( 1 + ( r , ) ) , = _ o ( r ) ( 1 + ( r , ) ) , where the terms @xmath41 and @xmath42 are understood to be of @xmath38 and we have introduced the convention that terms with an `` @xmath43 '' subscript are either contributions from the non - rotating background or quantities that are evaluated on the non - rotating background ( e.g. @xmath44 etc . ) . one can show furthermore that the metric corrections @xmath35 , @xmath36 , and @xmath37 can be decomposed into `` @xmath45 '' and `` @xmath46 '' terms using legendre polynomials for the angular dependence , i.e. @xmath47 where @xmath48 . as well we can write for the matter corrections = _ 0(r ) + _ 2(r ) p_2 ( ) , = _ 0(r ) + _ 2(r ) p_2 ( ) . lastly , andersson and comer @xcite have introduced a coordinate transformation @xmath49 that maps constant energy surfaces ( i.e. the level surfaces of @xmath50 ) for the non - rotating background into the rotationally modified constant energy surfaces . the mapping @xmath51 is given as _ o _ o _ 0 + _ o _ o _ 0 + _ o _ o _ o ( _ - _ ) ^2 = ^_o _ 0 [ coortrans0 ] for @xmath45 and for @xmath46 _ o _ o _ 2 + _ o _ o _ 2 - _ o _ o _ o ( _ - _ ) ^2 = ^_o _ 2 [ coortrans2 ] where @xmath52 and @xmath53 . no different in number , but different in form from the ordinary fluid case considered by hartle , the einstein / superfluid field equations reduce to four sets : ( i ) the non - rotating background that determines @xmath54 , @xmath55 , @xmath56 , and @xmath57 , ( ii ) the linear order that calculates the `` frame - dragging '' @xmath1 , ( iii ) the @xmath45 second order equations that yield @xmath58 , @xmath59 , @xmath60 , @xmath61 , and @xmath62 , and ( iv ) the @xmath46 set at second order that determines @xmath63 , @xmath64 , @xmath65 , @xmath66 , @xmath67 , and @xmath68 . for convenience we list the equations here and recall where appropriate the free parameters of the system and how they are specified during a numerical integration . the background equations are @xmath70 where the @xmath71 , @xmath72 , and @xmath73 coefficients are obtained from the master function and will be discussed in much greater detail below ( cf . [ anexp ] , [ meanfield ] and the appendix ) . throughout a prime `` @xmath74 '' will denote differentiation with respect to @xmath75 . regularity at the origin implies that @xmath76 , @xmath77 , @xmath78 , @xmath79 and @xmath80 all vanish . the radius @xmath28 is obtained from the condition that the pressure vanish on the surface ( i.e. @xmath81 ) and the background mass is given by m = - 4 _ 0^r _ o(r ) r^2 d r . the parameters @xmath82 and @xmath83 for the matter on the background are not independent because chemical equilibrium is imposed between the neutrons , protons , and a highly degenerate gas of relativistic electrons ( cf . the appendix ) . for @xmath34 there is only the frame - dragging @xmath84 , which is determined from ( r^4 e^- ( + ) /2 _ ^)^ - 16 e^(- ) /2 ( _ o - _ o ) _ = 16 e^(- ) /2 _ o _ o ( _ - _ ) . [ frmdrg ] the key difference with the frame - dragging equation for an ordinary fluid is the `` source '' term on the right - hand - side . any solution must be such that it joins smoothly to the exterior vacuum solution . continuity implies that _ ( r ) = - _ + , where @xmath85 is the total angular momentum of the system . using also that the derivative at the surface is to be continuous implies _ ( r ) = - _ - ^_(r ) . [ con ] when the frame - dragging equation is solved numerically , one searches for a central value @xmath86 that allows eq . ( [ con ] ) to be satisfied . with a solution in hand , the neutron and proton angular momenta , @xmath87 and @xmath88 , respectively , can then be calculated using j _ = - _ 0^r dr r^4 e^(- ) /2 and j _ = - _ 0^r dr r^4 e^(- ) /2 . the total angular momentum @xmath85 likewise follows since @xmath89 . one should first note that at this order there will exist both a @xmath45 and @xmath46 set of equations . the @xmath38 , @xmath45 set consists of eq . ( [ coortrans0 ] ) as well as @xmath90 + \cr & & \cr & & 8 \pi \lambda^{\prime}_{\rm o } \xi_0 - \frac{2}{r^2 } \left ( \frac{r}{e^{\lambda } } v_0\right)^{\prime } + \frac{r^2}{6 e^{\nu + \lambda } } \left(\tilde{l}_{\n}^{\prime}\right)^2 \ , \\ \label{000 } & & \cr 0 & = & \frac{2}{r e^{\lambda } } h^{\prime}_0 - \frac{2}{r e^{\lambda } } \left(\nu^{\prime } + \frac{1}{r}\right ) v_0 + \frac{r^2}{6 e^{\nu + \lambda } } \left(\tilde{l}_{\n}^{\prime}\right)^2 - 8 \pi \left[\mu_{\rm o } \n_{\rm o } \gamma_{\n } + \chi_{\rm o } \p_{\rm o } \gamma_{\p } - \left(\psi_{\rm o } - \lambda_{\rm o } \right ) h_0 + \right . \cr & & \cr & & \left.\frac{r^2}{3 e^{\nu } } \left(\mu_{\rm o } \n_{\rm o } \tilde{l}_{\n}^2 + \chi_{\rm o } \p_{\rm o } \tilde{l}_{\p}^2 \right ) - \frac{r^2}{3 e^{\nu } } \n_{\rm o } \p_{\rm o } \a_{\rm o } \left(\omega_{\n } - \omega_{\p}\right)^2\right ] \ . \label{110}\end{aligned}\ ] ] here @xmath91 and @xmath92 are integration constants , which are determined from data given at the center of the star : @xmath93 it is convenient to define the new variable @xmath94 . because we must have @xmath95 , then @xmath96 and @xmath97 are related via .(_o _ o _ 0)|_r = 0 + .(_o _ o _ 0)|_r = 0 = 0 . regularity at the center of the star also requires that @xmath98 , but @xmath99 and @xmath96 , say , are freely specified . as for the @xmath38 , @xmath46 equations , in addition to eq . ( [ coortrans2 ] ) , we have @xmath100 \ , \label{rem } \\ & & \cr 0 & = & \frac{1}{r } \left(v_2 + h_2\right ) - \left(k_2 + h_2 \right)^{\prime } - \frac{\nu^{\prime}}{2 } \left(h_2 - v_2 \right ) \ , \label{122 } \\ & & \cr 0 & = & \frac{2}{r e^{\lambda } } h^{\prime}_2 - \frac{6}{r^2 } h_2 - \frac{2}{r e^{\lambda } } \left(\nu^{\prime } + \frac{1}{r}\right ) v_2 + \frac{1}{e^{\lambda } } \left(\nu^{\prime } + \frac{2}{r } \right ) k^{\prime}_2 - \frac{4}{r^2 } k_2 - \frac{r^2}{6 e^{\nu + \lambda } } \left(\tilde{l}_{\n}^{\prime}\right)^2 + \cr & & \cr & & 8 \pi \left[\left(\psi_{\rm o } - \lambda_{\rm o}\right ) h_2 + \frac{r^2}{3 e^{\nu } } \left(\mu_{\rm o } \n_{\rm o } \tilde{l}_{\n}^2 + \chi_{\rm o } \p_{\rm o } \tilde{l}_{\p}^2 \right ) - \frac{r^2}{3 e^{\nu } } \n_{\rm o } \p_{\rm o } \a_{\rm o } \left(\omega_{\n } - \omega_{\p}\right)^2\right ] \ . \label{112}\end{aligned}\ ] ] in contrast to the @xmath45 case , the condition that @xmath101 at the center of the star leads to @xmath102 . also note that eq . ( [ rem ] ) can be used to remove @xmath67 in terms of the functions @xmath66 and @xmath103 . it is furthermore convenient to work with the new variable @xmath104 @xcite . regularity at the center of the star implies that @xmath105 and @xmath106 as @xmath107 , where the constants @xmath108 and @xmath109 are related by c_1 + 2 ( _ o(0 ) - _ o(0 ) ) c_2 = 0 . we can take advantage of an overall scale invariance of the field equations to remove the explicit dependence on both @xmath19 and @xmath20 by dividing through by @xmath20 , the net result being that only the ratio @xmath110 needs to be specified in advance . as for the other parameters , they are chosen so that the solutions obtained join smoothly to a vacuum exterior . for @xmath45 , after @xmath110 and @xmath96 are specified , this means a search over @xmath99 is performed until a smooth solution is achieved . for @xmath46 , the situation is slightly more involved , requiring that homogeneous and then particular solutions be obtained for the set @xmath111 ( see @xcite for more details ) . by searching over @xmath108 , say , and also trying different linear combinations of homogeneous and particular solutions eventually smoothness with the vacuum exterior can be achieved . having obtained a complete solution , the rotationally induced change to the mass can be determined using m = ( r - 2 m ) v_0(r ) + . also the rotationally induced changes to the neutron and proton particle numbers are obtained , respectively , from n_n = _ 0^r d r r^2 e^/2 _ o ( _ 0 + v_0 + _ 0 + ^2 _ ) [ baryn ] and n_p = _ 0^r d r r^2 e^/2 _ o ( _ 0 + v_0 + _ 0 + ^2 _ ) . [ baryp ] finally , the kepler frequency @xmath112 is calculated ( cf . @xcite and @xcite ) using the slow - rotation form of _ k = + where the metric variables are those of eq . ( [ finmetric ] ) and v = + is the orbital velocity according to a zero - angular momentum observer at the equator ( where all quantities are to be evaluated ) . in the slow rotation approximation , the kepler limit is the solution to _ k = - + \ { + ^2 - + } _ ^2 [ kep3 ] where we have made the scaling with @xmath113 explicit by introducing @xmath114 , etc . , and have also defined = ( 1 - ) + . in solving eq . [ kep3 ] we fix first the ratio @xmath110 . if @xmath115 , then we insert @xmath116 and solve for @xmath20 , otherwise we set @xmath117 and then again solve for @xmath20 . before proceeding to the next section that describes the mean field approach , one final observation needs to be made . we see in the equations above the appearance of not only the entrainment , via the @xmath13 coefficient , but also the relative velocity - dependent part of the entrainment , through the term @xmath118 . a priori one needs a formalism for the matter that will determine this term , if the effects of a relative rotation between the neutrons and protons are to be consistently incorporated . the mean field formalism presented in sec . [ meanfield ] can do just that . however , before embarking on a full - fledged calculation , it is very worthwhile to consider next the implications of the slow rotation expansion for the master function . the term @xmath118 would seem to imply , at least in principle , that we need to know the master function to @xmath119 . consider , then , regions of the matter ( i.e. fluid elements ) that are local enough that their spacetime can be taken to be that of minkowski , so that we can write out the @xmath120 term explicitly as x^2 = ( ) , where @xmath121 and @xmath122 are the neutron and proton , respectively , three - velocities in the local minkowski frame . we see from this that when the individual three - velocities @xmath123 satisfy @xmath124 , then @xmath125 to leading order in the ratios @xmath126 and @xmath127 . with this as motivation we write an analytic expansion for the master function of the form @xcite ( n^2,p^2,x^2 ) = _ i = 0^ _ i(n^2,p^2 ) ( x^2 - ) ^i . the @xmath13 , @xmath128 , etc . coefficients that appear in the field equations can now be written as @xmath129 - \left[i - 1\right ] \p \lambda_i\right ) \left(x^2 - \n \p \right)^{i - 2 } \ , \\ & & \cr \frac{\partial \a}{\partial \p } & = & - \frac{\partial \lambda_1 } { \partial \p } - \sum_{i = 2}^{\infty } i \left(\frac{\partial \lambda_i}{\partial \p } \left[x^2 - \n \p\right ] - \left[i - 1\right ] \n \lambda_i\right ) \left(x^2 - \n \p \right)^{i - 2 } \ , \\ \frac{\partial \a}{\partial x^2 } & = & - 2 \lambda_2 - \sum_{i = 3}^{\infty } i \left(i - 1\right ) \lambda_i \left(x^2 - \n \p\right)^{i - 2 } \ .\end{aligned}\ ] ] we conclude from this expansion that the master function coefficients @xmath130 , @xmath131 , and @xmath132 uniquely determine the background values for all the matter coefficients i.e . @xmath133 , @xmath134 , etc that enter the field equations . in particular , if we need to know @xmath132 then we need to know the master function to at least @xmath119 . however , the particular combinations that contain @xmath118 that enter the field equations are seen to reduce to @xmath135 that is , these combinations involve only the @xmath131 coefficient when evaluated on the background . this , in fact , makes perfect sense since the slow rotation approximation is to @xmath38 , which naturally corresponds to the term in the master function proportional to @xmath136 . the master function to be used has been obtained using a relativistic @xmath137 mean field model of the type described by glendenning @xcite . the lagrangian for this system is given by l = l_b + l _ + l_int , where @xmath138 here @xmath139 is the baryon mass , @xmath140 is an 8-component spinor with the proton components as the top 4 and the neutron components as the bottom 4 , the @xmath141 are the corresponding @xmath142 block diagonal dirac matrices , and @xmath143 . the coupled set of field equations obtained from this lagrangian are to be solved in each fluid element of the neutron star . the main approximations of the mean field approach are to assume that the nucleons can be represented as plane - wave states and that all gradients of the @xmath0 and @xmath144 fields can be ignored . the coupling constants @xmath145 and @xmath146 and field masses @xmath147 and @xmath148 are determined , for instance , from properties of nuclear matter at the nuclear saturation density . fortunately , in what follows , we only need to give the ratios @xmath149 and @xmath150 . of course , the main consideration is to produce a master function that incorporates the entrainment effect . consider again fluid elements somewhere in the neutron star . the fermionic nature of the nucleons means that they are to be placed into the various energy levels ( obtained from the mean field calculation ) until their respective ( local ) fermi spheres are filled . the fermi spheres are best visualized in momentum space , as in fig . [ cartoon ] . the entrainment is incorporated by displacing the center of the proton fermi sphere from that of the neutron fermi sphere , also illustrated in fig . [ cartoon ] . the neutron sphere is centered on the origin , and has a radius @xmath151 . displaced an amount @xmath152 from the origin is the center of the proton sphere , which has a radius @xmath153 . the fermi sphere radii and displacement @xmath154 are functions of the local neutron and proton number densities and the local , relative velocity of the protons , say , with respect to the neutrons . , whereas the proton sphere has radius @xmath153 and its center has been displaced an amount @xmath152 from the origin.,height=283 ] introducing the definitions _ g_^z , _ _ + k , then the master function takes the form @xcite = - ( k^3_+ k^3 _ ) ^2 + ^2_- ( m^2 - m_*^2 ) - 3 | ^x k_x , where @xmath155 ^ 2 + m_*^2 \right)^{3/2}\right ] + \int_{- k_\p}^{k_\p } d k_z \left [ \left(k^2_\p - 2 m^2 _ * - 2 \phi_\p^2 - 3 k^2_z - 4 \phi_\p k_z\right ) \right.\right.\cr & & \cr & & \left.\left.\left(k_\p^2 + \phi_\p^2 + m_*^2 + 2 \phi_\p k_z\right)^{1/2 } + 2 \left(\left[k_z + \phi_\p\right]^2 + m_*^2\right)^{3/2}\right]\right ) \ , \\ & & \cr m _ * & = & m - \frac{c^2_{\sigma}}{2 \pi^2 } m _ * \left ( \int_{- k_\n}^{k_\n } d k_z \left[k_\n^2 + \phi_\n^2 + m_*^2 + 2 \phi_\n k_z\right]^{1/2 } + \right . \cr & & \cr & & \left.\int_{- k_\p}^{k_\p } d k_z \left[k_\p^2 + \phi_\p^2 + m_*^2 + 2 \phi_\p k_z\right]^{1/2 } - \int_{- k_\n}^{k_\n } d k_z \left[\left(k_z + \phi_\n\right)^2 + m_*^2\right]^{1/2 } - \right . \cr & & \cr & & \left.\int_{- k_\p}^{k_\p } d k_z \left[\left(k_z + \phi_\p \right)^2 + m_*^2\right]^{1/2}\right ) \ , \label{mstar } \\ & & \cr \n^{0 } & = & \frac{1}{3 \pi^2 } k_{\n}^3 \ , \label{n0 } \\ & & \cr \p^{0 } & = & \frac{1}{3 \pi^2 } k_{\p}^3 \ , \label{p0 } \\ & & \cr \n^{z } & = & \frac{1}{2 \pi^2 } \int_{- k_{\n}}^{k_{\n } } d k_{z } \left(k_{z } + \phi_\n\right ) \left(\left[k_{\n}^2 + m_*^2 + \phi_\n^2 + 2 \phi_\n k_{z}\right]^{1/2 } - \right.\cr & & \cr & & \left.\left[\left(k_{z } + \phi_\n\right)^2 + m_*^2 \right]^{1/2}\right ) \ , \label{nz } \\ & & \cr \p^{z } & = & \frac{1}{2 \pi^2 } \int_{- k_{\p}}^{k_{\p } } d k_{z } \left(k_{z } + \phi_\p\right ) \left(\left[k_{\p}^2 + m_*^2 + \phi_\p^2 + 2 \phi_\p k_{z}\right]^{1/2 } - \right.\cr & & \cr & & \left.\left[\left(k_{z } + \phi_\p\right)^2 + m_*^2 \right]^{1/2}\right ) \ , \label{pz } \\ & & \cr \phi_\n & = & - c_{\omega}^2 \left(\n^z + \p^z\right ) \ . \label{phin}\end{aligned}\ ] ] as applied to this system , the slow - rotation approximation means that @xmath152 is small with respect to @xmath156 . in particular , we should keep terms up to and including @xmath157 . implementation of the mean - field approach is hampered by the fact that the most natural variables for it are the momenta @xmath154 , whereas the general relativistic superfluid formalism relies on @xmath158 . the two sets are related to each other via the kinematic relations ^2 = ( ^0)^2 - ( ^z)^2 , ^2 = ( ^0)^2 - ( ^z)^2 , x^2 = ^0 ^0 - ^z ^z , [ npx ] where @xmath159 , @xmath160 , @xmath161 , and @xmath162 are given above in eqs . ( [ n0 ] ) , ( [ p0 ] ) , ( [ nz ] ) , and ( [ pz ] ) . in principle one would specify values for the set @xmath158 , and then use eqs . ( [ n0 ] ) , ( [ p0 ] ) , ( [ nz ] ) , and ( [ pz ] ) in eq . ( [ npx ] ) to determine the set @xmath163 . one would then solve eq . ( [ mstar ] ) for the dirac effective mass @xmath164 . in practice we rewrite the field equations so as to solve directly for @xmath154 . earlier we discussed the slow - rotation approximation as it applies to the master function , and determined that we need to know the @xmath130 and @xmath131 coefficients . comer and joynt @xcite have shown that @xmath165 \sqrt{k_\p^2 + m^2_e } - \right.\cr & & \cr & & \left.m^4_e { \rm ln}\left[\frac{k_\p + \sqrt{k^2_\p + m^2_e}}{m_e}\right]\right ) \ , \\ & & \cr \lambda_1 & = & - c_{\omega}^2 - \frac{c^2_{\omega}}{5 \mu^2_{\rm o } } \left(2 k^2_\p \frac{\sqrt{k^2_\n + \ms0 ^ 2 } } { \sqrt{k^2_\p + \ms0 ^ 2 } } + \frac{c^2_{\omega}}{3 \pi^2 } \left[\frac{k^2_\n k^3_\p}{\sqrt{k^2_\n + \ms0 ^ 2 } } + \frac{k^2_\p k^3_\n}{\sqrt{k^2_\p + \ms0 ^ 2}}\right ] \right ) - \cr & & \cr & & \frac{3 \pi^2 k^2_{\p}}{5 \mu^2_{\rm o } k^3_\n } \frac{k^2_\n + \ms0 ^ 2}{\sqrt{k^2_\p + \ms0 ^ 2 } } \ , \end{aligned}\ ] ] where @xmath166 is the solution to the transcendental equation @xmath167 - \frac{1}{2 } \ms0 ^ 2 { \rm ln } \left[\frac{k_\p + \sqrt{k_\p^2 + \ms0 ^ 2}}{- k_\p + \sqrt{k_\p0 ^ 2 + \ms0 ^ 2 } } \right]\right ) \label{diracmass}\end{aligned}\ ] ] and we have added to @xmath130 the contributions due to the electrons ( @xmath168 ) . this is necessary to obtain a central proton fraction @xmath169 of about @xmath170 , which is considered typical for neutron stars . in order to solve for the background , we note that _ o = , _ o = . that is , we replace everywhere the background neutron and proton number densities with their respective fermi momenta . once the fermi momenta are specified , we can then determine the background value for the dirac effective mass , i.e. @xmath166 , from eq . ( [ diracmass ] ) . however , comer and joynt found that numerical solutions were easier to obtain by turning eq . ( [ diracmass ] ) into a differential equation using the identity 0^ = k^_+ k^ _ , [ msprime ] and where @xmath171 and @xmath172 are to be obtained simultaneously using eq . ( [ bckgrnd ] ) . the partial derivatives of the dirac effective mass can be found in the appendix , along with the other matter coefficients . in this section we present numerical solutions to the slow rotation equations . the two canonical background configurations ( see table [ table1 ] ) that will be used repeatedly are the same as those used by comer and joynt @xcite . they are such that the neutron and proton number densities vanish on the same surface . they are also such that chemical equilibrium has been imposed , which implies that the proton number density at the center of the star is no longer a free parameter . we do not apply chemical equilibrium for the rotating configurations since andersson and comer @xcite have shown it is not consistent with rigid rotation . the two sets of parameter values for the mean field model given in table [ table1 ] represent the two extremes discussed by glendenning @xcite . for all of the solutions we will set @xmath173 which implies also that @xmath174 . @xmath175 & @xmath176 @xmath177 & @xmath178 @xmath179 & @xmath180 @xmath181 & @xmath182 @xmath183 & @xmath184 @xmath185 & @xmath186 @xmath187 & @xmath188 @xmath173 & @xmath173 to begin we will give plots of the various field radial profiles for varying relative rotation rates of @xmath189 . these fall within the range of rates that were considered earlier by comer and joynt @xcite , who determined the frame - dragging and the angular momenta ( cf . their figs . 4 and 5 ) picking up at the second order , we present in fig . [ h0 ] a plot of @xmath190 vs @xmath75 for several relative rotation rates and for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . similarly in fig . [ m0 ] we show a plot of @xmath191 vs @xmath75 and in fig . [ ktilde ] we have @xmath192 vs @xmath75 . [ xi ] shows the radial profiles of @xmath193 and @xmath194 . figs . [ eta ] and [ phi ] show the radial profiles of @xmath195 and @xmath196 and @xmath197 and @xmath198 , respectively . ( we do not show plots of @xmath199 since it can be removed in terms of @xmath200 and @xmath201 . ) andersson and comer @xcite used a simplified equation of state ( i.e. a sum of polytropes ) and did not include entrainment and thus a direct comparison between our solutions and theirs would not be expected to yield the same details . that being said , there are some qualitative similarities . for instance , the signs of the @xmath45 fields versus the @xmath46 fields are in agreement , as are the number of zeroes that occur in the fields . the most pronounced qualitative differences are found in the functions @xmath197 and @xmath198 , especially near the surface of the star . this could perhaps be related to our use of an explicit term for the electrons in the equation of state . the 3 lower curves and @xmath200the 3 upper curves vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath202).,height=245 ] vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath203).,height=245 ] vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath203).,height=245 ] the 3 upper curves and @xmath194the 3 lower curves vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath202).,height=245 ] the 3 upper curves and @xmath204the 3 lower curves vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath202).,height=245 ] the 3 uppermost curves at @xmath205and @xmath206the 3 lowermost curves at @xmath205vs @xmath75 for models i ( left panel ) and ii ( right panel ) of table [ table1 ] . the relative rotation rates are varied ( @xmath202).,height=245 ] in fig . [ masses ] we have graphed the total mass for proton rotation rates that equal the fastest known pulsar ( i.e. @xmath207 ) versus the background central neutron number density . the left panel is for model i of table [ table1 ] and the right panel is for model ii . in both plots the relative rotation rate @xmath208 of the neutrons with respect to the protons is also varied . generally speaking , stars that are spun up without changing the total baryon number move upward and to the left in the figure . we find that we recover the so - called `` supramassive '' configurations first discussed by cook et al @xcite . these are stars whose central densities go beyond the maximum allowed for stable , non - rotating configurations and yet still remain on the stable ( upward ) branches of their particular mass vs central density curves . these configurations are stablized by their rotation . ( using @xmath209 ) . the relative rotation rates are also varied , i.e. @xmath210.,height=302 ] fig . [ elliptic ] gives plots of the ratio of the polar ( i.e. @xmath211 ) to equatorial ( i.e. @xmath212 ) radii , versus the relative rotation rate @xmath213 for the canonical backgrounds given in models i and ii of table [ table1 ] and assuming a proton rotation rate of the fastest known pulsar . up to and including @xmath38 we have = 1 + . we see a general overall quadratic behaviour in the relative rotation , which is expected for the slow rotation expansion . also expected is the limiting value of @xmath214 as @xmath215 . since the neutrons and protons have been constrained to vanish on the same surface , we do not find configurations like those of @xcite whereby the neutrons , say , can extend out further than the protons . as a function of the relative rotation rate @xmath110 for the canonical background configurations of models i and ii of table [ table1 ] and setting @xmath216.,height=302 ] fig . [ kepler ] is a graph of the kepler ( or mass - shedding ) limit @xmath112 versus the relative rotation rate for the canonical background configurations of table [ table1 ] . two competing factors are the amount of mass of the star that is in neutrons versus the relative rotation rate . for instance , for @xmath217 the kepler limit is seen to be essentially flat . this can be easily understood as a result of the fact that the neutrons represent nearly @xmath218 of the mass of the system , and so the kepler limit changes very little as the relative rotation is increased . such behavior has also been seen in the analytical solution of prix et al @xcite , and the numerical results of andersson and comer @xcite . the clear difference with the previous work is the general decrease in the kepler limit as the relative rotation is decreased below one . prix et al and andersson and comer found that the kepler limit increased monotonically as the relative rotation was decreased . they explained this as a result of the fact that most of the mass is in the neutrons , and as the relative rotation is decreased , the kepler limit is approaching the non - rotating value . as one can see in fig . [ kepler ] eventually a minimum is reached , and beyond that the kepler limit then starts to increase as the relative rotation is decreased . it just does not exhibit the monotonic increase of the earlier studies . the main differences with the two earlier studies are that here we use the mean field formalism for the equation of state the previous two studies used simplified polytropes and the entrainment is both relativistic and also has a non - trivial radial profile ( cf . figs . 2 and 3 of @xcite ) , whereas the relevant entrainment parameter is kept constant in prix et al and taken to be zero in andersson and comer . for the canonical background configurations of models i and ii of table [ table1].,height=302 ] the earlier work of comer and joynt @xcite laid the foundations for a fully relativistic approach to entrainment in superfluid neutron stars . they applied their formalism to first order in the rotational rates , and determined the impact of a relative rotation on the frame - dragging and the angular momenta of the two fluids . we have expanded on this work by extending the calculations to second order in the rotation rates , and have determined for the first time the maximum mass , shape , and kepler limit using a fully relativistic formulation for entrainment and allowing a relative rotation between the two fluids . an important extension of our work will be to include isotopic spin terms in the master function @xcite . these are important for incorporating symmetry energy effects which tend to force baryonic systems to have as many protons as neutrons . obviously nuclei tend to have equal numbers of neutrons and protons , and so this is why a symmetry energy term is an important addition to the equation of state . in terms of neutron stars , prix et al @xcite have shown clearly that symmetry energy impacts rotational equlibria ( e.g. ellipticity , the kepler limit , and moment of inertia ) as the relative rotation is varied . the next application we have in mind is to study quasinormal modes on slowly rotating backgrounds . andersson and comer @xcite ( see also @xcite ) have shown that a suitably advanced gravitational wave detector should be able to see gravitational waves emitted during vela or crab - like glitches . such observations could provide potentially unique information on the supranuclear equation of state and the parameters that govern entrainment . another important application will be to study further the recently discovered two - stream instability @xcite . it is the direct analog for superfluids of the instability of the same name known to exist in plasmas @xcite . finally , our results should also find some application in studies of vortex structure in neutron stars . link @xcite has suggested that free precession in neutron stars is not compatible with the protons behaving as a type ii superconductor . this should be explored in more detail by allowing for the effect of entrainment . i thank n. andersson for useful discussions and with providing aid with the code that served as the starting point for the research presented here . i also thank bob joynt for useful discussions and for pointing me in the direction of mean field theory for these entrainment calculations . the research presented here received support from nsf grant phys-0140138 . for convenience , we list here the various matter coefficients that are required as input in the field equations : @xmath219\right ) + \cr & & \cr & & \frac{3 \pi^2 k^2_{\p}}{5 \mu^2_{\rm o } k^3_\n } \frac{k^2_\n + \ms0 ^ 2}{\sqrt{k^2_\p + \ms0 ^ 2 } } \ , \\ & & \cr \left.\b\right|_{\rm o } & = & \frac{3 \pi^2 \mu_{\rm o}}{k^3_\n } - c_{\omega}^2 \frac{k^3_\p}{k^3_\n } - \frac{c^2_{\omega } k^3_\p } { 5 \mu^2_{\rm o } k^3_\n } \left(2 k^2_\p \frac{\sqrt{k^2_\n + \ms0 ^ 2}}{\sqrt{k^2_\p + \ms0 ^ 2 } } + \right.\cr & & \cr & & \left.\frac{c^2_{\omega}}{3 \pi^2 } \left[\frac{k^2_\n k^3_\p}{\sqrt{k^2_\n + \ms0 ^ 2 } } + \frac{k^2_\p k^3_\n}{\sqrt{k^2_\p + \ms0 ^ 2 } } \right]\right ) - \frac{3 \pi^2 k^5_{\p}}{5 \mu^2_{\rm o } k^6_\n } \frac{k^2_\n + \ms0 ^ 2}{\sqrt{k^2_\p + \ms0 ^ 2 } } \ , \\ & & \cr \left.\c\right|_{\rm o } & = & \frac{3 \pi^2 \chi_{\rm o}}{k^3_\p } - c_{\omega}^2 \frac{k^3_\n}{k^3_\p } - \frac{c^2_{\omega } k^3_\n } { 5 \mu^2_{\rm o } k^3_\p } \left(2 k^2_\p \frac{\sqrt{k^2_\n + \ms0 ^ 2}}{\sqrt{k^2_\p + \ms0 ^ 2 } } + \right.\cr & & \cr & & \left.\frac{c^2_{\omega}}{3 \pi^2 } \left[\frac{k^2_\n k^3_\p}{\sqrt{k^2_\n + \ms0 ^ 2 } } + \frac{k^2_\p k^3_\n}{\sqrt{k^2_\p + \ms0 ^ 2 } } \right]\right ) - \frac{3 \pi^2}{5 \mu^2_{\rm o } k_\p } \frac{k^2_\n + \ms0 ^ 2 } { \sqrt{k^2_\p + \ms0 ^ 2 } } + \cr & & \cr & & \frac{3 \pi^2}{k^3_\p } \sqrt{k^2_{\p } + m^2_e } \ , \\ & & \cr \left.\a00\right|_{\rm o } & = & - \frac{\pi^4}{k^2_\n k^2_\p } \frac{\partial^2 \lambda_0}{\partial k_\p \partial k_\n } = c_\omega^2 + \frac{\pi^2}{k^2_\p } \frac{\ms0 \frac{\partial \ms0 } { \partial k_\p}}{\sqrt{k^2_\n + \ms0 ^ 2}}\ , \\ & & \cr \left.\b00\right|_{\rm o } & = & \frac{\pi^4}{k^5_\n } \left(2 \frac{\partial \lambda_0}{\partial k_\n } - k_\n \frac{\partial^2 \lambda_0}{\partial k^2_\n}\right ) = c_\omega^2 + \frac{\pi^2 } { k^2_\n } \frac{k_\n + \ms0 \frac{\partial \ms0}{\partial k_\n } } { \sqrt{k^2_\n + \ms0 ^ 2 } } \ , \\ & & \cr \left.\c00\right|_{\rm o } & = & \frac{\pi^4}{k^5_\p } \left(2 \frac{\partial \lambda_0}{\partial k_\p } - k_\p \frac{\partial^2 \lambda_0 } { \partial k^2_\p}\right ) = c_\omega^2 + \frac{\pi^2}{k^2_\p } \frac{k_\p + \ms0 \frac{\partial \ms0}{\partial k_\p } } { \sqrt{k^2_\p + \ms0 ^ 2 } } + \frac{\pi^2}{k_\p } \frac{1 } { \sqrt{k^2_\p + m^2_e } } \ , \end{aligned}\ ] ] where @xmath220\right)^{- 1 } \ , \\ & & \cr \frac{\partial \ms0}{\partial k_\p } & = & - \frac{c_\sigma^2}{\pi^2 } \frac{\ms0 k^2_\p}{\sqrt{k^2_\p + \ms0 ^ 2 } } \cr & & \cr & & \left(\frac{3 m - 2 \ms0}{\ms0 } - \frac{c_\sigma^2}{\pi^2 } \left[\frac{k^3_\n}{\sqrt{k^2_\n + \ms0 ^ 2 } } + \frac{k^3_\p } { \sqrt{k^2_\p + \ms0 ^ 2}}\right]\right)^{- 1 } \ .\end{aligned}\ ] ] the two functions @xmath221 and @xmath222 are the background values for the two chemical potentials : _ o = ( k_^3 + k_^3 ) + , _ o = ( k_^3 + k_^3 ) + . chemical equilibrium for the background means @xmath223 must be imposed . given a value for @xmath224 then @xmath225 can be determined . using these then the initial value for the dirac effective mass can be obtained from eq . ( [ diracmass ] ) so that eq . ( [ msprime ] ) can be integrated . v. radhakrishnan and r.n . manchester , nature ( london ) * 244 * , 228 ( 1969 ) . lyne , in _ pulsars as physics laboratories _ , eds . r.d . blandford , a. hewish , a.g . lyne , and l. mestel , ( oxford university press inc . , new york , 1993 ) . p.e . reichley and g.s . downs , nature * 222 * , 229 ( 1969 ) . g. baym , c. pethick , d. pines , and m. ruderman , nature * 224 * , 872 ( 1969 ) . anderson and n. itoh , nature * 256 * , 25 ( 1975 ) . alpar , p.w . anderson , d. pines , and j. shaham , ap . j. * 276 * , 325 ( 1984 ) . alpar , p.w . anderson , d. pines , and j. shaham , ap . j. * 278 * , 791 ( 1984 ) . g.l . comer and r. joynt , phys . d * 68 * , 023002 ( 2003 ) . h. dimmelmeier , j.a . font , and e. mller , astron . astrophys . * 393 * , 523 ( 2002 ) . bruenn , k.r . de nisco , and a. mezzacappa , ap . j. * 560 * , 326 ( 2001 ) . f. weber , _ pulsars as astrophysical laboratories for nuclear and particle physics _ , ( institute of physics publishing , bristol and philadelphia , 1999 ) . r. prix , g.l . comer , and n. andersson , astron . astrophys . * 381 * , 178 ( 2002 ) . n. andersson , g.l . comer , and d. langlois , phys . d * 66 * , 104002 ( 2002 ) . r. prix , j. novak , g.l . comer,``stationary structure of relativistic superfluid neutron stars , '' in _ proceedings of the 26th spanish relativity meeting : gravitation and cosmology _ , eds . a. lobo , f. fayos , j. garriga , e. gaztaaga , and e. verdaguer ( university of barcelona , 2003 ) , pgs . . j. d. walecka , _ theoretical nuclear and subnuclear physics _ , ( oxford university press , new york , 1995 ) . glendenning , _ compact stars : nuclear physics , particle physics , and general relativity _ , ( springer - verlag , new york , 1997 ) . comer , d. langlois , and l.m . lin , phys . d * 60 * , 104025 ( 1999 ) . n. andersson and g. l. comer , class . and quant . grav . * 18 * , 969 ( 2001 ) . a.f . andreev and e.p . bashkin , sov . jetp * 42 * , 164 ( 1976 ) . j. sauls , in _ timing neutron stars _ , eds . h. gelman and e.p.j . van den heuvel , ( dordrecht , kluwer , 1989 ) , pp . 457 - 490 . m. borumand , r. joynt , and w. klzniak , phys . c * 54 * , 2745 ( 1996 ) . misner , k. thorne , and j.a . wheeler , _ gravitation _ ( freeman , san francisco , 1973 ) . b. carter , `` relativistic fluid dynamics '' , a. anile and m. choquet - bruhat , eds . , springer - verlag ( 1989 ) . g. l. comer and d. langlois , class . and quant . grav . * 11 * , 709 ( 1994 ) . b. carter and d. langlois , phys . d * 51 * , 5855 ( 1995 ) . b. carter and d. langlois , nucl . b * 454 * , 402 ( 1998 ) . b. carter and d. langlois , nucl . b * 531 * , 478 ( 1998 ) . d. langlois , a. sedrakian , and b. carter , mon . not . r. astron . soc . * 297 * , 1189 ( 1998 ) . r. prix , phys . d * 62 * , 103005 ( 2000 ) . comer , found . phys . * 32 * , 1903 ( 2002 ) . hartle , ap . j. * 150 * , 1005 ( 1967 ) . friedman , j.r . ipser and l. parker , ap . j. * 304 * , 115 ( 1986 ) . cook , s.l . shapiro , and s.a . teukolsky , ap . j. * 422 * , 227 ( 1994 ) . n. andersson and g. l. comer , phys . * 24 * , 241101 ( 2001 ) . n. andersson , g. l. comer , and r. prix , phys . letts . * 90 * , 091101 ( 2003 ) . r. prix , g.l . comer , and n. andersson , mon . not . r. astron . , in press ( 2004 ) . farley , phys . * 10 * , 279 ( 1963 ) . o. buneman , phys . * 10 * , 285 ( 1963 ) . b. link , phys . lett . * 91 * , 101101 ( 2003 ) .

we prove a theorem which bounds heegaard genus from below under special kinds of toroidal amalgamations of-manifolds . as a consequence , we conclude for any pair of knots , where denotes the tunnel number of . the tunnel number of a knot can be defined by the equation , where denotes heegaard genus and is an open regular neighborhood of . in more intuitive terms , the tunnel number of a knot is the minimal number of `` tunnels '' that must be drilled through in order to make the resulting manifold a handlebody . the behavior of under the operation of connected sum has been studied extensively . it is not difficult to see that , although it takes some work to find examples where equality is achieved in this bound , , . it is also known that for some pairs of knots , , , and that the degeneration can be arbitrarily large , . perhaps most difficult is the task of finding lower bounds on . it is known that for any pair of non - trivial knots in , and more generally that . in the case that and are small , it is known that ( the hard part of this is to show that ) , in fact this equation holds even under the assumption that and are meridionally small . another impressive lower bound is , which holds for any pair of non - trivial knots in ( in fact a more general analogue involving iterated connected sums is derived in ) . in this paper , we prove that for any pair of knots . this bound was previously unknown , although there are many examples which show it to be best possible , including those of and . the proof is centered around the rather involved construction of so - called _ doppelgnger _ surfaces . in rough outline , the strategy of our proof is as follows . suppose without loss of generality that and let be realized via the satellite construction with as the companion and as the pattern . this means that lies in , and if is the standard unknotted embedding of the solid torus , then . if is a thin generalized heegaard surface of , then can be isotoped to intersect in a particularly nice way . taking into account certain information contained in the intersection , we can then construct the previously mentioned _ doppelgnger surface _ inside of a solid torus which , in certain important respects , imitates the placement of the surface in . as a result , forms a generalized heegaard surface of which amalgamates to a surface of lower genus than the amalgamation of . this yields the desired lower bound . in section 1 we introduce the notion of a _ generalized compression body _ , which form the basic pieces of and , and prove a series of essential cutting and pasting lemmas about them . section 2 then describes and works out the basic topology of so - called _ spoke graphs _ and _ spoke surfaces _ , which form the building blocks of the doppelgnger surface . section 3 then constructs in detail and proves that it has the desired properties , culminating in the main technical result of the paper , theorem 3.23 . in section 4 the bound is proved ( theorem 4.1 ) , and some topics related to it are briefly discussed . throughout this paper , denotes a _ closed _ regular neighborhood of in , , and , or equivalently , . we assume throughout that behaves well with respect to intersection , so that . if is a topological space , denotes the number of components of . an embedding of manifolds is said to be _ proper _ if is transverse to and . a _ proper isotopy _ is a homotopy through proper embeddings ( note that this does _ not _ imply that the boundary remains fixed ) . as an informal aid to the reader , topological spaces which are allowed to have multiple connected components will usually be denoted in calligraphic font , e.g. , , , whereas connected topological spaces will usually be denoted in standard font , e.g. , , .

1407.3554

let @xmath56 be a compact orientable surface , and let @xmath57 , where the @xmath58-handles are attached along essential , non - boundary parallel curves in @xmath59 , and @xmath0-handles are attached along all spherical components of @xmath60 which are disjoint from @xmath61 . then @xmath62 is called a _ generalized compression body over @xmath56 _ , or simply a _ generalized compression body_. let @xmath63 , @xmath64 , and @xmath65 . if @xmath62 is connected and @xmath66 , @xmath62 is a _ compression body_. if @xmath62 is connected and @xmath67 , @xmath62 is a _ handlebody_. suppose @xmath62 is a generalized compression body , @xmath68 are disjoint components of @xmath69 , and @xmath70 is an orientation reversing homeomorphism which preserves @xmath71 . then @xmath72 is a generalized compression body over @xmath73 . if @xmath62 is a generalized compression body and @xmath74 is obtained by compressing @xmath62 along a properly embedded disk @xmath75 such that @xmath76 , then @xmath74 is again a generalized compression body . going the other way , if @xmath74 is obtained from @xmath62 by attaching an oriented @xmath77-handle along @xmath71 , then @xmath74 is a generalized compression body . let @xmath62 be a generalized compression body and let @xmath78 and @xmath79 be disjoint unions of annuli embedded in @xmath71 satisfying @xmath80 . let @xmath81 be a disjoint union of compressing disks for @xmath62 such that @xmath82 . if @xmath83 consists of a single spanning arc in one of @xmath84 or @xmath85 for all @xmath86 , and @xmath87 if @xmath88 , then @xmath89 is said to be a _ primitive disk set _ for @xmath90 , and the component of @xmath91 which meets @xmath92 is said to be _ dual _ to @xmath92 . the above orderings of the components of @xmath90 and @xmath89 will be called the _ primitive ordering _ associated with @xmath89 . see figure 1 . the choice of primitive ordering is essential to definition 1.4 , and some fixed choice is always assumed to be present when we are dealing with a primitive disk set @xmath89 for a paired union @xmath90 of annuli . for the most part , however , the primitive ordering will only be specified explicitly when necessary . let @xmath62 be a generalized compression body and let @xmath78 and @xmath79 be disjoint unions of annuli embedded in @xmath71 satisfying @xmath80 . let @xmath93 be an orientation reversing homeomorphism such that @xmath94 for all @xmath86 , and suppose that @xmath90 admits a primitive disk set . then @xmath72 is a generalized compression body over @xmath73 . we proceed by induction on @xmath95 . in the base case @xmath96 there is nothing to prove . if @xmath97 , suppose without loss of generality that @xmath98 is dual to @xmath99 ( the argument is the same if @xmath100 is dual @xmath99 ) . by observation 1.3 , @xmath101 is again a generalized compression body , and the result of reattaching @xmath102 to @xmath101 via the map @xmath103 again results in a generalized compression body @xmath104 , since this amounts to trivially attaching a ball to @xmath105 along a disk on its boundary . since @xmath106 is obtained from @xmath104 via a @xmath77-handle attachment along @xmath107 , observation 1.3 tells us that @xmath106 is a generalized compression body . since @xmath108 was disjoint from @xmath109 and @xmath99 , it remains a primitive disk set for @xmath110 and @xmath111 in @xmath106 , and the desired conclusion follows by induction . suppose @xmath62 is a generalized compression body , let @xmath112 be a disjoint union of annuli embedded in @xmath71 , and let @xmath113 be a disjoint union of disks properly embedded in @xmath62 such that @xmath114 . if @xmath115 consists of a single spanning arc in @xmath84 for all @xmath86 , and @xmath116 whenever @xmath117 , then manifold @xmath74 obtained by attaching 2-handles along @xmath53 is again a generalized compression body . the proposition is well known in the case the @xmath62 is a compression body . the proof in the general case here is essentially the same as that of proposition 1.6 . an annulus @xmath55 properly embedded in a generalized compression body @xmath26 is said to be _ spanning _ if one component of @xmath118 lies on @xmath119 and the other lies on @xmath120 . @xmath55 is said to be _ horizontal _ if @xmath121 . if @xmath122 is a compact surface and @xmath53 is a disjoint union of incompressible spanning annuli embedded in @xmath123 , then @xmath123 can be re - parameterized so that @xmath124 for some disjoint union of essential simple closed curves @xmath125 . this is a well known fact which often appears in the literature , so the following proof is merely a sketch . if @xmath126 , then @xmath127 is another union of spanning annuli @xmath128 such that @xmath129 is isotopic to @xmath130 in @xmath131 ( this follows from the @xmath132-injectivity of @xmath53 and @xmath128 ) . this allows @xmath128 to be properly isotoped so that @xmath133 and @xmath134 are parallel and disjoint in @xmath135 . since @xmath53 and @xmath128 are both incompressible , and @xmath123 is irreducible , any simple closed curves in @xmath136 which are trivial in either of @xmath128 or @xmath53 can be eliminated via further isotopy of @xmath128 using standard inner - most disk arguments . again , since each component of @xmath90 is @xmath132-injective , any remaining components of @xmath137 must come from pairs of annuli @xmath138 with isotopic boundaries on @xmath139 . thus these components of @xmath137 can also be removed , uppermost ones first , using the fact that any incompressible horizontal annulus in @xmath123 with parallel boundary components will co - bound a solid torus with an annulus on @xmath140 . once @xmath128 has been made disjoint from @xmath53 , the components of @xmath90 will co - bound solid tori with annuli in @xmath141 , so that @xmath128 can finally be isotoped onto @xmath53 . extending this proper isotopy of @xmath128 to an ambient isotopy of @xmath142 yields the desired reparameterization . let @xmath62 be a compression body and let @xmath143 be a disjoint union of incompressible annuli properly embedded in @xmath26 , so that every component of @xmath144 is spanning and every component of @xmath145 is horizontal . then @xmath146 is a generalized compression body @xmath104 such that @xmath147 and @xmath148 . moreover , for an appropriate ordering of the components of @xmath149 , if we set @xmath150 , @xmath151 , and @xmath152 , then the collection @xmath153 admits a primitive disk set in @xmath104 . lemma 2 of @xcite tells us that @xmath154 is a union of compression bodies . the annuli @xmath144 remain spanning in @xmath74 , thus are disjoint from some union of disks @xmath155 properly embedded in @xmath74 such that @xmath156 ( see , e.g. , lemma 3.15 of @xcite ) . by proposition 1.9 , we may assume that @xmath157 has been parameterized so that @xmath144 has the form @xmath158 in @xmath159 , where @xmath160 is a disjoint union of simple closed curves . thus @xmath161 is a generalized compression body satisfying @xmath162 , as claimed . we prove the second part by induction on @xmath95 . there is nothing to prove if @xmath96 . if @xmath97 , then lemma 9 of @xcite implies that @xmath145 is boundary - compressible in @xmath62 via some disk @xmath163 . we may choose @xmath163 so that it is disjoint from @xmath144 , and we assign the label @xmath164 to the component of @xmath145 which has been boundary - compressed by @xmath163 . by the first part of this lemma , @xmath165 is a generalized compression body and so by induction @xmath166 admits a primitive disk set @xmath167 in @xmath168 with respect to an appropriate choice of numbering for @xmath169 , with @xmath170 as the lowest indexed annulus . we may also choose ( appropriately indexed ) @xmath167 so that @xmath171 , and since @xmath172 , it follows that @xmath173 is a primitive disk set for @xmath174 in @xmath104 , as required . a graph @xmath47 embedded in a manifold @xmath175 is said to be _ properly embedded _ if @xmath47 is transverse to @xmath175 and @xmath176 is a union of elements from the set of univalent vertices of @xmath47 . @xmath47 is said to be _ unknotted _ if it can be isotoped into @xmath177 via an isotopy @xmath178 such that the function @xmath179 is a proper embedding for all @xmath180 . if @xmath42 is an unknotted tree properly embedded in the @xmath0-ball @xmath181 , and @xmath122 is the surface which results from removing an open collar of @xmath182 from @xmath183 , then there is homeomorphism @xmath184 such that @xmath185 and @xmath186 . let @xmath42 be a graph embedded in a handlebody @xmath26 such that @xmath187 . then @xmath42 is called a _ spine _ of @xmath26 . suppose @xmath42 is a graph embedded in the interior of a handlebody @xmath26 , and that there is a disjoint union of compressing disks @xmath89 properly embedded in @xmath26 so that the following are true : 1 . @xmath188 is a union of balls , 2 . @xmath189 is a properly embedded , unknotted tree for each component @xmath181 of @xmath188 , 3 . for each component @xmath75 of @xmath89 , @xmath190 is a single point on an edge of @xmath42 . then @xmath42 is a spine of @xmath26 . for each component @xmath181 of @xmath188 , let @xmath191 be the set of `` scars '' left behind by cutting along @xmath89 , i.e. @xmath192 , and let @xmath193 . then hypotheses ( 2 ) and ( 3 ) give us a homeomorphism @xmath194 as per observation 1.12 , where we may take @xmath195 . these homeomorphisms can then be pasted together to form a homeomorphism between @xmath196 and @xmath197 . suppose @xmath26 is a handlebody with spine @xmath42 , and that @xmath41 is a subgraph of @xmath42 without simply connected components . then @xmath198 is a compression body . @xmath198 is obtained from @xmath199 by attaching @xmath58-handles and @xmath0-handles along @xmath200 . a _ heegaard splitting _ @xmath201 of a manifold @xmath175 is a decomposition @xmath202 where each of @xmath203 is a compression body and @xmath204 . @xmath205 is called a _ heegaard surface_. the _ heegaard genus _ @xmath206 of a manifold is the minimal genus of a heegaard surface for @xmath175 . a _ generalized heegaard splitting _ @xmath207 of a manifold @xmath175 is a decomposition @xmath208 such that @xmath209 forms a heegaard splitting for the submanifold @xmath210 , @xmath211 for @xmath212 , and @xmath213 whenever else @xmath214 ( we include the case @xmath215 corresponding to standard heegaard splittings ) . the surface @xmath216 is called a _ generalized heegaard surface_. given a generalized heegaard surface @xmath28 , we let @xmath217 denote the _ thick surfaces _ of @xmath28 , and @xmath218 denote the _ thin surfaces _ of @xmath28 . a heegaard splitting @xmath201 is said to be : * _ stabilized _ if there exist compressing disks @xmath219 , @xmath220 such that @xmath221 is a single point . * _ reducible _ if there exist compressing disks @xmath219 , @xmath220 such that @xmath222 . * _ weakly reducible _ if it is not stabilized or reducible , but there are compressing disks @xmath219 , @xmath220 such that @xmath223 * _ strongly irreducible _ if it is not stabilized and , for all compressing disks @xmath219 , @xmath220 , @xmath224 there is process of _ untelescoping _ a weakly reducible heegaard splitting @xmath201 whereby it is changed into a generalized heegaard splitting @xmath207 satisfying @xmath225 . conversely , given a generalized heegaard splitting @xmath226 for @xmath175 , one can always use the process of _ amalgamation _ to change it into a standard heegaard splitting @xmath201 of @xmath175 satisfying the same equation . the interested reader is referred to @xcite for the details of these processes and a proof of the following lemma . @xcite if @xmath201 is a weakly reducible heegaard splitting of @xmath175 , then @xmath201 can be untelescoped to a generalized heegaard splitting @xmath227 @xmath228 such that @xmath209 is a strongly irreducible splitting of @xmath210 for each @xmath229 , and the thin surfaces @xmath230 are incompressible in @xmath175 for each @xmath231 . in this case the generalized splitting is said to be _ fully untelescoped_. a standard heegaard splitting that is strongly irreducible will also be considered _ fully untelescoped_. @xcite if @xmath28 is the union of the thick and thin surfaces of a fully untelescoped heegaard splitting of @xmath175 , and @xmath232 is an incompressible surface properly embedded in @xmath175 , then @xmath28 can be isotoped so that it meets @xmath232 only in simple closed curves which are non - trivial in both @xmath28 and @xmath232 . throughout this section , we set @xmath233 and parameterize it using polar coordinates @xmath234 , @xmath235 , @xmath236 . let @xmath237 be an embedded graph with one central vertex @xmath238 at @xmath239 , a finite number of outer vertices @xmath240 , one radial edge connecting each outer vertex @xmath241 to the central vertex @xmath238 , and one longitunidal edge @xmath242 connecting @xmath241 to itself , @xmath86 . then @xmath42 is said to be a _ connected spoke graph _ in @xmath243 . a finite , disjoint union of connected spoke graphs @xmath47 is simply called a _ spoke graph _ , @xmath244 is called a _ spoke surface _ , @xmath245 is a _ spoke chamber_. suppose @xmath42 is a connected spoke graph with central vertex at @xmath239 . let @xmath89 be a disjoint union of disks embedded in @xmath246 such that for each component @xmath75 of @xmath89 , * @xmath190 is a connected subarc of a radial edge of @xmath42 , * @xmath247 . then @xmath248 is called a _ stabilized _ spoke graph with stabilizing disk set @xmath89 , @xmath249 is a stabilized spoke surface , and @xmath250 is a stabilized spoke chamber . moreover , @xmath251 is called the set of _ stabilizing arcs _ of @xmath42 . let @xmath42 be a connected stabilized spoke graph with stabilizing disk set @xmath89 and central vertex at @xmath239 . let @xmath252 , and let @xmath253 , where @xmath254 . then the _ standard disk set _ of @xmath42 is the union of disks @xmath255 . the standard disk set of @xmath42 cuts @xmath256 into a union of balls . thus @xmath256 is a handlebody . let @xmath42 be a connected , stabilized spoke graph with stabilizing disk set @xmath89 , and whose central vertex has @xmath257-coordinate @xmath258 . let @xmath259 be another connected , stabilized spoke graph disjoint from @xmath42 with stabilizing disk set @xmath167 , whose central vertex has @xmath257-coordinate @xmath260 , and suppose the following properties are also satisfied : * the set of longitudinal edges of @xmath259 is precisely the set of core curves of the annuli @xmath261 . * @xmath262 . * every component of @xmath75 of @xmath89 meets precisely one component @xmath263 of @xmath167 in a single arc which has one endpoint on @xmath264 , and the other on @xmath265 , and conversely each component of @xmath167 meets precisely one component of @xmath89 in this way . * let @xmath266 be the projection @xmath267 . then @xmath268 is circle - valued morse function without singularities , and for every stabilizing arc @xmath269 of @xmath42 , @xmath270 is morse with only one critical point occurring at @xmath271 . likewise for the stabilizing disks and arcs of @xmath259 . then @xmath42 and @xmath259 are said to be _ dual _ to one another . see figure 2 . one way to obtain a dual graph @xmath259 is to rotate a copy of @xmath42 slightly in the @xmath257 and @xmath272 directions , and then isotope the stabilizing arcs of @xmath259 slightly so that they clasp those of @xmath42 in one to one fashion . however , we use the morse condition on the stabilizing arcs because it allows us complete flexibility in the choice of radial edges of @xmath259 at which to base its stabilizing arcs , while still avoiding knottedness . let @xmath42 and @xmath259 be a dual pair of connected spoke surfaces in @xmath243 , and let @xmath181 be the component of @xmath273 which contains the central vertex of @xmath259 , where @xmath274 is the standard disk set of @xmath42 from definition 2.4 . then @xmath181 is a ball , and @xmath275 is an unknotted tree properly embedded in @xmath181 . it is clear that @xmath181 is a ball , we show that @xmath275 is unknotted . we retain the notation of definition 2.6 throughout . if @xmath276 is the set of stabilizing arcs of @xmath259 , then the morse condition on @xmath276 ensures that @xmath277 is morse without singularities , and since @xmath268 is also morse without singularities , we can slide the endpoints of @xmath278 off of @xmath279 without introducing any further singularities . the components of @xmath280 can then be `` pushed in '' so that @xmath281 is level with respect to @xmath266 , and the arcs of @xmath278 can then be properly isotoped horizontally with respect to @xmath266 until they are vertical . after these isotopies , it is clear that @xmath275 is unknotted . if @xmath42 and @xmath259 are connected , dual , stabilized spoke graphs , then @xmath282 is homeomorphic to @xmath283 via a map sending @xmath284 to @xmath285 and @xmath286 to @xmath287 . take the double of @xmath243 to obtain @xmath288 , let @xmath289 be the double of of @xmath42 , and let @xmath290 be the double of @xmath259 . then @xmath291 is a handlebody , since the double @xmath155 of the standard disk set @xmath274 cuts @xmath291 into balls . moreover , @xmath291 , @xmath155 , and @xmath290 satisfy the hypotheses of proposition 1.14 ( lemma 2.8 handles the only subtle aspect of this ) . thus @xmath290 is a spine of @xmath291 , and we obtain a parameterization @xmath292 . by proposition 1.9 , the spanning annuli @xmath293 can be assumed to be vertical with respect to this parameterization , and the result follows . besides being a steppingstone to proposition 2.18 below , the significance of lemma 2.9 is that it allows us to isotope a connected , stabilized spoke surface @xmath294 back and forth between small neighborhoods of dual spoke graphs lying on opposite sides of @xmath295 in @xmath243 . this kind of isotopy will play an essential role in the final doppelgnger construction . let @xmath47 be a disjoint union of stabilized spoke graphs embedded in @xmath243 . then two components @xmath296 and @xmath297 of @xmath47 are said to be : * _ @xmath257-adjacent _ if there is a subarc @xmath298 such that the @xmath299 meets @xmath47 only in the central vertices of @xmath296 and @xmath297 , on its endpoints . in this case we call @xmath300 the _ spanning arc of @xmath257-adjacency_. * _ @xmath272-adjacent _ if the closure @xmath55 of some component of @xmath301 meets @xmath302 in one boundary component and @xmath303 in the other . in this case @xmath55 is said to be a _ spanning annulus of @xmath272-adjacency_. let @xmath47 be a disjoint union of stabilized spoke graphs whose components are ordered @xmath304 so that @xmath305 is @xmath257-adjacent to @xmath306 , and let @xmath307 be the spanning arc of @xmath257-adjacency for @xmath305 , @xmath306 , @xmath212 . then @xmath308 is said to be the _ binding arc _ of @xmath47 with respect to the given ordering of its components . suppose further that @xmath305 is @xmath272-adjacent to @xmath306 , and let @xmath84 be a spanning annulus of @xmath272-adjacency for @xmath305 , @xmath306 , @xmath212 . then @xmath309 is said to form an _ adjacency chain _ for @xmath47 with respect to the given ordering of its components . let @xmath47 be a disjoint union of stabilized spoke graphs ( possibly with detached longitudes ) , and let @xmath42 be a connected stabilized spoke graph ( possibly with detached longitudes ) obtained from @xmath47 by rotating each component of @xmath47 in the @xmath257-direction so that all of their central vertices coincide at a single vertex @xmath238 . then @xmath47 is said to be a _ decomposition _ of @xmath42 . if @xmath310 , for every central vertex @xmath311 occurring in a component of @xmath47 , where @xmath312 is the flat metric , then @xmath47 is said to be an _ @xmath313-small _ decomposition of @xmath42 . if @xmath47 is an @xmath313-small decomposition of @xmath42 , and @xmath314 , then there are exactly two orderings @xmath315 of the components of @xmath47 with the property that @xmath305 is @xmath257-adjacent to @xmath306 for all @xmath212 , and such that the binding arc @xmath269 has length less than @xmath316 ( and these two orderings are just the reverse of one another ) . if , moreover , @xmath305 is @xmath272-adjacent to @xmath306 for all @xmath212 for one ( and therefore both ) of these orderings , we shall say that @xmath47 is a _ good _ @xmath313-small decomposition of @xmath42 . let @xmath47 be an @xmath313-small decomposition of @xmath42 , let @xmath269 be the binding arc of @xmath47 . then @xmath317 is isotopic to @xmath286 , and @xmath245 is obtained from @xmath318 via @xmath58-handle attachments to @xmath319 along meridians of @xmath269 . let @xmath296 and @xmath297 be a pair of components in @xmath47 which are @xmath257- and @xmath272-adjacent , with spanning arc of @xmath257-adjacency @xmath269 and spanning annulus of @xmath272-adjacency @xmath55 . let @xmath75 be disk embedded in @xmath243 such that @xmath320 , where @xmath321 is the radial edge of @xmath305 nearest to @xmath55 , @xmath322 , @xmath323 is an arc joining @xmath324 to @xmath325 which spans @xmath55 , and @xmath326 . then @xmath75 is said to be a _ spanning disk _ of @xmath296 and @xmath297 . see figure 3 . let @xmath327 be the dilation @xmath328 . let @xmath42 be a stabilized spoke graph , let @xmath329 be a union of longitudinal edges of @xmath42 , let @xmath155 be the union of those radial edges of @xmath42 which meet @xmath329 , and let @xmath330 be the union of stabilizing arcs attached to @xmath155 . then the graph @xmath259 obtained by removing @xmath331 from @xmath42 and attaching @xmath332 is said to be a _ spoke graph obtained by @xmath313-small detachments of the longitudes @xmath333_. suppose @xmath42 and @xmath259 are dual stabilized spoke graphs embedded in @xmath243 , that @xmath314 , and that @xmath334 ( as usual @xmath335 is the flat metric ) . suppose @xmath47 is an @xmath336-small , good decomposition of @xmath42 , and that @xmath337 is an adjacency chain of annuli for @xmath47 . let @xmath41 be a spoke subgraph of @xmath259 which does not meet @xmath53 , and suppose @xmath54 is obtained from an @xmath336-small longitudinal detachments @xmath41 followed by an @xmath336-small decomposition ( which need not be `` good '' ) . let @xmath128 be the subset of @xmath338 consisting of those annuli which meet @xmath244 on one boundary component , and @xmath339 on the other . then @xmath340 is a generalized compression body @xmath26 such that @xmath341 and @xmath342 . we have chosen the various stabilizations and detachments small enough to ensure that they can all be carried simultaneously without creating new intersections . as in the proof of lemma 2.9 , we double @xmath243 to obtain @xmath288 , let @xmath289 denote the double of @xmath42 , and let @xmath290 be the double of @xmath259 . then if @xmath343 is the graph obtained by detaching some of the longitudes of @xmath41 , it remains isotopic to a subgraph of @xmath290 , which is a spine of @xmath291 . thus @xmath344 is a compression body with negative boundary @xmath345 . if @xmath54 is any @xmath336-small decomposition of @xmath343 , then @xmath346 is also compression body because observation 2.15 tells us that it is obtained from @xmath344 via @xmath58-handle attachments along @xmath347 . the annuli @xmath128 are the spanning annuli in the collection @xmath348 of incompressible annuli properly embedded in @xmath346 , and thus form the vertical boundary of the generalized compression body @xmath349 cut off by @xmath350 . order the components of @xmath351 so that @xmath84 meets @xmath352 and @xmath353 for all @xmath354 . the hypothesis that @xmath41 does not meet @xmath355 implies the existence of a spanning disk @xmath92 for each pair @xmath305 , @xmath306 which is disjoint from @xmath54 . if @xmath307 is the spanning arc of @xmath257-adjacency for @xmath305 , @xmath306 , then the meridian disk @xmath356 of @xmath307 in @xmath357 meets the disk @xmath358 in a single point . thus by proposition 1.7 , @xmath359 , which is obtained by attaching @xmath360 to @xmath361 for each @xmath354 , is also a generalized compression body . throughout this section , @xmath175 is a compact orientable @xmath0-manifold , @xmath28 is a generalized heegaard surface of @xmath175 , @xmath232 is a separating essential torus properly embedded in @xmath175 , and @xmath362 . @xmath28 and @xmath232 are said to be _ well - configured _ with respect to @xmath363 if the following conditions hold : 1 . @xmath364 consists only of simple closed curves which are essential in @xmath232 and @xmath28 , 2 . each component of @xmath365 separates @xmath363 , 3 . for each component @xmath26 of @xmath366 , @xmath367 consists only of annuli which are spanning or horizontal . for the remainder of the section , we assume that @xmath28 and @xmath232 are well - configured with respect to @xmath363 . let @xmath368 , which is a disjoint union of compression bodies , let @xmath369 , @xmath370 , and @xmath371 . then conditions ( 1 ) and ( 3 ) of definition 3.3 , together with proposition 1.10 , imply that @xmath372 is a generalized compression body satisfying @xmath373 , where @xmath144 is the union of spanning annuli in @xmath53 . moreover , if @xmath374 is the subset of horizontal annuli in @xmath53 , and @xmath375 , @xmath322 , then there is a primitive disk set @xmath89 for @xmath62 with respect to some ordering of @xmath376 the notation of observation 3.4 is fixed for the remainder of the section . moreover , we fix a choice of a primitive disk set @xmath377 , which imposes the primitive orderings @xmath378 , @xmath322 . here it is understood that , for all @xmath379 , @xmath380 is the frontier of a single component of @xmath381 . let @xmath26 be a component of @xmath382 . let @xmath383 , let @xmath384 , @xmath385 , @xmath386 , and let @xmath387 consist of those components of @xmath388 which are dual to some component of @xmath389 . for every component @xmath26 of @xmath62 , @xmath390 is connected . in the case @xmath391 there is nothing to prove , so assume @xmath392 . order the components of @xmath393 so that @xmath117 if @xmath92 has lower index than @xmath394 with respect to the primitive ordering of @xmath89 . similarly , order @xmath395 so that @xmath117 if @xmath84 has lower index than @xmath396 with respect to the primitive ordering of @xmath145 ( so @xmath92 is dual @xmath84 for all @xmath354 ) . the fact that @xmath89 is primitive implies that @xmath397 meets @xmath387 only in a single spanning arc of @xmath164 , so that @xmath164 meets a single component of @xmath398 . likewise , @xmath399 is disjoint from @xmath400 and meets @xmath170 only in a single spanning arc . since @xmath399 does not change the component of @xmath390 on which it lies when it passes through @xmath164 , it follows that @xmath170 also meets the same component of @xmath401 on each side . continuing in this way for the remaining components of @xmath387 , we see that every component of @xmath387 meets a single component of @xmath398 . since @xmath119 is connected , this implies that @xmath390 is also connected . let @xmath26 be a component of @xmath62 , and index the annuli @xmath402 so that @xmath117 implies that @xmath84 has lower index than @xmath396 with respect to the primitive ordering on @xmath145 . a set @xmath403 of components of @xmath404 is said to be _ connective _ if @xmath405 is connected . moreover , @xmath406 is _ minimal _ if ( 1 ) @xmath407 is minimal among all connective sets of @xmath26 and , ( 2 ) for every other connective set of annuli @xmath408 satisfying @xmath409 , @xmath410 for all @xmath411 . note that conditions ( 1 ) and ( 2 ) define a unique minimal connective set with respect to any given ordering . for the remainder of the section , let @xmath412 be a solid torus , parameterized as in convention 2.1 . furthermore , let @xmath413 , @xmath322 and let @xmath414 be a homeomorphism such that each component of @xmath415 is a longitude of the form @xmath416 . let @xmath417 be the projection which collapses @xmath418 onto @xmath419 along the @xmath420-fibers of @xmath421 ( we assume that @xmath422 ) . we let @xmath423 for the remainder of the section . if a component @xmath26 of @xmath424 meets @xmath419 at all , then @xmath119 must meet @xmath419 , since the annuli of @xmath425 are all either horizontal or spanning . however , it is possible that @xmath426 consists entirely of spanning annuli , so that @xmath120 does not meet @xmath419 , and this is a case which requires special treatment at certain points in our construction . for each component @xmath26 of @xmath424 which meets @xmath419 , let @xmath427 denote @xmath428 which is a union of annuli . for each component @xmath26 of @xmath424 , and each component @xmath181 of @xmath427 , both components of @xmath429 lie on the same component of @xmath430 . if the curves of @xmath429 lie on distinct components @xmath431 of @xmath432 , then we can construct an embedded curve in @xmath363 which is the union of a spanning arc @xmath269 of @xmath433 and an arc @xmath434 properly embedded in @xmath26 with @xmath435 . this curve would then intersect the surface @xmath436 in a single point , which contradicts the assumption that each component of @xmath365 ( and hence each component of @xmath430 ) , is separating in @xmath363 . let @xmath26 be a component of @xmath424 , and let @xmath122 be a component of @xmath430 that meets @xmath419 . let @xmath437 be the union of those components @xmath181 of @xmath427 such that @xmath438 . if @xmath42 is a connected , stabilized spoke graph , possibly with detached longitudes , whose non - detached longitudinal edges are the core curves of @xmath437 , then @xmath42 is said to be a _ doppelgnger spoke graph _ for @xmath122 . let @xmath26 be a component of @xmath424 , and let @xmath439 be a spoke graph constructed as follows : 1 . let @xmath440 be the union of those annuli in @xmath427 which appear in @xmath437 for some component @xmath122 of @xmath441 , let @xmath42 be a connected stabilized spoke graph whose longitudinal edges are the core curves of @xmath440 , and let @xmath259 be its dual . suppose @xmath442 , where @xmath335 is the flat metric on @xmath243 as in lemma 2.17 . suppose @xmath443 is the minimal connective set of components of @xmath404 defined in definition 3.8 . for some pair of components of @xmath441 , label them @xmath444 and @xmath445 , @xmath446 has one component in @xmath447 and the other in @xmath448 . since @xmath406 was chosen to be minimal , @xmath444 will be distinct from @xmath449 whenever @xmath450 , ( otherwise we could remove an element of @xmath406 and still have a connective set ) . on the other hand , since @xmath406 is connective , @xmath451 . thus there exists an @xmath336-small decomposition @xmath47 of @xmath42 such that @xmath452 , where @xmath453 is a doppelgnger spoke graph of @xmath444 , and moreover @xmath47 can be chosen so that @xmath454 forms an adjacency chain for @xmath47 , where here @xmath455 denotes the component of @xmath456 which contains @xmath457 ( see definition 2.12 ) . 3 . let @xmath458 , which is the union of those annuli in @xmath427 which appear in @xmath437 for some component @xmath122 of @xmath120 that meets @xmath419 . every component of @xmath459 contains at most one component of @xmath460 , for if two components of @xmath460 both lied in the same component of @xmath459 , this would imply the existence of a component of @xmath426 whose boundary components both lie in @xmath120 , contrary to part ( 3 ) of definition 3.2 . thus we may assume that the core curves of @xmath460 form a subset of the longitudinal edges of the dual @xmath259 of @xmath42 . 4 . define the _ prohibited _ longitudinal edges of @xmath259 to be those which lie in the same component of @xmath459 as a component of @xmath461 ( there is at most one component of this set lying inside each component of @xmath459 ) . a subgraph @xmath343 of @xmath259 is said to be _ admissable _ if it possesses every core curve of @xmath460 as a longitudinal edge , but no prohibited longitudinal edges . it is possible that @xmath462 , in which case we set @xmath463 . otherwise , let @xmath464 be the components of @xmath120 which meet @xmath419 . let @xmath343 be an admissible subgraph of @xmath259 , and let @xmath41 be obtained from @xmath343 via an @xmath336-small detachment of those longitudes of @xmath343 which are not core curves of @xmath460 . finally , let @xmath465 be an @xmath336-small decomposition of @xmath41 , where @xmath466 is a doppelgnger spoke graph for @xmath467 , @xmath468 . then @xmath469 is said to be a _ doppelgnger spoke graph _ of @xmath26 , and it is said to be _ perfect _ if @xmath470 and @xmath471 for all @xmath472 , @xmath473 . the construction of definition 3.14 was tailored to the hypotheses of proposition 2.18 . it implies that if @xmath474 is a doppelgnger spoke graph associated with @xmath26 , then @xmath475 is a generalized compression body satisfying @xmath476 . @xmath477 is then the union of @xmath478 with those components of @xmath479 whose boundary components both lie in @xmath478 . moreover , the minimality of @xmath406 stipulated in part ( 2 ) of definition 3.14 allow us to deduce the existence of a primitive disk set @xmath480 in @xmath481 which will serve as a substitute for the disk set @xmath389 of @xmath26 , as the following lemma shows . let @xmath26 be a component of @xmath424 , let @xmath439 be a doppelgnger spoke graph of @xmath26 , and let @xmath482 . suppose @xmath402 is ordered as in definition 3.8 , and that @xmath483 . then there is an ordered , disjoint collection of disks @xmath484 properly embedded in @xmath481 with the following properties : 1 . @xmath485 , 2 . @xmath486 is a single spanning arc for all @xmath487 , 3 . @xmath488 for all @xmath487 , @xmath489 . for each component @xmath490 of @xmath388 , let @xmath491 denote the closure of the component of @xmath492 which contains @xmath493 . let @xmath403 be the minimal connective set for @xmath388 , and order the components of @xmath351 so that the components of @xmath494 lie in @xmath495 and @xmath496 for all @xmath472 . in the next two paragraphs we describe the components @xmath497 of @xmath498 . suppose first that @xmath499 lies on a single component @xmath500 of @xmath244 , where @xmath501 has central vertex @xmath502 . in the notation of part ( 4 ) of definition 3.13 , @xmath343 must be disjoint from @xmath503 , since the longitude of @xmath259 that lies in @xmath503 is prohibited ( @xmath503 contains a component of @xmath504 by definition ) . hence @xmath54 is disjoint from the component of @xmath505 that meets @xmath503 . we will let @xmath497 denote this component , which is a disk properly embedded @xmath340 . call this kind of disk a _ simple disk_. if the components of @xmath499 lie on distinct components @xmath500 and @xmath506 of @xmath244 , @xmath507 , then the disk we need to construct is a bit more complex . first let @xmath205 be the union of the spanning disks ( recall definition 2.16 ) for the annuli @xmath508 , and let @xmath75 be the spanning disk of @xmath503 . then let @xmath497 be the disk @xmath509 , which again is disjoint from @xmath54 by part ( 4 ) of definition 3.14 , and is thus properly embedded in @xmath340 . call this kind of disk a _ disk with flaps _ , see figure 4 . a simple disk is disjoint from all the other disks defined above , and any disks with flaps which overlap one another can also be isotoped slightly to be disjoint from one another . the resulting collection of disks is our collection @xmath480 . obviously @xmath480 satisfies conclusions ( 1 ) and ( 2 ) . conclusion ( 3 ) is also clear in the case that @xmath497 is simple , since in fact the only annulus of the form @xmath493 which it meets is @xmath510 . in the case that @xmath497 has flaps , condition ( 2 ) on minimality for @xmath406 given in definition 3.8 ensures that the index of @xmath497 is higher than that of every element element @xmath511 such that @xmath512 . otherwise we could replace @xmath490 with @xmath513 in @xmath406 to obtain another connective set which violates the minimality of @xmath406 . since @xmath497 meets no other annuli of the form @xmath493 except for @xmath510 , the final condition is satisfied . for every component @xmath26 of @xmath424 , there is a perfect doppelgnger spoke graph embedded in @xmath243 . for any doppelgnger spoke graph @xmath514 of a component @xmath122 of @xmath430 , the surface @xmath515 will have the same number of boundary components as @xmath122 . moreover , the genus of @xmath515 is the same as the total number of stabilizing arcs and detached longitudes that occur in @xmath514 . in particular , if @xmath514 is unstabilized and has no detached longitudes , then @xmath515 will be planar . thus a doppelgnger spoke graph @xmath514 can always be found which satisfies @xmath516 after attaching a sufficient number of stabilizing arcs and/or detached longitudes . in part ( 1 ) of definition 3.14 , we have the flexibility to stabilize @xmath42 as often as we need , with stabilizing arcs based on radial edges of our choosing . this allows us to choose the number of stabilizing arcs that will eventually occur in the components of the spoke graph @xmath47 defined in part ( 2 ) of definition 3.14 . it follows from this and the previous paragraph that we may choose @xmath47 so that each of its components @xmath453 satisfies @xmath517 . as noted in remark 2.7 , the morse condition of definition 2.6 grants us enough flexibility to choose the radial edges on which the stabilizing arcs of @xmath259 ( the dual of @xmath42 ) will be based , and this in turn allows us to control the component of @xmath54 on which they will eventually occur in part ( 5 ) of definition 3.14 . likewise , we can choose the components of @xmath54 on which the detached longitudes of @xmath41 shall occur after the decomposition described in part ( 5 ) of definition 3.14 . so , similar to the case with @xmath47 , we may distribute detached longitudes and stabilizing arcs among the components of @xmath54 however we please . but there is an important difference : the total number of stabilizing arcs and detached longitudes that can occur in @xmath54 is bounded above by @xmath518 , where @xmath519 denotes the number of stabilizing arcs that occur in @xmath42 , and @xmath520 denotes the maximal number of detached longitudes that can occur on an admissible subgraph of @xmath259 ( as defined in part ( 4 ) of definition 3.14 ) . therefore , to complete the proof we must show that a total of @xmath518 stabilizing arcs and detached longitudes is always sufficient to create a spoke graph @xmath54 which satisfies the equation @xmath521 for each of its components @xmath466 . as in part ( 5 ) of definition 3.14 , let @xmath522 be the union of those components of @xmath120 which meet @xmath419 . by the first paragraph of this proof , the total number of stabilizing arcs and detached longitudes necessary to ensure that @xmath523 for all @xmath468 is equal to @xmath524 , where @xmath525 is the genus of @xmath467 . since @xmath526 for any connected , compact surface @xmath122 , we obtain @xmath527 the fact that this quantity is less than @xmath518 is ultimately derived from the inequality @xmath528 the truth of ( 2 ) can be seen as follows : @xmath529 is a generalized compression body with the same negative boundary as @xmath26 . furthermore , @xmath530 is connected since @xmath531 is connected , as can be seen using essentially the same proof as that of lemma 3.7 . now @xmath532 is obtained from @xmath530 via surgeries along disks , and there must be at least @xmath533 such surgeries since @xmath530 is connected . thus @xmath534 . inequality ( 2 ) now follows from the fact that @xmath535 , and the fact that @xmath536 . since no component of @xmath120 is a disk or sphere , and @xmath537 , @xmath538 . thus from ( 2 ) we easily obtain the analogue @xmath539 . in conjunction with equation ( 1 ) , we obtain @xmath540 our choice of @xmath47 has ensured that @xmath541 . we then compute @xmath542 , and so deduce @xmath543 the detached longitudes of @xmath41 , as described in part ( 5 ) of definition 3.14 , all come from core curves of the annuli of @xmath459 which contain a component of @xmath544 . on the other hand , the number of prohibited annuli ( part ( 4 ) of definition 3.14 ) is equal to @xmath545 ( where @xmath406 denotes the set of annuli defined in part ( 2 ) of definition 3.14 ) . hence there will be at most @xmath546 detached longitudes which may occur in @xmath54 . however , @xmath547 by the minimality of @xmath407 . plugging this into inequality ( 4 ) yields @xmath548 the lemma now follows . let @xmath26 be a component of @xmath424 , let @xmath549 denote the union of those components @xmath205 of @xmath365 such that @xmath550 . let @xmath474 be a perfect doppelgnger spoke graph for @xmath26 , and let @xmath551 be the result of isotoping @xmath552 so that @xmath553 , via an isotopy supported in @xmath554 ( here @xmath555 is the union of annuli described in definition 3.11 ) . then @xmath551 is called the _ doppelgnger surface _ of @xmath26 , and the closure of the component of @xmath556 which does _ not _ contain @xmath439 is the _ doppelgnger chamber _ of @xmath26 . there is little difference between the doppelgnger chamber @xmath481 of @xmath26 and @xmath340 , outside of the fact that the boundary components @xmath557 line up well with respect to our gluing map . in particular , observation 3.15 and lemma 3.16 apply in the case that @xmath481 is a doppelgnger chamber . let @xmath42 be a connected , unstabilized spoke graph and let @xmath259 be its dual . suppose @xmath41 is obtained from @xmath42 by attaching a total of @xmath95 stabilizing arcs and detached longitudes , and that @xmath343 is obtained from @xmath259 by attaching a total of @xmath95 stabilizing arcs and detached longitudes , in any fashion . then @xmath558 is properly isotopic to @xmath559 in @xmath243 . if @xmath41 has detached longitudes , then there is a sequence of edge slides which change @xmath41 into a connected stabilized spoke graph @xmath560 without detached longitudes ( see figure 5 ) , and these correspond to an isotopy of @xmath558 to @xmath561 , one which we can choose to be supported outside of a small open collar of @xmath562 . we may then isotope @xmath561 onto @xmath563 via a `` sweepout '' isotopy , where @xmath564 is the dual of @xmath560 . we will call this the _ sweepout stage _ of the isotopy between the surfaces . note that it is the only stage of the isotopy during which the boundary of our surface is not fixed . finally , since @xmath564 is obtained from @xmath259 by attaching @xmath95 stabilizing arcs , using ( a reversed version of ) the same kind of isotopy described in the first paragraph , we may slide the stabilizing arcs of @xmath564 along @xmath259 so that they coincide with the stabilizing arcs and detached longitudes of @xmath343 . this corresponds to an isotopy of @xmath563 onto @xmath559 which fixes @xmath565 . composing these istopies yields the result . let @xmath26 and @xmath566 be a pair of components of @xmath424 , and let @xmath549 , @xmath567 be defined as in definition 3.18 . then @xmath26 and @xmath566 are said to be _ adjacent _ if @xmath568 . suppose @xmath26 and @xmath566 are an adjacent pair of components of @xmath424 . then @xmath569 consists of a single component , call it @xmath205 . moreover , if @xmath570 and @xmath571 are the component of @xmath551 and @xmath572 , respectively , which satisfy @xmath573 , then @xmath570 is isotopic to @xmath571 in @xmath243 via an isotopy which fixes @xmath574 . if @xmath575 had at least two components @xmath576 and @xmath577 , then we could embed a simple closed curve @xmath578 which meets each of @xmath26 , @xmath566 , @xmath579 and @xmath580 in a single arc . since @xmath581 meets @xmath576 in a single point , this contradicts our assumption that each component of @xmath365 is separating in @xmath363 . thus there is exactly one component , @xmath205 , of @xmath575 . let @xmath582 and @xmath583 . then @xmath122 is a component of @xmath430 , @xmath584 is a component of @xmath585 , and we let @xmath514 , @xmath586 denote the corresponding perfect doppelgnger spoke graphs which give rise to @xmath570 and @xmath571 . setting @xmath587 and @xmath588 in lemma 3.20 , we see that there is an isotopy of @xmath515 onto @xmath589 . outside of a small open collar @xmath590 of @xmath562 , the isotopy we require from @xmath570 to @xmath571 is identical to the isotopy described in the proof of lemma 3.20 . the only difference is that , during the sweepout stage , @xmath591 will be taken onto @xmath592 via an isotopy which leaves @xmath574 fixed , as shown in fig . ? ? if the generalized heegaard surface @xmath593 amalgamates to a minimal genus heegaard surface of @xmath175 , then @xmath594 ( here @xmath595 is the manifold obtained by gluing @xmath243 to @xmath596 as described in convention 3.9 ) . our strategy is to construct a surface @xmath597 so that @xmath598 ( see convention 3.9 ) forms a generalized heegaard surface of @xmath595 , one which amalgamates to a heegaard surface of at most the same genus as the heegaard surface that @xmath28 amalgamates to in @xmath175 . this will yield the desired result . our assumption that @xmath232 and @xmath28 are well - configured does not eliminate the possibility that @xmath599 . but in this case @xmath232 will be parallel to a component of @xmath600 , the thin part of @xmath28 , which implies the stronger conclusion @xmath601 . so we assume @xmath602 . let @xmath26 be a component of @xmath424 which meets @xmath419 , and let @xmath551 be its doppelgnger surface . by lemma 3.22 , for each component @xmath566 which is adjacent to @xmath26 , there is exactly one component @xmath570 of @xmath551 which is isotopic to a component @xmath571 of @xmath572 . since this isotopy fixes @xmath574 , it may be extended to an ambient isotopy of @xmath243 which fixes @xmath562 , and this ambient isotopy will push the remaining components of @xmath551 and the doppelgnger chamber @xmath481 of @xmath26 , into the complement of the doppelgnger chamber @xmath603 of @xmath566 in @xmath243 , thereby allowing @xmath603 and the remaining components of @xmath572 to be embedded in @xmath243 without meeting the deformed copies of @xmath551 and @xmath481 . call this ambient isotopy a _ flipping isotopy_. since any flipping isotopy leaves @xmath562 fixed , every disk embedded in the original version of the doppelgnger @xmath481 of @xmath26 is deformed to a disk which meets @xmath604 in the same way as the original . thus the deformed copy of @xmath481 still satisfies observation 3.15 and lemma 3.16 . thus , by performing and reversing flipping isotopies across the components of @xmath551 , we may simultaneously embed deformed , but topologically equivalent , versions of the doppelgnger surfaces and chambers of every component of @xmath424 that is adjacent to @xmath26 . this process of embedding can then be repeated for each of the components of @xmath424 that are adjacent to @xmath26 , and then to the components of @xmath424 which are adjacent to the components of @xmath424 that are adjacent to @xmath26 , and so on . this process will eventually terminate with the desired surface @xmath32 , and @xmath605 will be the disjoint union of ( harmlessly deformed versions of ) the doppelgnger chambers of those components of @xmath424 which meet @xmath419 . observation 3.15 then implies that @xmath606 is a generalized compression body . moreover , if @xmath155 is the union of all the ( deformed versions of ) the disk sets @xmath480 defined in lemma 3.16 , then lemma 3.16 implies that @xmath607 admits an ordering which makes it a primitive disk set for the ordered union of annuli @xmath608 ( see definition 1.4 and convention 3.5 ) with respect to the map @xmath609 ( see convention 3.9 ) . thus , if we let @xmath610 , and @xmath611 ( here @xmath612 as in observation 3.14 ) , then by proposition 1.6 , @xmath613 is also a generalized compression body . moreover , @xmath614 ( here @xmath615 ) satisfies the requirements of observation 1.2 , hence @xmath616 is also a generalized compression body . but if @xmath617 , then @xmath618 , so in fact @xmath619 is a generalized heegaard splitting of @xmath595 . moreover , since the doppelgnger surface @xmath32 was constructed from _ perfect _ spoke graphs as per definition 3.18 , the components of @xmath619 will all have exactly the same genus as the corresponding components of @xmath28 . it follows that @xmath619 will amalgamate to a heegaard surface of genus no higher than the surface which results from amalgamating @xmath28 . for the sake of clarity is worth remarking that in the proof above we can not necessarily conclude that @xmath619 amalgamates to a surface of the _ same _ genus as the one which @xmath28 amalgamates to , because @xmath619 will have no components which correspond to any components of @xmath28 which lie entirely inside of @xmath363 . in the event that such components of @xmath28 do exist , it is easy to verify that @xmath619 will amalgamate to a surface of strictly lower genus . it is also perhaps worth pointing out that , since the core @xmath620 of @xmath243 can be embedded in @xmath621 , we can stabilize @xmath619 once ( if necessary ) to obtain a generalized heegaard splitting of @xmath622 of genus at most @xmath623 . this allows us to deduce the following corollary . if @xmath28 amalgamates to a minimal genus heegaard surface of @xmath175 , then @xmath624 . if @xmath17 and @xmath18 are knots in @xmath15 , then @xmath1 . the proposition is trivial if one of @xmath17 or @xmath18 is the unknot , so suppose that both are non - trivial knots . assume also that @xmath22 . let @xmath232 be the `` swallow - follow '' torus in @xmath625 which swallows the @xmath18 summand and follows the @xmath17 summand ( see figure 6 ) . we apply theorem 3.23 by setting @xmath626 , noting that one component of @xmath627 is homeomorphic to @xmath628 , which will correspond to @xmath363 . what needs to be shown is that the untelescoped minimal splitting @xmath28 can be isotoped so that it meets @xmath232 only in essential simple closed curves , and such that each component of @xmath629 is separating . by its definition as a swallow - follow torus , @xmath232 is isotopic to @xmath630 , where @xmath55 is the decomposing annulus of the connected sum in @xmath626 , and @xmath181 the sub - annulus of @xmath631 which lies in the component of @xmath632 corresponding to @xmath628 . by proposition 1.21 , @xmath28 can be isotoped to intersect @xmath55 only in essential simple closed curves , and since each boundary component of @xmath629 is then a standard meridional curve of @xmath633 , every component of @xmath629 is separating in @xmath628 ( otherwise we could obtain a non - separating surface in @xmath15 ) . the hypotheses of theorem 3.23 ( which assume conventions 3.1 , 3.3 , and 3.9 ) can then be satisfied by isotoping @xmath232 sufficiently close to @xmath630 . now @xmath596 is the component of @xmath627 which is _ not _ homeomorphic to @xmath628 , but is instead homeomorphic to @xmath634 , where @xmath635 is the link in @xmath15 which has @xmath18 as one component , and a meridian @xmath636 of @xmath18 as its other component , and @xmath637 under this correspondence ( see figure 6 ) . furthermore , the slope in which @xmath28 has been made to intersect @xmath638 is the standard longitudinal slope determined by the meridian disk @xmath639 with @xmath640 and @xmath641 . thus the slope of the trivial dehn filling of @xmath642 which yields @xmath643 meets each component of @xmath644 exactly once , and theorem 3.23 applies to @xmath645 , yielding @xmath646 ( in yellow ) is seen as situated in @xmath647 , and @xmath364 is indicated with dashed lines lying on the `` swallow - follow '' torus @xmath232 . in the second diagram , @xmath596 is reimbedded in @xmath648 , and in the final diagram we see @xmath596 and @xmath364 as they look after inverting the image of @xmath232 under this reimbedding.,scaledwidth=95.0% ] this proof also works if the knots @xmath17 and @xmath18 are embedded in homology spheres ( or any pair of compact @xmath0-manifolds in which every closed embedded surface is separating ) . in general , however , it is important to keep in mind the delicacy theorem 3.23 ( and corollary 3.25 ) , whose assumptions are encoded in conventions 3.1 , 3.3 and 3.9 . in particular , the assumption of convention 3.3 that @xmath232 and @xmath28 are well - configured can not always be satisfied , as can be shown using straightforward examples in @xmath649 , where @xmath122 is a closed genus @xmath650 surface . thus corollary 3.25 can not be applied to prove the following plausible conjecture in the case @xmath651 in any obvious way . suppose @xmath175 is a compact @xmath0-manifold and @xmath232 is a separating , incompressible , orientable , genus @xmath652 surface properly embedded in @xmath175 . if @xmath363 and @xmath596 are the components of @xmath627 , then @xmath653 . similarly , the need for @xmath232 and @xmath28 to be well - configured is what keeps us from applying theorem 3.23 and corollary 3.25 to prove the analogue of theorem 4.1 for satellite knots . theorem 4.1 has some relation to the `` rank - genus conjecture '' for knot complements in @xmath15 . if we define @xmath654 to be the minimal number of generators for @xmath655 , then the rank - genus conjecture states : for all knots @xmath5 , @xmath656 . since a genus @xmath652 heegaard splitting of a knot complement induces a @xmath652-generator presentation of @xmath655 , it is clear that @xmath657 , but it remains unknown whether it is possible for this inequality to be strict . the rank - genus conjecture for closed @xmath0-manifolds is known to be false @xcite , even when restricted to the class of hyperbolic @xmath0-manifolds @xcite . this suggests that the rank - genus conjecture also fails for knot complements , and a pair of knots in @xmath15 whose tunnel number degenerated enough to violate theorem 4.1 would have given a counterexample , since the following analogue of theorem 4.1 for rank is trivial ( thanks to richard weidmann for pointing out the simple line of proof below ) . @xmath658 for any knots @xmath2 . @xmath659 is an amalgamated free product @xmath660 which retracts onto each of its factors . the fact that theorem 4.1 is true indicates that the class of knot pairs which experience high tunnel number degeneration is not a good place to look for counterexamples to the rank - genus conjecture after all . it might even be seen as a small vote in favor of the possibility that the rank - genus conjecture is valid for knot complements in @xmath15 . i would like to thank maggy tomova and charlie frohman for their support and guidance throughout my career . i would like to thank jesse johnson for the same , as well as for many helpful conversations and observations about this paper in particular .

we study compactified pure gauge / gravitational theories with gauge - fixing terms and show that these theories possess quantum mechanical susy - like symmetries between unphysical degrees of freedom . these residual symmetries are global symmetries and generated by quantum mechanical supercharges . also , we establish new one - parameter family of gauge choices for higher - dimensional gravity , and calculate as a check of its validity one graviton exchange amplitude in the lowest tree - level approximation . we confirm that the result is indeed-independent and the cancellation of the-dependence is ensured by the residual symmetries . we also give a simple interpretation of the vdvz - discontinuity , which arises in the lowest tree - level approximation , from the supersymmetric point of view .

0801.0845

one of the salient features in compactified higher - dimensional field theories is the geometric `` higgs '' mechanism resulting from the dimensional reduction . if gauge particles can propagate to extra dimensions , the corresponding higher - dimensional gauge fields should be decomposed into normal modes of extra dimensions and then , from the four - dimensional point of view , they can be recast into an infinite tower of massive gauge bosons whose longitudinal degree of freedom is provided by absorbing one of extra spatial components of the field . this gauge boson mass generation is realized without invoking any scalar higgs fields ; it is essentially geometrical in nature . it also occurs in compactified higher - dimensional gravity theories the same mass generation mechanism without invoking any fundamental higgs fields . there , it seems that there is no explicit symmetry breaking . however , it was shown in the kaluza - klein ( kk ) theory that the spontaneous symmetry breaking certainly occurs @xcite : there exists an infinite - dimensional kac - moody - like symmetry at the 4d lagrangian level , but it is broken down to the 4d translation and internal u(1 ) symmetry by the vacuum configuration @xmath2 , and then the non - zero graviton modes become massive . though it would be the best to describe the geometric `` higgs '' mechanism by the argument along this line , it seems to be hard to extend the analysis to other more complicated compactified gravity theories . it is less obvious such analysis is applicable to compactified gauge theories . recently , an alternative view of this mass generation mechanism is given by @xcite : it is best described by using the hidden quantum mechanical supersymmetry in 4d mass spectrum . the point is that mass eigenfunction for each non - zero kk - mode has its own superpartner . these functions are defined as eigenfunctions of @xmath3 matrix super - hamiltonian . therefore it can be explained that @xmath4th kk - mode of the 4d component of the gauge field can absorb as its longitudinal degrees of freedom its partner @xmath4th kk - mode of the extra component of the field , and then @xmath4th gauge boson becomes massive . in ordinary spontaneously broken gauge theories such as the abelian higgs model , if we work in the @xmath5 gauge it is inevitable to introduce the fictitious particles , i.e. the would - be nambu - goldstone ( ng ) bosons , so as to maintain the unitarity of the @xmath6-matrix . the scalar component of the massive gauge field , the would - be ng boson , the faddeev - popov ( fp ) ghost and the anti fp ghost are all in degenerate with the same mass - squared @xmath7 . these spurious degrees of freedom consist of a multiplet , known as the brs quartet @xcite , and do not contribute to the physical amplitude by canceling each other , and then the unitarity of the @xmath6-matrix is maintained . however , there is no such scalar higgs fields in pure gauge theory with extra compact dimensions , nor in compactified higher - dimensional pure gravity . thus the unphysical polarization states of higher - dimensional gauge / gravitational fields must be canceled among themselves . furthermore , in order to realize this cancellation these spurious degrees of freedom must be in degenerate with the same gauge - dependent mass - squared . in view of this , we have to guarantee at least two things : the first is the same number of degrees of freedom for the unphysical components of 4d gauge / gravitational field and the would - be ng bosons . these degrees of freedom have to appear in pairs . the second is the degeneracy of mass , that is , the same @xmath1-dependent pole of the propagators between these pairs . in view of this , it is sufficient to consider pure gauge / gravitational theories up to quadratic order and compute the propagators in @xmath5 gauges . it should be noted at this stage that , from the knowledge of the supersymmetric structure in the 4d mass spectrum , the mass matrices of these pairs should be given by the @xmath8 super - hamiltonians . therefore it is reasonable to expect that these free field theories would be invariant under some susy - like transformations generated by supercharges to rotate these unphysical degrees of freedom with different spins . this is the main subject of this paper . we study the compactified pure abelian gauge theory and pure gravity to quadratic order of gravitational fluctuations . we show that these free field theories exhibit residual global susy - like symmetries between the unphysical components of 4d gauge / gravitational fields and the would - be nambu - goldstone bosons , which are one of the extra space components of the higher - dimensional fields . also , we establish new @xmath5 gauges in five - dimensional gravity . we check its validity by computing the lowest tree - level graviton exchange amplitude . of course the result does not depend on the gauge parameter @xmath1 . note that we will not argue with unitarity bounds of some specific models as discussed in @xcite . in this paper we will restrict ourselves to free field theories . the rest of this paper is organized as follows . in section [ sec : abelian ] we study the pure abelian gauge theory with extra @xmath9-dimensions compactified on a riemannian manifold and show that in the @xmath5 gauge the scalar component of the four - dimensional gauge field and the would - be scalar ng boson are in degenerate with the same gauge - dependent mass - squared , and they form a multiplet under the susy - like transformation generated by the supercharge found in the analysis of the 4d mass spectrum . in section [ sec : gravity ] we extend the analysis to five - dimensional gravity with the randall - sundrum background and obtain the similar results . in this section we establish the one - parameter family of gauge choices analogous to the @xmath5 gauge in spontaneously broken gauge theories . as a check of validity of this @xmath5 gauge , we compute the one graviton exchange amplitude in the lowest tree - level and show that the result is indeed @xmath1-independent . from the result obtained by this computation , we give a simple interpretation of the appearance of the van dam - veltman - zakharov ( vdvz ) discontinuity from supersymmetric viewpoint . since it seems to be unfamiliar to particle theorists , we devote appendix [ appendix : vdvz ] to a brief review of the vdvz - discontinuity . we conclude in section [ sec : summary ] . in this section we show that any compactified pure abelin gauge theories intrinsically possess hidden quantum mechanical susy - like symmetry between a scalar component of 4d gauge field and one of scalar components of higher - dimensional gauge field , which should be regarded as a would - be scalar ng boson . let us consider the pure abelian gauge theory with extra @xmath9-dimensions compactified on a riemannian manifold @xmath10 without boundary . we denote the @xmath11-dimensional gauge field by @xmath12 @xmath13 , @xmath14 are the coordinates of the ordinary four - dimensional minkowski spacetime @xmath15 and @xmath16 are the coordinates of @xmath10 . the bulk metric is given by @xmath17 where @xmath18 is the flat spacetime metric whose signature is @xmath19 . the action we consider is @xmath20 where @xmath21 and @xmath22 . the action is invariant under the @xmath23 gauge transformation @xmath24 where @xmath25 is an arbitrary function . for the following discussion it is highly convenient to use the elegant mathematical tool of the differential forms . notice that on the @xmath9-dimensional manifold @xmath10 , @xmath26 and @xmath27 behave as a scalar and vector fields such that they can be regarded as 0-forms and a 1-form on @xmath10 , respectively . for further discussion it is also convenient to introduce the inner product of differential forms on the manifold @xmath10 . the inner product of two @xmath28-forms on @xmath10 is defined by @xmath29 where @xmath30 is the hodge star operator and @xmath31 @xmath32 and @xmath33 denote k - tuples of indices . with this definition , and can be written as the following form : @xmath34 and @xmath35 where @xmath36 and @xmath37 . to clarify the supersymmetric structure , we will follow the method of ref.@xcite . the hodge decomposition theorem tells us that any @xmath28-form on @xmath10 can be uniquely decomposed into the sum of the harmonic @xmath28-form , the exact @xmath28-form and the coexact @xmath28-form on @xmath10 . noting that there is no exact 0-form , we can write @xmath38 where @xmath39^{-1/2}$ ] and @xmath40 is the first betti number . as we shall see immediately , @xmath40 gives the number of 4d massless scalar bosons , which are identified as higgs fields in the gauge - higgs unification scenario . for example , @xmath41 for @xmath42 , @xmath43 for @xmath44 , @xmath45 for @xmath46 ( knneth formula ) , etc . @xmath47 and @xmath48 ( @xmath49 ) are coexact @xmath28-form and exact @xmath50-form and defined as the orthonormal sets of the eigenmodes of the positive semi - definite isospectral hamiltonians @xmath51 and @xmath52 : @xmath53 where the relative phase is chosen to be @xmath54 now we introduce the following @xmath3 matrix operators @xmath55 which satisfy the @xmath8 supersymmetry algebra @xmath56 = [ h , ( -1)^{f } ] = 0 , \nonumber\\ & \{(-1)^{f } , q_{a}\ } = 0 , \quad a , b = 1,2 , \label{eq : abelian12}\end{aligned}\ ] ] where @xmath57 . @xmath58 is the `` fermionic '' number operator which assigns to the `` bosonic '' state @xmath47 a quantum number @xmath59 and to the `` fermionic '' state @xmath60 a quantum number @xmath61 . from the orthonormality of the eigenmodes ] ] , it is easy to write down the 4d reduced lagrangian @xmath62 where @xmath63 . it is worth mentioning at this point that each kk - mode transforms under the u(1 ) gauge transformation as @xmath64 , @xmath65 , @xmath66 and the others are equal to zero . hence , from the higher - dimensional gauge invariant point of view , @xmath67 has to appear to @xmath68 in the combination @xmath69 , which is realized in as it should be . by ( partially ) fixing the gauge @xmath70 ( with @xmath71 left arbitrary ) all the would - be ng bosons @xmath67 are gauged away . this is the unitary gauge of the theory . the four - dimensional particle content of the theory is now obvious : the massless photon @xmath72 , the infinite tower of massive gauge bosons @xmath73 with mass @xmath74 , the @xmath40 massless scalar bosons @xmath75 and the infinite tower of massive scalar bosons @xmath76 with mass @xmath77 . as already mentioned in section [ sec : intro ] , in this paper we are interested in the @xmath5 gauge . since there is an undesirable quadratic term mixing @xmath26 and @xmath78 in ( or @xmath73 and @xmath67 in ) , we have to get rid of this mixing term by correctly choosing a gauge - fixing function . such a gauge - fixing function is already introduced , for example , in @xcite in the context of five - dimensional gauge theory . the @xmath9-dimensional version of such a gauge - fixing function should be chosen as @xmath79 = \partial^{\mu}a_{\mu } - \xi{\rm d}^{\dagger}a^{(1 ) } = \partial^{\mu}a_{\mu } - \xi{\rm d}^{\dagger}h . \label{eq : abelian14}\ ] ] by adding a gauge - fixing term @xmath80^{2}$ ] to and integrating by parts we obtain the gauge - fixed lagrangian @xmath81a_{\nu}\bigr ) \nonumber\\ & \hspace{1em } + \frac{1}{2}\bigl(\varphi , \box\varphi\bigr ) + \frac{1}{2}\bigl(\phi , [ \box - { \rm d}^{\dagger}{\rm d}]\phi\bigr ) \nonumber\\ & \hspace{1em } + \frac{1}{2}\bigl(h , [ \box - \xi{\rm d}{\rm d}^{\dagger}]h\bigr ) , \label{eq : abelian15}\end{aligned}\ ] ] where @xmath82 . inverting the kinetic term operator we obtain the gauge boson @xmath73 propagator containing a term proportional to @xmath83 with the gauge - dependent pole at @xmath84 , which is exactly the same simple pole position of the would - be ng boson s ( @xmath67 s ) , as it should . in any free field theories when two different fields are in degenerate and have the same degrees of freedom , there obviously exists a symmetry to rotate them . now this is indeed the case . the term proportional to @xmath83 with the gauge - dependent pole at @xmath84 in the gauge boson propagator suggests that the scalar component of the gauge field has the same mass - squared @xmath85 to the would - be ng boson s . indeed , if we decompose @xmath26 into the transverse and longitudinal ( or @xmath86 ) and longitudinal parallel to @xmath87 ( or @xmath86 ) . ] ( or pure gauge ) part @xmath88 the lagrangian becomes @xmath89a^{({\rm t})\mu}\bigr ) - \frac{1}{2\xi}\bigl(a^{({\rm l } ) } , \box[\box - \xi{\rm d}^{\dagger}{\rm d}]a^{({\rm l})}\bigr ) \nonumber\\ & \hspace{1em } + \frac{1}{2}\bigl(\varphi , \box\varphi\bigr ) + \frac{1}{2}\bigl(\phi , [ \box - { \rm d}^{\dagger}{\rm d}]\phi\bigr ) \nonumber\\ & \hspace{1em } + \frac{1}{2}\bigl(h , [ \box - \xi{\rm d}{\rm d}^{\dagger}]h\bigr ) . \label{eq : abelian17}\end{aligned}\ ] ] to canonicalize the lagrangian we redefine the field as ( @xmath90 . ) ] @xmath91 so that @xmath92-sector can then be recast into the following @xmath3 matrix form @xmath93 we have thus separated the lagrangian into the gauge - independent and -dependent parts . the gauge - dependent part can be written in terms of @xmath3 matrix form and the mass matrix is exactly the hamiltonian in , as expected . it is now obvious the lagrangian is invariant under the transformation @xmath94 this symmetry transformation is generated by the supercharge @xmath95 in , and hence susy - like . several comments are now in order : \(1 ) this is a symmetry between the scalar component of the gauge field @xmath26 and the would - be ng boson @xmath96 , both are unphysical degrees of freedom of the theory . thus if we add to the lagrangian the gauge invariant interaction terms or extend to non - abelian gauge theories , and then calculate the @xmath6-matrix elements , these spurious degrees of freedom must be canceled so as to maintain the unitarity of the @xmath6-matrix . we may therefore conclude that these cancellations are compensated for the symmetry , though it is much less obvious whether residual symmetry such as might exist in a fully interacting theory . \(2 ) as already mentioned in section [ sec : intro ] , is an expected symmetry . from the unitarity point of view , these unphysical gauge - dependent part must have its partners to ensure the cancellation of each contributions to the @xmath6-matrix elements . in view of this , these partners must be in degenerate with the same @xmath1-dependent mass - squared . since the super - hamiltonian @xmath97 is only the candidate for the mass - matrix and @xmath97 is written by product of the supercharges , it is quite natural these spurious degrees of freedom are mutually related by the global transformation generated by @xmath95 . \(3 ) since the residual symmetry is a spacetime symmetry between a spin-1 and a spin-0 particle , one might suspect that would conflict with the coleman - mandula theorem @xcite . however , the residual symmetry rotates only the unphysical degrees of freedom . hence this is not the case . \(4 ) this residual symmetry can be regarded as a relic of higher - dimensional gauge symmetry by using the gauge - fixing condition @xmath98 = 0 $ ] . it is easy to see that the transformation is indeed the higher - dimensional gauge transformation with an identification @xmath99 . in view of this fact , one might expect that compactified gauge theories with gauge invariant interaction terms would also possess some residual symmetries such as . however , this is an open question . we would like to further extend the previous analysis to compactified higher - dimensional pure gravity to quadratic order of gravitational fluctuations . because of authors inability to analyze the gravity in general dimensions with general background metric , we shall restrict ourselves to consider the perturbation to the well - known randall - sundrum metric @xcite as a less trivial example . let us consider the five - dimensional gravity with single extra dimension compactified on an interval @xmath100 with two 3-branes which are located at the end points of the interval . the action respecting this configuration is @xmath101 where @xmath102 is the five - dimensional planck mass , @xmath103 is the ricci scalar ] ] evaluated by the 5d metric @xmath104 and @xmath105 ( @xmath106 ) is the metric induced on the uv ( ir ) brane . the bulk cosmological constant @xmath107 and the brane cosmological constants @xmath108 and @xmath109 are tuned to give a warped background solution with a slice of @xmath110 : @xmath111 with @xmath112 and @xmath113 , where @xmath28 is the ads curvature scale defined as @xmath114 . it is just a matter of choice which coordinates ( physical or conformal ) one will use , however , the conformal coordinate will turn out to be advantageous to evaluate various quantities , especially the gravity action because it utilizes the very useful formula of the conformal transformation . thus in what follows we will use the conformal coordinate @xmath115 for computational convenience . we investigate the gravitational fluctuations around the background @xmath116 , \quad a(z ) = -\ln(1 + kz ) . \label{eq : gravity03}\end{aligned}\ ] ] it should be noted that the explicit expression for the warp factor @xmath117 is valid on the _ open _ interval @xmath118 : @xmath119 , \label{eq : gravity04}\end{aligned}\ ] ] where prime ( @xmath120 ) indicates the derivative with respect to @xmath115 . in the following analysis , however , the delta functions on the right hand side of eq . can safely be ignored because of the boundary conditions for the metric fluctuations consistent with the general coordinate invariance . therefore we will use the relation @xmath121 in the whole of this paper instead of . the fluctuations @xmath122 should be parameterized as @xmath123 which enables us to identify the spectrum of the linearized theory . notice that the limit @xmath124 gives @xmath125 and @xmath126 and hence the background geometry just reduces to @xmath127 , as it should . the quadratic action takes the form @xmath128 where @xmath129 @xmath130 these expressions to canonicalize the operator @xmath131 . ] are also found in @xcite and consistent with the einstein equations @xcite . one can easily verify that the action is invariant under infinitesimal general coordinate transformations @xmath132 which transform the fluctuations at the linearized level as @xmath133 * supersymmetric structure : * for the later discussion we here briefly summarize the quantum mechanical supersymmetric structure of the theory . as shown in ref.@xcite the fluctuations are expanded as @xmath134 where we have excluded the vector zero - mode @xmath135 from the expansion , since as we shall see immediately it will be incompatible with the supersymmetric structure . the orthonormal sets @xmath136 @xmath137 and @xmath138 are defined as the eigenfunctions of the schrdinger - like equations @xmath139 with the boundary conditions @xmath140 where @xmath141 or @xmath142 . note that the background solution enables us to factorize the hamiltonian for the vector - mode @xmath143 in two different ways ( except on the boundaries ) ; @xmath144 . these eigenfunctions are mutually related through @xmath145 now it is clear that there is no nontrivial solution of the vector zero - mode @xmath135 which has to satisfy the two different linear differential equations @xmath146 we have already included this fact into the expansion . it should be noted that the eigenfunctions satisfy the identity @xmath147 which plays an essential role in the proof of gauge - independence of one graviton exchange amplitude in the @xmath5 gauge ; see section [ sec : amplitude ] . now we introduce the following @xmath3 matrix operators @xmath148 which satisfy the _ two _ @xmath8 superalgebras @xmath149 = [ h , ( -1)^{f } ] = 0 , \nonumber\\ & \{(-1)^{f } , q_{a}\ } = 0 , \label{eq : gravity16a}\\ & \{{\bar q}_{a } , { \bar q}_{b}\ } = 2\delta_{ab}{\bar h } , \quad [ { \bar h } , { \bar q}_{a } ] = [ { \bar h } , ( -1)^{\bar f } ] = 0 , \nonumber\\ & \{(-1)^{\bar f } , { \bar q}_{a}\ } = 0 , \quad a , b = 1,2 , \label{eq : gravity16b}\end{aligned}\ ] ] where @xmath57 and @xmath150 . we emphasize that is the _ unique _ boundary conditions to ensure the hamiltonians @xmath97 and @xmath151 to be isospectral ( except for the zero modes ) , or the supersymmetries in the mass spectrum . indeed the hermiticity of the hamiltonians and supercharges is ensured if and only if the boundary conditions are imposed ( see ref.@xcite ) . 2ex * unitary gauge * : next we will identify the particle content of the theory . to this end we first note that using the mode decompositions the linearized general coordinate transformations read @xmath152 where the gauge freedom @xmath153 are assumed to be expanded as @xmath154 if we move on to the coordinate frame by choosing @xmath155 @xmath156 and @xmath157 with @xmath158 are all gauged away , leaving only @xmath159 so that the quadratic action becomes ] ] @xmath160 where @xmath161 is the ( 5d averaged ) 4d planck mass scale defined as @xmath162 which reduces to @xmath163 as it should , and @xmath164 notice that the graviton mass appears as the so - called fierz - pauli @xcite form , which is the only form that does not introduce ghosts @xcite . the four - dimensional field contents are now obvious : the massless graviton @xmath165 , the infinite tower of massive gravitons @xmath166 with increasing masses @xmath167 and the massless scalar boson @xmath168 known as the radion . it is worth mentioning at this point that @xmath161 is _ not _ directly related to the newton constant @xmath169 ; @xmath170 is just introduced as the coefficient of the massless graviton lagrangian , whereas @xmath169 has to be determined as the coefficient of the inverse square law part of the static gravitational force , which now includes the massless radion s contribution as well as the massless graviton s one . once fixed the action to the unitary gauge , all the fictitious massive vector- and scalar - modes would disappear from the dimensional reduced action so that we could not see the cancelation mechanism of the would - be ng bosons which appear in the other gauges , nor the expected susy - like symmetries between the unphysical degrees of freedom with the same gauge - dependent mass - squared . in view of this we have to establish the one - parameter family of gauge choices analogous to the @xmath5 gauge in the previous section . however , it is not straightforward to find such a gauge due to the appearance of three undesirable quadratic mixing terms . our strategy to find out the appropriate gauge - fixing function similar to the @xmath5 gauge in spontaneously broken gauge theories is based on the observations : \(1 ) in 4d general relativity the so - called harmonic gauge , which is the counterpart of the lorentz gauge in ordinary unbroken gauge theories , is just the linearized version of @xmath171 . therefore we could expect that the harmonic gauge conditions for the warped randall - sundrum metric would also be given by something like the conditions @xmath172 to linear order of @xmath122 . \(2 ) in order to fix the gauge arbitrariness @xmath153 we have to impose five independent conditions . the gauge - fixing function for @xmath173 must be expanded by the orthonormal set with the same boundary condition to @xmath173 , i.e. @xmath174 , whereas that for @xmath175 must be expanded by @xmath176 . as we shall see immediately , this is a key observation to select the terms adaptable to the gauge - fixing function from the linearized @xmath177 . in what follows we first investigate the harmonic gauge ( or t hooft - feynman gauge from the four - dimensional spontaneously broken point of view ) for the warped gravity and then establish the one - parameter family of gauge choices analogous to the @xmath5 gauge in spontaneously broken gauge theories . the analysis presented below is based on the knowledge of gauge - fixing in 4d general relativity , which is briefly summarized in appendix [ appendix : vdvz ] together with a brief review of the vdvz - discontinuity . 2ex * t hooft - feynman gauge * : as in 4d general relativity , we will first calculate @xmath178 to linear order : @xmath179 \nonumber\\ & \hspace{1em } + o({\bar h}^{2 } ) , \label{eq : gravity23a}\\ \partial_{m}\left(\sqrt{- g}g^{m5}\right ) & = { \rm e}^{3a } \biggl[3a ' + \frac{3}{2}a'\left(h - \phi\right ) + \frac{1}{2}\partial_{z}h - \partial_{\mu}h^{\mu 5 } \nonumber\\ & \hspace{1em } - \frac{3}{2}\left(\partial_{z } + 2a'\right)\phi\biggr ] + o({\bar h}^{2 } ) , \label{eq : gravity23b}\end{aligned}\ ] ] where @xmath180 . first look at the right hand side ( rhs ) of eq .. the three terms in the brace can be expanded in only @xmath181-modes so that it seems to be adopted as a gauge - fixing term . next look at rhs of eq .. since we are now interested in gauge - fixing for quadratic part of the action , the first term @xmath182 should be ignored , because it gives 0th and 1st order of the fields . the last three terms in the brace can be expanded in @xmath143-modes , however , the second term can not . thus , for the present , we will adopt only the last three terms for gauge - fixing and examine whether it works well or not . our modified harmonic gauge conditions are therefore @xmath183 & = - \partial^{\lambda}h_{\lambda\mu } + \frac{1}{2}\partial_{\mu}h - \left(\partial_{z } + 3a'\right)h_{\mu5 } = 0 , \label{eq : gravity24a}\\ f_{5}[h ] & = \frac{1}{2}\partial_{z}h - \partial^{\mu}h_{\mu 5 } - \frac{3}{2}\left(\partial_{z } + 2a'\right)\phi = 0 . \label{eq : gravity24b}\end{aligned}\ ] ] let us first examine how to implement these gauge - fixing conditions . to see this , it is sufficient to calculate @xmath184 $ ] and @xmath185 $ ] . using the transformation law of the fluctuations at kk - mode level we get @xmath186 & = \sum_{n=0}^{\infty } \left [ \left(\box - m_{n}^{2}\right)\xi_{\mu}^{(n ) } \right]f^{(n ) } , \label{eq : gravity25a}\\ \delta f_{5}[h ] & = \sum_{n=1}^{\infty } \left [ \left(\box - m_{n}^{2}\right)\xi_{5}^{(n ) } \right]g^{(n)}. \label{eq : gravity25b}\end{aligned}\ ] ] it is obvious that one can always impose the conditions and by choosing @xmath187 as one of solutions to the following differential equations for each @xmath4 : @xmath188 these are the analogue of eq .. notice that when one impose the gauge - fixing conditions and , however , there still remains gauge freedom generated by on - shell parameters @xmath189 and @xmath190 which satisfy @xmath191 and @xmath192 . this is also the analogue of residual on - shell gauge freedom in 4d general relativity . adding to the action a gauge - fixing term @xmath193\bigr)^{2 } - \frac{1}{2}\bigl(f_{5}[h]\bigr)^{2}\right\ } , \label{eq : gravity27}\ ] ] we get @xmath194 h^{\rho\sigma}. \nonumber\\ & \hspace{9em } + \frac{1}{2}h^{\mu5}\left [ \eta_{\mu\nu}\bigl(\box + \partial_{z}(\partial_{z } + 3a')\bigr)\right ] h^{\nu5 } + \frac{3}{8}\phi\left [ \box + ( \partial_{z } + a ' ) ( \partial_{z } + 2a')\right ] \phi\biggr\}. \label{eq : gravity28}\end{aligned}\ ] ] the dimensional reduced action is @xmath195 h^{\rho\sigma(n ) } \nonumber\\ & \hspace{5em } + \sum_{n=1}^{\infty } \frac{1}{2}h^{\mu5(n ) } [ \eta_{\mu\nu}(\box - m_{n}^{2 } ) ] h^{\nu5(n ) } + \sum_{n=0}^{\infty } \frac{3}{8}\phi^{(n ) } [ \box - m_{n}^{2 } ] \phi^{(n ) } \biggr\}. \label{eq : gravity29}\end{aligned}\ ] ] it should be emphasized the first line suggests that the residue of the graviton propagator is common to _ any _ @xmath196 so that one may expect in the t hooft - feynman gauge it may be possible to evade the vdvz discontinuity , which arises in the unitary gauge . for further discussion , see section [ sec : amplitude ] . * @xmath197 gauge * : so far we have used the notation @xmath198 and @xmath199 , which implies that , as in gauge theories , our t hooft - feynman gauge just corresponds to the case @xmath200 in somewhat underlying @xmath5 gauge , however , no mention has been made of how we have inserted the gauge parameter @xmath1 into the action . actually , it is highly non - trivial to insert @xmath1 into the gauge - fixing lagrangian . to see the difficulty , consider the arithmetic @xmath201 since we already know the t hooft - feynman gauge , which corresponds to @xmath202 , we now want to choose the coefficients @xmath203 and @xmath204 to make the three cross terms independent of @xmath1 : @xmath205 . but obviously it is not possible . at first glance , it seems to be impossible to eliminate the three undesirable mixing terms by a single gauge parameter @xmath1 . however , we can construct the one - parameter family of gauge choices analogous to the @xmath5 gauge in spontaneously broken gauge theories with the help of the boundary conditions . one of the crucial keys is that we have two squared brackets in , @xmath206\bigr)^{2}$ ] and @xmath207\bigr)^{2}$ ] , the former fixes the four - dimensional gauge arbitrariness @xmath173 and the latter the extra - dimensional one @xmath175 . since the scalar field @xmath208 only appears in the latter bracket , we have to insert @xmath1 as a coefficient of @xmath208 to cancel out by the overall factor @xmath209 . then the cross term between @xmath96 and @xmath210 in the latter bracket come to depend on @xmath1 . however , we can make all the cross terms independent of @xmath1 by inserting it into the former bracket as following combination : @xmath211\bigr)^{2 } - \frac{1}{2}\bigl(f^{r_{\xi}}_{5}[h]\bigr)^{2}\right\ } , \label{eq : gravity31}\end{aligned}\ ] ] with @xmath212 & = -\partial^{\lambda}h_{\lambda\mu } + \frac{1}{2}\left(2 - \frac{1}{\xi}\right)\partial_{\mu}h - \xi(\partial_{z } + 3a')h_{\mu5 } , \label{eq : gravity32a}\\ f^{r_{\xi}}_{5}[h ] & = \frac{1}{2}\partial_{z}h - \partial^{\mu}h_{\mu5 } - \frac{3\xi}{2}(\partial_{z } + 2a')\phi . \label{eq : gravity32b}\end{aligned}\ ] ] upon integrating by parts , one can easily check that the cross term between @xmath96 and @xmath210 is just the same as with the help of the boundary conditions . thus correctly eliminates the undesirable mixing terms . * gauge - fixed action to quadratic order * : adding to the action a gauge - fixing term we get @xmath213 where @xmath214 - \left(1 - \frac{1}{\xi}\right ) \partial_{\mu}\partial_{\rho } , \label{eq : gravity34b}\\ k_{\phi ; \phi}^{r_{\xi } } & = \box + ( 3\xi - 2)(\partial_{z } + a')(\partial_{z } + 2a ' ) . \label{eq : gravity34c}\end{aligned}\ ] ] now it is straightforward to compute the propagators in the @xmath5 gauge . the results are summarized in appendix [ appendix : propagator ] . as in a massive gauge boson propagator of spontaneously broken gauge theories , the massive graviton propagator can be written into the sum of the gauge - independent and -dependent part separately . the gauge - independent part corresponds to the transverse - traceless part of @xmath215 , as it should . the gauge - dependent part can further be decomposed into the sum of the three independent parts ; the transverse gauge freedom part @xmath216 , the longitudinal gauge freedom part @xmath217 and the trace part @xmath218 . @xmath216 has the same simple pole position to the transverse part of the vector propagator . @xmath217 and @xmath218 have the same simple pole positions to the longitudinal part of the vector propagator and the scalar propagator , respectively . in view of this fact we expect that as in the case of the pure abelian gauge theory , these spurious degrees of freedom have its partners with the same mass - squared , and consist of multiplets under susy - like transformations generated by the supercharges . to this end we decompose the graviton and vector fields into the following polarization states @xmath219 where @xmath220 is the transverse - traceless mode ( 5 dof ) , @xmath221 the transverse gauge freedom ( 3 dof ) , @xmath222 the longitudinal gauge freedom ( 1 dof ) , @xmath223 the `` trace '' piece ( 1 dof ) , @xmath221 the transverse mode ( 3 dof ) and @xmath224 the longitudinal mode ( 1 dof ) . notice that these decompositions are justified only for the case that @xmath215 and @xmath225 are both massive . indeed the massless tensor decomposition are more complicated ; see for example appendix of @xcite . however , since we are now interested in the gauge - dependent mass terms of the would - be ng bosons , we do not need the massless tensor and vector decompositions . inserting the decompositions , into we get @xmath226 { \hat h}^{({\rm tt})}_{\mu\nu } \nonumber\\ & \hspace{1em } + \frac{1}{2 } \raisebox{2.5ex}{$t$}\ ! \begin{bmatrix } { \hat h}^{({\rm t})\mu } \\ { \hat h}^{({\rm t})\mu5 } \end{bmatrix } \begin{bmatrix } -1 & 0 \\ 0 & 1 \end{bmatrix } \left(\box - \xi \begin{bmatrix } -(\partial_{z } + 3a')\partial_{z } & 0 \\ 0 & -\partial_{z}(\partial_{z } + 3a ' ) \end{bmatrix}\right ) \begin{bmatrix } { \hat h}^{({\rm t})}_{\mu } \\ { \hat h}^{({\rm t})}_{\mu5 } \end{bmatrix } \nonumber\\ & \hspace{1em } + \frac{1}{2 } \raisebox{2.5ex}{$t$}\ ! \begin{bmatrix } { \hat h}^{({\rm l } ) } \\ { \hat h}^{({\rm l})}_{5 } \end{bmatrix } \begin{bmatrix } 1 & 0 \\ 0 & -1 \end{bmatrix } \left(\box - \xi^{2 } \begin{bmatrix } -(\partial_{z } + 3a')\partial_{z } & 0 \\ 0 & -\partial_{z}(\partial_{z } + 3a ' ) \end{bmatrix}\right ) \begin{bmatrix } { \hat h}^{({\rm l } ) } \\ { \hat h}^{({\rm l})}_{5 } \end{bmatrix } \nonumber\\ & \hspace{1em } + \frac{1}{2 } \raisebox{2.5ex}{$t$}\ ! \begin{bmatrix } { \hat h}^{({\rm tr } ) } \\ { \hat \phi } \end{bmatrix } \begin{bmatrix } -1 & 0\\ 0 & 1 \end{bmatrix } \left(\box - ( 3\xi - 2 ) \begin{bmatrix } -(\partial_{z } + 3a')\partial_{z } & 0 \\ 0 & -(\partial_{z } + a')(\partial_{z } + 2a ' ) \end{bmatrix}\right ) \begin{bmatrix } { \hat h}^{({\rm tr } ) } \\ { \hat \phi } \end{bmatrix}\biggr\ } , \label{eq : gravity36}\end{aligned}\ ] ] where ( @xmath90 . ) ] @xmath227 this is our final expression for the quadratic action . we have separated the action to the gauge - independent and -dependent part , just as done in the pure abelian gauge theory . it is now obvious that is invariant under the following three independent global transformations @xmath228 as expected , the transformations and are generated by supercharge @xmath95 in . the transformation is not , however , expressed by a first - order differential operator . this is because is the symmetry transformation between ( the unphysical component of ) @xmath215 and @xmath208 , which should be generated by product of two supercharges . several comments are in order at this stage : \(1 ) it can be regarded that is the so - called higher - derivative supersymmetric transformation . indeed , if we introduce the following @xmath229 matrix operators @xmath230 it is easy to check that these operators satisfy the so - called higher - derivative supersymmetry algebra ( hsusy algebra ) : @xmath231 = [ { \hat h } , ( -1)^{\hat f } ] = 0 , \nonumber\\ & \{(-1)^{\hat f } , { \hat q}_{a}\ } = 0 , \quad a , b = 1,2 , \label{eq : gravity40}\end{aligned}\ ] ] where @xmath232 . this is an extended nonlinear superalgebra discussed in @xcite . as is different from the standard superalgebra , , the supercharges are now the second order differential operators , as they should from the mention above . the point is once again the refactorization of the hamiltonian for the vector - mode , @xmath144 , with the help of the background solution @xmath233 ( except on the boundaries ) . \(2 ) there is no residual symmetry generated by @xmath234 . one may wonder about the absence of the symmetry generated by @xmath234 . however , it is reasonable from the unitarity point of view . in ordinary spontaneously broken gauge theories , the @xmath1-dependent part of the gauge boson propagator must have as its partners the would - be ng bosons and the faddeev - popov ghosts with the same @xmath1-dependent pole so as to cancel out the @xmath1-dependence of any physical amplitudes , and then maintain the unitarity of the theory . in this case , since we do not need the faddeev - popov ghosts , the unphysical polarization states of the massive graviton field ( 5 dof ) must have as its partners the would - be ng vector ( 4 dof ) and scalar ( 1 dof ) bosons . thus , we do not have any degrees of freedom to relate by the symmetry transformation generated by @xmath234 . \(3 ) in special cases @xmath235 , the four polarization states in and are happened to be in degenerate with the same gauge - dependent mass - squared . hence the action comes to exhibit the following four additional global symmetries : @xmath236 it is interesting to note that the @xmath237-symmetry is generated by @xmath234 in . \(4 ) the residual symmetries , , and are spacetime symmetries between a spin-2 and a spin-1 particles , a spin-2 and a spin-1 particles and a spin-2 and a spin-0 particles , respectively . because of the same reason in the abelian gauge theory with extra compact dimensions in section [ sec : abelian ] , it does not contradict with the coleman - mandula theorem . in this section we will calculate the lowest tree - level amplitude for the process of a gravitational interaction between two ( conserved ) external energy - momentum tensors residing on the 3-branes by using the previously derived @xmath5 gauge , and show that the result is actually independent of the gauge parameter @xmath1 . the coupling of the fluctuations @xmath238 to the energy - momentum tensor @xmath239 is @xmath240 where @xmath241 or @xmath142 . the last equality follows from the canonically normalized fields @xmath242 and @xmath243 . the scattering amplitude for our process is given by ( see fig.[fig : amp ] ) @xmath244 @xmath245 note that , as shown in appendix [ appendix : propagator ] , the massive graviton propagator can be separated into the transverse - traceless part @xmath246 , the transverse gauge freedom part @xmath247 , the longitudinal gauge freedom part @xmath248 and the trace part @xmath249 . since @xmath247 and @xmath248 do not couple to the conserved energy momentum tensor , the cancellation of the @xmath1-dependence for @xmath158 must take place between the trace part of @xmath215 and the would - be ng boson @xmath208 . its cancellation certainly occurs in a bit surprising way : @xmath250 = 0 , \label{eq : amplitude03}\end{aligned}\ ] ] where the last equality follows from the identity and the boundary conditions . thus the resulting amplitude does not depend on a gauge choice thanks to the supersymmetries between the eigenfunctions . we finally arrive at the expression @xmath251 it is worth mentioning that in our normalization the zero - mode functions are @xmath252 so that if we set @xmath253 and @xmath254 , @xmath255 and @xmath256 give the same value and are equal to one . this means that for the process of a gravitational interaction between two energy - momentum tensors sitting on the uv and ir branes , the massless limit @xmath257 is continuous even in the lowest tree - level approximation . thus , in this special case it may be possible to evade the vdvz - discontinuity without invoking ( anti-)de sitter backgrounds @xcite nor nonperturbative effects @xcite . on the other hand , if we set @xmath258 or @xmath259 , the massless radion s contribution could not coincide with the massless graviton s one . in these cases the massless limit is not continuous and hence the vdvz - discontinuity appears in the amplitude . as already mentioned , it is known that there exists a discontinuity between the massless graviton and the massive graviton theory in the lowest tree - level approximation with minkowski background @xcite . as summarized in appendix [ appendix : vdvz ] , the discontinuity is due to the fact that the helicity 0 state dose not decouple to the trace of a conserved energy - momentum tensor even in the massless limit . in the @xmath5 gauge the fictitious scalar fields propagate on the branes , however , as mentioned before , we have the identity @xmath260 for @xmath158 on the branes , which is nothing but the consequence of the supersymmetries in the mass spectrum , such that the massive scalars contributions can be converted into the massive gravitons one . however , the zero modes @xmath261 and @xmath262 do not mutually relate so that the massless graviton s contribution differs from the massive graviton s one , and then there arises the discontinuity . in this paper we have studied higher - dimensional gauge / gravity theories with compact extra dimension(s ) . in section [ sec : abelian ] we have investigated the pure abelian gauge theory with extra compact @xmath9-dimensions in the @xmath5 gauge , and found that the gauge - fixed action possesses the residual global symmetries between unphysical component of the gauge fields and the would - be scalar ng bosons . the @xmath1-dependent mass - matrix of this multiplet is exactly the super - hamiltonian and the residual global symmetry is generated by the supercharge found in the analysis of the mass spectrum . in section [ sec : gravity ] we have extended the analysis to five - dimensional pure gravity to quadratic order as perturbation to the randall - sundrum background . we have established the one - parameter family of gauge choices for 5d gravity analogous to the @xmath5 gauge in spontaneously broken gauge theories . in this @xmath5 gauge we have shown that the gauge - fixed action exhibits the three independent residual global symmetries . as in the pure abelian gauge theory with extra dimensions , it is the symmetries between unphysical component of the graviton field and the would - be vector and scalar ng bosons . however , as opposed to the pure abelian gauge theories with extra dimensions , one of these residual symmetry transformations can not be expressed in terms of the first - order differential supercharges but the second - order differential supercharge . in section [ sec : amplitude ] , to see the validity of our @xmath5 gauge we have computed the one graviton exchange amplitude in the lowest tree - level approximation , and confirmed that it is indeed @xmath1-independent . using this result we also found that for the process of a gravitational interaction between two conserved energy - momentum tensors residing on the uv and ir branes , the famous vdvz - discontinuity does not arise even in the lowest tree - level approximation . c.s.l . and m.s . are supported in part by the grant - in - aid for scientific research ( no.18204024 and no.18540275 ) by the japanese ministry of education , science , sports and culture . the authors would like to thank k.ghoroku , n.maru , k.sato , h.sonoda , m.tachibana and k.takenaga for valuable discussions . in the massive 4d graviton theory with the fierz - pauli mass term , the extra scalar degree of freedom corresponding to the helicity 0 state does not decouple to a trace of a stress - energy tensor even in the limit @xmath263 . this is the famous van dam - veltman - zakharov ( vdvz ) no - go theorem , that is , the helicity 0 degree of freedom causes the discontinuity between the massless graviton theory and the massless limit of the massive graviton theory . in this appendix we briefly review a massless and massive graviton theory in four - dimensional minkowski background and discuss how the extra helicity 0 degree of freedom comes in to the propagator and couples to the conserved energy - momentum tensor . we start out with the action @xmath264 where @xmath265 @xmath102 is the fundamental mass scale which is determined in terms of the newton constant @xmath169 later . our flat metric @xmath18 has the signature @xmath266 . the action is invariant under the linearized general coordinate transformations : @xmath267 where @xmath173 are four arbitrary functions . the equation of motion ( eom ) is @xmath268 where @xmath269 . we apply @xmath270 to eq . with the result @xmath271 eq . then reduces to @xmath272 to see the physical degrees of freedom , we impose the harmonic gauge conditions : @xmath273 notice that eq . is the linearized version of @xmath274 . this is one of the crucial keys to construct a gauge - fixing term for 5d gravity ; see section [ subsec : gravity_gauge - fix ] . the conditions are implemented by choosing the gauge parameters @xmath173 as solutions to the following differential equations : @xmath275 eom now becomes @xmath276 note that there still remains gauge freedom generated by @xmath173 which satisfy @xmath277 . if we decompose on - shell parameters @xmath173 into transverse part ( 3 dof ) and longitudinal part ( 1 dof ) , @xmath278 , the combination @xmath279 becomes @xmath280 where @xmath281 . when we choose @xmath282 as a solution to the equation @xmath283 , the fluctuations become traceless and then , because of the harmonic gauge conditions , automatically satisfy transverse conditions @xmath284 . now that @xmath285 is transverse - traceless and has five degrees of freedom , however , 3 out of 5 dof can be eliminated by @xmath286 . thus we see that @xmath215 has two physical degrees of freedom . this is the helicity degrees of freedom for massless graviton . next gauge - fix the action . by adding a gauge - fixing term @xmath287 , we can eliminate the first two lines in eq . so that the action becomes @xmath288h^{\rho\sigma } \nonumber\\ & = \int{\rm d}^{4}x \frac{1}{2}{\hat h}^{\mu\nu } \left[\frac{1}{2 } \left(\eta_{\mu\rho}\eta_{\nu\sigma } + \eta_{\mu\sigma}\eta_{\nu\rho } - \eta_{\mu\nu}\eta_{\rho\sigma}\right ) \box\right]{\hat h}^{\rho\sigma}\label{eq : massless11}\end{aligned}\ ] ] where we have rescaled the field @xmath289 to be the canonical action . noting that the symmetric unit bitensor is expressed as @xmath290 , it is easy to check that the inverse of a bitensor @xmath291 is actually itself . thus , in the harmonic gauge , the free propagator for massless graviton is given by @xmath292 next determine the mass scale @xmath102 to give the newton constant @xmath169 by computing the static gravitational potential between two masses , @xmath293 and @xmath294 . the corresponding amplitude is @xmath295 which gives the static potential @xmath296 so that it should be chosen that @xmath297 . next examine the tensor structure of the residue of the propagator . in @xmath298 frame , we have two independent polarization tensors @xmath299 where we have normalized them with respect to the inner product @xmath300 as explicitly constructed in the next subsection , is given by the tensor product of the polarization vectors corresponding to the helicity @xmath301 states as in a manner similar to the addition of angular momentum in quantum mechanics . one can verify the following relation : @xmath302 indeed , @xmath303 while @xmath304 , @xmath305 while @xmath306 , and so on . from this , we see that the residue of the propagator has the correct polarization structure . next consider the massive graviton theory by adding a fierz - pauli @xcite mass term @xmath307 : @xmath308 where @xmath309 @xmath310 is the corresponding mass scale for the massive graviton theory . the resulting eom is @xmath311 by contracting and taking the trace of eq . we have @xmath312 since @xmath313 , these equations can farther be reduced to the following constraints : @xmath314 which provide @xmath315 conditions , so that the eom now becomes @xmath316 now it is obvious that 5 out of 10 dof are eliminated by the traceless - transverse conditions so that @xmath317 has only @xmath318 dof , corresponding to the helicity degrees of freedom of a massive spin-2 particle . this is why the fierz - pauli mass term @xmath319 gives the graviton a mass in a natural way . the massive graviton propagator @xmath320 is defined as the green function inverse to the differential operator @xmath321 : @xmath322 after tedious calculations we get @xmath323 , \label{eq : massive08}\end{aligned}\ ] ] in the momentum space . let us compute the static gravitational potential via exchange of a massive graviton between two masses , @xmath293 and @xmath294 . the corresponding amplitude is @xmath324 which gives the static potential @xmath325 . if we require @xmath326 with an extremely tiny but non - vanishing mass @xmath327 gives the same result as that of the exchange of a massless graviton , the newton constant @xmath169 should be related as @xmath328 , where @xmath329 is the coupling constant for the massive graviton theory . this result that the coupling constant for the massive graviton theory @xmath330 is 25% smaller than that of the massless graviton theory @xmath169 yields the different prediction for the bending of light ( photon , whose energy - momentum tensor is traceless ) passing near the sun , whose experimental test ( measurement of deflection angle ) has been performed with the accuracy @xmath3311% to the prediction for the standard general relativity , while the massive graviton theory s prediction is about 25% smaller than that of the massless graviton theory , which contradicts with the measurement . this difference is caused by the helicity 0 degree of freedom which couples to the trace of an energy - momentum tensor even in the massless limit . to see this let us finally examine the polarization structure of the residue of the massive graviton propagator . to this end , we move on to the rest frame for the massive graviton : @xmath332 . in this frame with the cartesian coordinate basis , the five independent polarization tensors are given by @xmath333 where @xmath334 and @xmath335 are the polarization vectors for a massive vector boson with mass @xmath327 : @xmath336 and @xmath337 denotes the tensor product defined as @xmath338 one can easily verify that @xmath339 indeed , @xmath340 while @xmath341 , @xmath342 while @xmath343 , and so on . it should be noted that eq . is different from eq . even if @xmath344 . this `` 2/3-discontinuity '' is due to the existence of the helicity 0 state @xmath345 , which was first pointed out in 1970 independently by iwasaki @xcite , and by van dam and veltman @xcite , and by zakharov @xcite . here we give the expressions for the propagators in the @xmath5 gauge . * vector propagator * : @xmath347 \nonumber\\ & = { \tilde \delta}^{({\rm t})}_{\{\mu , n\ } ; \{\nu , m\}}(k ) + { \tilde \delta}^{({\rm l})}_{\{\mu , n\ } ; \{\nu , m\}}(k ) , \label{eq : propagator02}\end{aligned}\ ] ] where @xmath348 , \label{eq : propagator03a}\\ { \tilde \delta}^{({\rm l})}_{\{\mu , n\ } ; \{\nu , m\}}(k ) & = \frac{i\xi\delta_{nm}}{-k^{2 } - \xi^{2}m_{n}^{2 } } \left[- \frac{k_{\mu}k_{\nu}}{\xi^{2 } m_{n}^{2}}\right ] . \label{eq : propagator03b}\end{aligned}\ ] ] * graviton propagator * : @xmath349 \nonumber\\ & = { \tilde \delta}^{({\rm tt})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) + { \tilde \delta}^{({\rm t})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) + { \tilde \delta}^{({\rm l})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) + { \tilde \delta}^{({\rm tr})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) , \label{eq : propagator04}\end{aligned}\ ] ] where @xmath350 , \label{eq : propagator05a}\\ { \tilde \delta}^{({\rm t})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) & = \frac{(i\xi/2)\delta_{nm}}{-k^{2 } - \xi m_{n}^{2 } } \biggl [ \left(\eta_{\mu\rho } + \frac{k_{\mu}k_{\rho}}{\xi m_{n}^{2}}\right ) \left(- \frac{k_{\nu}k_{\sigma}}{\xi m_{n}^{2}}\right ) + \left(\eta_{\mu\sigma } + \frac{k_{\mu}k_{\sigma}}{\xi m_{n}^{2}}\right ) \left(- \frac{k_{\nu}k_{\rho}}{\xi m_{n}^{2}}\right ) \nonumber\\ & \hspace{6em } + \left(\eta_{\nu\rho } + \frac{k_{\nu}k_{\rho}}{\xi m_{n}^{2}}\right ) \left(- \frac{k_{\mu}k_{\sigma}}{\xi m_{n}^{2}}\right ) + \left(\eta_{\nu\sigma } + \frac{k_{\nu}k_{\sigma}}{\xi m_{n}^{2}}\right ) \left(- \frac{k_{\mu}k_{\rho}}{\xi m_{n}^{2}}\right ) \biggr ] , \label{eq : propagator05b}\\ { \tilde \delta}^{({\rm l})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) & = \frac{(i\xi^{3}/2)\delta_{nm}}{-k^{2 } - \xi^{2}m_{n}^{2 } } \left [ 4\frac{k_{\mu}k_{\nu}k_{\rho}k_{\sigma}}{\xi^{4}m_{n}^{4 } } \right ] , \label{eq : propagator05c}\\ { \tilde \delta}^{({\rm tr})}_{\{\mu\nu , n\ } ; \{\rho\sigma , m\}}(k ) & = \frac{(i/6)\delta_{nm}}{-k^{2 } - ( 3\xi - 2)m_{n}^{2 } } \left[- \left(\eta_{\mu\nu } - 2\frac{k_{\mu}k_{\nu}}{m_{n}^{2}}\right ) \left(\eta_{\rho\sigma } - 2\frac{k_{\rho}k_{\sigma}}{m_{n}^{2}}\right)\right ] . \label{eq : propagator05d}\end{aligned}\ ] ] 1 . these propagators are calculated with the canonically normalized fields , @xmath351 , @xmath352 and @xmath353 . 2 . by taking the limit @xmath354 these propagators just reduces to those calculated in the unitary gauge . the unitary gauge limit @xmath354 and the massless limit @xmath257 do not commute with each other , just as in ordinary spontaneously broken gauge theories . @xmath355 is nothing but the propagator for the 4d massive graviton with the fierz - pauli mass term . 30 natexlab#1#1bibnamefont # 1#1bibfnamefont # 1#1citenamefont # 1#1url # 1`#1`urlprefix[2]#2 [ 2][]#2 , * * , ( ) . , * * , ( ) . , * * , ( ) . , * * , ( ) . , * * , ( ) . , , , , * * , ( ) , . , , , * * , ( ) , . , , , , , . , * * , ( ) . , , , * * , ( ) , . , , , , * * , ( ) , . , * * , ( ) . , * * , ( ) , . , * * , ( ) . , * * , ( ) , . , , , * * , ( ) , . , , , * * , ( ) , . , * * , ( ) . , , , * * , ( ) , . , , , , * * , ( ) , . , , , * * , ( ) . , * * , ( ) , . , , , * * , ( ) , . , , , * * , ( ) , . , , , * * , ( ) , . , * * , ( ) , . , * * , ( ) , . , , , * * , ( ) , . , , , * * , ( ) , . , * * , ( ) , . , * * , ( ) . , , , , * * , ( ) , . , * * , ( ) . , * * , ( ) . , * * , ( ) .

we consider a model of galaxy clusters in which the hot gas is in hydrostatic equilibrium and maintains energy balance between radiative cooling and heating by thermal conduction . we analyze the thermal stability of the gas using a lagrangian perturbation analysis . for thermal conductivity at the level of% of spitzer conductivity , consistent with previous estimates for cluster gas , we find that the growth rate of the most unstable global radial mode is times lower than the growth rate of local isobaric modes at the cluster center in the absence of conduction . the growth time in typical clusters is gyr , which is comparable to the time since the last major merger episode , when the gas was presumably well mixed . thus , we suggest that thermal instability is not dynamically significant in clusters , provided there is an adequate level of thermal conduction . on the other hand , if the heating of the gas is not the result of thermal conduction or any other diffusive process such as turbulent mixing , then the thermal instability has a growth time under a gyr in the central regions of the cluster and is a serious threat to equilibrium . we also analyze local nonradial modes and show that the lagrangian technique leads to the same dispersion relation as the eulerian approach , provided that clusters are initially in strict thermal equilibrium . because cluster gas is convectively stable , nonradial modes always have a smaller growth rate than equivalent radial modes .

astro-ph0303097

galaxy clusters contain a large amount of hot diffuse gas radiating prolifically in thermal x - rays . in the absence of heat sources , the radiative cooling due to this emission should induce a subsonic inflow of gas in the central regions , leading to substantial condensation of cold gas in rich clusters ( e.g. , @xcite and references therein ) . however , recent high - resolution x - ray data from _ xmm - newton _ and _ chandra _ reveal no evidence for such cooling flows , nor any significant evidence for mass dropout . in particular , there is no evidence for gas at temperatures below about a third of the average temperature @xcite , suggesting that there must be some heat source ( or sources ) that prevents the gas from cooling below this limit . candidate heating mechanisms include radiative and mechanical power from active galactic nuclei ( agn ) @xcite , thermal conduction from the hotter outer regions of the cluster to the center ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * hereafter zn03 ) , or perhaps both ( e.g. , zn03 , @xcite ) . the effects of conduction in clusters have been widely discussed by many authors ( e.g. , @xcite ; see @xcite and references therein ) . it was shown that thermal conduction , when unimpeded , can significantly reduce inferred mass deposition rates in cooling flow clusters @xcite , although such systems may end up being almost isothermal @xcite . nevertheless , thermal conduction was considered to be unimportant in galaxy clusters as it was believed that , in the presence of magnetic fields , the cross - field diffusion coefficient would be negligibly small . since magnetic fields are ubiquitous in clusters ( e.g. , @xcite ) , it appeared that the effective isotropic conduction coefficient @xmath3 in the presence of magnetic fields would be smaller by orders of magnitude than the classical @xcite value , @xmath4 , of an unmagnetized plasma . @xcite recently showed that cross - field diffusion of electrons is quite efficient if magnetic fields are fully turbulent and have a wide range of coherence length scales . extending earlier work by @xcite , @xcite , @xcite , and @xcite , and adopting the model of @xcite for mhd turbulence , @xcite estimated that @xmath5 in a turbulent magnetized plasma . very recently , @xcite have used direct numerical simulations to show that the turbulent diffusion of a scalar field in mhd turbulence is as efficient as , or perhaps even more efficient than , the prediction of @xcite . these studies have led to the revival of the idea that conduction might be an important heating source in clusters . the viability of thermal conduction as a heating mechanism has found support in the work of @xcite , @xcite , and zn03 , who showed that the level of conductivity needed to fit the observed temperature distributions of x - ray clusters is consistent with the theoretical estimate of @xcite . in particular , zn03 explicitly solved the equations for hydrostatic equilibrium and energy balance between radiative cooling and conductive heating and showed that conduction with @xmath6 provides reasonable profiles of density and temperature for several clusters . a similar result was also obtained by @xcite using numerical simulations . although other heating sources , including agn jets , can not be ruled out , these studies suggest that an energy balance based purely on conduction is not unreasonable . while the above studies show that it is possible to construct equilibrium cluster models with conductive and/or agn heating , there is no guarantee that the equilibrium will be stable . this is because hot , optically - thin , x - ray - emitting gas is well - known to be thermally unstable @xcite . since the growth time of the thermal instability is comparable to the cooling time ( e.g. , @xcite ) , one might expect rapid mass dropout as a result of the instability , even when there is a source of heat to eliminate the classic cooling flow . on the other hand , the absence of any evidence for gas below a few kev in clusters implies that the instability is either absent or very slow . a likely reason for the lack of instability is thermal conduction , which is known to suppress thermal instability on small scales ( see , e.g. , @xcite ; zn03 ) . however , all previous analyses of this process have been limited to local wkb - type perturbations , where the wavelength of the perturbations is much smaller than the local radius of the system . zn03 found from their wkb analysis the intriguing result that perturbations with wavelengths up to almost the radius are stable , but that longer wavelengths are probably unstable . such large - scale variations can be analyzed only through a full global mode analysis , which has not been done so far . in this paper we present a detailed and formal analysis of local and global modes of thermal instability in equilibrium galaxy clusters with conduction . in 2 , we describe the basic equations we solve and present equilibrium solutions for our fiducial model cluster . in 3 , we carry out a linear stability analysis of the assumed equilibrium and calculate the global radial modes of the system . we show that there is a single unstable mode , whose growth rate is much lower than the usual thermal instability growth rate in the absence of conduction . in 4 , we confirm the main results by means of numerical simulations which show the development of the thermal instability under various assumptions . in 5 , we consider the stability of nonradial modes and clarify a few points on which there has been confusion in the literature . we conclude in 6 with a brief summary of the results and a discussion of the implications . we consider the thermodynamic evolution of a hot plasma subject to radiative cooling and thermal conduction . we do not consider any dynamical effects of magnetic fields . we also neglect the self - gravity of the gas which is generally weaker than the gravity of the dark matter . % from the non - self - gravitating values . ] the governing hydrodynamic equations are @xmath7 @xmath8 @xmath9 here , @xmath10 is the lagrangian time derivative , @xmath11 is the mass density , @xmath12 is the velocity , @xmath13 is the temperature , @xmath14 is the gravitational potential , @xmath15 is the adiabatic index of the gas , @xmath16 is the energy loss rate per unit volume due to thermal bremsstrahlung ( @xcite ; zn03 ) , and @xmath17 is the thermal pressure for which we adopt the equation of state of an ideal gas , @xmath18 where @xmath19 is the atomic mass unit , @xmath20 is the electron number density , and @xmath21 and @xmath22 denote the mean molecular weight per hydrogen atom and per electron , respectively . we use @xmath23 and @xmath24 , corresponding to a fully ionized gas with hydrogen fraction @xmath25 and helium fraction @xmath26 ( zn03 ) . in equation ( [ eq_engy ] ) , the conductive heat flux @xmath27 is determined by @xmath28 where the conductivity @xmath3 is a fraction @xmath29 of the classical @xcite conductivity @xmath4 , @xmath30 with the coulomb logarithm @xmath31 . we do not consider nonlocal heat transport as in the model of @xcite . in an unmagnetized plasma , conduction is isotropic and @xmath29 is unity . the presence of a magnetic field generally reduces @xmath29 by resisting motions of thermal elections across the field lines . although the suppression of @xmath3 would be very high if magnetic fields are uniform or only moderately tangled ( e.g. , @xcite ) , @xcite showed that @xmath32 in a fully turbulent plasma medium in which magnetic fields are chaotic over a wide range of length scales . in this paper , we assume that @xmath29 is constant in both space and time . we use the nfw form of the dark matter distribution @xmath33 to determine the gravitational potential @xmath14 through @xmath34 with a softened core radius @xmath35 , a scale radius @xmath36 , and a characteristic mass @xmath37 . utilizing the mass - temperature relation of @xcite and the mass - scale relation of @xcite , we determine @xmath37 and @xmath36 for a given cluster from the observed temperature in the outer regions of the cluster ( see zn03 ) . for @xmath35 , which determines the shape of the potential in the very inner parts , we adopt the best - fit values ( either @xmath38 or @xmath39/20 ) recommended by zn03 . without any heating source to compensate for x - ray cooling , a cluster would lose its thermal energy at a rate @xmath40 . assuming that cooling occurs at fixed pressure , we may define from equation ( [ eq_engy ] ) the isobaric cooling time times longer than the conventional definition , @xmath41 , with @xmath42 ( cf . @xcite ) . at constant pressure , equation ( [ eq_engy ] ) gives @xmath43 . ] @xmath44 similarly , we define the conduction time as @xmath45 where @xmath46 is the length scale over which the temperature changes appreciably . by comparing equations ( [ eq_tcool ] ) and ( [ eq_tcond ] ) , we see that conductive heating becomes comparable to radiative cooling at @xmath47 where @xmath48 is the field length @xcite . this corresponds to the length scale below which thermal conduction erases temperature perturbations completely @xcite . since the length scale for temperature variations is typically @xmath49 kpc , we conclude that thermal conduction with @xmath50 can potentially support galaxy clusters against radiative cooling ( see @xcite ; zn03 ) . it is straightforward to find initial equilibrium solutions of equations ( [ eq_cont])-([eq_poss ] ) . we assume that the equilibrium is spherically symmetric , static , and time - independent . working in spherical coordinates ( @xmath51 , @xmath52 , @xmath53 ) , we simplify equations ( [ eq_mome ] ) , ( [ eq_engy ] ) , and ( [ eq_flux ] ) to @xmath54 @xmath55 @xmath56 where @xmath57 denotes the radial heat flux . full numerical solutions of the differential equations ( [ eq_poss ] ) and ( [ eq_hse])-([eq_fbal ] ) were presented by zn03 . they varied @xmath29 to find the solutions that give the best fits to the observed density and temperature profiles of ten clusters . they found that thermal conduction with @xmath58 explains the density and temperature distributions of five clusters fairly well ( a1795 , a1835 , a2199 , a2390 , rxj1347.5 - 1145 ) . however , five other clusters require unphysically large values of @xmath29 , indicating that those clusters are incompatible with a pure conduction model and require other heat sources . in this paper we take the best - fit parameters obtained by zn03 for the five clusters that are consistent with conduction , and we analyze their stability to global radial modes . table [ table1 ] lists the model parameters and various time scales . we adopt the cluster a1795 as our fiducial model . figure [ back_a1795 ] illustrates an equilibrium solution for a1795 . the temperature is minimum at @xmath59 and begins to rise sharply at @xmath60 kpc . expansion of the variables near the center gives @xmath61 , where the subscript `` 0 '' indicates the values at the center . therefore , stronger cooling or smaller conductivity would cause the temperature to increase faster . notice that the temperature is always an increasing function of the radius . radial heat influx is largest at @xmath62 kpc beyond which the low density makes cooling as well as conductive heating unimportant . figure [ back_a1795]@xmath63 plots the local isobaric cooling time @xmath64 defined in equation ( [ eq_tcool ] ) ( solid line ) and the growth time , @xmath65 ( see eq . [ [ eq_tinf ] ] ) , of local isobaric thermal perturbations without conduction ( dashed line ) , which are compared with the growth time of the global radial mode in the presence of conduction ( @xmath66 ; dotted line ) : we discuss these instability time scales in the next section . the radial equilibrium profiles of a1795 presented in figure [ back_a1795 ] are close to but not exactly the same as those derived by @xcite . since a1795 is known to contain a cool filament near the center @xcite , the real density and temperature distributions are somewhat non - axisymmetric . our axisymmetric equilibrium model of a1795 should , therefore , be regarded as an idealized version of what is a much more complex distribution of the intracluster medium in a1795 . as we will show below , however , the lowest - order radial mode of thermal instability is the fastest growing mode , so that neglecting the non - axisymmetric parts in the background profiles probably does not affect the results significantly . we linearize equations ( [ eq_cont])-([eq_flux ] ) in the lagrangian framework . a lagrangian perturbation , represented by an operator @xmath67 , is related to an eulerian perturbation @xmath68 in the usual way , @xmath69 where the vector @xmath70 measures the lagrangian displacement of a fluid element from its unperturbed location ( e.g. , @xcite ) . one of the advantages of adopting the lagrangian approach is that it simplifies the perturbed equations greatly . for instance , the perturbed continuity equation ( [ eq_cont ] ) becomes @xmath71 nevertheless , both lagrangian and eulerian descriptions should give the same results , especially if the initial state is static and in complete equilibrium . various properties of @xmath67 and the commutation relations associated with @xmath67 and @xmath68 can be found in @xcite . we assume that the perturbations are all radial . applying @xmath67 to equations ( [ eq_mome])-([eq_flux ] ) and using equation ( [ eq_lcont ] ) , we obtain @xmath72 @xmath73 @xmath74 where @xmath75 , @xmath76 , @xmath77 is the radial component of @xmath70 , and @xmath78 is the radial heat luminosity . we use @xmath77 , @xmath79 , and @xmath80 as independent variables . we seek linear eigenmodes that behave as @xmath81 with time . we then rewrite equations ( [ eq_lmome])-([eq_flux_r ] ) as @xmath82 @xmath83,\ ] ] @xmath84 + \frac{{\delta l_r}}{4\pi r^2},\ ] ] which are our desired perturbation equations . this set of ordinary differential equations forms an eigenvalue problem for global modes , with @xmath85 as the eigenvalue , which can be solved numerically subject to appropriate boundary conditions . before considering the full problem , we study local modes that do not depend on boundaries to gain some physical insight . let us consider local wkb perturbations of the form @xmath86 and let us assume that the radial wavenumber @xmath87 satisfies @xmath88 , @xmath89 . we may then neglect the spherical geometry and ignore local gradients of the background quantities relative to spatial gradients of the perturbed variables . we also assume that @xmath90 , corresponding to slowly evolving perturbations ; this eliminates sound waves from consideration . in this local approximation , equation ( [ eq_f1 ] ) simplifies to @xmath91 while equations ( [ eq_f2 ] ) and ( [ eq_f3 ] ) may be combined to give @xmath92 eliminating @xmath77 and @xmath93 from equations ( [ eq_loc1 ] ) and ( [ eq_loc2 ] ) , we find @xmath94 where @xmath95 , defined by @xmath96 is the growth rate of isobaric thermal perturbations without conduction @xcite . equation ( [ ldisp ] ) is a local dispersion relation for thermal fluctuations . ) can be obtained by taking the limit of @xmath97 in the dispersion relation ( 15 ) of @xcite for a uniform medium . note , however , that the condition @xmath98 in his initial equilibrium state gives @xmath99(\partial{\mathcal{l}}/\partial t)_p$ ] instead of equation ( [ eq_siginf ] ) . ] in the absence of conduction , @xmath100 , so that instability develops if the cooling function satisfies the generalized field criterion @xmath101 for isobaric thermal instability @xcite . for x - ray emitting clusters of galaxies with @xmath102 , the condition ( [ eq_cri ] ) is easily met . the corresponding growth time amounts to @xmath103 suggesting that local radial disturbances will grow slightly faster than the isobaric cooling time of the system . figure [ back_a1795]@xmath63 plots @xmath65 for a1795 as a dashed line . it is well known that thermal conduction stabilizes short - wavelength perturbations against thermal instability ( e.g. , @xcite ) . it not only reduces growth rates but also suppresses thermal instability completely if @xmath104 , where @xmath105 . the existence of unstable modes , therefore , requires large wavelength perturbations . zn03 showed that the clusters they studied are thermally stable for all wavelengths up to approximately the radius . the fate of modes with longer wavelengths was unclear in their study since a local analysis is no longer valid . a global analysis is required to study such large scale perturbations . this is the topic of the next subsection . to analyze the global stability problem , we solve equations ( [ eq_f1])-([eq_f3 ] ) numerically . since these equations are equivalent to four first - order differential equations , we need to specify four boundary conditions . the two inner boundary conditions are @xmath106 the first condition guarantees that the solutions are regular , while the second condition corresponds to a zero gradient in the perturbed temperature at the center . since @xmath107 at @xmath59 for radial modes , we fix @xmath108 as a normalization condition . the two other boundary conditions come from the outer boundary which we arbitrarily locate at @xmath109 mpc . since all eigenfunctions that we have obtained are found to decay rapidly with increasing radius , the solutions are quite insensitive to the particular choice of the outer boundary conditions as well as the location of the outer boundary . this is because the cooling time well exceeds the hubble time in the outer parts of clusters ( fig . [ back_a1795]@xmath63 ) , and so the thermal instability develops very slowly near the outer boundary . the results presented in this paper correspond to the specific boundary conditions @xmath110 we have tried various other outer boundary conditions such as a fixed pressure , a fixed entropy , etc . the results are unchanged . the numerical calculations proceed as follows . we first fix @xmath85 and integrate equations ( [ eq_f1])-([eq_f3 ] ) from @xmath59 to @xmath111 , setting @xmath112 to an arbitrary value . at the outer boundary , we check the first condition in equation ( [ outerb ] ) , update @xmath113 using the newton - raphson method , and continue iterating until convergence is attained . then , we scan @xmath85 in the range @xmath114 , where @xmath115 is the adiabatic sound speed at the cluster center , and use the second condition in equation ( [ outerb ] ) as a discriminant for solutions . by this procedure , we do not miss any eigensolution . we allow for both real and complex @xmath85 . when @xmath85 is real and positive , we find that a cluster with @xmath116 has only one unstable global mode . in figure [ eig_all ] , we plot the eigenfunctions of the unstable mode for all five clusters listed in table [ table1 ] . the perturbations have largest amplitude near the center ( @xmath117 kpc ) and decay rapidly with increasing radius . this is consistent with the notion that the central regions have a much shorter cooling time and are thus more prone to thermal instability . note that @xmath118 has the same sign as @xmath93 , indicating that the thermal instability leads to a mass inflow . we will show in 4 that the mass inflow rate driven by the instability is , however , much lower than that in conventional cooling flows . as the last two columns of table [ table1 ] shows , the growth time of the most unstable global radial modes is typically @xmath119 gyr , which is about 6 to 9 times longer than that of local isobaric modes at the cluster center in the absence of conduction . this time scale is not much shorter than the age of the system since the last major merger . therefore , thermal instability in the presence of conduction is unlikely to be a serious threat to clusters . why is the growth time @xmath66 of the mode so much longer than the growth time @xmath120 of the local isobaric mode at the cluster center ? the reason is that the eigenfunction extends over a considerable range of @xmath51 , of order tens of kpc ( fig . [ eig_all ] ) . over this range , @xmath65 increases from @xmath120 at the center to much larger values ( fig . [ back_a1795]@xmath63 ) . since the growth time of the mode corresponds to an appropriate eigenfunction - weighted volume - average of @xmath65 , and since the averaging is dominated by @xmath121 tens of kpc , this causes @xmath66 to be significantly larger than @xmath120 . how effectively does conduction stabilize thermal instability ? to address this question , we have solved equations ( [ eq_f1])-([eq_f3 ] ) using different values of @xmath29 for the perturbations . we adopt the same density and temperature distributions as shown in figure [ back_a1795 ] , which means that we take the same @xmath122 for the initial unperturbed state . however , we allow @xmath29 to vary in the perturbations . in figure [ f_a1795_pos ] , we show the resulting mode growth rates @xmath85 relative to @xmath123 , the central value of the growth rate of local thermal instability without conduction ( see eq . [ [ eq_siginf ] ] ) . when @xmath124 ( although unphysical ) , we find that conduction quenches thermal instability completely ; that is , there is no unstable mode . for @xmath125 , there exists a single unstable mode that has no node in the eigenfunction for @xmath126 . the case shown in figure [ eig_all ] , @xmath122 , corresponds to this range . when @xmath29 is below 0.14 , a new mode appears and now the system has two unstable modes . because the new mode has a node in @xmath118 somewhere inside the cluster , its growth rate is smaller than the zero - node mode , as equation ( [ ldisp ] ) suggests . as @xmath29 keeps decreasing , new modes having lower growth rates successively emerge . in the limit of zero conduction , there is an infinite number of unstable modes , with the maximum growth rate equal to @xmath123 . figure [ f_a1795_pos ] shows that over much of the range of @xmath29 , the growth rate increases very slowly as @xmath29 decreases . as before , the reason may be traced to the fact that the eigenfunction extends over a range of @xmath51 and @xmath85 represents a suitable average of @xmath95 over this volume . thus , even for @xmath127 , the fundamental @xmath128 mode has a growth rate about 4 times smaller than @xmath123 . eigenfunctions become more centrally concentrated as @xmath29 decreases , and thus a mode with smaller @xmath29 generally has a larger growth rate , though the increase of @xmath85 with decreasing @xmath29 is rather slow ( fig . [ f_a1795_pos ] ) . in addition to unstable modes for which thermal conduction plays a stabilizing role , clusters also contain decaying modes with real and negative @xmath85 . figure [ f_a1795_neg ] displays examples of eigenfunctions for a few selected decaying modes and shows the dependence of @xmath85 on the reduction factor @xmath29 . in contrast to unstable modes , which have smaller frequencies as the number of nodes increases , higher - order decaying modes have larger ( negative ) frequencies . this is consistent with equation ( [ ldisp ] ) for decaying modes , equation ( [ ldisp ] ) , which assumes @xmath129 , does not describe higher - order modes very accurately . ] , which implies that the presence of decaying modes is a simple manifestation of diffusion ironing out temperature perturbations . we have searched for global solutions with complex @xmath85 , i.e. , overstable modes , but found no overstable radial mode with re(@xmath130 under the imposed boundary conditions . physically , thermal overstability occurs when sound waves are amplified by absorbing ( losing ) heat during the compression ( rarefaction ) phase of oscillations @xcite . since the cooling function for x - ray emitting clusters does not satisfy the isentropic instability criterion of @xcite , the absence of radial overstability is in fact guaranteed this leaves pure unstable modes as the sole growing modes . although some underdamping modes ( complex @xmath85 with negative real parts ) were found , they are of no interest and we do not discuss them further . as a check of the global linear stability analyses described in the previous section , we have solved the full dynamical equations ( [ eq_cont])-([eq_poss ] ) using a time - dependent approach . starting with the equilibrium density and temperature profiles of a1795 calculated in 2.2 , we have run three models : ( model a ) a cluster without any heating ; ( model b ) a cluster with no conduction , but constantly heated ( by some fictitious agency ) at a rate such as to maintain thermal balance in the initial equilibrium ; ( model c ) a cluster with self - consistent conductive heating with @xmath122 . model a simulates the standard cluster cooling flow problem , whereas models b and c are intended to explore the growth of thermal instability from an initial equilibrium state , with and without thermal conduction . note that our adopted cooling function for x - ray free - free emission becomes invalid when the gas in models a - c cools below @xmath131 kev . nevertheless , these simulations allow us to confirm the growth rates of thermal instability that we computed in 3 and to estimate mass inflow rates resulting from either radiative cooling or thermal instability . we follow the nonlinear evolution of our model clusters using the zeus hydrodynamic code @xcite . we construct a logarithmically - spaced radial grid , with 500 zones , from @xmath132 kpc to 1 mpc , and carry out a one - dimensional simulation . we implement a fully implicit algorithm for thermal conduction in the energy equation @xcite . all the variables at both the inner and outer boundaries are assigned to have a vanishing gradient across the boundaries , except for the radial velocity . we fix the radial velocity to be zero at the outer boundary , while allowing it to vary as a linear function of radius at the inner boundary . on the initial equilibrium , we add as a perturbation the most unstable global eigenfunction for density that we found in 3.3 , with an amplitude equal to 0.1% of the background density at the cluster center . figure [ f_nevol ] shows the evolutionary histories of the maximum density and the mass inflow rate @xmath133 at @xmath134 kpc for models a - c . the two dotted lines in figure [ f_nevol]@xmath135 represent the analytic estimates for the growth rates of thermal instability with and without conduction ; the predictions are in excellent agreement with the results of the numerical simulations . we have also compared the perturbed density and velocity profiles in the linear regime of model c with the corresponding eigenfunctions obtained from the linear theory . even though the radial grid does not cover the inner 1 kpc , they agree to within @xmath136% . radial profiles of electron number density , radial velocity , and temperature for a few selected epochs are plotted in figure [ f_profiles ] . model a , which is not in equilibrium and is subject to strong radiative cooling , immediately experiences radial inflow of mass everywhere . although the temperature of the cluster decreases steadily as the gas cools , the thermal pressure does not drop owing to adiabatic compression , which in turn maintains the radial velocity at the level of about @xmath137% of the central sound speed until @xmath138 gyr . during this period , we find @xmath139 so that the cooling flow is nearly isobaric . as the insert in figure [ f_nevol]@xmath135 shows , the maximum density follows the prediction of isobaric cooling fairly well . when the cooling modifies the density and temperature distributions significantly ( @xmath140 gyr ) , the increased cooling rate in the central parts induces larger mass inflows , causing the central density to increase in a runaway fashion . the change of density and temperature is so rapid that there is no chance for thermal instability to grow . we stop the simulation of model a at @xmath141 gyr when the central temperature becomes vanishingly small . the mass inflow rate depends rather sensitively on time and is a linear function of radius for @xmath142 kpc , with a slope of @xmath143 yr@xmath144 kpc@xmath144 , at the end of the run ( see figure [ f_nevol]@xmath145 ) . models b and c are initially in equilibrium with heating balancing cooling , but they both experience thermal instability . without conduction , perturbations in model b grow at a rate of @xmath146 gyr)@xmath144 and become fully unstable within @xmath147 gyr . although the growth time of thermal instability in model b is slighter shorter than the isobaric cooling time in model a ( see eq . [ [ eq_tinf ] ] ) , model b takes a longer time for eventual runaway than model a. this is simply because the former requires the growth of perturbations from a small amplitude , while the latter changes its background state immediately . model c clearly illustrates the stabilizing effect of thermal conduction . with @xmath122 , conduction causes the maximum growth rate of thermal instability to be about six times smaller than in the nonconducting model b , thereby delaying runaway growth until @xmath148 gyr ( for the particular amplitude of initial perturbations ) . the mass inflow rate is correspondingly very small throughout its entire evolution . had we started the evolution with a higher initial perturbation amplitude , it would of course take less time to reach a fully nonlinear state . small - scale perturbations in real clusters may well have high amplitudes , but they are readily erased by conduction . since it seems unlikely that the amplitudes of perturbations that are spherically symmetric and coherent over @xmath149 kpc are more than @xmath150 of background quantities , we believe that galaxy clusters with @xmath116 are not likely to exhibit a strong thermal instability for many gyr . we now carry out a local analysis of nonradial lagrangian perturbations in a static , stratified medium . we do not attempt to solve the full eigenvalue problem for nonradial modes , but instead focus on the local behavior of nonradial modes in the presence of density and/or temperature stratification . similar work has been reported by @xcite and @xcite who used eulerian perturbations and showed that a restoring buoyancy force from a stable entropy gradient can change thermal instability into an overstability , confirming the previous results of @xcite . using lagrangian perturbations , on the other hand , @xcite showed that the negative radial gradient of a net cooling function leads to convectively unstable flows in a non - equilibrium system . he further claimed that the eulerian and lagrangian approaches give different results , although in a subsequent paper @xcite showed that the lagrangian approach of thermal instability in dynamical media produces results that are apparently consistent with those from the eulerian analysis ( balbus 2003 , private communication ) . we show below that the lagrangian approach in fact gives exactly the same dispersion relation for nonradial thermal instability as the eulerian analyses referred to above , provided the background is in initial thermal equilibrium . we use the lagrangian technique and linearize equations ( [ eq_cont])-([eq_flux ] ) assuming that perturbations have small amplitudes of the form @xmath151 , where @xmath152 is the usual spherical harmonic . the detailed procedure to derive a local dispersion relation is given in the appendix : we write only the final result here , @xmath153 where @xmath154^{1/2}/r$ ] is the tangential wavenumber , @xmath155 , and @xmath156 is the brunt - vsil frequency for convective motions in a radially stratified atmosphere . when the conductivity @xmath157 , equation ( [ nonrdisp ] ) reduces to equation ( 3.17 ) of @xcite in an initial state , it is unclear how to compute an effective adiabatic index @xmath158 in his notation . applying @xmath67 directly to his equation ( 2.1c ) , one obtains @xmath159 instead of his equation ( 2.8 ) . plugging this expression into his equation ( 3.17 ) would produce exactly the same local dispersion relation as our equation ( [ nonrdisp ] ) when @xmath157 . then , the condition @xmath160 in his equation ( 3.28 ) is replaced by @xmath161 . ] . if @xmath162 in the last term , equation ( [ nonrdisp ] ) becomes exactly the same as the dispersion relation of the eulerian approach @xcite . it is the last term in equation ( [ nonrdisp ] ) that has led to the discrepancy between the eulerian and lagrangian perturbation analyses . this term vanishes if the background is in thermal balance , either in the presence of thermal conduction in which case @xmath163 , or with some other sources of heating in which case @xmath164 everywhere . if one perturbs equation ( [ eq_engy ] ) using the eulerian operator @xmath68 and lagrangian operator @xmath67 , respectively , and takes the difference of the resulting equations , the residual term is @xmath165 this explains the origin of the last term in equation ( [ nonrdisp ] ) and the apparent discrepancy of the eulerian and lagrangian approaches . the two approaches are consistent with each other only if the background state is in strict thermal equilibrium . one may argue that the assumption of initial thermal equilibrium is very special and a local analysis can be carried out even in a non - equilibrium system . however , such an analysis would be misleading unless the change of the background state either is allowed for in doing the perturbations or occurs slowly compared to the growth of perturbations of interest . as we showed in 3 , the cooling time in galaxy clusters is comparable to the growth time of thermal instability . moreover , in 4 , we performed a numerical simulation of a cluster that is initially out of thermal equilibrium ( model a ) and showed that the resulting cooling flow quickly dominates the cluster evolution , leaving no time for the development of a local thermal instability . since we assume the system is in initial thermal balance , we rewrite equation ( [ nonrdisp ] ) as @xmath166 which was already derived and discussed by @xcite . compared to equation ( [ ldisp ] ) for radial perturbations , equation ( [ fdisp ] ) has an extra term that arises from the local buoyancy force . since @xmath167 and @xmath168 , therefore @xmath169 , and galaxy clusters are stable to convective instability . the buoyancy term has then a stabilizing effect , reducing the growth rate of the thermal instability and even altering the character of the instability to an overstability when @xmath170 is high enough @xcite . if a medium is thermally overstable , a bubble that is displaced radially outward ( inward ) becomes more ( less ) dense compared to the adiabatic case , thus experiencing an increase of its oscillation amplitude with time . from equations ( [ ldisp ] ) and ( [ fdisp ] ) , it is trivial to show that for a given wave - vector @xmath171 , re(@xmath85 ) of the local nonradial mode is always smaller than that of the local radial mode ( e.g. , @xcite ) , implying that the latter is more unstable . this suggests that global nonradial modes , if they exist , would have smaller growth rates compared to global radial modes . @xcite and @xcite have shown that thermal conduction is quite efficient in a fully turbulent , magnetized plasma . the diffusion coefficient in such a medium is approximately a fraction @xmath172 a few tenths of the spitzer coefficient ( the full spitzer coefficient applies to an unmagnetized medium ) . conduction at this level is sufficient to provide significant heat to the cooling gas in a galaxy cluster and to maintain energy balance in the cluster center ( @xcite ; zn03 ) . this is an attractive explanation for the suppression of mass dropout in cooling cores of clusters . however , the demonstration that one can build equilibrium models with heating equal to cooling is not sufficient , since the hot gas is likely to be subject to a thermal instability @xcite . to be fully consistent with the new x - ray data on clusters , which show the absence of gas below a few kev , it is necessary to demonstrate that the thermal instability is either fully suppressed or is at least much weaker than expected . it is well - known that thermal conduction suppresses thermal instability in a hot gas for perturbations with short wavelengths . indeed , using a local wkb - like analysis , zn03 have shown that perturbations with wavelengths up to a fraction of the radius are thermally stable in clusters . however , this still leaves open the possibility that one or more long - wavelength global modes may be unstable . in this paper , we have analyzed in some detail the stability of such global modes . following zn03 , we set up the equilibrium density and temperature profiles of clusters , assuming strict hydrostatic and energy equilibrium ( 2 ) . by applying lagrangian perturbations , we derive a set of differential equations that describe the eigenvalue problem for global radial modes , with the mode growth rate @xmath85 acting as the eigenvalue ( eqs . [ [ eq_f1]][[eq_f3 ] ] , 3 ) . in the absence of conduction ( @xmath173 ) , we find as expected that a cluster has an infinite number of unstable modes , all with rapid growth rates @xmath174 ( eqs . [ [ ldisp ] ] and [ [ eq_tinf ] ] ) , where @xmath65 is the standard growth time scale associated with the thermal instability . for typical cooling flow clusters , the modes should grow in less than a gyr , implying a serious threat to stability . however , when we set @xmath122 , roughly the value recommended by @xcite and zn03 , we find that all radial modes except one become stable . the one residual unstable mode is the fundamental nodeless ( in @xmath118 ) mode , i.e. the mode with the longest possible `` wavelength '' that can fit within the system . this mode is weakly unstable , with a growth time scale @xmath175 that is 6 to 9 times longer than @xmath65 . @xcite has recently posted a paper in which he describes a quasi - global stability analysis of clusters . however , both the initial equilibrium he assumes and the perturbations he considers are very approximate , and it is hard to compare his results with ours . for the clusters we have studied , @xmath175 for the lone unstable mode is @xmath2 gyr ( table 1 ) . this is an interestingly long time scale , which is probably comparable to the elapsed time since the last major merger event in a hierarchical clustering scenario ( typically @xmath176 gyr for massive clusters , e.g. , @xcite ) . we imagine that major mergers ( and perhaps to a lesser extent even minor mergers ) leave the cluster gas in a highly mixed and turbulent state . if the merger drives the whole system well out of thermal equilibrium , mass dropout would occur very rapidly ( @xmath177 gyr ) at the central parts ( as in model a in 4 ) , which may in turn allow the remaining gas to achieve a new thermal equilibrium that is similar to its present state . in a sense , the merger acts to reset initial conditions . after the merger , the unstable global radial mode would grow . however , if the growth time for the mode is comparable to the effective age of the cluster since the last merger , as appears to be the case for our models , then we do not expect the thermal instability to be a major problem . we have confirmed the results of 3 by means of numerical simulations of radial perturbations in a model cluster ( 4 , fig . 5 ) . three models are discussed . model a considers a cluster without any heating to balance cooling . this model exhibits the classic cooling flow catastrophe , a violent and very rapid runaway at the cluster center . the runaway is so rapid that there is no possibility for the thermal instability to do anything . model c considers a cluster in which cooling is exactly balanced by conductive heating in equilibrium . the numerical simulation shows that this model has a slow instability that grows at precisely the rate calculated in 3 . for the particular initial conditions selected , the mode grows to the non - linear runaway stage only after 24 gyr , which means that the model is for all practical purposes stable . model b is an interesting in - between case in which heating and cooling are balanced , except that the heating is not from conduction but from some other local ( unspecified ) agency . this model shows the classic thermal instability . the perturbations grow quite rapidly , on the time scale @xmath65 of the @xcite instability . in 4 gyr , the model is completely destroyed by nonlinear runaway . we thus reach the following important conclusion from the simulations : apart from balancing heating and cooling , it is necessary also to make sure that the heating is of the right kind to control the thermal instability . specifically , the heating must involve diffusive transport of energy so that short - wavelength perturbations are smoothed out and not allowed to grow , and even long - wavelength perturbations are partially stabilized . thermal conduction is diffusive in nature and is quite effective in this respect , as we have shown in this paper . turbulent mixing would be similarly effective . agn heating via a jet might behave diffusively if the heat is transferred to the cluster gas via mechanical turbulence . heating through mass infall in minor mergers might also be effective since in this case again the heat is likely to spread through the gas via turbulence . however , radiative heating @xcite or cosmic ray heating @xcite from an agn are unlikely to control the thermal instability , since these heating mechanisms are not diffusive in nature . in this connection , we note that not all solutions to the cooling flow problem involve equilibrium models . @xcite have described an interesting model involving agn heating in which the gas goes through a limit cycle . for most of the time , the agn at the center is in quiescence and the gas is not significantly heated . as the gas undergoes a cooling catastrophe , the gas density around the agn increases , causing the agn to switch on , to eject a powerful jet and to heat the cluster gas . the heated gas expands , the agn switches off , and the gas starts the cycle again . the analysis we have carried out in this paper does not include such scenarios , since we have explicitly assumed that heating and cooling are balanced . in 5 , we analyze the properties of nonradial perturbations in a cluster . this problem has been studied by @xcite , @xcite , and @xcite using an eulerian approach , and by @xcite using a lagrangian approach . we show that the lagrangian analysis leads to the same local dispersion relation ( see eq . [ [ fdisp ] ] ) as the previous eulerian analyses , provided that the background is in thermal equilibrium . for a cluster that is convectively stable ( as all clusters are ) , @xmath169 ( eq . ) , and the buoyancy force associated with the entropy gradient plays a stabilizing role . as a result , if we consider radial perturbations with a given wave - vector that are thermally unstable , then non - radial perturbations with the same wave - vector always have a lower growth rate . this means that the most unstable mode is always a radial mode . although this result is obtained via a local analysis , it is expected to be true for global modes as well . we have , therefore , not analyzed global non - radial modes . is it an accident that the mode growth time that we have estimated is comparable to the age of the system as measured since the last major merger ? we suggest that perhaps it is not . imagine a cluster that is formed with sufficient gas such that several modes are initially thermally unstable with fairly short growth times . if the cluster were to start from a well - stirred initial state ( say immediately after a merger ) , the various modes would grow and when the age of the system exceeds the growth time of any particular mode , that mode would go non - linear and cause a certain amount of mass to cool and drop out from the hot medium . the mass dropout will cause the amount of gas in the hot phase to decrease , and as a result , the other modes would become less unstable . with continued mass dropout , perhaps only one unstable mode would be left finally . this mode would remain in the system so long as its growth time is not much shorter than the current age . with increasing age , presumably this mode too will cause some mass dropout , but always in such a manner that the gas that is left will have a growth time comparable to the age . in this picture , mass dropout acts as a safety valve that enables the system to be always in a state of marginal equilibrium . if this scenario is valid , then it is of course natural that the clusters we observe have unstable modes with growth times of order several gyr . in this paper , we have assumed the parameter @xmath29 to be constant over the entire volume of a cluster . this may not be a good approximation in some clusters that exhibit sharp discontinuities in temperature and density ( e.g. , @xcite ) . these cold fronts , located at about @xmath178 kpc , probably result from cluster mergers . @xcite argued that strong magnetic fields parallel to cold fronts may be responsible for the low conductivity . if so , the existence of cold fronts is unlikely to affect the strong stabilizing role played by conduction for the bulk of the cluster gas . the cold fronts are presumably transient features , surviving for only a dynamical time ( @xmath177 gyr ) before being disrupted by kelvin - helmholtz instability @xcite or by merger shocks @xcite . from spatial variations of temperature detected in the cluster a754 , @xcite argued that , in addition to cold fronts , the bulk of the gas in this cluster may have conductivity much smaller than the spitzer value . the critical question is how recently were the observed temperature inhomogeneities formed and how much longer will they last . if the cluster is undergoing a merger , for instance , the density and temperature fluctuations are all transient . although conduction tries to smooth out local temperature fluctuations , these structures may be continuously created by subsequent mergers and/or the sloshing motion of the gas in the dark matter potential @xcite . acoustic motions of the gas may also induce temporary fluctuations . since a large fraction of galaxy clusters exhibit powerful extended radio sources at their centers , many authors have considered outbursts from the central agn to be the source of heat to balance the radiative cooling in clusters . as noted above , this works best if the heating is done via some turbulent agency so that short wavelength thermal instability can be controlled . in addition , the agn heating can not be too strong since it would smooth out and erase the radial variations of metallicity that have been observed in some clusters ( e.g. , @xcite ) . a very real possibility is that both agn heating and conductive heating work together in clusters , though their relative importance may vary from cluster to cluster ( zn03 ) . it is even possible that agn heating dominates in the inner regions , while conduction plays a more important role farther out . recently , @xcite and @xcite demonstrated that simultaneous heating by agn and thermal conduction produces quasi - static density and temperature profiles that are similar to those of observed clusters . in particular , @xcite showed that clusters with agn heating alone exhibit either too high mass accretion rates or unrealistic temperature distributions , while conductive heating with @xmath179 alone or together with agn heating gives reasonable fits to observations . although the effects of their computational methods ( e.g. , removal of cold gas out of the computational domain ) and prescriptions for the agn heating are uncertain , their simulations suggest that the stabilization of the hot gas is most likely achieved by conduction . therefore , even in clusters where agn heating is dominant , the role of thermal conduction as a stabilizing agent should not be ignored ( zn03 ) . we gratefully acknowledge helpful discussions with s. balbus , l. david , g. field , e. ostriker , and n. yoshida . we also thank an anonymous referee for useful comments . this work was supported in part by nasa grant nag5 - 10780 . in this appendix , we linearize equations ( [ eq_cont])-([eq_flux ] ) for nonradial modes of perturbations and derive a local dispersion relation for thermal instability in a stratified medium . as in 3 for radial modes , our approach is lagrangian and includes the effects of thermal conduction . similar analyses have been performed by @xcite for eulerian perturbations , and by @xcite for lagrangian perturbations , but without conduction . the effects of a radial mass flow and magnetic fields were studied by @xcite and @xcite using lagrangian perturbations . we apply @xmath67 to equations ( [ eq_mome ] ) and ( [ eq_engy ] ) and obtain @xmath180 @xmath181 where @xmath182 denotes the tangential component of @xmath70 and @xmath183 is the tangential gradient @xmath184 with @xmath185 and @xmath186 denoting unit vectors in the polar and azimuthal directions , respectively . equation ( [ eq_flux_r ] ) linking @xmath79 and @xmath80 is valid for both radial and nonradial perturbations . following @xcite , we assume that the perturbations are of the form ( see also @xcite ) @xmath187 y_{lm}(\theta,\phi)e^{\sigma t},\ ] ] where @xmath152 is the spherical harmonic , satisfying an identity @xmath188 we assume the same angular and temporal dependences in the other perturbed variables @xmath79 and @xmath80 ; in what follows , we omit @xmath189 from all the perturbation variables . the tangential component of equation ( [ app_mome ] ) is integrated to give @xmath190 r\xi_t = \frac{1}{r^2}\frac{dr^2\xi_r}{dr } - \frac{{\delta t}}{t } + \frac{d\ln p}{dr}\xi_r.\ ] ] from equations ( [ sph ] ) and ( [ tan_xi ] ) , we thus write @xmath191,\end{aligned}\ ] ] where the dimensionless parameter @xmath192 , defined by @xmath193 measures approximately the square of the ratio of the growth time to the sound crossing time across a tangential wavelength . substituting equations ( [ sph ] ) and ( [ div_xi ] ) into the radial component of equation ( [ app_mome ] ) , we obtain @xmath194 \frac{\xi_r}{r } \nonumber \\ & = & \frac{1}{r } \left[\frac{d}{dr}\left(\frac{{\delta t}}{t}\right ) + \left(d\frac{d\ln p}{dr } - \frac{d\ln d}{dr}\right)\frac{{\delta t}}{t } \right].\end{aligned}\ ] ] similarly , equation ( [ app_engy ] ) becomes @xmath195.\end{aligned}\ ] ] note that when @xmath196 , corresponding to pure radial modes , equations ( [ rad_p ] ) and ( [ eng_p ] ) are reduced to equations ( [ eq_f1 ] ) and ( [ eq_f2 ] ) , respectively . subject to proper boundary conditions , equations ( [ rad_p ] ) , ( [ eng_p ] ) , and ( [ eq_f3 ] ) may be integrated to yield solutions for global nonradial modes , but this is beyond the scope of the present paper . let us define the local tangential wavenumber @xmath154^{1/2}/r$ ] and the local radial wavenumber @xmath197 , where @xmath198 refers to any perturbed variable . let us also focus on local modes which vary rapidly in both radial and tangential directions and grow slowly compared to the sound crossing time across their wavelengths , i.e.@xmath88 , @xmath89 , and @xmath199 ( or @xmath200 ) . using equation ( [ eq_f3 ] ) in the following alternative form , @xmath201 we simplify equations ( [ rad_p ] ) and ( [ eng_p ] ) to @xmath202 @xmath203\xi_r = \left[\frac{\gamma p}{\gamma-1}(\sigma -\sigma_\infty ) + \kappa tk^2\right]\frac{{\delta t}}{t},\ ] ] where @xmath204 is the amplitude of the total wavenumber and @xmath95 is defined by equation ( [ eq_siginf ] ) . combining equations ( [ simp1 ] ) and ( [ simp2 ] ) , we finally obtain equation ( [ nonrdisp ] ) as a local dispersion relation for nonradial thermal perturbations . afshordi , n. , & cen , r. 2002 , 564 , 669 allen , s. w. , ettori , s. , & fabian , a. c. 2001 , , 324 , 877 balbus , s. a. 1986 , , 303 , l79 balbus , s. a. 1988 , , 328 , 395 balbus , s. a. 1991 , , 372 , 25 balbus , s. a. , & soker , n. 1989 , , 341 , 611 bertschinger , e. , & meiksin , a. 1986 , , 306 , l1 binney , j. , & cowie , l. l. 1981 , , 247 , 464 bhringer , h. , belsole , e. , kennea , j. , matsushita , k. , molendi , s. , worrall , d. m. , mushotzky , r. f. , ehle , m. , guainazzi , m. , sakelliou , i. , stewart , g. , vestrand , w. t. , dos santos , s. 2001 , , 365 , l181 bregman , j. n. , & david , l. p. 1988 , , 326 , 639 brighenti , f. , & mathews , w. g. 2002 , , 573 , 542 brighenti , f. , & mathews , w. g. 2003 , , 587 , 580 brggen , m. , & kaiser , c. r. 2002 , nature , 418 , 301 carilli , c. l. , taylor , g. b. 2002 , , 40 , 319 chandran , b. d. g. , & cowley , s. c. 1998 , phys . lett . , 80 , 3077 chandran , b. d. g. , cowley , s. c. , ivanushkina , m. , & sydora , r. 1999 , , 525 , 638 cho , j. , lazarian , a. , honein , a. , knaepen , b. , kassinos , s. , & moin , s. 2003 , , 589 , l77 chun , e. , & rosner , r. 1993 , , 408 , 678 churazov , e. , sunyaev , r. , forman , w. , & bhringer , h. 2002 , , 332 , 729 ciotto , l. , & ostriker , j. p. 2001 , , 551 , 131 cox , j. p. 1980 , theory of stellar pulsations ( princeton : princeton univ . press ) david , l. p. , hughes , j. p. , & tucker , w. h. 1992 , , 394 , 452 david , l. p. , nulsen , p. e. j. , mcnamara , b. r. forman , w. , jones , c. , ponman , t. , robertson , b. , & wise , m. 2001 , , 557 , 546 defouw , r. j. 1970 , , 160 , 659 dupke , r. , & white , r. e. , iii . 2003 , , 583 , l13 ettori , s. , fabian , a. c. , allen , s. w. , & johnstone , r. m. 2002 , , 331 , 635 fabian , a. c. 1994 , , 32 , 277 fabian , a. c. , mushotzky , r. f. , nulsen , p. e. j. , & peterson , j. r. 2001a , , 321 , l20 fabian , a. c. , sanders , j. s. , ettori , s. , taylor , g. b. , allen , s. w. , crawford , c. s. , iwasawa , k , & johnstone , r. m. 2001b , , 321 , l33 fabian , a. c. , voigt , l. m. , & morris r. g. 2002 , , 335 , l71 field , g. b. 1965 , , 142 , 531 gaetz , t. j. 1989 , , 345 , 666 goldreich , p. , & sridhar , s. 1995 , , 438 , 763 goldreich , p. , & sridhar , s. 1997 , , 485 , 680 gruzinov , a. 2002 , astro - ph/0203031 johnstone , r. m. , allen , s. w. , fabian , a. c. , & sanders , j. s. 2002 , , 336 , 299 kaiser , c. r. , binney , j. 2003 , , 338 , 837 kitayama , t. , & suto , y. 1996 , , 469 , 480 loewenstein , m. , zweibel , e. g. , & begelman , m. c. 1991 , , 377 , 392 malagoli , a. , rosner , r. , & bodo , g. 1987 , , 319 , 632 malyshkin , l. , & kulsrud , r. 2001 , , 549 , 402 maoz , d. , rix , h .- w . , gal - yam , a. , & gould , a. 1997 , , 486 , 75 markevitch , m. et al . 2000 , , 541 , 542 markevitch , m. et al . 2003 , , 586 , l19 matsushita , k. , belsole , e. , finoguenov , a. , & bhringer , h. 2002 , , 386 , 77 mazzotta , p. , vikhlinin , a. , fusco - femiano , r. , & markevitch , m. 2002 , , 569 , l31 mckee , c. f. , & begelman , m. c. 1990 , , 358 , 392 molendi , s. , & pizzolato , f. 2001 , , 560 , 194 nagai , d. , & kravtsov , a. v. 2003 , , 587 , 514 narayan , r. , & medvedev , m. v. 2001 , , 562 , l129 peterson , j. r. , paerels , f. b. s. , kaastra , j. s. , arnaud , m. , reiprich , t. h. , fabian , a. c. , mushotzky , r. f. , & jernigan , j. g. , sakelliou , i. 2001 , , 365 , l104 peterson , j. r. , kahn , s. m. , paerels , f. b. s. , kaastra , j. s. , tamura , t. , bleeker , j. a. m. , ferrigo , c. , & jernigan , j. g. 2003 , , 590 , 207 pistinner , s. , & shaviv , g. 1996 , , 459 , 1 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes in fortran ( cambridge : cambridge univ . press ) , 838 rechester , a. b. , & rosenbluth , m. n. 1978 , phys . lett . , 40 , 38 reynolds , c. s. , heinz , s. , & begelman , m. c. 2002 , , 332 , 271 roettiger , k. , stone , j. m. , & mushotzky , r. f. 1998 , , 493 , 62 rosati , p. , borgani , s. , & norman , c. 2002 , , 40 , 539 rosner , r. , & tucker , w. h. 1989 , , 338 , 761 ruszkowski , m. , & begelman , m. c. 2002 , , 581 , 223 rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : wiley ) sarazin , c. l. 1988 , x - ray emissions from clusters of galaxies ( cambridge : cambridge university press ) shapiro , s. l. , & teukolsky , s. a. 1983 , black holes , white dwarfs , and neutron stars ( new york : wiley - interscience ) , pp 127 - 147 soker , n. 2003 , , 342 , 463 spitzer , l. 1962 , physics of fully ionized gases ( new york : interscience ) stone , j. m , & norman , m. l. 1992 , , 80 , 753 tamura , t. , kaastra , j. s. , peterson , j. r. , paerels , f. b. s. , mittaz , j. p. d. , trudolyubov , s. p. , stewart , g. , fabian , a. c. , mushotzky , r. f. , lumb , d. h. , & ikebe , y. 2001 , , 365 , l87 tucker , w. h. , & rosner , r. 1983 , , 267 , 547 white , r. e. , iii , & sarazin , c. l. 1987 , , 318 , 612 vikhlinin , a. , markevitch , m. , & murray , s. s. 2001a , , 549 , l47 vikhlinin , a. , markevitch , m. , & murray , s. s. 2001b , , 551 , 160 voigt , l. m. , schmidt , r. w. , fabian , a. c. , allen , s. w. , johnstone , r. m. 2002 , , 335 , l7 zakamska , n. l. , & narayan , r. 2003 , , 582 , 162 ( zn03 ) ccccccc abell 1795 & 2 & 0.049 & 0.2 & 0.98 & 4.1 & 6.3 + abell 1835 & 5 & 0.17 & 0.4 & 0.45 & 1.9 & 6.3 + abell 2199 & 1.6 & 0.074 & 0.4 & 0.58 & 3.5 & 9.0 + abell 2390 & 4 & 0.069 & 0.3 & 0.98 & 4.6 & 7.1 + rxj 1347.5 - 1145 & 6 & 0.11 & 0.3 & 0.76 & 3.3 & 6.2 +

we present the first comprehensive search for submillimeter and radio emission from the host galaxies of twenty well - localized-ray bursts ( grbs ) . with the exception of a single source , all observations were undertaken months to years after the grb explosions to ensure negligible contamination from the afterglows . we detect the host galaxy of grb000418 in both the sub - mm and radio , and the host galaxy of grb000210 only in the sub - mm . these observations , in conjunction with the previous detections of the host galaxies of grb980703 and grb010222 , indicate that about of grb host galaxies are ultra - luminous ( l ) and have star formation rates of about m yr . as an ensemble , the non - detected hosts have a star formation rate of about m yr ( ) based on their radio emission . the detected and ensemble star formation rates exceed the optically - determined values by an order of magnitude , indicating significant dust obscuration . in the same vein , the ratio of bolometric dust luminosity to uv luminosity for the hosts detected in the sub - mm and radio ranges from , and follows the known trend of increasing obscuration with increasing bolometric luminosity . we also show that , both as a sample and individually , the grb host galaxies have bluer colors as compared with galaxies selected in the sub - mm in the same redshift range . this possibly indicates that the stellar populations in the grb hosts are on average younger , supporting the massive stellar progenitor scenario for grbs , but it is also possible that grb hosts are on average less dusty . beyond the specific results presented in this paper , the sub - mm and radio observations serve as an observational proof - of - concept in anticipation of the upcoming launch of the swift grb mission and sirtf . these new facilities will possibly bring grb host galaxies into the forefront of star formation studies .

astro-ph0210645

one of the major thrusts in modern cosmology is an accurate census of star formation and star - forming galaxies in the universe . this endeavor forms the backbone for a slew of methods ( observational , analytical , and numerical ) to study the process of galaxy formation and evolution over cosmic time . to date , star - forming galaxies have been selected and studied mainly in two observational windows : the rest - frame ultraviolet ( uv ) , and rest - frame radio and far - infrared ( fir ) . for galaxies at high redshift these bands are shifted into the optical and radio / sub - mm , allowing observations from the ground . still , the problem of translating the observed emission to star formation rate ( sfr ) involves large uncertainty . this is partly because each band traces only a minor portion of the total energy output of stars . moreover , the optical / uv band is significantly affected by dust obscuration , thus requiring order of magnitude corrections , while the sub - mm and radio bands lack sensitivity , and therefore uncover only the most prodigiously star - forming galaxies . the main result that has emerged from star formation surveys over the past few years is exemplified in the so - called madau diagram . namely , the sfr volume density , @xmath10 , rises steeply to @xmath11 , and seemingly peaks at @xmath12 . there is still some debate about the how steep the rise is , with values ranging from @xmath13 @xcite to @xmath14 ( e.g. @xcite ) . the evolution beyond @xmath15 is even less clear since optical / uv observations indicate a decline @xcite , while recent sub - mm observations argue for a flat @xmath10 to higher redshift , @xmath16 @xcite . consistency with this trend can be obtained by invoking large dust corrections in the optical / uv @xcite . for general reviews of star formation surveys we refer the reader to @xcite , @xcite , and @xcite . despite the significant progress in this field , our current understanding of star formation and its redshift evolution is still limited by the biases and shortcomings of current optical / uv , sub - mm , and radio selection techniques . in particular , despite the fact that the optical / uv band is sensitive to galaxies with modest star formation rates ( down to a fraction of a m@xmath3 yr@xmath5 ) at high redshift , these surveys may miss the most dusty , and vigorously star - forming galaxies . moreover , it is not clear if the simple , locally - calibrated prescriptions for correcting the observed _ un - obscured _ sfr for dust extinction ( e.g. @xcite ) , hold at high redshift ; even if they do , these prescriptions involve an order of magnitude correction . finally , the optical / uv surveys are magnitude limited , and therefore miss the faintest sources . sub - mm surveys have uncovered a population of highly dust - extincted galaxies , which are usually optically faint , and have star formation rates of several hundred m@xmath3 yr@xmath5 ( e.g. @xcite ) . however , unlike optical / uv surveys , sub - mm surveys are severely sensitivity limited , and only detect galaxies with @xmath17 l@xmath3 . more importantly , current sub - mm bolometer arrays ( such as scuba ) have large beams on the sky ( @xmath18 arcsec ) making it difficult to unambiguously identify optical counterparts ( which are usually faint to begin with ) , and hence measure the redshifts @xcite ; in fact , of the @xmath19 sub - mm galaxies identified to date , only a handful have a measured redshift . finally , translating the observed sub - mm emission to a sfr requires significant assumptions about the temperature of the dust , and the dust emission spectrum ( e.g. @xcite ) . surveys at decimeter radio wavelengths also suffer from low sensitivity , but the high astrometric accuracy afforded by synthesis arrays such as the vla allows a sub - arcsec localization of the radio - selected galaxies . as a result , it is easier to identify the optical counterparts of these sources . recently , this approach has been used to pre - select sources for targeted sub - mm observations resulting in an increase in the sub - mm detection rate @xcite and redshift determination @xcite . however , this method is biased toward finding luminous ( high sfr ) sources since it requires an initial radio detection . an additional problem with radio , even more than with sub - mm , selection is contamination by active galactic nuclei ( agn ) . an examination of the x - ray properties of radio and sub - mm selected galaxies reveals that of the order of @xmath1 can have a significant agn component @xcite . the most significant problem with current star formation studies , however , is that the link between the optical and sub - mm / radio samples is still not well understood . the hubble deep field provides a clear illustration : the brightest sub - mm source does not appear to have an optical counterpart @xcite , and only recently a detection has been claimed in the near - ir ( @xmath20 mag ; @xcite ) . along the same line , sub - mm observations of the optically - selected lyman break galaxies have resulted in very few detections @xcite , and even the brightest lyman break galaxies appear to be faint in the sub - mm band @xcite . in addition , there is considerable diversity in the properties of optical counterparts to sub - mm sources , ranging from galaxies which are faint in both the optical and near - ir ( nir ) to those which are bright in both bands @xcite . as a result of the unclear overlap , and the sensitivity and dust problems in the sub - mm and optical surveys , the fractions of global star formation in the optical and sub - mm / radio selected galaxies is not well constrained . it is therefore not clear if the majority of star formation takes place in ultra - luminous galaxies with very high star formation rates , or in the more abundant lower luminosity galaxies with star formation rates of a few m@xmath3 yr@xmath5 . given the difficulty with redshift identification of sub - mm galaxies , the redshift distribution of dusty star forming galaxies remains highly uncertain . one way to alleviate some of these problems is to study a sample of galaxies that is immune to the selection biases of current optical / uv and sub - mm / radio surveys , and which may draw a more representative sample of the underlying distribution of star - forming galaxies . the host galaxies of @xmath0-ray bursts ( grbs ) may provide one such sample . the main advantages of the sample of grb host galaxies are : ( i ) the galaxies are selected with no regard to their emission properties in any wavelength regime , ( ii ) the dust - penetrating power of the @xmath0-ray emission results in a sample that is completely unbiased with respect to the global dust properties of the hosts , ( iii ) grbs can be observed to very high redshifts with existing missions ( @xmath21 ; @xcite ) , and as a result volume corrections for the star formation rates inferred from their hosts are trivial , ( iv ) the redshift of the galaxy can be determined via absorption spectroscopy of the optical afterglow , or x - ray spectroscopy allowing a redshift measurement of arbitrarily faint galaxies ( the current record - holder is the host of grb990510 with @xmath22 mag and @xmath23 ; @xcite ) , and ( v ) since there is excellent circumstantial evidence linking grbs to massive stars ( e.g. @xcite , the sample of grb hosts is expected to trace global star formation @xcite . of course , the sample of grb hosts is not immune from its own problems and potential biases . the main problem is the relatively small size of the sample in comparison to both the optical and sub - mm samples ( although the number of grb hosts with a known redshift exceeds the number of sub - mm galaxies with a measured redshift ) . as a result , at the present it is not possible to assess the sfr density that is implied by grb hosts , or its redshift evolution . a bias towards sub - solar metallicity for grb progenitors ( and hence their environments ) has been discussed @xcite , but it appears that very massive stars ( e.g. @xmath24 m@xmath3 ) should produce black holes even at solar metallicity . the impact of metallicity on additional aspects of grb formation ( e.g. angular momentum , loss of hydrogen envelope ) is not clear at present . finally , given the observed dispersion in metallicity within galaxies ( e.g. @xcite ) , it is likely that even if grbs require low metallicity progenitors , this does not imply that the galaxy as a whole has a lower than average metallicity . to date , grb host galaxies have mainly been studied in the optical and nir bands . with the exception of one source ( grb020124 ; @xcite ) , every grb localized to a sub - arcsecond position has been associated with a star - forming galaxy @xcite . these galaxies range from @xmath25 mag , have a median redshift , @xmath26 , and are generally typical of star - forming galaxies at similar redshifts in terms of morphology and luminosity @xcite , with star formation rates from optical spectroscopy of @xmath27 m@xmath3 yr@xmath5 . at the same time , there are hints for higher than average ratios of [ neiii]3869 to [ oii]3727 , possibly indicating the presence of massive stars @xcite . only two host galaxies have been detected so far in the radio ( grb980703 ; @xcite ) and sub - mm ( grb010222 ; @xcite ) . here we present sub - mm and radio observations of a sample of @xmath28 grb host galaxies , ranging in redshift from about @xmath29 to @xmath30 ( [ sec : obs ] ) ; one of the 20 sources is detected with high significance in both the sub - mm and radio bands , and an additional source is detected in the sub - mm ( [ sec : res ] ) . we compare the detected sub - mm and radio host galaxies to local and high-@xmath31 ultra - luminous galaxies in [ sec : sed ] , and derive the sfrs in [ sec : sfr ] . we then compare the inferred sfrs of the detected host galaxies , and the ensemble of undetected hosts , to optical estimates in [ sec : opt ] . finally , we compare the optical properties of the grb host galaxies to those of sub - mm and radio selected star - forming galaxies ( [ sec : comp ] ) . at the time we conducted our survey , the sample of grb host galaxies numbered 25 , twenty of which had measured redshifts . these host galaxies were localized primarily based on optical afterglows , but also using the radio and x - ray afterglow emission . of the 25 host galaxies we observed eight in both the radio and sub - mm , seven in the radio , and five in the sub - mm . the galaxies were drawn from the list of 25 hosts at random , constrained primarily by the availability of observing time . thus , the sample presented here does not suffer from any obvious selection biases , with the exception of detectable afterglow emission in at least one band . sub - mm observations of grb afterglows , and a small number of host galaxies have been undertaken in the past . starting in 1997 , @xcite and @xcite have searched for sub - mm emission from the afterglow of thirteen grbs . while they did not detect any afterglow emission , these authors used their observations to place constraints on emission from eight host galaxies , with typical @xmath32 rms values of 1.2 mjy . since these were target - of - opportunity observations , they were not always undertaken in favorable observing conditions . more recently , @xcite reported targeted sub - mm observations of the host galaxies of four optically - dark grbs ( i.e. grbs lacking an optical afterglow ) . they focused on these particular sources since one explanation for the lack of optical emission is obscuration by dust , which presumably points to a dusty host . none of the hosts were detected , with the possible exception of grb000210 ( see [ sec:0210 ] ) , leading the authors to conclude that the hosts of dark bursts are not necessarily heavily dust obscured . thus , the observations presented here provide the most comprehensive and bias - free search for sub - mm emission from grb host galaxies , and the first comprehensive search for radio emission . observations in the sub - mm band were carried out using the sub - millimeter common user bolometer array ( scuba ; holland et al . 1999 ) on the james clerk maxwell telescope ( jcmt ) . we observed the positions of thirteen well - localized grb afterglows with the long ( 850 @xmath33 m ) and short ( 450 @xmath33 m ) arrays . the observations , summarized in table [ tab : submm ] , were conducted in photometry mode with the standard nine - jiggle pattern using the central bolometer in each of the two arrays to observe the source . in the case of grb000301c we used an off - center bolometer in each array due to high noise levels in the central bolometer . to account for variations in the sky brightness , we used a standard chopping of the secondary mirror between the on - source position and a position 60 arcsec away in azimuth , at a frequency of 7.8125 hz . in addition , we used a two - position beam switch ( nodding ) , in which the beam is moved off - source in each exposure to measure the sky . measurements of the sky opacity ( sky - dips ) were taken approximately every two hours , and the focus and array noise were checked at least twice during each shift . the pointing was checked approximately once per hour using several sources throughout each shift , and was generally found to vary by @xmath34 arcsec ( i.e. less than one quarter of the beam size ) . all observations were performed in band 2 and 3 weather with @xmath35 . the data were initially reduced with the scuba data reduction facility ( surf ) following the standard reduction procedure . the off - position pointings were subtracted from the on - position pointings to account for chopping and nodding of the telescope . noisy bolometers were removed to facilitate a more accurate sky subtraction ( see below ) , and the data were then flat - fielded to account for the small differences in bolometer response . extinction correction was performed using a linear interpolation between skydips taken before and after each set of on - source scans . in addition to the sky subtraction offered by the nodding and chopping , short - term sky contributions were subtracted by using all low - noise off - source bolometers ( sky bolometers ) . this procedure takes advantage of the fact that the sky contribution is correlated across the array . as a result , the flux in the sky bolometers can be used to assess the sky contribution to the flux in the on - source bolometer . this procedure is especially crucial when observing weak sources , since the measured flux may be dominated by the sky . we implemented the sky subtraction using surf and our own routine using matlab . we found that in general the surf sky subtraction under - estimated the sky contribution , and as a result over - estimated the source fluxes . we therefore used the results of our own analysis routine . for this purpose we calculated the median value of the two ( three ) outer rings of bolometers in the 850 @xmath33 m ( 450 @xmath33 m ) array , after removing noisy bolometers ( defined as those whose standard deviation over a whole scan deviated by more than 5@xmath36 from the median standard deviation of all sky bolometers ) . following the sky subtraction , we calculated the mean and standard deviation of the mean ( sdom ) for each source in a given observing shift . noisy data were eliminated in two ways . first , the data were binned into 25 equal time bins , and the sdom was calculated step - wise , i.e. at each step the data from an additional bin were added and the mean and sdom were re - calculated . in an ideal situation where the data quality remains approximately constant , the sdom should progressively decrease as more data are accumulated . however , if the quality of the data worsens ( due to deteriorating weather conditions for example ) the sdom will increase . we therefore removed time bins in which the sdom increased . following this procedure , we recursively eliminated individual noisy data points using a sigma cutoff level based on the number of data points ( chauvenet s criterion ; @xcite ) until the mean converged on a constant value . typically , two or three iterations were required , with only a few data points rejected each time . typically , only a few percent of the data were rejected by the two procedures . finally , flux conversion factors ( fcfs ) were applied to the resulting voltage measurements to convert the signal to jy . using photometry observations of mars and uranus , and/or secondary calibrators ( oh231.8 + 4.2 , irc+10216 , and crl618 ) , we found the fcf to vary between @xmath37 jy / v at 850 @xmath33 m , consistent with the typical value of @xmath38 . at 450 @xmath33 m , the fcfs varied between @xmath39 jy / v . _ very large array ( vla ) : _ we observed @xmath40 grb afterglow positions with the vla from april 2001 to february 2002 . all sources were observed at 8.46 ghz in the standard continuum mode with @xmath41 mhz bands . in addition , was observed at 1.43 and 4.86 ghz , and grb0010222 was observed at 4.86 ghz . in table [ tab : rad ] we provide a summary of the observations . in principle , since the median spectrum of faint radio sources between 1.4 and 8.5 ghz is @xmath42 @xcite , the ideal vla frequency for our observations ( taking into account the sensitivity at each frequency ) is 1.43 ghz . however , we chose to observe primarily at 8.46 ghz for the following reason . the majority of our observations were taken in the bnc , c , cnd , and d configurations , in which the typical synthesized beam size is @xmath43 arcsec at 1.43 ghz , compared to @xmath44 arcsec at 8.46 ghz . the large synthesized beam at 1.43 ghz , combined with the larger field of view and higher intrinsic brightness of radio sources at this frequency , would result in a significant decrease in sensitivity due to source confusion . thus , we were forced to observe at higher frequencies , in which the reduced confusion noise more than compensates for the typical steep spectrum . we chose 8.46 ghz rather than 4.86 ghz since the combination of @xmath1 higher sensitivity and @xmath45 lower confusion noise , provide a more significant impact than the typical @xmath46 decrease in intrinsic brightness . the 1.43 ghz observations of were undertaken in the a configuration , where confusion does not play a limiting role . for flux calibration we used the extragalactic sources 3c48 ( j0137 + 331 ) , 3c147 ( j0542 + 498 ) , and 3c286 ( j1331 + 305 ) , while the phases were monitored using calibrator sources within @xmath47 of the survey sources . we used the astronomical image processing system ( aips ) for data reduction and analysis . for each source we co - added all the observations prior to producing an image , to increase the final signal - to - noise . _ australia telescope compact array ( atca ) : _ we observed the positions of four grb afterglows during april 2002 , in the 6a configuration at @xmath48 and @xmath49 mhz . using the 6-km baseline resulted in a significant decrease in confusion noise , thus allowing observations at the most advantageous frequencies . the observations are summarized in table [ tab : rad ] . we used j1934@xmath50638 for flux calibration , while the phase was monitored using calibrator sources within @xmath47 of the survey sources . the data were reduced and analyzed using the multichannel image reconstruction , image analysis and display ( miriad ) package , and aips . the flux measurements at the position of each grb are given in tables [ tab : submm ] and [ tab : rad ] , and are plotted in figure [ fig : fluxes ] . of the @xmath28 sources , only was detected in both the radio and sub - mm with @xmath51 ( [ sec:0418 ] ) . one additional source , grb000210 , is detected with @xmath51 when combining our observations with those of @xcite . two hosts have radio fluxes with @xmath52 ( grb000301c and grb000926 ) , but as we show below this is due in part to emission from the afterglow . the typical @xmath53 thresholds are about @xmath54 mjy , @xmath28 @xmath33jy , and @xmath55 @xmath33jy in the scuba , vla , and atca observations , respectively . in figure [ fig : fluxes ] we plot all sources with @xmath51 as detections , and the rest as @xmath53 upper limits . in addition , for the sources observed with the atca we plot both the 1.4 ghz upper limits , and the inferred upper limits at 8.46 ghz assuming a typical radio spectrum , @xmath42 @xcite . one obvious source for the observed radio and sub - mm fluxes ( other than the putative host galaxies ) is emission from the afterglows . to assess the possibility that our observations are contaminated by flux from the afterglows we note that the observations have been undertaken at least a year after the grb explosion . on this timescale , the sub - mm emission from the afterglow is expected to be much lower than the detection threshold of our observations . in fact , the brightest sub - mm afterglows to date have only reached a flux of a few mjy ( at 350 ghz ) , and typically exhibited a fading rate of @xmath56 after about one day following the burst @xcite . thus , on the timescale of our observations , the expected sub - mm flux from the afterglows is only @xmath57 @xmath33jy , well below the detection threshold . the radio emission from grb afterglows is more long - lived , and hence posses a more serious problem . however , on the typical timescale of the radio observations the 8.46 ghz flux is expected to be at most a few @xmath33jy ( e.g. @xcite ) . in the following sections we discuss the individual detections in the radio and sub - mm , and also provide an estimate for the radio emission from each afterglow . a source at the position of is detected at four of the five observing frequencies with @xmath51 . the scuba source , which we designate smm12252 + 2006 , has a flux density of @xmath58 mjy , and @xmath59 mjy . these values imply a spectral index , @xmath60 ( @xmath61 ) , consistent with a thermal dust spectrum as expected if the emission is due to obscured star formation . the radio source ( vla122519.26 + 200611.1 ) , is located at @xmath62(j2000)= , @xmath63(j2000)= , with an uncertainty of 0.1 arcsec in both coordinates . this position is offset from the position of the radio afterglow of grb000418 ( berger et al . 2001 ) by @xmath64 arcsec and @xmath65 arcsec ( figure [ fig : lmap ] ) . in comparison , the offset measured from keck and _ hubble space telescope _ images is smaller , @xmath66 arcsec and @xmath67 arcsec . vla122519.26 + 200611.1 has an observed spectral slope @xmath68 , flatter than the typical value for faint radio galaxies , @xmath69 @xcite , and similar to the value measured for the host of grb980703 ( @xmath70 ; @xcite ) . the source appears to be slightly extended at 1.43 and 8.46 ghz , with a size of about @xmath71 arcsec ( @xmath72 kpc at @xmath73 ) . the expected afterglow fluxes at 4.86 and 8.46 ghz at the time of our observations are about @xmath74 and @xmath75 @xmath33jy , respectively @xcite . at 1.43 ghz the afterglow contribution is expected to be about @xmath75 @xmath33jy based on the 4.86 ghz flux and the afterglow spectrum @xmath76 . thus , despite the contribution from the afterglow , the radio detections of the host galaxy are still significant at better than @xmath77 level . correcting for the afterglow contribution we find an actual spectral slope @xmath78 , consistent with the median @xmath69 for 8.46 ghz radio sources with a similar flux @xcite . as with all scuba detections , source confusion arising from the large beam ( @xmath79 arcsec at 350 ghz and @xmath80 arcsec at 670 ghz ) raises the possibility that smm12252 + 2006 is not associated with the host galaxy of . fortunately , the detection of the radio source , which is located @xmath81 arcsec away from the position of the radio afterglow of , indicates that smm12252 + 2006 and vla122519.26 + 200611.1 are in fact the same source the host galaxy of . besides the positional coincidence of the vla and scuba sources , we gain further confidence of the association based on the spectral index between the two bands , @xmath82 . this spectral index is redshift dependent as a result of the different spectral slopes in the two regimes @xcite . we find @xmath83 , in good agreement with the @xcite value of @xmath84 ( for the redshift of grb000418 , @xmath73 ) . we also detect another source , slightly extended ( @xmath85 arcsec ) , approximately 1.4 arcsec east and 2.7 arcsec south of the host of ( designated vla122519.36 + 200608.4 ) , with @xmath86 @xmath33jy and @xmath87 @xmath33jy ( see figure [ fig : lmap ] ) . this source appears to be linked by a bridge of radio emission ( with @xmath88 at both frequencies ) to the host of . the physical separation between the two sources , assuming both are at the same redshift , @xmath73 , is 25 kpc . there is no obvious optical counterpart to this source in _ hubble space telescope _ images down to about @xmath89 mag . based purely on radio source counts at 8.46 ghz @xcite , the expected number of sources with @xmath90 @xmath33jy in a 3 arcsec radius circle is only about @xmath91 . thus , the coincidence of two such faint sources within 3 arcsec is highly suggestive of an interacting system , rather than chance superposition . interacting radio galaxies with separations of about @xmath28 kpc , and joined by a bridge of radio continuum emission have been observed locally @xcite . in addition , optical surveys ( e.g. @xcite ) show that a few percent of galaxies with an absolute @xmath92-band magnitude similar to that of the host of , have companions within about @xmath93 kpc . the fraction of interacting systems is possibly much higher , @xmath94 , in ultra - luminous systems ( such as the host of ) , both locally @xcite and at high redshift ( e.g. @xcite ) . we note that with a separation of only 3 arcsec , the host of grb000418 and the companion galaxy fall within the scuba beam . thus , it is possible that smm12252 + 2006 is in fact a superposition of both radio sources . this will change the value of @xmath95 to about @xmath96 . the host galaxy of grb980703 has been detected in deep radio observations at 1.43 , 4.86 , and 8.46 ghz @xcite . the galaxy has a flux @xmath97 @xmath33jy , and a radio spectral slope @xmath98 . in addition , the radio emission is unresolved with a maximum angular size of @xmath99 arcsec ( @xmath100 kpc ) . based on the @xcite value of @xmath101 ( for the redshift of grb980703 , @xmath102 ) , the expected flux at 350 ghz is @xmath103 mjy . the observed ( @xmath53 ) flux limit @xmath104 mjy , is consistent with the expected value . grb010222 has been detected in scuba and iram observations with a persistent flux of about 3.5 mjy at 350 ghz and 1 mjy at 250 ghz @xcite . the persistent emission , as well as the steep spectral slope , @xmath105 , indicated that while the detected emission was partially due to the afterglow of grb010222 , it was dominated by the host galaxy . in fact , accounting for the expected afterglow emission , we find that the host galaxy has a flux , @xmath106 mjy . the radio flux predicted from the sub - mm emission @xcite is @xmath107 @xmath33jy ( for @xmath108 , the redshift of grb010222 ) , which corresponds to @xmath109 @xmath33jy , and @xmath110 @xmath33jy ( assuming @xmath111 ) . therefore , our measured values , @xmath112 @xmath33jy and @xmath113 @xmath33jy are consistent with the observed sub - mm emission . the expected afterglow fluxes at 4.86 and 8.46 ghz are @xmath114 and @xmath115 @xmath33jy , respectively , significantly lower than the measured values . thus , the observed flux mainly arises from the host . recently , @xcite measured a flux of @xmath116 mjy for grb000210 , in good agreement with our value of @xmath117 mjy . a weighted - average of the two measurements gives @xmath118 mjy , similar to the sub - mm flux from the host galaxies of grb000418 and grb010222 . the radio flux at the position of grb000210 is @xmath119 @xmath33jy . based on a redshift of 0.846 @xcite and the sub - mm detection , the expected radio flux from this source @xcite is @xmath120 @xmath33jy , consistent with the measured value . the expected flux of the afterglow at the time of the radio observations is less than 1 @xmath33jy @xcite . following the localization of grb980329 , @xcite observed the afterglow position with scuba and claimed the detection of a source with a 350 ghz flux of about @xmath121 mjy on 1998 , apr . subsequent observations indicated a fading trend , with a decline to @xmath122 mjy on apr . 6.2 , and @xmath123 mjy ( @xmath53 ) on apr . 11.2 . based on a comparison to the radio flux of the afterglow , @xcite concluded that the detected sub - mm flux was in excess of the emission from the afterglow itself , and therefore requires an additional component , most likely a host galaxy . recently , @xcite re - analyzed the scuba data and showed that the initial sub - mm flux was in fact only about @xmath124 mjy , and perfectly consistent with the afterglow flux . as a result , it is not clear that an additional persistent component is required . we also re - analyzed the data from apr . 1998 using the method described in [ sec : submm ] . we find the following fluxes : @xmath125 mjy ( apr . 5 ) , @xmath126 mjy ( apr . 6 ) , @xmath127 mjy ( apr . 7 ) , @xmath128 mjy ( apr . 8) , and @xmath129 mjy ( apr . a comparison to the results in @xcite reveals that , with the exception of the last epoch , they over - estimate the fluxes by about @xmath130 mjy . our observations of grb980329 from september and october of 2001 reveal a flux , @xmath131 mjy , indicating that the flattening to a value of about @xmath132 mjy in the late epochs of the apr . 1998 observations may indicate emission from the host galaxy . the radio observations are similarly inconclusive , with @xmath133 @xmath33jy . we estimate that the flux of the afterglow at 8.46 ghz at the time of our observations is only @xmath134 @xmath33jy @xcite . since the redshift of grb980329 is not known , we can not assess the expected ratio of the radio and sub - mm fluxes . this source is detected in the vla observations with a flux of @xmath135 @xmath33jy ( @xmath136 ) . the expected flux from the afterglow at the time of the observations , @xmath137 days after the burst , is @xmath75 @xmath33jy @xcite . thus , the observed emission exceeds the afterglow flux by @xmath138 . in the calculations below we use a host flux of @xmath139 @xmath33jy . the vla observations of this grb position reveal a source with @xmath140 @xmath33jy ( @xmath141 ) . the flux of the afterglow at the time of the observations is about @xmath74 @xmath33jy ( berger et al . 2000 ) . thus , the excess emission is significant at the @xmath142 level . the sub - mm emission predicted based on the @xcite relation is @xmath143 mjy ( for @xmath144 , the redshift of grb000301c ) . this value is in agreement with the measured flux of @xmath145 mjy . in figure [ fig : sed ] we plot the radio - to - uv spectral energy distributions ( seds ) of the detected host galaxies of grb980703 , grb000418 , and grb010222 , as well as that of arp220 , a proto - typical local ultra - luminous ir galaxy ( ulirg ; @xcite ) , and ero j164502 + 4626.4 ( hur10 ) , a high-@xmath31 analog of arp220 @xcite . the luminosities are plotted as a function of rest - frame frequencies , to facilitate a direct comparison . the detected grb hosts are brighter than arp220 ( @xmath146 l@xmath3 , sfr@xmath147 m@xmath3 yr@xmath5 ) , and are similar in luminosity to hur10 ( @xmath148 l@xmath3 , sfr@xmath149 m@xmath3 yr@xmath5 ; @xcite ) . as such , we expect the host galaxies to have star formation rates of a @xmath150 m@xmath3 yr@xmath5 , and luminosities in excess of @xmath151 l@xmath3 . on the other hand , the optical / nir properties of the detected grb hosts are distinctly different than those of arp220 and hur10 ( as well as other local and high-@xmath31 ulirgs ) . in particular , from figure [ fig : sed ] it is clear that , while the grb host galaxies are similar to hur10 and arp220 in the radio and sub - mm bands , their optical / nir colors ( as defined for example by @xmath9 ) are much bluer . moreover , while there is a dispersion of a factor of few in the radio and sub - mm bands between the grb hosts , hur10 , and arp220 , the dispersion in the optical / nir luminosity is about two orders of magnitude . this indicates that there is no simple correlation between the optical / nir properties of grb hosts ( and possibly other galaxies ) and their fir and radio luminosities . in the following sections we expound on both points to evaluate the star formation rates that are implied by the sub - mm and radio measurements , we use the following expression for the observed flux as a function of sfr @xcite : @xmath152 } { { \rm exp}(0.00083\nu_0)-1 } \ } \frac{(1+z){\rm sfr}}{d_l^2}\,\,{\rm jy}. \label{eqn : sfr}\ ] ] here , @xmath153 ghz , sfr is the star formation rate in m@xmath3 yr@xmath5 , @xmath154 is the luminosity distance in mpc , and @xmath155 is a scaling factor ( of order unity ) which accounts for the difference in the conversion between synchrotron flux and sfr in the milky way and other galaxies . the first term on the right - hand - side accounts for the fact that non - thermal synchrotron emission arising from supernova remnants is proportional to the sfr , while the second term is the contribution of free - free emission from hii regions . these two flux terms dominate in the radio regime . the last term in equation . [ eqn : sfr ] is the dust spectrum , which dominates in the sub - mm and fir regimes . in this case , the parameters that have been chosen to characterize the spectrum are a dust temperature , @xmath156 k , and a dust emissivity , @xmath157 , based on a sample of 23 ir - selected starburst galaxies with @xmath158 l@xmath3 @xcite . we note that other authors ( e.g. @xcite ) favor a lower dust temperature , @xmath159 k , which would result in somewhat different inferred star formation rates . to calculate @xmath154 we use the cosmological parameters @xmath160 , @xmath161 , and @xmath162 km s@xmath5 mpc@xmath5 . we also use the typical value @xmath69 for the radio measurements @xcite . in figure [ fig : fluxes ] we plot contours of constant sfr overlaid on the sub - mm and radio flux measurements . our radio observations are sensitive to galaxies with @xmath163 m@xmath3 yr@xmath5 at @xmath164 , and @xmath165 m@xmath3 yr@xmath5 at @xmath166 . the sub - mm flux , on the other hand , is relatively constant for a given sfr , independent of @xmath31 . this is due to the large positive k - correction resulting from the steep thermal dust spectrum . therefore , at the typical limit of our sub - mm observations we are sensitive to galaxies with sfr@xmath167 m@xmath3 yr@xmath5 . for the host galaxies that are detected with @xmath51 in the sub - mm and radio , as well as those detected in the past ( i.e. grb980703 and grb010222 ) we calculate the following star formation rates : grb000418 @xmath168 m@xmath3 yr@xmath5 , @xmath169 m@xmath3 yr@xmath5 ; grb000210 @xmath170 m@xmath3 yr@xmath5 ; grb010222 @xmath171 m@xmath3 ; grb980703 @xmath172 m@xmath3 yr@xmath5 . here @xmath173 and @xmath174 are the sfrs derived from the sub - mm and radio fluxes , respectively . the detections and upper limits from this survey , combined with the detections and upper limits discussed in the literature @xcite indicate that about @xmath1 of all grbs explode in galaxies with star formation rates of @xmath175 m@xmath3 yr@xmath5 . at the same time , it is clear that @xmath176 of grb host galaxies have more modest star formation rates , @xmath177 m@xmath3 yr@xmath5 . despite the fact that the majority of the survey sources are not detected , we can ask the question of whether the grb host galaxies exhibit a significant sub - mm and/or radio emission _ on average_. the weighted average emission from the non - detected sources ( @xmath178 ) is @xmath179 mjy , and @xmath180 @xmath33jy . this average radio flux is possibly contaminated by flux from the afterglows at the level of about @xmath114 @xmath33jy , so we use @xmath181 @xmath33jy ( @xmath182 ) . therefore , as an ensemble , the grb host galaxies exhibit radio emission , but no significant sub - mm emission . using the median redshift , @xmath183 , for the non - detected sample , the average radio flux implies an average @xmath184 m@xmath3 yr@xmath5 , while the sub - mm @xmath53 upper limit on @xmath185 is about @xmath186 m@xmath3 yr@xmath5 . the average sub - mm flux can be compared to @xmath187 mjy for the non - detected sub - mm sources in a sample of radio pre - selected , optically faint ( @xmath188 mag ) galaxies @xcite , @xmath189 mjy for lyman break galaxies @xcite , or @xmath190 mjy for optically - selected starbursts in the hubble deep field @xcite . thus , it appears that grb host galaxies trace a somewhat fainter population of sub - mm galaxies compared to the radio pre - selected sample , but similar to the lyman break and hdf samples . this is not surprising given that the radio pre - selection is naturally biased in favor of luminous sources . we can further extend this analysis by calculating the average sub - mm and radio fluxes in several redshift bins . here we include both detections and upper limits . from the sub - mm ( radio ) observations we find : @xmath191 mjy ( @xmath192 @xmath33jy ) for @xmath193 , @xmath194 mjy ( @xmath195 @xmath33jy ) for @xmath196 , and @xmath197 mjy ( @xmath198 @xmath33jy ) for @xmath199 . these average fluxes are marked in figure [ fig : fluxes ] . in the sub - mm there is a clear increase in the average flux from @xmath200 to @xmath12 , and a flattening or decrease beyond @xmath15 . in the radio , on the other hand , the average flux is about the same in all three redshift bins . the average radio fluxes translate into the following star formation rates : for @xmath200 the inferred average sfr is @xmath201 m@xmath3 yr@xmath5 , for @xmath202 it is @xmath19 m@xmath3 yr@xmath5 , and for @xmath199 it is @xmath203 m@xmath3 yr@xmath5 ( with @xmath204 significance in each bin ) . the sub - mm observations on the other hand , indicate a rise from a value of @xmath205 m@xmath3 yr@xmath5 for @xmath200 to @xmath206 m@xmath3 yr@xmath5 for @xmath202 , followed by a decline to @xmath207 m@xmath3 yr@xmath5 for @xmath199 . the typical _ un - obscured _ star formation rates inferred from optical spectroscopy are of the order of @xmath208 m@xmath3 yr@xmath5 ( e.g. @xcite ) . in particular , the host galaxy of grb980703 has an optical sfr of about @xmath75 m@xmath3 yr@xmath5 @xcite , compared to about @xmath209 m@xmath3 yr@xmath5 from the radio observations . similarly , the host of has an optical sfr of about @xmath210 m@xmath3 yr@xmath5 @xcite , compared to about @xmath211 m@xmath3 yr@xmath5 based on the radio and sub - mm detections , while the host of grb000210 has an optical sfr of @xmath212 m@xmath3 yr@xmath5 compared to about @xmath213 m@xmath3 yr@xmath5 from the sub - mm observations . finally , the average radio sfr for the non - detected sources , @xmath214 m@xmath3 yr@xmath5 , significantly exceeds the average optical sfr . the discrepancy between the optical and radio / sub - mm star formation rates indicates that the majority of the star formation in the grb host galaxies that are detected in the sub - mm and radio is obscured by dust . it is possible that the same is true for the sample as a whole , but this relies on the less secure average sfr in the non - detected hosts . the significant dust obscuration is not surprising given that a similar trend has been noted in high-@xmath31 starburst galaxies , for which the typical dust corrections ( based on the uv slope technique ) are an order of magnitude @xcite . in this case we find similar correction factors . we can also assess the level of obscuration by comparing the uv luminosity at @xmath215 , @xmath216 , to the bolometric dust luminosity , @xmath217 . the ratio of these two quantities provides a rough measure of the obscuration , while the sum provides a rough measure of the total star formation rate @xcite . we estimate @xmath216 using the following host magnitudes : @xmath218 mag ( grb980703 ; @xcite ) , @xmath219 mag ( grb0000210 ; piro et al . 2002 ) , @xmath220 mag ( grb000418 ; @xcite ) , and @xmath221 mag ( grb010222 ; @xcite ) , and assuming a spectrum @xmath222 . we calculate the following values of @xmath216 : @xmath223 l@xmath3 ( grb980703 ) , @xmath224 l@xmath3 ( grb000210 ) , @xmath225 l@xmath3 ( grb000418 ) , and @xmath226 l@xmath3 ( grb010222 ) . we estimate @xmath217 from the radio and sub - mm observations , using the conversion factors of @xcite . the resulting values are : @xmath227 l@xmath3 ( grb980703 ) , @xmath228 l@xmath3 ( grb000210 ) , @xmath229 l@xmath3 ( grb000418 ) , and @xmath230 l@xmath3 ( grb010222 ) . thus , @xmath231 evaluates to : @xmath28 ( grb980703 ) , @xmath232 ( grb000210 ) , @xmath233 ( grb000418 ) , and @xmath234 ( grb010222 ) . these results , as well as the sample of starbursts and ulirgs at @xmath164 taken from @xcite are plotted in figure [ fig : lbol ] . we note that the grb hosts are within the scatter of the @xmath164 sample , with a general trend of increasing value of @xmath231 ( i.e. inceasing obscuration ) with increasing @xmath235 ( i.e. inceasing sfr ) . at the same time , the particular lines of sight to the grbs within the sub - mm / radio bright host galaxies do not appear to be heavily obscured . for example , an extinction of @xmath236 mag has been inferred for grb980703 @xcite , @xmath237 mag has been found for grb000418 @xcite , and @xmath238 mag has been found for grb010222 . the optically - dark grb000210 suffered more significant extinction , @xmath239 mag . in addition , the small offset of grb980703 relative to its radio host galaxy ( @xmath240 arcsec ; @xmath241 kpc at the redshift of the burst ) , combined with the negligible extinction , indicates that while the burst probably exploded in a region of intense star formation , it either managed to destroy a large amount of dust in its vicinity , or the dust distribution is patchy . it is beyond the scope of this paper to evaluate the potential of dust destruction by grbs ( see e.g. @xcite ) , but it is clear that the grbs that exploded in the detected sub - mm and radio host galaxies , did not occur in the most heavily obscured star formation sites . as we noted in [ sec : sed ] , the optical / nir colors of the detected grb host galaxies are bluer than those of arp220 ( @xmath242 mag ) and hur10 ( @xmath243 mag ; @xcite ) . in this section we compare the @xmath9 color of grb hosts to the @xmath9 colors of radio pre - selected sub - mm galaxies @xcite and sub - mm selected galaxies with a known optical counterpart and a redshift @xcite . in figure [ fig : color ] we plot @xmath9 color versus redshift for grb hosts and radio pre - selected sub - mm galaxies . before comparing the two populations , we note that the mean @xmath9 color and redshift for the entire grb sample are @xmath244 mag and @xmath245 , respectively , and for the hosts that are detected in the sub - mm and radio they are @xmath246 mag and @xmath247 , respectively . thus , there is no clear correlation between the optical / nir colors of the grb hosts and their sub - mm / radio luminosity . for the sub - mm sample the mean @xmath9 color and redshift are @xmath248 mag and @xmath249 , respectively . to facilitate a more direct comparison with the grb sample we calculate the mean values for the sub - mm sample in the same redshift range as the grb hosts : @xmath250 mag and @xmath251 . moreover , if we examine only the host galaxies that were detected in the radio and sub - mm with high significance we find @xmath9 colors of : 2.2 mag ( grb000418 ) , 2.8 mag ( grb980703 ) , 2.1 mag ( grb010222 ) , and 2.6 ( grb000210 ) . the bluest sub - mm galaxies , on the other hand , have @xmath252 mag . in general , the grb hosts that are detected with @xmath51 in the sub - mm and radio have a distribution of @xmath9 that is indistinguishable from the general grb host galaxy sample . the obvious difference in @xmath9 color indicates that the grb and radio / sub - mm selections result in a somewhat different set of galaxies . the red colors of the sub - mm selected galaxies are not surprising since these sources are expected to be dust obscured . on the other hand , the mean color of the grb hosts is bluer by about 2.5 mag ( @xmath253 significance ) compared to sub - mm galaxies in the same redshift range , indicating a bias towards less dust obscuration . a more patchy dust distribution , or intrinsically bluer colors . it is possible that there is a bias toward less dust obscuration in the general grb host sample because the bursts that explode in dusty galaxies would have obscured optical afterglows , and hence no accurate localization . however , this is not a likely explanation since the grbs which exploded in the sub - mm and radio bright hosts are not significantly dust obscured ( [ sec : opt ] ) . moreover , it does not appear that the hosts of dark grbs are brighter in the sub - mm as expected if the dust obscuration is global @xcite . finally , the localization of afterglows in the radio and x - rays allows the selection of host galaxies even if they are dusty . in particular , the only two grbs in which significant obscuration of the optical afterglow has been inferred ( grb970828 : @xcite ; grb000210 : piro et al . 2002 ) , have been localized thanks to accurate positions from the radio and x - ray afterglows , and have host galaxies with @xmath9 colors of 3.7 and 2.6 mag , not significantly redder than the general population of grb hosts . therefore , a bias against dust obscured host galaxies is not the reason for the bluer color of the sample . an alternative explanation is that the distribution of dust in grb hosts is different than in the radio pre - selected and sub - mm selected galaxies . this may be in terms of a spatially patchy distribution , which will allow more of the uv light to escape , or a different distribution of grain sizes ( i.e. a different extinction law ) , possibly due to a different average metallicity . however , in both cases it is not clear why there should be a correlation between the dust properties of the galaxy and the occurence of a grb . finally , it is possible that grb host galaxies are preferentially in an earlier stage of the star formation ( or starburst ) process . in this case , a larger fraction of the shorter - lived massive stars would still be shining , and the overall color of the galaxy would be bluer relative to a galaxy with an older population of stars . one way to examine the age of the stellar population is to fit population synthesis models to the broad - band optical / nir spectra of the host galaxies . this approach has recently been used by @xcite who find some evidence that the age of the stellar population in some grb host galaxies ( including the host of grb980703 ) is relatively young , of the order of @xmath254 myr . this result is also expected if grbs arise from massive stars , as indicated by recent observations ( e.g. @xcite ) , since in this case grbs would preferentially select galaxies with younger star formation episodes . regardless of the exact reason for the preferential selection of bluer galaxies relative to the radio pre - selected sub - mm population , two results seem clear : ( i ) the grb host galaxies detected in the sub - mm and radio are likely drawn from a population that is generally missed in current sub - mm surveys , and ( ii ) grb host galaxies may not be a completely bias - free sample . the first point is particularly interesting in light of the fact that optical observations of these host galaxies do not identify them as particularly exceptional in terms of sfr . therefore , while similar galaxies are not necessarily missed in optical surveys , their star formation rates are likely under - estimated . we presented the most comprehensive scuba , vla , and atca observations of grb host galaxies to date . the host galaxy of is the only source detected with high significance in both the sub - mm and radio , while the host galaxy of grb000210 is detected with @xmath255 in the sub - mm when we combine our observations with those of @xcite . when taken in conjunction with the previous detections of grb980703 in the radio @xcite and grb010222 in the sub - mm @xcite , these observations point to a @xmath256 detection rate in the radio / sub - mm . this detection rate confirms predictions for the number of sub - mm bright grb hosts , with @xmath257 mjy , based on current models of the star formation history assuming a large fraction of obscured star formation @xcite . the host galaxies detected in the sub - mm and radio have star formation rates from about @xmath258 to @xmath259 m@xmath3 yr@xmath5 , while statistically the non - detected sources have an _ average _ sfr of about @xmath6 m@xmath3 yr@xmath5 . these star formation rates exceed the optically - inferred values by over an order of magnitude , pointing to significant dust obscuration within the grb host galaxies detected in the sub - mm and radio , and possibly the sample as a whole . still , the optical afterglows of the bursts that exploded in the sub - mm / radio bright host galaxies did not suffer significant extinction , indicating that : ( i ) the grbs did not explode in regions where dust obscuration is significant , or ( ii ) the uv and x - ray emission from the afterglow destroys a significant amount of dust in the local vicinity of the burst . we have also shown that grb host galaxies , even those detected in the sub - mm / radio , have bluer @xmath9 colors compared to galaxies selected in the sub - mm or radio bands in the same redshift range . this is not the result of an observational bias against dusty galaxies in the grb host sample since the afterglows of grbs which exploded in the radio / sub - mm bright hosts were not significantly obscured . more likely , this is the result of younger stellar populations in these galaxies , or possibly a patchy dust distribution . if the reason is younger stellar population then this provides additional circumstantial evidence in favor of massive ( and hence short - lived ) stars as the likely progenitors of grbs . a potential bias of the grb host galaxy sample is that the popular `` collapsar '' model of grbs calls for high mass , low metallicity stellar progenitors @xcite . this may result in preferential selection of low metallicity ( and hence less dusty ) host galaxies . however , it appears that grb progenitors can even have solar metallicity , and that a very low metallicity is unfavored by the required initial conditions for a grb explosion . moreover , studies of the milky way ( see @xcite for a recent review ) , local galaxies ( e.g @xcite ) , and high-@xmath31 galaxies ( e.g. @xcite ) , indicate that there are considerable variations in metallicity within galaxies . this may be especially true if several independent episodes of star formation have occured within the galaxy . thus , even if there is a bias towards low metallicity for grb progenitors ( and hence their immediate environments ) it is not obvious that this introduces a bias in the host galaxy sample . nonetheless , while the observations presented in this paper clearly indicate the potential of grb selection of high-@xmath31 galaxies for the study of star formation , a much larger sample is required to complement existing optical and sub - mm surveys . this may become possible in the near future with the upcoming launch ( sep . 2003 ) of swift . with an anticipated rapid ( @xmath260 minute ) and accurate localization of about 150 bursts per year , the grb - selected sample will probably increase to several hundred galaxies over the next few years . the rapid localization would most likely result in a large fraction of redshift measurements thanks to the bright optical afterglows . in addition to the localization of a large number of grb hosts , the study of these galaxies ( as well as those in other samples ) would greatly benefit from the advent of new facilities , such as sirtf , alma , evla , and the ska . in figure [ fig : sirtf ] we again plot the rest - frame seds of arp220 and the sub - mm / radio bright grb hosts . overplotted on these seds are the @xmath32 sensitivities of sirtf , alma , and the evla for 200-sec exposures at redshifts 1 and 3 , as well as the sensitivities of current instruments ( vla and scuba ) . the contributions of these new facilities to star formation studies are threefold : ( i ) increased sensitivity , ( ii ) increased resolution , and ( iii ) increased frequency coverage . these improvements will serve to ameliorate the main limitations of present radio , sub - mm , and ir observations ( [ sec : intro ] ) , by allowing the detection of more representative star forming galaxies at high redshift , in addition to a better constraint on the total dust bolometric luminosity and accurate localizations , which would facilitate follow - ups at optical wavelengths . in conjunction with increasingly larger samples of galaxies selected in the optical , the radio / sub - mm / ir , and by grbs , the future of star formation studies is poised for great advances and new discoveries . we thank k. adelberger , a. blain , and a. shapley for helpful discussions , and g. moriarty - schieven for help with the data reduction . we also thank s. chapman for providing us with the optical / nir colors and redshifts of radio pre - selected sub - mm galaxies prior to publication . , j. r. 1982 , an introduction to error analysis . the study of uncertainties in physical measurements , : a series of books in physics , oxford : university press , and mill valley : university science books , 1982 ) . lclrrr grb970228 & 0.695 & nov . 1 , 2001 & @xmath261 & @xmath262 & + & & nov . 2 , 2001 & @xmath263 & @xmath264 & @xmath265 + grb970508 & 0.835 & sep . 9 , 2001 & @xmath266 & @xmath267 & + & & sep . 10 , 2001 & @xmath268 & @xmath269 & + & & sep . 12 , 2001 & @xmath270 & @xmath271 & @xmath272 + grb971214 & 3.418 & nov . 2 , 2001 & @xmath273 & @xmath274 & @xmath273 + grb980329 & & sep . 13 , 2001 & @xmath275 & @xmath276 & + & & oct . 29 , 2001 & @xmath277 & @xmath278 & @xmath279 + grb980613 & 1.096 & nov . 1 , 2001 & @xmath280 & @xmath281 & + & & nov . 2 , 2001 & @xmath282 & @xmath283 & + & & dec . 7 , 2001 & @xmath284 & @xmath285 & @xmath286 + grb980703 & 0.966 & sep . 10 , 2001 & @xmath287 & @xmath288 & + & & sep . 12 , 2001 & @xmath289 & @xmath290 & @xmath291 + grb991208 & 0.706 & dec . 6 , 2001 & @xmath292 & @xmath293 & + & & dec . 7 , 2001 & @xmath294 & @xmath295 & @xmath296 + grb991216 & 1.020 & oct . 31 , 2001 & @xmath297 & @xmath298 & + & & nov . 3 , 2001 & @xmath299 & @xmath300 & + & & nov . 4 , 2001 & @xmath301 & @xmath302 & @xmath303 + grb000210 & 0.846 & sep . 12 , 2001 & @xmath304 & @xmath305 & + & & sep . 13 , 2001 & @xmath306 & @xmath307 & + & & sep . 14 , 2001 & @xmath308 & @xmath309 & @xmath310 + grb000301c & 2.034 & dec . 29 , 2001 & @xmath311 & @xmath312 & + & & dec . 30 , 2001 & @xmath313 & @xmath314 & @xmath315 + grb000418 & 1.119 & oct . 30 , 2001 & @xmath316 & @xmath317 & + & & oct . 31 , 2001 & @xmath318 & @xmath319 & + & & nov . 1 , 2001 & @xmath320 & @xmath321 & @xmath322 + grb000911 & 1.058 & sep . 13 , 2001 & @xmath323 & @xmath324 & + & & sep . 14 , 2001 & @xmath325 & @xmath326 & + & & oct . 31 , 2001 & @xmath327 & @xmath328 & + & & nov . 3 , 2001 & @xmath329 & @xmath330 & + & & nov . 4 , 2001 & @xmath331 & @xmath332 & @xmath333 + grb011211 & 2.140 & dec . 29 , 2001 & @xmath334 & @xmath335 & + & & dec . 30 , 2001 & @xmath336 & @xmath337 & + & & dec . 31 , 2001 & @xmath338 & @xmath339 & @xmath340 lcclcr 0.1 in grb970828 & 0.958 & vla & jun . 47 , 2001 & 8.46 & @xmath341 + grb980329 & & vla & jul . 22 sep . 10 , 2001 & 8.46 & @xmath342 + grb980613 & 1.096 & vla & may 1826 , 2001 & 8.46 & @xmath343 + grb981226 & & vla & jul . 24 oct . 15 , 2001 & 8.46 & @xmath344 + grb991208 & 0.706 & vla & apr . 14 jul . 20 , 2001 & 8.46 & @xmath345 + grb991216 & 1.020 & vla & jun . 8 jul . 13 , 2001 & 8.46 & @xmath346 + grb000210 & 0.846 & vla & sep . 16 oct . 12 , 2001 & 8.46 & @xmath347 + grb000301c & 2.034 & vla & jun . 15 jul . 22 , 2001 & 8.46 & @xmath348 + grb000418 & 1.119 & vla & jan . 14 feb . 27 , 2002 & 1.43 & @xmath349 + & & vla & dec . 8 , 2001 jan . 10 , 2002 & 4.86 & @xmath350 + & & vla & may 28 jun . 3 , 2001 & 8.46 & @xmath351 + grb000911 & 1.058 & vla & mar . 21 apr . 2 , 2001 & 8.46 & @xmath352 + grb000926 & 2.037 & vla & jun . 11 jul . 12 , 2001 & 8.46 & @xmath353 + grb010222 & 1.477 & vla & sep . 29 oct . 13 , 2001 & 4.86 & @xmath354 + & & vla & jun . 24 aug . 27 , 2001 & 8.46 & @xmath355 + grb990510 & 1.619 & atca & apr . 28 , 2002 & 1.39 & @xmath356 + grb990705 & 0.840 & atca & apr . 2122 , 2002 & 1.39 & @xmath357 + grb000131 & 4.5 & atca & apr . 28 , 2002 & 1.39 & @xmath358 + grb000210 & 0.846 & atca & apr . 2728 , 2002 & 1.39 & @xmath359 lccc 0.1 in grb970228 & @xmath360 & & 1 + grb970508 & @xmath361 & & 1 + grb970828 & & @xmath362 & 1.2 + grb971214 & @xmath363 & & 3 + grb980329@xmath364 & @xmath365 & @xmath366 & + grb980613 & @xmath367 & @xmath368 & + grb980703 & @xmath361 & @xmath369 & 10 + grb981226@xmath370 & & @xmath371 & + grb990510 & & @xmath372 & + grb990705 & & @xmath373 & + grb991208 & @xmath374 & @xmath375 & 20 + grb991216 & @xmath376 & @xmath377 & + grb000131 & & @xmath378 & + grb000210 & @xmath379 & @xmath380 & 3 + grb000301c & @xmath381 & @xmath382 & + grb000418 & @xmath383 & @xmath384 & 55 + grb000911 & @xmath385 & @xmath386 & 2 + grb000926 & & @xmath387 & + grb010222 & @xmath388 & @xmath389 & 1.5 + grb011211 & @xmath390 & &

we present a new algorithm for computing motorcycle graphs that runs in time for any , improving on all previously known algorithms . the main application of this result is to computing the straight skeleton of a polygon . it allows us to compute the straight skeleton of a non - degenerate polygon with holes in expected time . if all input coordinates are-bit rational numbers , we can compute the straight skeleton of a ( possibly degenerate ) polygon with holes in expected time . in particular , it means that we can compute the straight skeleton of a simple polygon in expected time if all input coordinates are-bit rationals , while all previously known algorithms have worst - case running time .

1303.5958

the straight skeleton of a polygon @xmath8 is a straight line graph embedded in @xmath8 , formed by the traces of the vertices of @xmath8 when it is shrunk , each edge moving at the same speed and remaining parallel to its original position . ( see [ fig : mgdef ] . ) .3 , title="fig : " ] .3 , title="fig : " ] .3 , title="fig : " ] it has been known since at least the 19th century ; for instance , figures representing the straight skeleton can be found in the book by von peschka @xcite . aichholzer et al . @xcite gave the first efficient algorithms for computing the straight skeleton , and presented it as an alternative to the medial axis having only straight - line edges . the straight skeleton has found numerous applications in computer science , for instance to city model reconstruction @xcite , architectural modeling @xcite , polyhedral surface reconstruction @xcite , biomedical image processing @xcite . it also has a direct application to cad , as it allows to compute offset polygons @xcite . the straight skeleton has become a standard tool in geometric computing , and thus fast and robust software has been developed to compute it @xcite . the complexity of straight skeleton computation , however , is still very much open . the previously best known algorithms were the @xmath9-time algorithm by eppstein and erickson @xcite , and the @xmath10-time randomized algorithm by cheng and vigneron @xcite . the only known lower bound is @xmath11 , by a reduction from sorting @xcite . in this paper , we give new subquadratic algorithms for computing straight skeletons . in particular , if all input coordinates are @xmath4-bit rational numbers , we give an @xmath12-time randomized algorithm for computing the straight skeleton of a polygon with @xmath2 holes . it is the first near - linear time algorithm for computing the straight skeleton of a simple polygon . eppstein and erickson @xcite introduced _ motorcycle graphs _ so as to model the main difficulty of straight skeleton computation . we are given a set of @xmath13 motorcycles , each motorcycle having a starting point and a velocity . each motorcycle moves at constant velocity until it reaches the track left by another motorcycle , in which case it crashes . the resulting graph is called a motorcycle graph . ( see [ fig : mgss]a . ) [ fig : mgss],title="fig : " ] .4 . [ fig : mgss],title="fig : " ] the motorcycle graph is a special case of the straight skeleton , where each motorcycle is modeled by a small and thin triangle . conversely , a polygon induces a motorcycle graph , where each motorcycle starts at a reflex vertex and moves with the same velocity as this vertex moves during the shrinking process . ( see [ fig : mgss]b . ) cheng and vigneron @xcite showed that computing the straight skeleton of a non - degenerate polygon reduces to computing this induced motorcycle graph , and a lower envelope computation ; huber and held extended this proof to degenerate cases @xcite . the lower envelope computation can be done in @xmath14 expected time if @xmath8 has @xmath2 holes . previously , the bottleneck of straight skeleton computation was the induced motorcycle graph computation . this is our main motivation for designing a faster motorcycle graph algorithm . in this paper , we give an algorithm for computing a motorcycle graph that runs in @xmath0 time , for any @xmath1 , improving on all previously known algorithms . here is a brief description of our algorithm . for each motorcycle , we maintain a tentative track , which may be longer than its actual track in the motorcycle graph . we also maintain a set of target points , which contains the endpoints of the tentative tracks that have been created earlier for this motorcycle , and that it has not reached yet . initially , the tentative tracks are empty , and then we try to extend them one by one , all the way to the destination point . if two tentative tracks cross , we retract them , by roughly halving the number of possible crossing points on each of them . after performing this halving , the tentative tracks do not intersect , and we can safely move the motorcycle that reaches the end of its tentative track first . then we try to extend the tentative track of this motorcycle to its next target point , and repeat the process . an example is given in appendix [ sec : example ] . apart from obtaining better time bounds for straight skeleton computation , there are at least two other reasons for studying motorcycle graphs . first , huber and held @xcite used the idea of computing the straight skeleton from its induced motorcycle graph to design and implement a practical straight skeleton algorithm . so it is important , even in practice , to get a better understanding of motorcycle graph computation . another motivation for studying motorcycle graphs is a direct application to computer graphics , for quad mesh partitioning @xcite . some of our results make no particular assumptions on the input , but we also present a few results where we assume that the input coordinates are @xmath4-bit rational numbers . we believe that this assumption is sufficient for most applications . for instance , in the applications mentioned above , it is hard to imagine that the input polygons would have features smaller than 1 nm , and size larger than 1000 km , so 64-bit integers should be more than sufficient . the main novelty in this paper is our algorithm for computing motorcycle graphs ( section [ sec : algorithm ] ) . this algorithm is essentially different from the two previous algorithms @xcite that both simulate the construction in chronological order . our algorithm , on the other hand , does not construct the motorcycle tracks in chronological order : it may move some motorcycle to its position at time @xmath15 , and then later during the execution of the algorithm , move another motorcycle to its position at an earlier time @xmath16 . ( we give one such example in appendix [ sec : example ] . ) this answers an open question by eppstein and erickson ( * ? ? ? * end of section 5 ) , who asked whether the running time can be improved by relaxing the chronology of the events . our algorithm uses two auxiliary data structures , one for ray shooting , and another for halving queries . given a query segment on the supporting line of a motorcycle , a halving query returns a splitting point on this segment such that there are roughly the same number of intersections with other supporting lines on both sides . ( see section [ sec : notation ] . ) the implementation of these data structures in different settings lead to different time bounds . for all our results , we use the standard real - ram model @xcite , that allows to perform arithmetic operations exactly on arbitrary real numbers . but for some of our results , we make the assumption that all input coordinates are @xmath4-bit rational numbers . it has two advantages : it yields better time bounds , and allows us to handle the straight skeleton of degenerate polygons . this improvement comes from the fact that , for bounded precision input , two distinct crossing points between the supporting lines of two pairs of motorcycles are at distance @xmath17 from each other . it allows us to use a simpler halving scheme : instead of halving a segment according to the number of intersection points , we use the midpoint according to the euclidean distance . ( see section [ sec : bpmg ] and [ sec : prelss ] . ) [ [ arbitrary - precision - input . ] ] arbitrary precision input . + + + + + + + + + + + + + + + + + + + + + + + + + + for our first set of results , the input coordinates are arbitrary real numbers , on which we can perform exact arithmetic operations . in this case , our new algorithm computes a motorcycle graph in @xmath0 time ( theorem [ th : general ] ) . this improves on the two subquadratic algorithms that were known before : the @xmath9-time algorithm by eppstein and erickson @xcite , which was first published in 1998 , and the @xmath18-time algorithm by cheng and vigneron @xcite , which first appeared in 2002 . we also give , in section [ sec : coriented ] , an @xmath19-time algorithm for the case of @xmath20-oriented motorcycles , where the velocities take only @xmath20 different directions . this improves on the algorithm by eppstein and erickson @xcite , which runs in @xmath0 time when @xmath21 . our last result with arbitrary precision input is an @xmath22 expected time algorithm for computing the straight skeleton of a polygon with @xmath13 vertices and @xmath2 holes . ( this result does not hold for a degenerate polygon where two reflex vertices may collide during the shrinking process , as in [ fig : degenerate ] . ) .3 , title="fig : " ] .3 , title="fig : " ] .3 , title="fig : " ] it improves on the algorithm by cheng and vigneron @xcite which runs in @xmath23 expected time . it also improves on the @xmath9 time bound of the algorithm by eppstein and erickson @xcite , but their algorithm is deterministic and applies to degenerate cases . [ [ bounded - precision - input . ] ] bounded precision input . + + + + + + + + + + + + + + + + + + + + + + + + the following results hold when all input coordinates are @xmath4-bit rational numbers . there has been recent interest in studying computational geometry problems under a bounded precision model ( the word ram ) , for instance the computation of delaunay triangulations , convex hulls , polygon triangulation and line segment intersections @xcite . we first show in section [ sec : bpmg ] that a motorcycle graph can be computed in @xmath24 time if the motorcycles move within a simple polygon , starting from its boundary . the only other non - trivial cases where we know how to compute a motorcycle graph in near - linear time seem to be the case where all velocities have positive @xmath25-coordinate , which can be solved in @xmath26 time by plane sweep , the case of a constant number of different velocity vectors @xcite , or a constant number of directions ( section [ sec : coriented ] ) . then in section [ sec : bpss ] , we show that the straight skeleton of a polygon with @xmath13 vertices and @xmath2 holes can be computed in @xmath12 expected time . this result still holds in degenerate cases . so with bounded - precision input , and if we allow randomization , it improves on the @xmath9-time algorithm by eppstein and erickson @xcite . when @xmath27 , it also improves on the @xmath10-time algorithm by cheng and vigneron @xcite , which can not handle all degenerate cases . in particular , our algorithm runs in expected @xmath28 time when @xmath29 , so it is the first near - linear time algorithm for computing the straight skeleton of a simple polygon . the previously best known algorithms run in @xmath7 time in the worst case @xcite . for any two points @xmath30 , we denote by @xmath31 the line segment between @xmath32 and @xmath33 . unless specified otherwise , @xmath31 is a closed segment . the _ relative interior _ of @xmath31 is the open segment @xmath34 . we say that two segments _ cross _ if their relative interiors intersect . the motorcycles are numbered from @xmath35 to @xmath13 . each motorcycle @xmath36 has a _ starting point _ @xmath37 , moves with constant velocity @xmath38 , and has a _ destination point _ @xmath39 that lies in the ray @xmath40 . ( see [ fig : notation]a . ) .55 , title="fig : " ] .35 , title="fig : " ] when @xmath41 , we denote by @xmath42 the time when motorcycle @xmath36 reaches @xmath32 , so @xmath43 . the _ supporting line _ @xmath44 of motorcycle @xmath36 is the line through @xmath37 with direction @xmath38 . each motorcycle @xmath36 starts at @xmath37 at time 0 , and moves at velocity @xmath38 until it meets the track left by another motorcycle and crashes , or it reaches @xmath39 and stops . so motorcycle @xmath36 crashes if it reaches a point @xmath32 such that @xmath45 , for some motorcycle @xmath46 that has not crashed or stopped earlier than @xmath47 . if motorcycle @xmath36 crashes , we denote by @xmath48 the point where it crashes , called the _ crashing point_. ( see [ fig : notation]b . ) otherwise , @xmath36 reaches @xmath39 , and we set @xmath49 . the _ trajectory _ of @xmath36 is the segment @xmath50 ; in other words it is the track of @xmath36 in the motorcycle graph . in the original motorcycle graph problem , the destination point @xmath39 is at infinity in direction @xmath38 . we can handle this case by computing a bounding box that includes all the vertices of the arrangement of the supporting lines @xmath44 , @xmath51 , and choosing as destination points the intersections of the rays @xmath40 with the bounding box . the bounding box can be computed in @xmath52 time as any extreme vertex in the arrangement is the intersection of two lines with consecutive slopes . unless specified otherwise , we make the following general position assumptions . no two motorcycles share the same supporting line , or have parallel supporting lines . no three supporting lines are concurrent . no point @xmath53 lies on @xmath54 if @xmath55 . no two motorcycles reach the same point at the same time . ( we make these assumptions so as to simplify the description of the algorithm and the proofs , but our results still hold in degenerate cases . ) the _ crossing point _ @xmath56 is the intersection between @xmath44 and @xmath54 , and thus @xmath57 . the _ size _ @xmath58 of a segment @xmath31 is the number of crossing points @xmath56 that lie in @xmath31 . ( see [ fig : size ] . ) is @xmath59 . point @xmath2 is a possible result @xmath60 of a halving query @xmath61 with @xmath62 . [ fig : size ] ] we will need a data structure to answer _ halving queries _ : given a query @xmath63 where @xmath30 are points on the supporting line @xmath44 , find a point @xmath64 such that @xmath65 and @xmath66 , for a constant @xmath67 . in addition , we require that @xmath2 is not a crossing point , and that both @xmath68 and @xmath69 are strictly smaller than @xmath58 if @xmath70 . in this section , we present our algorithm for computing motorcycle graphs , as well as its proof of correctness and analysis . an example of the execution of this algorithm on a set of 4 motorcycles is given in appendix [ sec : example ] . our algorithm maintains , for each motorcycle @xmath36 , a _ confirmed track _ @xmath71 , and a _ tentative track _ @xmath72 , such that @xmath73 and @xmath74 . so the tentative track is at least as long as the confirmed track . as we will show in the next section , the confirmed track is a subset of the trajectory , so we have @xmath75 at any time during the execution of the algorithm . the tentative track , however , may go beyond @xmath48 . ( see appendix [ sec : example ] . ) our algorithm will ensure that no two tentative tracks cross . we keep all the tentative tracks in a ray shooting data structure , so that we can enforce this invariant by checking for intersection each time we try to extend a tentative track . this data structure returns the first tentative track hit by a query ray @xmath76 , if any . we also build a data structure to answer halving queries , which will be used to shorten tentative tracks and keep them disjoint . our algorithm builds the motorcycle graph by extending the confirmed tracks until they form the whole motorcycle graph . we may also update the tentative track of a motorcycle when we extend its confirmed track . a set of _ target points _ is associated with each motorcycle @xmath36 . in particular , we maintain in a stack @xmath77 the set of target points that lie beyond the confirmed track of motorcycle @xmath36 , thus @xmath78 . in other words , @xmath77 records the target points that motorcycle @xmath36 has not reached yet . ( see [ fig : tracks ] . ) , ( m ) . the confirmed tracks are solid , and the tentative tracks are dashed . for motorcycle 1 , the confirmed track and the tentative track go to @xmath79 . the stack @xmath80 only records @xmath81 . for motorcycle 2 , the confirmed track ends at @xmath82 , the tentative track ends at @xmath83 , and @xmath84 . for motorcycle 3 , we have @xmath85 , @xmath86 , and @xmath87 . for motorcycle 4 , we have @xmath88 , @xmath89 , and @xmath90 . [ fig : tracks ] ] the stack @xmath77 is ordered from @xmath91 to @xmath39 . we denote by @xmath92 its first element , so @xmath92 is the target point in @xmath77 that is closest to @xmath91 . at the beginning , we set @xmath93 for all @xmath36 . new target points will be created in case ( 3b ) of our algorithm , as described below . if motorcycle @xmath36 has neither crashed nor stopped , then its tentative track ends at the first target point in @xmath77 , so @xmath94 . otherwise , the tentative track and the confirmed track are the same , thus @xmath95 . so after a motorcycle has crashed or stopped , the ray shooting data structure records its confirmed track . an _ event _ @xmath96 happens when a motorcycle @xmath36 reaches a target point @xmath32 we process events one by one , and while an event is being processed , new events may be generated . after an event has been processed , we process the earliest available event . as @xmath94 is the closest target point to @xmath36 in @xmath77 , it means that we always process the event @xmath97 such that @xmath98 is smallest . note that it does not imply that our simulation is done in chronological order : when we process an event @xmath97 , we may create a new event @xmath99 such that @xmath100 . ( see appendix [ sec : example ] . ) we record in a priority queue @xmath101 the event @xmath97 for each motorcycle @xmath36 that has not crashed or stopped . an event with earlier time @xmath98 has higher priority . as @xmath94 , we can update the event queue @xmath101 in @xmath4 time each time a stack @xmath77 is updated . so we can find the next available event in @xmath4 time . the first @xmath13 events are the events @xmath102 , @xmath51 , and occur at time @xmath103 . we process these @xmath13 events in an arbitrary order . we now explain how to process an event @xmath97 . to avoid confusion , for any motorcycle @xmath46 , we use the notation @xmath104 to denote the endpoints of its confirmed and tentative track just before processing this event , and we use the notation @xmath105 for their position just after processing this event . we first extend the confirmed track of motorcycle @xmath36 to @xmath106 , thus @xmath107 . we also delete @xmath106 from @xmath77 . we are now in one of the following cases : * if @xmath108 , then motorcycle @xmath36 stops . in order to avoid processing irrelevant events in the future , we remove @xmath77 from @xmath101 . * if @xmath106 is a crossing point @xmath56 that lies in the confirmed track of @xmath46 ( that is , @xmath109 ) , then @xmath36 crashes at @xmath106 . so we remove @xmath77 from @xmath101 . * otherwise , we try to extend the tentative track to the next target point @xmath110 . so we perform a ray shooting query with ray @xmath111 , which gives us the first track intersected by @xmath112 , if any . * * if @xmath112 does not cross any track , then @xmath113 , and we do not need to do anything else to handle this event . * * otherwise , let @xmath46 be the result of the ray - shooting query , so @xmath114 is the first track hit by segment @xmath112 , starting from @xmath106 . we shorten the tentative track of @xmath36 , which means that we insert the new target point @xmath56 into @xmath77 , as well as the point @xmath115 obtained by a halving query on @xmath116 . if the crossing point @xmath56 does not lie in the confirmed track of @xmath46 , that is , if @xmath117 , then we also shorten the tentative track of @xmath46 , so we insert @xmath56 into @xmath118 , and we insert @xmath119 into @xmath118 . after applying the rules above , we update the ray shooting data structure ( if needed ) , and we move to the next available event . initially , we create the target points @xmath53 for @xmath51 . after this , we create new target points only in case ( 3b ) of our algorithm . there are two types of such target points : the crossing points @xmath56 obtained by ray - shooting , and the points obtained by halving queries . we call @xmath120-targets the first type of target points , and @xmath2-targets the latter . by our assumption that the result of a halving query is not a crossing point , a target point can not be both a @xmath2-target and a @xmath120-target . we need the following lemma . remember that we say that two segments cross if their relative interiors intersect . [ lem : invariant ] during the course of the algorithm , no two tentative tracks cross . for sake of contradiction , assume that two tentative tracks cross during the course of the algorithm . let @xmath97 be the first event that generates such a crossing , so just before processing this event , the tentative tracks @xmath114 , @xmath121 do not cross , and there is a crossing among the tracks @xmath122 . we must be in case ( 3 ) , because we do not extend any tentative track in cases ( 1 ) and ( 2 ) . besides , we only extend the track of motorcycle @xmath36 in case ( 3 ) . so there must be another motorcycle @xmath123 such that @xmath124 crosses @xmath125 . in case ( 3a ) , the segment @xmath112 obtained by ray shooting does not cross any tentative track @xmath114 , @xmath126 , and since @xmath113 , then the new portion @xmath127 of the track does not cross any other tentative track . the same is true in case ( 3b ) , because @xmath128 is in @xmath116 , where track @xmath46 is the first track hit by @xmath112 . so we just proved that , in any case , the new portion @xmath127 of the track does not cross any track @xmath114 , @xmath126 , and in particular , @xmath127 does not cross @xmath125 . by our assumption , we also know that @xmath125 can not cross @xmath72 . so the only remaining possibility is that @xmath125 crosses @xmath129 at @xmath106 . then @xmath106 is the crossing point @xmath130 . this point @xmath131 can not be in the confirmed track @xmath132 , because that would be case ( 2 ) of our algorithm , and we showed that we are in case ( 3 ) . since @xmath130 is a @xmath120-target of @xmath36 , and it does not lie in the confirmed track of @xmath133 , then it must have been inserted at the same time in @xmath77 and @xmath134 while processing a previous event . since @xmath130 is not on the confirmed track of @xmath133 , then it must still be in @xmath134 . so the tentative track @xmath125 can not contain @xmath130 in its relative interior , a contradiction . we want to argue that our algorithm computes the motorcycle graph correctly . so assume it is not the case . as our algorithm moves motorcycles forward until they either reach their destination point or crash , it could only fail if during the execution of our algorithm , the confirmed track of at least one motorcycle @xmath36 goes beyond the point @xmath48 where it is supposed to crash in the motorcycle graph . let us consider the event @xmath97 that is first processed by our algorithm , such that motorcycle @xmath36 goes beyond @xmath48 . so @xmath48 is in the segment @xmath135 . let @xmath46 denote the motorcycle that @xmath36 crashes into , in the ( correct ) motorcycle graph , so @xmath136 . when we process @xmath97 , in the current graph constructed by our algorithm , motorcycle @xmath46 can not have reached @xmath137 , because it would mean that the tentative tracks @xmath72 and @xmath138 are crossing at @xmath136 , which is impossible by lemma [ lem : invariant ] . we now rule out the case where , when our algorithm processes @xmath97 , motorcycle @xmath46 has already crashed into some motorcycle @xmath133 in the graph constructed by the algorithm . ( see [ fig : correctness2 ] . ) for sake of contradiction , assume it did happen . ] * if we had @xmath139 , then @xmath56 would have been created as a @xmath120-target for @xmath46 earlier . at this point , @xmath36 had not gone past @xmath56 , because @xmath97 is the first such event . as @xmath140 , the algorithm would have moved @xmath46 to @xmath56 before @xmath36 moves further , and thus @xmath46 would not crash at @xmath56 , a contradiction . * thus we must have @xmath141 . as @xmath106 is beyond @xmath136 , and tentative tracks can not cross , we must have @xmath142 . so @xmath46 crashed into @xmath133 at @xmath143 . as in the correct motorcycle graph , @xmath46 does not crash into @xmath133 , it means that the algorithm has already moved @xmath133 past its ( correct ) crashing point , which contradicts our assumption that @xmath97 was the first such event . we just proved that @xmath46 has not stopped or crashed when the algorithm processes event @xmath97 , so at this point there should be an event @xmath144 in the queue . by lemma [ lem : invariant ] , the tracks @xmath72 and @xmath114 can not cross , so we must have @xmath145 . ( see [ fig : correctness ] . ) ] it implies that @xmath146 . but since @xmath36 crashes into @xmath46 in the ( correct ) motorcycle graph , we must have @xmath147 , thus @xmath148 . as @xmath149 , we have @xmath150 , thus @xmath151 . but this is impossible , because our algorithm always processes the earliest available event , so it would have processed @xmath144 rather than @xmath97 . our algorithm uses two auxiliary data structures : for answering halving queries , and for ray shooting . the running time of our algorithm depends on their preprocessing time and query time . let @xmath152 denote an upper bound on the preprocessing time of these two data structures , and let @xmath153 denote an upper bound on the time needed for a query or update so we can answer a ray - shooting query or a halving query in time @xmath153 , and we can update the ray shooting data structure in time @xmath153 . we now prove the following result : [ th : analysis ] we can compute a motorcycle graph of size @xmath13 in time @xmath154 . each time we handle an event , we perform at most two halving queries , one ray - shooting query , and we may update two tentative tracks in the ray - shooting data structure . we also pay an @xmath4 time overhead to update the priority queue @xmath101 . so after preprocessing , the running time will be at most the number of events times @xmath155 . thus we only need to argue that our algorithm processes a total of @xmath52 events . in fact , at each event we process , a motorcycle reaches a target point , so we only need to show that @xmath52 target points are created during the course of the algorithm . initially , we create @xmath156 target points , which are @xmath53 for @xmath51 . after this , we only create new target points in case ( 3b ) of the algorithm . in this case , we create one @xmath120-target , and at most two @xmath2-targets obtained by halving . thus we only need to bound the number of @xmath120-targets . at the end of the algorithm , some of these @xmath120-targets @xmath56 correspond to an actual crash , with motorcycle @xmath36 crashing into @xmath46 , or @xmath46 crashing into @xmath36 . in any case , there are at most @xmath13 such @xmath120-targets . we need to consider the other @xmath120-targets , that do not correspond to an actual crash . in this case , either motorcycle @xmath36 or @xmath46 does not reach @xmath56 , so at the end of the computation , @xmath56 must appear in the stack @xmath77 or @xmath118 of target points that have not been reached by motorcycle @xmath36 or @xmath46 , respectively . thus , in order to complete the proof of theorem [ th : analysis ] , we only need the following lemma . [ lem : targets ] at the end of the execution of our algorithm , for any motorcycle @xmath36 , the number of @xmath120-targets in @xmath77 is @xmath4 . in this proof , we only consider the status of the stack @xmath77 at the end of the algorithm , and we assume that it contains more than one @xmath120-target . we denote by @xmath157 the @xmath120-targets in @xmath77 , in reverse order , so @xmath158 is a subsequence of @xmath77 , where @xmath159 is closest to @xmath48 and @xmath160 is closest to @xmath39 . each target @xmath161 was created in case ( 3b ) of our algorithm . at the same time , an @xmath2-target @xmath162 was created by a halving query using another target point @xmath163 . as the points @xmath161 , @xmath164 are in @xmath77 , motorcycle @xmath36 never reaches these points during the course of the algorithm , so @xmath160 and @xmath165 must have been created first , then @xmath166 and @xmath167 and finally @xmath159 and @xmath168 . for any @xmath169 , as @xmath161 is created after @xmath170 , and these two points are created when motorcycle @xmath36 reaches @xmath163 and @xmath171 , respectively , it implies that @xmath171 is in @xmath172 . we also know that @xmath170 lies in @xmath173 , because @xmath170 appears after @xmath161 in @xmath77 . , @xmath164 is a sequence of nested segments , that is , we have @xmath175 for all @xmath176 . more precisely : * if @xmath177 is in @xmath77 , then @xmath178 , because @xmath161 is created after @xmath177 , and motorcycle @xmath36 never reaches @xmath177 . ( see [ fig : targets1 ] . ) * if @xmath177 is not in @xmath77 , then @xmath179 . ( see [ fig : targets2 ] . ) it can be proved as follows . as @xmath177 is created at the same time as @xmath170 , then @xmath161 is created after @xmath177 . so @xmath161 must have been created after motorcycle @xmath36 reaches @xmath177 , otherwise we would have @xmath180 , and since motorcycle @xmath36 reaches @xmath177 later , @xmath161 would not be in @xmath77 . as @xmath161 is created after motorcycle @xmath36 reaches @xmath177 , we must have @xmath181 . , first case.[fig : targets1 ] ] , second case.[fig : targets2 ] ] thus @xmath174 is contained in either @xmath182 or @xmath183 , and since @xmath184 , it follows that the size @xmath185 decreases exponentially when @xmath46 increases from 1 to @xmath186 . as @xmath187 contains @xmath188 and @xmath189 , we have @xmath190 . in addition , @xmath191 , so we must have @xmath192 . our algorithm , presented in section [ sec : algorithm ] , requires two auxiliary structures . the first one is simply a ray - shooting data structure . as ray shooting is a standard operation in computational geometry , we will be able to directly use known data structures . the second data structure we need is for answering halving queries . we show below how to construct efficient data structures for this type of queries , and the corresponding time bounds for our motorcycle graph algorithm . in this section , we present the auxiliary data structures for the most general case , as presented in section [ sec : notation ] . so motorcycles have arbitrary starting position , destination point and velocity . for ray shooting , we can directly use a data structure by agarwal and matouek @xcite , which requires preprocessing time @xmath0 , with update and query time @xmath193 , for any @xmath1 . for halving queries , we use known range searching data structures and parametric search , as in the work of agarwal and matouek on ray shooting : our problem is an optimization version of range counting in an arrangement of lines , so we obtain the same bounds ( * ? ? ? * section 3.1 ) . [ lem : halving ] given the @xmath13 supporting lines @xmath194 , we can construct a data structure with @xmath0 preprocessing time and @xmath193 query time that answers the following queries @xmath63 . assume there are @xmath133 crossing points @xmath56 on @xmath31 . then we return the median crossing point and the next : the @xmath195th and the @xmath196th such crossing point , in the ordering from @xmath32 to @xmath33 along @xmath31 . with the two auxiliary data structures above , theorem [ th : analysis ] yields the following result . [ th : general ] a motorcycle graph can be computed in @xmath0 time , for any @xmath1 . it should be possible to replace the @xmath197 factor in the bounds of lemma [ lem : halving ] with a polylogarithm using known range searching techniques @xcite , because we only need a static data structure for halving queries , but in any case we need a dynamic data structure for ray shooting queries , so it would not improve our overall time bounds . we consider the special case where motorcycles can only take @xmath20 different directions @xmath198 . eppstein and erickson gave an @xmath0-time algorithm when @xmath21 . we show that with appropriate auxiliary data structures , we can solve this case in time @xmath28 . in the following , we do not assume that @xmath21 , so our time bounds will also have a dependency on @xmath20 . [ cor : coriented ] we can compute a @xmath20-oriented motorcycle graph in @xmath199 time . we use the following data structures , and then the result follows from theorem [ th : analysis ] . [ [ ray - shooting - data - structures . ] ] ray shooting data structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a first approach to answer our ray shooting queries is to use @xmath20 instances of a data structure for vertical ray shooting in a planar subdivision . several data structures are known for this problem @xcite , we use a data structure by cheng and janardan @xcite that takes @xmath200 time per update and @xmath4 time per query . so overall , we get @xmath201 with the terminology of theorem [ th : analysis ] . alternatively , we can use @xmath202 instances of a data structure for vertical ray shooting among horizontal segments . each data structure is used to answer ray shooting queries with a given direction , into segments with another direction : we just need to change the two coordinate axis to these two directions . using a recent result by giyora and kaplan @xcite , we obtain @xmath203 . [ [ halving - queries . ] ] halving queries . + + + + + + + + + + + + + + + + our data structure for halving queries simply consists of a sorted list of motorcycles for each direction . so for each @xmath204 , we have an array @xmath205 of the motorcycles with direction @xmath206 , sorted according to the intercept of their supporting lines with a line orthogonal to @xmath206 . we now explain how to answer a halving query @xmath207 . without loss of generality , assume @xmath44 has direction @xmath208 . for each direction @xmath206 , @xmath209 , the subset of motorcycles whose supporting lines cross @xmath31 appear in consecutive positions in @xmath205 . we can find the first and the last index of these lines in @xmath4 time by binary search . so we obtain all the arrangement vertices in @xmath31 in @xmath210 sorted subarrays . we then compute the median @xmath211 of each such subarray @xmath212 , and the median of these points @xmath211 weighted by the number of points @xmath213 in the corresponding subarray . this gives a halving point @xmath60 with @xmath214 . the median of each subarray can be found in @xmath215 time , and their weighted median in @xmath216 time @xcite , so the query time is dominated by the @xmath20 binary searches . thus , we can answer halving queries in @xmath217 time . the data structure for answering halving queries in section [ sec : general ] is quite involved . in practice , one would rather implement halving queries by simply halving the euclidean length @xmath218 instead of approximately halving the number of crossing points . unfortunately , in the infinite precision model that is commonly used in computational geometry , this would cause the analysis of our algorithm in lemma [ lem : targets ] to break down , because a stack of target points @xmath77 may have size @xmath219 at the end of the algorithm . such a counterexample would require the distance between consecutive target points in @xmath77 to become exponentially small near the crashing point , which does not seem likely to happen in practice . to formalize this idea , we make the assumption that all input numbers ( the coordinates of the starting points , the destination points , and the velocities ) are rational numbers , whose numerator and denominator are in @xmath220 for some integer @xmath221 . in other words , the input numbers are @xmath221-bit signed integers . we still assume that arithmetic operations between two numbers can be performed in constant time . this model also allows us to handle the case where input coordinates are @xmath221-bit rational numbers , that is , rational numbers with @xmath221-bit numerator and denominator ; we just need to scale up each coordinate by a factor @xmath222 to obtain @xmath223-bit integers , losing only a constant factor in our time bounds . in the proofs below we assume the input numbers are integers , to simplify the presentation , but the results are stated with rational coordinates . as the input coordinates are @xmath221-bit integers , the coordinates of a crossing point @xmath56 are rational numbers obtained by solving a @xmath224 linear system , their denominator being the determinant @xmath225 . thus , the denominator is an integer between @xmath226 and @xmath227 . so any two distinct crossing points are at distance at least @xmath228 from each other . assume that we replace our halving operation , as defined in section [ sec : notation ] , with halving the euclidean length . so @xmath60 is the midpoint of @xmath31 , which can be computed in constant time . then any nested sequence of segments obtained by successive halving , as in the proof of lemma [ lem : targets ] , consists of @xmath229 such nested segments , because a segment @xmath174 of length smaller than @xmath228 can not contain another crossing point in its interior , and hence it can not be further subdivided . so the bound on the size of @xmath77 becomes @xmath229 , and we get the following result . [ th : bpmotorcycle ] if the input coordinates are @xmath221-bit rational numbers , we can compute a motorcycle graph in time @xmath230 , where @xmath231 is the time needed for updates or queries in the ray - shooting data structure . for bounded - precision input , the bottleneck of our algorithm has thus become the ray shooting data structure , whose update and query time bound is @xmath193 in the most general case . therefore , we obtain a faster motorcycle graph algorithm if we are in a special case where faster ray - shooting data structures are known . one such case is ray - shooting in a connected planar subdivision , which can be done in @xmath200-time per update and query using a data structure by goodrich and tamassia @xcite . we can use this data structure if , for instance , all motorcycles move inside a simple polygon @xmath8 , starting from its boundary . ( so for all @xmath36 , @xmath232 , and @xmath37 is on the boundary of @xmath8 . ) then we perform ray shooting in the union of the tentative tracks and the edges of @xmath8 , which form a connected subdivision . it yields the following time bound . [ th : bpsmotorcycle ] we can compute a motorcycle graph in time @xmath28 for @xmath13 motorcycles moving inside a simple polygon with @xmath156 vertices , starting on its boundary , and if the input has @xmath4-bit rational coordinates . in this section , we give new results on straight skeleton computation , using our new motorcycle graph algorithm . as we mentioned in the introduction , the straight skeleton problem and the motorcycle graph problem are closely related . we now explain it in more details . consider the reflex ( non - convex ) vertices of a polygon @xmath8 . when we construct the straight skeleton of @xmath8 , these vertices move inward and may collide into edges , or other vertices . these events , called split events and vertex events , are the difficult part of straight skeleton computation , because they affect the topology of the shrinking polygon by splitting it , and because they are non - local : a reflex vertex may affect a chain of edges on the other side of the polygon . the other type of events , called edge events , where an edge shrinks to a point , are easily handled with a priority queue . so the interaction between reflex vertices is a crucial part in straight skeleton computation , and the motorcycle graph presented below helps to determine these interactions . the motorcycle graph _ induced _ by a polygon @xmath8 is such that each motorcycle starts at a reflex vertex , moves as the same velocity as the corresponding reflex vertex when we shrink @xmath8 , and stops if it reaches the boundary @xmath233 of @xmath8 . ( see [ fig : mgss ] . ) if @xmath8 is degenerate , then two reflex vertices may collide and create a new reflex vertex . in this case we need to create a new motorcycle after the collision . ( see [ fig : degenerate ] . ) so when two motorcycles collide in the induced motorcycle graph , we may have to create a new motorcycle @xcite . our motorcycle graph algorithm , as described above , does not apply directly to this case , because the proof of lemma [ lem : targets ] breaks down . ( the reason is that @xmath77 may hold a linear number of target points at the end of the execution of the algorithm , due to the newly created motorcycles . see [ fig : counterexample ] . ) holds a linear number of target points at the end of execution of the algorithm . the speed of the motorcycle at the bottom are adjusted so that @xmath160 is created first , then @xmath166 [fig : counterexample ] ] in section [ sec : bpss ] , we will explain how to compute these generalized motorcycle graphs efficiently on bounded - precision input . but the following theorem still holds in degenerate cases . [ th : reduction ] the straight skeleton of a polygon @xmath8 with @xmath13 vertices and @xmath2 holes can be computed in expected time @xmath14 if we know the motorcycle graph induced by the vertices of @xmath8 . from the discussion above , and using our motorcycle graph algorithm from theorem [ th : general ] , we obtain the following result . [ cor : motorgeneral ] we can compute the straight skeleton of a non - degenerate polygon with @xmath13 vertices and @xmath2 holes in @xmath234 time for any @xmath1 . we use the same bounded precision assumptions as in section [ sec : bpmg ] , where the input coordinates are @xmath221-bit integers or , equivalently , @xmath221-bit rational numbers . similarly , to simplify the presentation , we use the integer model in the proofs , but we state the results in the rational model . thus , the coordinates of the vertices of the input polygon @xmath8 are @xmath221-bit integers . the supporting lines @xmath44 of the motorcycles are angle bisectors between two edges of the input polygon . in order to apply the same halving scheme as in section [ sec : bpmg ] , where the euclidean length is used instead of the number of arrangement vertices , we need to argue that the separation between two vertices in this arrangement of bisectors can not be too small . this distance can be shown to be at least @xmath235 , where @xmath236 , by applying the separation bound by burnikel et al . so we obtain a result for induced motorcycle graphs that is analogous to theorem [ th : bpmotorcycle ] . [ lem : bpinduced ] given a polygon @xmath8 whose input coordinates are @xmath221-bit rational numbers , we can compute the motorcycle graph induced by @xmath8 in time @xmath230 , where @xmath231 is the time needed for updates or queries in the ray - shooting data structure . in the lemma statement above , we do not exclude degenerate cases . this is another advantage of this bounded precision model . as the argument in our analysis only relies on the separation bound between two distinct crossing points , and not on the number of motorcycles crossing a given segment , a newly created motorcycle does not affect our analysis as it still obeys the same separation bound : a newly created motorcycle still lies on the bisector of two input edges , though these two edges are not adjacent in the input polygon @xcite . ( see [ fig : degenerate ] . ) we still need to describe an efficient ray - shooting data structure . as our input polygon has @xmath2 holes , the boundary @xmath233 of @xmath8 together with the tentative tracks form a collection of @xmath237 disjoint simple polygons . we could directly use known ray - shooting data structures @xcite , which can be made dynamic at the expense of an extra @xmath197 factor in the running time @xcite . in the following , we give a different approach , which leads to a better time bound when used as a subroutine of our algorithm . this approach takes advantage of the fact that the holes of @xmath8 are fixed ( only the tentative tracks are dynamic ) . we use a spanning tree with low crossing number , which is not a new idea in ray - shooting data structures @xcite . we pick one point on the boundary of each hole of @xmath8 , and on the boundary of @xmath8 . we connect these @xmath237 points using a spanning tree @xmath238 with low stabbing number @xcite , that is , a spanning tree such that any line crosses at most @xmath239 edges of @xmath240 . this tree can be computed in @xmath241 time ( * ? ? ? * section 8) . we maintain a polygonal subdivision which is the overlay of @xmath8 with @xmath238 and the tentative tracks . so at each intersection between an edge of @xmath238 and an edge of @xmath8 or a tentative track , we split the corresponding edges and tracks at the intersection point . this subdivision @xmath242 is connected and has @xmath243 edges , and we maintain it in the ray shooting data structure by goodrich and tamassia @xcite , which has @xmath200 update and query time . each time a tentative track is extended or retracted , as a tentative track intersects @xmath239 edge of @xmath238 , we can update the subdivision and the data structure by making @xmath239 updates in the ray shooting data structure . similarly , when our motorcycle graph algorithm tries to extend a tentative track , we can find the first tentative track being hit by a query ray in @xmath244 time : we first perform a ray shooting query in @xmath242 , which takes @xmath200 time . if we hit an edge of @xmath238 , we make another ray shooting query starting at the hitting point of the previous query , and in the same direction . we repeat this process as long as the result of the query is an edge of @xmath238 , and by the low - stabbing number property , it may only happen @xmath239 times . overall , our ray shooting data structure has update and query time @xmath245 . so by theorem [ th : reduction ] and lemma [ lem : bpinduced ] , we obtain the following result . note that it still holds for degenerate input . [ th : bpss ] the straight skeleton of a polygon with @xmath13 vertices and @xmath2 holes , whose coordinates are @xmath4-bit rational numbers , can be computed in @xmath12 expected time . 10 p. agarwal and j. erickson . geometric range searching and its relatives . in b. chazelle , j. e. goodman , and r. pollack , editors , _ advances in discrete and computational geometry _ , volume 23 , page 156 . american mathematical society , 1998 . p. agarwal and j. matouek . ray shooting and parametric search . , 22(4):794806 , 1993 . p. agarwal and m. sharir . applications of a new space - partitioning technique . , 9(1):1138 , 1993 . p. agarwal and m. sharir . ray shooting amidst convex polygons in 2d . , 21(3):508519 , 1996 . o. aichholzer , f. aurenhammer , d. alberts , and b. grtner . a novel type of skeleton for polygons . , 1(12):752761 , 1995 . l. arge , g. s. brodal , and l. georgiadis . improved dynamic planar point location . in _ proc 47th symposium on foundations of computer science _ , pages 305314 , 2006 . g. barequet , m. goodrich , a. levi - steiner , and d. steiner . straight - skeleton based contour interpolation . in _ proc . 14th acm - siam symposium on discrete algorithms _ , pages 119127 , 2003 . k. buchin and w. mulzer . delaunay triangulations in _ o_(sort(_n _ ) ) time and more . , 58(2):6 , 2011 . c. burnikel , r.fleischer , k. mehlhorn , and s. schirra . a strong and easily computable separation bound for arithmetic expressions involving square roots . in _ proc 8th acm - siam symposium on discrete algorithms _ , pages 702709 , 1997 . f. cacciola . a cgal implementation of the straight skeleton of a simple 2d polygon with holes . in _ 2nd cgal user workshop _ http://www.cgal.org/userworkshop/2004/straight_skeleton.pdf . t. chan . optimal partition trees . , 47(4):661690 , 2012 . t. chan and m. patrascu . transdichotomous results in computational geometry , i : point location in sublogarithmic time . , 39(2):703729 , 2009 . b. chazelle , h. edelsbrunner , m. grigni , l. guibas , j. hershberger , m. sharir , and j. snoeyink . ray shooting in polygons using geodesic triangulations . , 12(1):5468 , 1994 . cheng and r. janardan . new results on dynamic planar point location . , 21(5):972999 , 1992 . cheng and a. vigneron . motorcycle graphs and straight skeletons . , 47(2):159182 , 2007 . f. cloppet , j. oliva , and g. stamon . angular bisector network , a simplified generalized voronoi diagram : application to processing complex intersections in biomedical images . , 22(1):120128 , 2000 . t. cormen , c. leiserson , r.rivest , and c. stein . . the mit press , 3 edition , 2009 . d. eppstein and j. erickson . raising roofs , crashing cycles , and playing pool : applications of a data structure for finding pairwise interactions . , 22(4):569592 , 1999 . d. eppstein , m. goodrich , e. kim , and r. tamstorf . motorcycle graphs : canonical quad mesh partitioning . , 27(5):14771486 , 2008 . j. erickson . crashing motorcycles efficiently . http://www.cs.uiuc.edu/~jeffe/open/cycles.html , 1998 . p. felkel and . obdrlek . straight skeleton implementation . in _ proc . 14th spring conference on computer graphics _ , pages 210218 , 1998 . y. giyora and h. kaplan . optimal dynamic vertical ray shooting in rectilinear planar subdivisions . in _ proc . 18th acm - siam symposium on discrete algorithms _ , pages 1928 , 2007 . m. goodrich and r. tamassia . dynamic ray shooting and shortest paths in planar subdivisions via balanced geodesic triangulations . , 23(1):5173 , 1997 . j. hershberger and s. suri . a pedestrian approach to ray shooting : shoot a ray , take a walk . , 18(3):403431 , 1995 . s. huber and m. held . theoretical and practical results on straight skeletons of planar straight - line graphs . in _ proc . 27th symposium on computational geometry _ , pages 171178 , 2011 . t. kelly and p. wonka . interactive architectural modeling with procedural extrusions . , 30(2):14:114:15 , 2011 . r. laycock and a. day . automatically generating large urban environments based on the footprint data of buildings . in _ proc . 8th acm symposium on solid modeling and applications _ , pages 346351 , 2003 . j. matouek . efficient partition trees . , 8:315334 , 1992 . j. oliva , m. perrin , and s. coquillart . 3d reconstruction of complex polyhedral shapes from contours using a simplified generalized voronoi diagram . , 15(3):397408 , 1996 . p. palfrader , m. held , and s. huber . on computing straight skeletons by means of kinetic triangulations . in _ proc . 20th european symposium on algorithms _ , pages 766777 , 2012 . f. preparata and m. shamos . . springer , berlin , 1985 . g. von peschka . . buschak & irrgang , brnn , 1877 . in this section , we give the pseudocode of our algorithm . it is more detailed than the algorithm description in section [ sec : algorithm ] , and it can handle degenerate cases . the proof of correctness and the analysis are essentially the same as in section [ sec : main ] , but they require a more detailed case analysis . to deal with the degenerate cases where some supporting lines are concurrent , or two or more motorcycles reach a point at the same time , we record all the target points created so far in a dictionary data structure @xmath246 . we can implement @xmath246 as a balanced binary search tree , sorted in lexicographical order of the coordinates @xmath247 , which allows to retrieve a point in @xmath4 time . we associate three fields with each point @xmath32 stored in @xmath246 : * a list @xmath248 records the motorcycles @xmath36 such that @xmath249 . so @xmath248 records all the motorcycles @xmath36 that could possibly reach @xmath32 , at a given point of the execution of our algorithm . the set @xmath248 itself is stored in a dictionary data structure , so that we can decide in @xmath4 time whether a motorcycle @xmath36 is in @xmath248 . * two motorcycles @xmath250 of @xmath248 such that @xmath251 are smallest . it will allow us to find out whether two motorcycles crash simultaneously at @xmath32 . * a flag @xmath252 which is set to false initially , and is set to true as soon as a confirmed track has reached @xmath32 , implying that any other motorcycle that reaches @xmath32 must crash . after the initialization stage , our algorithm handles repeatedly the earliest available event , according to the four cases ( 1 ) , ( 2 ) , ( 3a ) and ( 3b ) from section [ sec : algorithm ] . lines [ line : check_0 ] and [ line : check_1 ] deal with case ( 1 ) and ( 2 ) . the condition @xmath253 corresponds to case ( 1 ) . the other two conditions check whether we are in case ( 2 ) . in particular , condition @xmath254 means that at least one other motorcycle has reached @xmath32 , thus motorcycle @xmath36 crashes . with degenerate input , it is possible that another ( or several other ) motorcycle reaches @xmath32 at the same time as @xmath36 , in which case @xmath255 if @xmath96 is the first event involving @xmath32 that has been processed . the condition at line [ line : check_1 ] checks whether we are in this situation . if so , @xmath36 must crash . as @xmath96 is the first event involving @xmath32 that we process , there is no earlier event in @xmath248 , so we can find another motorcycle that reaches @xmath32 at the same time in constant time using the second field associated with @xmath32 in @xmath246 . case ( 3a ) corresponds to a positive answer to the test at line [ line : check_2 ] . the condition @xmath256 detects whether the track of @xmath36 to the next target points hits any other track . the other condition @xmath257 and @xmath258 , checks for a boundary case , where @xmath92 falls on another tentative track . the test is positive if @xmath259 has already been identified as a target point of @xmath46 before . in this case we only extend the tentative track of @xmath36 , without doing any unnecessary halving . line [ line : check_3 ] branches to case ( 3b ) . similar to line [ line : check_2 ] , we do not perform an unnecessary halving operation when @xmath260 or @xmath261 . our pseudocode does not handle explicitly the case where two motorcycles have same supporting line . these cases can be easily handled by ad - hoc arguments @xcite . one way of doing it is to insert additional target points at initialization . for each supporting line shared by several motorcycles , between any two consecutive motorcycles @xmath262 along this line that go toward each other , we insert their potential collision point , that is , we insert into @xmath77 and @xmath118 the point @xmath32 such that @xmath263 . for each motorcycle @xmath36 along this line , if the first starting point @xmath264 in the ray @xmath40 is in @xmath265 , we also update @xmath39 to be @xmath264 . initialize the dictionary @xmath246 . set @xmath266 , @xmath267 and @xmath268 . insert @xmath37 and @xmath39 into @xmath246 ; insert motorcycle @xmath36 into @xmath269 and @xmath270 . initialize the event queue @xmath271 and the data structures for ray - shooting and halving . let @xmath96 be the earliest available event . set @xmath272 . pop @xmath92 from @xmath77 . [ line : check_1 ] motorcycle @xmath36 crashes . set @xmath273 . remove @xmath36 from @xmath274 for all @xmath275 . remove @xmath77 from @xmath271 . [ line : check_2 ] set @xmath277 . [ line : check_3 ] [ line : check_4 ] set @xmath277 . [ line : check_5 ] push @xmath259 into @xmath77 . push @xmath278 into @xmath77 . set @xmath279 . insert @xmath36 into @xmath280 and @xmath281 . [ line : check_6 ] push @xmath259 into @xmath118 . push @xmath282 into @xmath118 . set @xmath283 . insert @xmath46 into @xmath280 and @xmath281 . set @xmath284 . we give an example of the execution of our algorithm on a set of 4 motorcycles . ( confirmed tracks are solid , and tentative tracks are dotted . ) it demonstrates two features of our algorithm , that were mentioned above . * a tentative track may be longer than the final track in the motorcycle graph . for instance , the tentative track @xmath285 in ( b ) is longer than the final track @xmath286 in ( q ) . * our algorithm does not construct the motorcycle graph in chronological order . for instance , in ( i ) , motorcycle 2 is moved to @xmath287 , which is its position at time @xmath288 . then in ( k ) , motorcycle 3 is moved to @xmath289 , which is its position at time @xmath290 . the four motorcycles @xmath291 and @xmath292 start at time @xmath293 at initial points @xmath294 , @xmath295 , @xmath296 and @xmath297 . their velocities are @xmath298 , @xmath299 , @xmath300 , and @xmath301 . we use the halving scheme as specified in section [ sec : notation ] with @xmath62 . so for instance , we create @xmath289 in ( j ) by halving @xmath302 . there are three crossings along this segment : @xmath303 . then @xmath289 is created as a point between @xmath304 and @xmath305 , in this case we just use the midpoint .

we study the number of measurements required for quantum process tomography under prior information , such as a promise that the unknown channel is unitary . we introduce the notion of an interactive observable and we show that any unitary channel acting on a-level quantum system can be uniquely identified among all other channels ( unitary or otherwise ) with only interactive observables , as opposed to the required for tomography of arbitrary channels . this result generalizes , so that channels with at most kraus operators can be identified with only interactive observables . slight improvements can be obtained if we wish to identify such a channel only among unital channels or among other channels with kraus operators . these results are proven via explicit construction of large subspaces of hermitian matrices with various conditions on rank , eigenvalues , and partial trace . our constructions are built upon various forms of totally nonsingular matrices . process tomography , unitary channels , high - rank subspaces

1309.0840

the problem of deducing the action of an unknown quantum channel by gathering statistics from repeated measurement is called _ process tomography_. in general , one must measure a set of @xmath2 distinct observables in order to uniquely identify a given channel @xmath5 acting on a @xmath0-level quantum system , owing to the fact that the set of all such channels spans an affine space of dimension @xmath6 . however , it could be that far fewer observables are required if we are promised that @xmath5 belongs to some prescribed set @xmath7 of channels or if we only wish to identify @xmath5 among some specific subset @xmath8 of all possible channels . in other words , we are interested in the following question for various sets @xmath9 of channels : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ given an unknown channel @xmath10 , what is the minimum number of observables required to ensure that there is no other channel in @xmath8 with the same measurement statistics as @xmath5 ? _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we show that if @xmath5 is promised to be a _ unitary _ channel then it suffices to measure only @xmath1 observables in order to uniquely identify @xmath5 among all channels , and that slight improvements can be obtained if we wish to identify @xmath5 only among unital or unitary channels . in particular , 1 . [ it : result : unitary ] @xmath11 observables suffice to identify a unitary among other unitaries . if @xmath12 then only 6 observables suffice @xmath13 observables suffice to identify a unitary among all channels . @xmath14 observables suffice to identify a unitary among unital channels . these results generalize to the problem of identifying channels with at most @xmath3 kraus operators ; we show that it suffices to measure @xmath4 observables , with slight improvements if we wish to identify such a channel only among unital channels or among other channels with @xmath3 kraus operators . a detailed statement of these results is given in theorem [ thm : unitary_uda_udp ] of section [ sec : results : tomography ] . several potential extensions of these results are described in section [ sec : future ] . our work is inspired by previous results on the number of observables required for _ state _ tomography , the study of which was initiated by heinosaari , mazzarella , and wolf @xcite who showed that any pure state of a @xmath0-level quantum system can be uniquely identified among other pure states with only @xmath15 observables , as opposed to the @xmath1 required for tomography of an arbitrary state . in particular , @xmath16 observables suffice and this bound generalizes to the problem of identifying states with rank at most @xmath3 among other states with rank at most @xmath3 . chen , dawkins , ji , johnston , kribs , shultz , and zeng investigated the number of observables required in order to identify a given pure state among _ all _ states @xcite . those authors showed that @xmath15 observables suffice even in this more demanding setting . in particular , @xmath16 observables no longer suffice but @xmath17 observables is sufficient . as in ref . @xcite , this bound generalizes to the problem of identifying states with rank at most @xmath3 among all states . this work was continued by carmeli , heinosaari , schultz , and toigo @xcite , who slightly improved the number of observables that suffice in this setting in some cases . at the time of this writing we were aware of prior results on the number of observables required for process tomography only under certain sparsity assumptions @xcite , but not under the promise that the unknown channel is unitary or has few kraus operators . additional prior work was subsequently brought to our attention after the journal version of this article was published . most notably , flammia , gross , liu , and eisert used compressed sensing techniques to show that a channel with @xmath3 kraus operators can be identified with @xmath18 observables , which is already optimal to within a logarithmic factor of @xmath0 @xcite . their article contains a wealth of references on state tomography under prior information , and the idea for their approach to process tomography was described as early as 2009 @xcite . moreover , their interactive observables can be implemented by preparing only eigenstates of pauli operators and by measuring only pauli observables , which tend to admit relatively easy implementations in the laboratory . along similar lines , kimmel , da silva , ryan , johnson , and ohki showed that the unital part of any multi - qubit channel can be deduced from clifford measurements and that their procedure is robust in the presence of measurement error @xcite . by contrast to this prior work , our primary goal is to minimize the number of observables required for tomography without regard to how those observables might be implemented . however , in section [ sec : future : experimental ] we observe that the six observables required for tomography of qubit unitary channels are in fact products of clifford operators . the primary results of section [ sec : intro : results ] rest upon several supplementary results on interactive observables , on the construction of subspaces of hermitian matrices with various conditions on rank , eigenvalues , and partial trace , and on the construction of totally nonsingular matrices with various useful properties . before we can talk of the number of observables required for process tomography it is necessary to clarify the meaning of the word `` observable '' in this context . in section [ sec : observable ] we introduce the notion of an _ interactive observable _ for channels , a special case of which is an ordinary observable for states . the latter specifies a measurement for a state ; the former , both an input state for a channel and a measurement for the resulting output . we claim that every @xmath19 hermitian matrix can be viewed as an interactive observable for channels acting on a @xmath0-level quantum system in the same sense that every @xmath20 hermitian matrix can be viewed as an ordinary observable for states of a @xmath0-level quantum system . to this end we observe that an arbitrary @xmath19 hermitian matrix @xmath21 can be decomposed into an input - state measurement pair if and only if @xmath21 lies in the unit ball of the dual of the diamond norm ( also known as the completely bounded 1-norm or the completely bounded trace norm ) , from which the above claim follows via a simple scaling argument . ( see section [ sec : observable : operational ] for details and section [ sec : future : strategies ] for further generalization . ) by comparison , it is easily seen that an arbitrary @xmath20 hermitian matrix can be decomposed into a binary measurement if and only if it lies in the unit ball of the operator norm . in section [ sec : reduction ] we observe that questions about the number of interactive observables required for process tomography can be reduced to the problem of finding large subspaces of hermitian matrices with some combination of restrictions on rank , eigenvalues , and partial trace . detailed statements of our primary result on process tomography and supplementary results on hermitian subspaces can be found in section [ sec : results ] . explicit constructions of these subspaces appear in section [ sec : high_rank ] . our supplementary results on hermitian subspaces in section [ sec : high_rank ] rest upon the existence of various forms of totally nonsingular matrices , proofs of which appear in section [ sec : total_nonsing ] . we restrict attention throughout the paper to finite - dimensional linear algebra . the symbol @xmath22 denotes the @xmath23-dimensional complex vector space of @xmath20 matrices with complex entries , which is identified in the usual way with the space of linear operators acting on the @xmath0-dimensional complex euclidean space @xmath24 . it is helpful to give separate labels to the different factor spaces in tensor product spaces such as @xmath25 or @xmath26 . to this end the calligraphic letters @xmath27 are used to denote isomorphic copies of @xmath24 . tensor products such as @xmath28 are abbreviated to @xmath29 so that @xmath30 the symbol @xmath31 denotes the vector space of linear operators acting on @xmath32 so that @xmath33 and @xmath34 linear maps from matrices to matrices of the form @xmath35 are in one - to - one correspondence with elements of @xmath36 via the _ choi jamiokowski isomorphism _ : @xmath37 here @xmath38 denotes the standard orthonormal basis of @xmath24 written in the ket notation . a linear map @xmath35 is : ( a ) completely positive if and only if @xmath39 is positive semidefinite , ( b ) trace - preserving if and only if @xmath40 , and ( c ) unital if and only if @xmath41 . moreover , @xmath5 is completely positive if and only if there exist linear operators @xmath42 with @xmath43 operators with this property are called _ kraus operators _ for @xmath5 . the minimum number of kraus operators required in any such decomposition of @xmath5 is equal to the rank of the choi matrix @xmath39 . associated with each @xmath0-level quantum system is a @xmath0-dimensional complex euclidean space @xmath32 . the set of possible _ states _ of such a system is identified with the set of positive semidefinite matrices @xmath44 with trace equal to one . the set of possible _ channels _ acting on such a system is identified with the set of completely positive and trace - preserving linear maps @xmath35 . measurement _ of such a system with a finite number of outcomes indexed by @xmath45 is represented by a finite set @xmath46 of positive semidefinite matrices with @xmath47 . the probability with which a system in state @xmath48 yields outcome @xmath45 when measured according to @xmath49 is given by the inner product @xmath50 . the concept of an observable is familiar in the context of state tomography : every @xmath20 hermitian matrix @xmath21 specifies an _ observable _ for a @xmath0-level quantum system . if that system is in state @xmath48 then the _ expectation of measuring @xmath21 _ is the quantity @xmath51 . perhaps less widely known is the fact that every @xmath19 hermitian matrix @xmath52 can also specify an `` observable '' for a _ channel _ acting on a @xmath0-level quantum system . given such a channel @xmath35 the _ expectation of measuring @xmath21 _ is the quantity @xmath53the inner product between @xmath21 and the choi matrix @xmath39 associated with @xmath5 . in the context of process tomography , observables such as @xmath21 shall be called _ interactive observables _ in order to distinguish them from observables for states . what does it mean to measure an interactive observable @xmath21 for a channel @xmath5 ? to answer this question it is helpful to recall the meaning of measurement of an ordinary observable @xmath54 on a state @xmath44 . if @xmath55 then it is easy to construct a binary measurement @xmath56 with @xmath57 . suppose a system in state @xmath48 is measured according to @xmath58 and suppose that we assign the quantities @xmath59 to the measurement outcomes @xmath60 . then the expected value of this quantity is @xmath61the expectation of measuring @xmath21 . if @xmath62 then the expectation of measuring @xmath21 is simply @xmath63 times the expectation of measuring @xmath64 , which , again , is @xmath51 . for interactive observables on channels , if @xmath52 is sufficiently small then a more elaborate construction can be used to extract from @xmath21 a state @xmath65 and a binary measurement @xmath66 with the property that @xmath67 thus , if we apply @xmath5 to one of two @xmath0-level systems in joint state @xmath68 and measure the resulting state @xmath69 according to @xmath58 , assigning the quantities @xmath59 to the measurement outcomes @xmath60 as before , then the expected value of this quantity is precisely the inner product @xmath53 of @xmath21 with the choi matrix @xmath39the expectation of measuring @xmath21 . given an interactive observable @xmath21 , how does one compute the associated state @xmath68 and measurement @xmath58 ? suppose @xmath52 is small enough that it can be written @xmath70 for some positive matrices @xmath71 such that @xmath72 for some state @xmath44 . such a set @xmath73 has been called a _ one - round measuring co - strategy _ @xcite , a _ 1-tester _ @xcite , a _ process povm _ @xcite , and an _ interactive measurement _ @xcite . each of refs . @xcite offers a proof that any interactive measurement @xmath74 can be decomposed into a state @xmath65 and a measurement @xmath66 with @xmath75 for all channels @xmath35 . these proofs are constructive and the construction can be implemented efficiently on a computer . we are not aware of a succinct formula for @xmath76 in terms of an arbitrary interactive observable @xmath21 . but such a formula can be derived for the special case in which the state @xmath44 above is the completely mixed state @xmath77 . in this case we may take @xmath78 where @xmath79 denotes the canonical maximally entangled pure state . if we let @xmath80 denote the positive and negative parts of @xmath21 then the interactive measurement @xmath73 for @xmath21 can be written @xmath81 and the desired measurement @xmath58 is given by normalization and matrix transposition : @xmath82 our results on process tomography are not sensitive to scaling . specifically , the sets of interactive observables presented in the present paper for the purpose of identifying unitary channels can always be assumed to consist entirely of interactive observables @xmath21 with @xmath83 so that the formulas of section [ sec : observable : decompose ] applies for extracting @xmath76 from @xmath21 . in particular , each of our interactive observables can be implemented by applying @xmath5 to the maximally entangled pure state @xmath84the only difference among our observables is in how the resulting state @xmath85 is measured . in retrospect it is not surprising that a fixed maximally entangled pure input state suffices for process tomography : every channel @xmath5 is uniquely determined by its choi matrix @xmath39 , which is given by @xmath86 in section [ sec : observable : decompose ] we claimed that a small enough hermitian matrix @xmath52 can always be decomposed into a state @xmath65 and measurement @xmath66 . the curious reader might ask , `` how small is ` small enough ' ? '' the answer is that @xmath21 must lie in the unit ball of the _ dual of the diamond norm_. specifically , it was shown via semidefinite programming in ref . @xcite that @xmath21 is in the unit ball of the dual of the diamond norm if and only if it can be written @xmath70 for some interactive measurement @xmath73 , from which the claim follows . ( an alternate proof of this fact that does not use semidefinite programming follows almost immediately from ( * ? ? ? * theorem 2 ) . ) what is the dual of the diamond norm of a hermitian operator @xmath21 ? the familiar diamond norm is traditionally defined for linear maps of the form @xmath87 so let us begin by considering the dual of the diamond norm for linear maps of this form : @xmath88 here the inner product @xmath89 between two linear maps @xmath90 is defined in the natural way by @xmath91 the diamond norm and its dual can also be defined for operators @xmath52 via the choi jamiokowski isomorphism . to this end let @xmath92 be the unique linear map with @xmath93 . the diamond norm and its dual of @xmath21 are thus given by @xmath94 now that we have formally defined the concept of an interactive observable we may return to the question of the number of such observables required to uniquely identify a unitary or low - rank channel . previous works on state tomography began by reducing the problem to one of finding large subspaces of traceless hermitian matrices of high rank @xcite or with a prescribed number of positive eigenvalues @xcite . following this lead , we begin by reducing the problem of process tomography to one of finding large subspaces of hermitian matrices with zero _ partial _ trace ( as opposed to zero trace ) and having large rank or a prescribed number of positive eigenvalues . let @xmath95 be arbitrary @xmath19 hermitian matrices , which we view as interactive observables for channels acting on a @xmath0-level quantum system as discussed in section [ sec : observable ] , and let @xmath96 denote an arbitrary ordering of these observables . for any hermitian matrix @xmath97 we write @xmath98 so that for each channel @xmath35 the symbol @xmath99 denotes an ordered vector of expectations obtained by measuring each of @xmath100 with respect to @xmath5 . under this notation , a set @xmath101 of interactive observables uniquely identifies channels in @xmath7 among channels in @xmath8 if and only if @xmath102 for each choice of distinct @xmath10 and @xmath103 . let @xmath104 be hermitian subspaces of @xmath36 such that 1 . @xmath105 for each choice of @xmath10 and @xmath103 . 2 . @xmath106 for any choice of distinct channels @xmath10 and @xmath103 . a space @xmath107 with these properties called a _ discriminating subspace _ for @xmath108 . letting @xmath109 denote the orthogonal compliment of @xmath110 within @xmath111 , we claim that any spanning set @xmath101 of @xmath109 uniquely identifies channels in @xmath7 among channels in @xmath8 . to see this , let @xmath96 be any list of interactive observables such that the space @xmath112 is a discriminating subspace for @xmath108 . then for any distinct channels @xmath113 we have @xmath114 from which we conclude that @xmath101 uniquely identifies each channel @xmath7 among channels in @xmath8 . given spaces @xmath115 as above , the number @xmath116 of interactive observables required to identify channels in @xmath7 among channels in @xmath8 is given by @xmath117 this number is minimized when the dimension of @xmath118 is minimal and the dimension of @xmath110 is maximal . it appears that there is little flexibility in the choice of @xmath118 : in this paper we take @xmath118 to be one of the two spaces @xmath119 having @xmath120 the difficulty lies in exhibiting large discriminating subspaces @xmath121 for @xmath108 ; the remainder of this paper is devoted to this task for various choices of sets @xmath108 . we address the following three questions on process tomography : in section [ sec : reduction ] we reduced questions of this type to the problem of finding large discriminating subspaces . for each of these three questions we identify a potential class of discriminating subspaces and we assert the explicit constructibility of a subspace in that class . ( the constructions themselves are given in section [ sec : high_rank ] . ) answers to the above questions are given in section [ sec : results : tomography ] after we have enumerated our subspace constructions in sections [ sec : results : low - rank][sec : results : unital ] . the subspace @xmath122 of eq . contains the difference @xmath123 for every choice of channels @xmath90 . if @xmath124 each have at most @xmath3 kraus operators then their choi matrices @xmath125 each have rank at most @xmath3 and so it must be that @xmath126 . thus , any subspace of @xmath127 in which every element has rank at least @xmath128 is a discriminating subspace for question [ it : question : low - rank ] . [ thm : low - rank ] in the special case of question [ it : question : low - rank ] in which @xmath129 we are asked to identify unitary channels among other unitary channels . because every unitary channel is also a unital channel it holds that the difference @xmath123 is contained in the smaller subspace @xmath130 of eq . . in this case , a slight improvement can be obtained if we construct a discriminating subspace within @xmath130 . [ thm : unitary ] proposition [ thm : unitary ] serves to reduce the number of interactive measurements from @xmath131 to @xmath132 in the @xmath133 case of identifying unitary channels among other unitaries . otherwise , it reduces that number only by @xmath134 . as observed in section [ sec : results : low - rank ] , the subspace @xmath122 of eq . contains the difference @xmath123 for every choice of channels @xmath90 . if @xmath5 has at most @xmath3 kraus operators then its choi matrix @xmath39 has rank at most @xmath3 and so it must be that @xmath123 has at most @xmath3 positive eigenvalues for any choice of channel @xmath135 . thus , any subspace of @xmath127 in which every element has at least @xmath136 positive eigenvalues is a discriminating subspace for question [ it : question : all ] . [ thm : all ] the subspace @xmath130 of eq . contains the difference @xmath123 for every choice of unital channels @xmath90 . if @xmath5 has at most @xmath3 kraus operators then its choi matrix @xmath39 has rank at most @xmath3 and so it must be that @xmath123 has at most @xmath3 positive eigenvalues for any choice of channel @xmath135 . thus , any subspace of @xmath137 in which every element has at least @xmath136 positive eigenvalues is a discriminating subspace for question [ it : question : unital ] . [ thm : unital ] indeed , the bound on the dimension of @xmath138 in theorem [ thm : unital ] can be improved slightly when @xmath139 , but it is too cumbersome to write the precise quantity here : see section [ sec : high_rank : unital ] for details . our answers to questions [ it : question : low - rank][it : question : unital ] from the beginning of this section are as follows . [ thm : unitary_uda_udp ] the following hold for channels acting on a @xmath0-level quantum system : 1 . [ it : unitary_uda_udp : unitary ] @xmath140 interactive observables suffice to identify all channels with at most @xmath3 kraus operators among all other channels with at most @xmath3 kraus operators . + ( here @xmath141 is given in proposition [ thm : low - rank ] . ) + in the special case of identifying unitaries among unitaries ( @xmath129 ) , proposition [ thm : unitary ] further reduces this number to @xmath11 for @xmath142 and from 9 to 6 when @xmath12 . @xmath143 interactive observables suffice to identify all channels with at most @xmath3 kraus operators among all other channels . + ( here @xmath144 is given in proposition [ thm : all ] . ) 3 . @xmath145 interactive observables suffice to identify all unital channels with at most @xmath3 kraus operators among all other unital channels . + ( here @xmath146 is given in proposition [ thm : unital ] . ) the claims of section [ sec : intro : results ] are recovered from theorem [ thm : unitary_uda_udp ] in the special case @xmath129 . the remainder of the paper is devoted to the construction of the large subspaces of hermitian matrices satisfying the various partial trace , rank , and eigenvalue restrictions described in section [ sec : results ] . our key building block in the construction of such subspaces ( and indeed , one of the building blocks used in each of refs . @xcite ) is the notion of _ totally nonsingular matrices_matrices in which every square submatrix is nonsingular . one of the most well - known examples of a totally non - singular matrix is a _ vandermonde matrix _ @xcite @xmath147 in which @xmath148 . although only a few other explicit families of totally nonsingular matrices are known , total nonsingularity is a general phenomenon in the sense that the matrices that are not totally nonsingular form a set of measure zero . the ways in which totally nonsingular matrices are used to construct high - rank subspaces in refs . @xcite are all similar and involve placing the columns of a totally nonsingular matrix along diagonals of matrices . for example , one way to construct a real @xmath131-dimensional subspace @xmath149 of hermitian matrices such that every nonzero @xmath150 has @xmath151 is to construct matrices with columns of vandermonde matrices down various diagonals as follows ( we use @xmath152 to denote @xmath153 entries ) : @xmath154 then the @xmath131 hermitian matrices @xmath155 defined by @xmath156 @xmath157 , @xmath158 @xmath159 , and @xmath160 @xmath161 span a real subspace in which every nonzero matrix has rank at least @xmath162 . to see this , notice that in every real linear combination of the @xmath163 s , there is an upper - right - most diagonal that is nonzero , and that diagonal has at least @xmath162 nonzero entries . the corresponding @xmath164 submatrix is thus nonsingular , which implies that the matrix has rank at least @xmath162 . in order to generalize this argument to different dimensions and ranks , we need a way of generating columns ( to be placed down various diagonals of the matrices @xmath155 ) with the property that their linear combinations do not have `` too many '' zero entries . the following lemma , which is well - known , shows that totally nonsingular matrices serve this purpose . [ lem : tr0_sum_cols ] let @xmath165 be positive integers . if @xmath166 is such that all of its @xmath167 submatrices have full rank @xmath168 , then any linear combination of @xmath168 columns of @xmath169 contains at most @xmath170 zero entries . suppose for a contradiction that some linear combination of @xmath168 columns of @xmath169 contained @xmath171 or more zero entries . then the @xmath167 submatrix of @xmath169 whose columns correspond to the @xmath168 columns in the linear combination and whose rows correspond to @xmath171 of the zero entries must have rank at most @xmath172 , which is a contradiction . while totally nonsingular matrices are useful for the construction of subspaces of high - rank matrices , the subspaces we wish to construct must also satisfy certain partial trace conditions . thus we do nt need totally nonsingular matrices themselves , but rather matrices such that certain subsets of their rows sum to @xmath153 , yet their submatrices all have high rank . the following two lemmas serve this purpose the first lemma when there is one partial trace constraint , the second when there are two . the first lemma is intuitive enough ( albeit slightly technical ) that we only provide an `` intuitive '' and not terribly rigorous proof . the second lemma of this section is much less straightforward , so it is proved rigorously using algebraic geometry techniques . the same techniques could be directly adapted to rigorously prove the first lemma . [ lem : vander_tr0 ] for all @xmath173 , if we define @xmath174 then there exists @xmath175 with the following two properties : 1 . [ it : vander_tr0:zero - sum ] for all @xmath176 , the sum of rows @xmath177 of @xmath169 equals @xmath153 . [ it : vander_tr0:full - rank ] for all @xmath178 , every @xmath179 submatrix of @xmath169 has full rank @xmath180 . observe that property [ it : vander_tr0:full - rank ] is satisfied by a generic matrix with property [ it : vander_tr0:zero - sum ] . indeed , suppose first that @xmath181 and the entries of @xmath169 are selected uniformly at random from the interval @xmath182 $ ] subject to the constraint that every @xmath0th row is the negative of the sum of the previous @xmath183 rows . if @xmath184 then , generically , every @xmath167 submatrix of @xmath169 will have rank @xmath185 , where @xmath186 is the number of distinct values of @xmath187 such that this submatrix contains each of the @xmath0 rows @xmath177 of @xmath169 . since @xmath188 , the result in the @xmath181 case follows by letting @xmath189 . for the case of general @xmath116 , simply append @xmath190 randomly - generated rows to the bottom of @xmath169 . to help illustrate lemma [ lem : vander_tr0 ] , we note that in the @xmath191 case it says that there exist @xmath192 matrices such that the sum of the first two rows equals @xmath153 , the sum of the last two rows equals @xmath153 , every @xmath193 submatrix has rank @xmath134 , and every @xmath194 submatrix has rank @xmath162 ( it also says that every @xmath195 submatrix has rank @xmath134 and that every @xmath196 submatrix has rank @xmath162 , but these properties follow automatically from the @xmath193 and @xmath194 rank properties ) . such a matrix is easily constructed simply by randomly generating the first and third rows , setting the second and fourth rows equal to their negatives , and then verifying that the rank conditions are satisfied . for example , it is easily - verified that the following matrix satisfies all of the requirements : @xmath197 while lemma [ lem : vander_tr0 ] is useful for constructing subspaces of high - rank matrices with vanishing partial trace , we also need to construct subspaces of high - rank matrices with _ two _ vanishing partial traces . the following lemma serves this purpose . [ lem : vander_both_tr0 ] for all @xmath198 , if we define @xmath199 then there exists @xmath200 with the following three properties : 1 . [ it : vander - both - tr0:zero - sum ] for all @xmath201 , the sum of rows @xmath177 of @xmath169 equals @xmath153 . [ it : vander - both - tr0:zero - sum2 ] for all @xmath201 , the sum of rows @xmath202 of @xmath169 equals @xmath153 . [ it : vander - both - tr0:full - rank ] for all @xmath203 , every @xmath179 submatrix of @xmath169 has full rank @xmath180 . as in the proof of lemma [ lem : vander_tr0 ] , observe that a generic matrix that satisfies properties [ it : vander - both - tr0:zero - sum ] and [ it : vander - both - tr0:zero - sum2 ] also satisfies property [ it : vander - both - tr0:full - rank ] . this lemma is perhaps less intuitive than lemma [ lem : vander_tr0 ] though , so we prove it a bit more formally . we begin by noting that the set of matrices satisfying conditions [ it : vander - both - tr0:zero - sum ] and [ it : vander - both - tr0:zero - sum2 ] of the lemma forms an irreducible real algebraic variety @xmath204 ( the fact that it is irreducible follows from it being a subspace ) . let @xmath205 and @xmath206 be integers , and define @xmath207 to be the real variety consisting of matrices whose submatrix formed by rows @xmath208 and columns @xmath209 has rank strictly less than @xmath180 . our goal is to show that @xmath210 suppose for now , for each fixed @xmath205 and @xmath206 , that @xmath211 is a proper subset of @xmath212 . it then follows that it in fact has zero measure in @xmath212 ( since @xmath212 is irreducible ) . since a finite union of measure zero sets again has measure zero , this implies that @xmath213 has zero measure in @xmath212 , which implies equation . it thus suffices to show that for each fixed @xmath205 and @xmath206 there exists a matrix @xmath214 satisfying conditions [ it : vander - both - tr0:zero - sum ] and [ it : vander - both - tr0:zero - sum2 ] of the lemma such that the submatrix of @xmath214 corresponding to rows @xmath208 and columns @xmath209 has full rank @xmath180 . ( henceforth we denote this submatrix by @xmath215 . ) the existence of @xmath214 is straightforward , as the conditions [ it : vander - both - tr0:zero - sum ] and [ it : vander - both - tr0:zero - sum2 ] impose conditions only on @xmath216 of the rows of @xmath215 . thus we are free to set some @xmath217 submatrix of @xmath215 to be the identity matrix ( which ensures that @xmath214 has full rank @xmath180 ) and then fill in the remaining entries of @xmath214 arbitrarily , subject to constraints [ it : vander - both - tr0:zero - sum ] and [ it : vander - both - tr0:zero - sum2 ] . to help illustrate the above result , we present the following example of a matrix @xmath218 that satisfies all three conditions of lemma [ lem : vander_both_tr0 ] in the @xmath219 case : @xmath220 it is straightforward to verify that : ( a ) the sum of rows @xmath221 is zero , as is the sum of rows @xmath222 and the sum of rows @xmath223 ; ( b ) the sum of rows @xmath224 is zero , as is the sum of rows @xmath225 and the sum of rows @xmath226 ; and ( c ) every @xmath193 , @xmath164 , @xmath227 , and @xmath228 submatrix of @xmath169 is nonsingular . by combining lemma [ lem : vander_tr0 ] or [ lem : vander_both_tr0 ] with lemma [ lem : tr0_sum_cols ] , we are able to construct matrices whose rows sum to zero in ways that play nicely with the partial trace and are also such that linear combinations of their columns contain large numbers of nonzero entries . we make use of such matrices in the next section to construct high - rank subspaces with zero partial trace . in this section we invoke the preliminary results of section [ sec : total_nonsing ] on totally nonsingular matrices in order to prove propositions [ thm : low - rank][thm : unital ] of section [ sec : results ] on the existence of large subspaces of hermitian matrices that satisfy various partial trace , rank , and eigenvalue restrictions . we repeat the statements of those propositions here for readability . before proving these results , it will be useful to recall explicitly that elements of @xmath36 can be thought of as @xmath20 block matrices , and the maps @xmath229 and @xmath230 have very simple formulas in terms of the elements of such block matrices : @xmath231 since our goal in the proofs of the following propositions is to construct a basis of hermitian matrices that spans a space of high rank with zero partial trace , it suffices to find such a basis with either @xmath232 for all @xmath233 and all @xmath21 in the basis , or with @xmath234 for all @xmath21 in the basis , or both , depending on which subset of the partial traces we want to vanish . we now present our results that allow us to construct large subspaces of hermitian matrices with high rank and zero partial trace . throughout the proofs of both of these lemmas , we will discuss various diagonals of block matrices , only some of which have any effect on the partial trace . for example , in the @xmath235 block matrix @xmath236,\end{aligned}\ ] ] only the diagonals consisting of @xmath237 s , @xmath238 , and @xmath131 s affect @xmath239 ( in general , only the main diagonal and the @xmath240 other nearest diagonals affect @xmath239 ) , and only the diagonals consisting of @xmath241 s , @xmath132 s , and @xmath131 s affect @xmath242 ( in general , only the main diagonal and every @xmath0-th diagonal on either side of it affect @xmath242 ) . [ thm : low - rank][subspaces of high - rank matrices with vanishing partial trace ] we prove the statement by giving an explicit construction , which is built upon the construction given in ( * ? ? ? * proposition 4 ) ( i.e. , the construction that was roughly illustrated in section [ sec : total_nonsing_highrank ] ) . we begin by taking the basis @xmath243 of the subspace given without the partial trace condition and then impose the partial trace condition and see how much the dimensionality is reduced . ( however , we actually set @xmath244 rather than @xmath245 , as this results in a larger subspace one can then just swap the spaces @xmath246 to get @xmath245 . ) more specifically , we consider the @xmath247 matrices @xmath248 constructed along diagonals in the proof of ( * ? ? ? * proposition 4 ) and think of them as @xmath249 block matrices . in order to ensure that @xmath250 for all @xmath251 , if @xmath248 is defined along a diagonal that consists of the main diagonals of @xmath252 of its subblocks , we place along that diagonal the columns of a matrix described by lemma [ lem : vander_tr0 ] ( with @xmath181 ) rather than the columns of a totally nonsingular matrix . since we want any linear combination of the @xmath248 s to have at least @xmath128 nonzero entries , from lemma [ lem : tr0_sum_cols ] and the @xmath253 case of lemma [ lem : vander_tr0 ] we can construct @xmath254 such matrices along these `` problem '' diagonals , rather than the @xmath255 such matrices in the original proof . we thus lose @xmath256 matrices along each diagonal that goes through @xmath257 block diagonals , for a total of @xmath258 matrices . since the original subspace without the partial trace condition had dimension @xmath259 , the subspace @xmath260 described after imposing the partial trace condition has the dimensionality specified in the statement of the proposition . [ thm : unitary][subspaces of rank - three matrices with two vanishing partial traces ] we prove the result by giving an explicit construction that arises from modifying the basis given in the proof of proposition [ thm : low - rank ] . that is , we begin by taking the basis @xmath243 of the subspace given with only one partial trace restriction and then impose the other partial trace condition and see how much the dimensionality is reduced . as before , we think of each @xmath248 as a @xmath249 block matrix . in order to ensure that @xmath261 for all @xmath251 , if @xmath248 is defined along its main diagonal , we place along that diagonal the columns of a matrix described by lemma [ lem : vander_both_tr0 ] rather than the columns of a matrix described by lemma [ lem : vander_tr0 ] . similarly , if @xmath248 is defined along one of the other @xmath240 diagonals closest to the main diagonal , we place along that diagonal the columns of a matrix described by lemma [ lem : vander_tr0 ] ( with @xmath262 and the rows permuted accordingly so that @xmath261 ) rather than the columns of a totally nonsingular matrix . since we want any linear combination of the @xmath248 s to have at least @xmath241 nonzero entries , from lemmas [ lem : tr0_sum_cols ] and [ lem : vander_both_tr0 ] we can construct @xmath263 such matrices along the main diagonal ( i.e. , we lose @xmath264 matrices ) , and along the other @xmath240 diagonals nearest to the main diagonal we can construct @xmath265 matrices ( i.e. , we lose @xmath266 matrices along these diagonal ) . since the original subspace @xmath267 with just one partial trace constraint had dimension @xmath268 , the subspace @xmath269 described after imposing the second partial trace constraint has dimension @xmath270 . the claim about the case @xmath133 follows from observing the the following @xmath241 block matrices form a basis for a subspace that satisfies all of the given rank and partial trace requirements : @xmath271 , \quad \left[\begin{array}{cc|cc } 0 & 1 & 0 & 0 \\ 1 & 0 & 2 & 0 \\ \hline 0 & 2 & 0 & -1 \\ 0 & 0 & -1 & 0 \end{array}\right ] , \quad \left[\begin{array}{cc|cc } 0 & i & 0 & 0 \\ -i & 0 & 2i & 0 \\ \hline 0 & -2i & 0 & -i \\ 0 & 0 & i & 0 \end{array}\right ] . \end{aligned}\ ] ] we now present our results that allow us to construct large subspaces of hermitian matrices with many positive eigenvalues . unlike the proofs in the previous subsection , the matrices considered here are constructed along _ anti_-diagonals ( i.e. , diagonals that run from lower - left to upper - right ) rather than standard diagonals . that is , we now follow the approach of ( * ? ? ? * lemma 2 ) , rather than that of ( * ? ? ? * proposition 4 ) , but our constructions are again more complicated as a result of us requiring that the partial trace of every element of the subspace vanishes . note that , in the following propositions , we only consider the upper - triangular part of the anti - diagonals ( i.e. , the part of the anti - diagonals lying _ strictly _ above the main diagonal ) , as the hermiticity condition determines the lower - triangular entries and all entries along the main diagonal equal @xmath153 . [ thm : all][subspaces of matrices with many positive eigenvalues and vanishing partial trace ] we prove the statement by giving an explicit construction of such a subspace , which is built upon the construction given in ( * ? ? ? * lemma 2 ) . in fact , our construction is exactly the same , but with just one change in order to impose the partial trace condition . as usual , we think of each matrix @xmath272 as a @xmath249 block matrix . we construct a basis @xmath243 of @xmath273 as in @xcite by placing columns of a totally nonsingular matrix along the upper - triangular part of anti - diagonals . in order to ensure that @xmath261 for all @xmath251 , however , we subtract @xmath274 from the top - left block of @xmath248 . two examples of such matrices @xmath248 in the @xmath219 case are as follows : @xmath275 & & \left[\begin{array}{ccc|ccc|ccc } 0 & 0 & 0 & 2 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \hline 2 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \hline 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \end{array}\right ] \end{aligned}\ ] ] in the left matrix , we initially placed @xmath276 $ ] along one of the upper - triangular anti - diagonals , then we adjusted the other entries to enforce hermiticity and the partial trace constraint . in the right matrix , we similarly placed @xmath277 $ ] along one of the upper - triangular anti - diagonals , but in this case , when we enforced the hermiticity and partial trace constraints , the @xmath278- and @xmath279-entries of the matrix were forced to equal @xmath153 . the rightmost example above illustrates the one and only case where forcing @xmath261 has any effect on the proof that these matrices are linearly independent and every matrix in this subspace has at least @xmath136 positive and @xmath136 negative eigenvalues : when the anti - diagonal that defines @xmath248 goes through the top - left @xmath20 sub - block . in this case , all entries within that block of @xmath248 are forced to be @xmath153 , which reduces the number of distinct matrices @xmath248 that we can construct along that anti - diagonal ( for example , the matrix on the right above is _ not _ included in the basis of our subspace when @xmath280 , since the zeros at the @xmath278- and @xmath279-entries cause it to only have @xmath134 positive and @xmath134 negative eigenvalue ) . there are @xmath281 upper - triangular anti - diagonals that intersect the top - left block . however , @xmath282 of those anti - diagonals are of length @xmath283 and do not have an associated matrix @xmath248 . for each of the other anti - diagonals that intersect the top - left block , we lose @xmath284 matrices , where @xmath285 is the number of entries along that upper - triangular anti - diagonal within the top - left block and @xmath186 is the length of that anti - diagonal . thus we remove a total of @xmath286 matrices from the set @xmath243 in order to not violate the eigenvalue requirement . since the dimension of the subspace constructed in this way in @xcite _ without _ the partial trace restriction was @xmath287 , the result follows . throughout the proof of the following proposition , we will repeatedly refer to the quantity @xmath288 that is , @xmath289 is the length of the @xmath257th upper - triangular anti - diagonal of a @xmath249 matrix , just as in @xcite . note that we count the upper - triangular anti - diagonals from left to right , so that the @xmath290-entry is the first upper - triangular anti - diagonal , the @xmath291-entry is the second upper - triangular anti - diagonal , the @xmath292- and @xmath278-entries together make up the third upper - triangular anti - diagonal , and so on . for a @xmath249 matrix , there are @xmath293 distinct upper - triangular anti - diagonals . [ thm : unital][subspaces of matrices with many positive eigenvalues and two vanishing partial traces ] before proving this proposition , we note that the bound on @xmath146 that we actually prove is slightly better than stated above when @xmath294 . indeed , if we define @xmath295 then we show that there exists such a subspace @xmath138 with @xmath296 when @xmath294 , and it is straightforward to show that @xmath297 in this case . once again , we prove this result by explicitly constructing a basis of such a subspace . our construction uses the basis @xmath243 constructed in the proof of proposition [ thm : all ] as a starting point . as before , we think of these matrices as @xmath249 block matrices . to enforce the requirement that @xmath298 for all @xmath251 , we add to the top - left entry of each block of @xmath248 the negative of the trace of that block . applying this procedure to the same two matrices considered in the @xmath299 case in the proof of proposition [ thm : all ] results in the following two matrices : @xmath300 & & \left[\begin{array}{ccc|ccc|ccc } 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \hline 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \hline 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \end{array}\right ] . \end{aligned}\ ] ] in the left matrix , the only change is that the @xmath301- and @xmath302-entries of the matrix now equal @xmath303 . the right matrix , however , now consists entirely of zeroes . as before , there is only one case where this procedure affects the proof , and that is when the anti - diagonal that defines @xmath248 goes through the top - left corner of a block ( as in the example on the right above ) . in this case , that entry of @xmath248 is forced to be @xmath153 , which reduces the number of distinct matrices @xmath248 that we can construct along that anti - diagonal by @xmath162 ( one real matrix and one imaginary matrix ) . there are @xmath281 top - left corners of blocks in the upper - triangular portion of @xmath21 , however we only remove a matrix from our basis if the anti - diagonal going through that top - left corner has length @xmath304 . in other words , we remove @xmath305 matrices , where @xmath306 is as in . however , if @xmath307 then the anti - diagonals hitting the top - left corner of one of these blocks ( the @xmath290-block ) have already been removed by the requirement that @xmath261 from proposition [ thm : all ] , so we add back in @xmath162 matrices . furthermore , it is straightforward to verify that @xmath308 when @xmath307 and @xmath297 when @xmath294 , so the result follows . the questions considered in this work can be viewed as questions about state tomography on certain subsets of states via the choi jamiokowski isomorphism . for example , using process tomography to uniquely identify unitary channels among unitary channels is equivalent to using state tomography to uniquely identify maximally entangled bipartite pure states among maximally entangled bipartite pure states . in other words , we can think of process tomography as state tomography on the intersection of the set of pure states and the affine space of operators with completely mixed partial trace . this line of thinking raises a natural question : can we answer uniqueness questions for tomography when we restrict attention to affine spaces other than the space of operators with completely mixed partial trace ? we do not have a detailed answer for this question , but we can make some observations . first , some work in this area has already been done : it was shown in ref . * proposition 9 ) that for any manifold @xmath309 with real dimension @xmath310 almost all families of @xmath311 observables uniquely determine elements of @xmath309 among elements of @xmath309 . for example , the space of unitary channels acting on @xmath22 is a real manifold of dimension @xmath312 , so almost any family of @xmath313 interactive observables uniquely identifies unitary channels among unitary channels . remarkably , this quantity is smaller than the @xmath314 interactive observables that were shown to suffice in item [ it : unitary_uda_udp : unitary ] of theorem [ thm : unitary_uda_udp ] ! however , our proof of theorem [ thm : unitary_uda_udp ] is constructive ; no explicit set of @xmath313 interactive observables is known to uniquely identify unitary channels among other unitary channels . second , the above technique does not tell us anything about how many measurements can be used to uniquely determine elements of @xmath309 ( such as the set of unitary channels ) among a _ different _ set ( such as the set of _ all _ quantum channels ) . to get general results in this direction , we simply use the fact that if @xmath315 are two subspaces , then @xmath316 . to illustrate this approach , let @xmath107 be the subspace of operators with at least @xmath162 positive eigenvalues of dimension @xmath317 constructed in ref . @xcite and let @xmath318 be the subspace of operators with vanishing partial trace of dimension @xmath6 . using the fact that @xmath319 , we see that @xmath320 it then follows from the reduction of section [ sec : reduction ] that there is a set of @xmath321 interactive observables that uniquely identify unitary channels among all channels , which is slightly worse than the @xmath322 interactive observables that were shown in the present paper to suffice for this task . a channel @xmath323 acting on two @xmath0-level quantum systems is _ no - signaling _ if the output of the channel on each system is independent of the input to the channel on the other system . this condition is nicely characterized by the linear constraints @xmath324 on the choi matrix @xmath39 . thus , the reduction of section [ sec : reduction ] can be used to bound the number of interactive observables needed to identify channels among no - signaling channels by finding discriminating subspaces @xmath107 of @xmath325 where @xmath326 however , the task of constructing large subspaces of high - rank matrices within @xmath325 is considerably more complicated than that of constructing similar spaces within @xmath122 or @xmath130 . whereas @xmath327 have only one or two vanishing partial trace constraints , @xmath325 has several additional partial trace constraints . our experience is that each additional constraint adds considerably to the complication of constructing high - rank subspaces based on totally nonsingular matrices . while this method could in principle yield a significant improvement in the number of interactive observables needed to identify no - signaling channels , the question seems to call out for a different approach . a _ quantum strategy _ is a specification of the actions of one party in a multi - party interaction involving the exchange of multiple rounds of quantum messages among the parties . channels arise as a special case of strategies in which only one round of messages is exchanged . the strategy formalism was introduced in refs . @xcite wherein it was shown that a matrix @xmath328 represents an @xmath171-round strategy for input spaces @xmath329 and output spaces @xmath330 if and only if @xmath331 is positive semidefinite and @xmath332 quantum strategies are therefore intimately related to no - signaling channels in that the above constraints also characterize the choi matrices of channels of the form @xmath333 that act on @xmath171 distinct @xmath0-level quantum systems and are no - signaling from system @xmath187 to system @xmath251 for each @xmath334 . an important difference between strategies and no - signaling channels ( or any other type of channel ) is that channel tomography is achieved by preparing an input state and measuring the output . by contrast , tomography for @xmath171-round quantum strategies introduces the need to prepare @xmath171 input systems and process @xmath171 output systems in sequence , with future inputs possibly depending on previous outputs . in other words , the `` observables '' required for strategy tomography are themselves full - blown , multi - round strategies . at first glance this complication might seem intractable . in order to perform tomography on a multi - round strategy one must deduce not only how the strategy acts on various input states , but also how it acts when various _ channels _ are applied to its outputs and then subsequently returned to it as inputs in future rounds . however , strategies that measure other strategies have appeared previously in the literature under the names _ @xmath171-round measuring co - strategy _ @xcite and _ @xmath171-tester _ @xcite . the discussion of section [ sec : observable ] generalizes readily in light of this prior work . specifically , one can show that : 1 . an arbitrary @xmath335 hermitian matrix @xmath21 can be viewed as an `` @xmath171-round interactive observable '' for a strategy that exchanges a @xmath0-level quantum system @xmath171 times in sequence . 2 . if @xmath21 is small enough then one can extract from @xmath21 a collection of objects an initial state , a sequence of channels with memory , and a measurement for the final system that describe the behaviour of the observable operationally . such an extraction is possible if and only if @xmath21 lies in the unit ball of the _ dual of the strategy @xmath171-norm _ presented in ref . @xcite . thus , as with no - signaling channels in section [ sec : future : ns ] , one can employ the reduction of section [ sec : reduction ] and search for large discriminating subspaces of @xmath336 where @xmath337 in short : the task of formalizing the notion of an @xmath171-round interactive observable is not the bottleneck that thwarts attempts to bound the number of observables needed for strategy tomography . rather , the bottleneck lies in the same place as with no - signaling channels and other types of channel in the ability ( or lack thereof ) to construct large subspaces of high - rank matrices that also satisfy a myriad of partial trace constraints . one of the drawbacks of our approach to process tomography is that the interactive observables obtained from theorem [ thm : unitary_uda_udp ] do not necessarily have a nice form they might not be clifford operators , they could be non - local , and so on . hence , they might be difficult to implement in the laboratory . it would be interesting to consider the number of interactive observables required to reconstruct unitary channels under certain additional `` niceness '' restrictions such as these . in order to illustrate the type of result that would be desirable we consider the @xmath338 case of theorem [ thm : unitary_uda_udp].1 , which says that six interactive observables suffice to uniquely determine unitary qubit channels among unitary qubit channels . we now show that we can in fact find six interactive observables for this purpose that are both local and clifford . to this end our goal is to find six linearly independent hermitian local clifford operators @xmath339 with two vanishing partial traces that are each orthogonal to the three operators presented in the proof of proposition [ thm : unitary ] . the following are six such operators : @xmath340 where @xmath341 are the pauli operators and @xmath342 are the clifford operators with standard basis representations @xmath343 this construction of @xmath344 is _ ad hoc _ , and it is not clear whether it can be generalized to the other cases of theorem [ thm : unitary_uda_udp ] . as described in section [ sec : observable : decompose ] , each of these interactive observables specifies a measurement on a maximally entangled input state , which is not nice . it is not difficult to see that these interactive observables could be measured using only experimentally - friendly product state inputs . however , this product state implementation essentially consists of measuring all twelve degrees of freedom and then throwing away half of the resulting information in order to compile the results into the six observables required for tomography . obviously , nothing is gained by such an exercise . the authors thank steven t. flammia , marcus da silva , john watrous , and bei zeng for helpful conversations . research at the perimeter institute is supported by the government of canada through industry canada and by the province of ontario through the ministry of research and innovation . gg also acknowledges support from cryptoworks21 . nj is supported by the natural sciences and engineering research council of canada . jianxin chen , hillary dawkins , zhengfeng ji , nathaniel johnston , david w. kribs , frederic shultz , and bei zeng . uniqueness of quantum states compatible with given measurement results . , 88:012109 , 2013 . arxiv:1212.3503 [ quant - ph ] . steven t. flammia , david gross , yi - kai liu , and jens eisert . quantum tomography via compressed sensing : error bounds , sample complexity , and efficient estimators . , 14:095022 , 2012 . arxiv:1205.2300 [ quant - ph ] . rahul jain , sarvagya upadhyay , and john watrous . two - message quantum interactive proofs are in pspace . in _ proceedings of the 50th ieee symposium on foundations of computer science ( focs 2009 ) _ , pages 534543 , 2009 . arxiv:0905.1300 [ quant - ph ] . a. shabani , r. l. kosut , m. mohseni , h. rabitz , m. a. broome , m. p. almeida , a. fedrizzi , and a. g. white . efficient measurement of quantum dynamics via compressive sensing . , 106:100401 , 2011 . arxiv:0910.5498 [ quant - ph ] .

we consider a yule process until the total population reaches size , and assume that neutral mutations occur with high probability ( in the sense that each child is a new mutant with probability , independently of the other children ) , where . we establish a general strategy for obtaining poisson limit laws for the number of subpopulations exceeding a given size and apply this to some mutation regimes of particular interest . finally , we give an application to subcritical bernoulli bond percolation on random recursive trees with percolation parameter tending to zero . * key words : * branching processes , mutation , percolation , random increasing trees . 60j27 ; 60j80 ; 60k35 .

1603.06564

we consider a system of branching processes with mutations specified as follows . the underlying total population process is modeled by a standard yule process @xmath4 , that is a continuous - time birth process started from one individual with unit birth rate per unit population size . we superpose independent mutations , by declaring that a new - born child is a clone of its parent with probability @xmath5 , and a mutant otherwise . being a mutant means that the individual obtains a new genetic type which was not present before . we observe the process @xmath4 at the instant when the @xmath6th individual is born and group individuals of the same genetic type into subpopulations . in this paper , we are interested in questions concerning the ( asymptotic ) sizes of these subpopulations under strong mutations , the sense that @xmath7 . by approximating the population system from below and above by two different processes , where sub - populations are independent and have an explicit distribution , we develop a general strategy to obtain non - trivial ( poisson ) weak limits for the number of subpopulations exceeding a given size ( which might grow with @xmath6 as well ) . we then discuss our strategy in the context of three qualitatively different mutation regimes . for fixed @xmath8 , we identify first @xmath9 , @xmath10 fixed , as the regime in which , in the limit @xmath11 , the largest subpopulations have size @xmath12 . for its number , we obtain a poisson limit law and show that the number of subpopulations of size @xmath13 for @xmath14 tends to infinity ( theorem [ thm : finitesubp ] and corollary [ cor:1 ] ) . secondly , we discuss the regime @xmath15 . since the size of the subpopulation containing the ancestor is of order @xmath16 , see proposition [ prop : ancestralsubp ] , this is the border - line case between a bounded and an unbounded size for the ancestral subpopulation . we show that the sizes of the largest subpopulations are concentrated around @xmath17 , where @xmath18 and @xmath19 are positive constants depending on @xmath20 ( theorem [ thm : lognsubp2 ] ) . for the exact choice @xmath21 and @xmath22 , we find a correction @xmath23 such that , with @xmath24 , the number of subpopulations greater than @xmath25 converges along subsequences @xmath26 with converging fractional part to a poisson@xmath27-distributed random variable , where @xmath28 is expressed in terms of @xmath29 and @xmath30 ( theorem [ thm : lognsubp1 ] ) . thirdly , we study the case @xmath31 . here , it turns out that the sizes of the largest subpopulations are to first order given by @xmath32 . for @xmath33 and given @xmath22 , we compute a precise barrier such that the number of subpopulations exceeding this barrier follows in the limit the poisson - law with parameter @xmath34 ( theorem [ thm : greatsubp ] ) . this work originates from questions about bernoulli bond percolation on so - called random recursive trees , when their size @xmath6 tends to infinity and the percolation parameter @xmath35 satisfies @xmath36 . the connection to branching processes stems from the fact that the genealogical tree built from the first @xmath6 individuals in a standard yule process can be interpreted as a random recursive tree @xmath37 on @xmath38 . mutations in the yule process can naturally be modeled on its genealogical tree , by cutting the edges that connect mutants to its parent . then the connected subsets of vertices form the subpopulations of the same genetic type . to put it differently , the connected components ( clusters ) on a random recursive tree @xmath37 that arise from a bernoulli bond percolation , where each edge is erased with probability @xmath1 independently of each other , can be viewed as the subpopulations in a yule process with mutation rate @xmath1 , observed at the instant when there are @xmath6 individuals in total in the system . the strategy we develop here in terms of yule processes allows a concise analysis of cluster sizes , for any choice of @xmath3 tending to zero . for sequences of @xmath3 such that @xmath39 or @xmath40 remains constant , similar connections between systems of branching processes and percolation on increasing tree families have been utilized before in , e.g. , @xcite . the precise definition of a random recursive tree , its connection to yule processes and more references to existing results on percolation will be discussed in section [ sec : app - percolation ] . the rest of this paper is organized as follows . in the next section [ sec : yule ] , we properly define the population system and provide some heuristics for regimes of interest and poisson limits . then , in section [ sec : strategy ] , we explain our strategy for obtaining poisson limit laws for the number of subpopulations ( or clusters ) greater than a given size . section [ sec : limitresults ] contains our main results ; we exemplify our strategy by proving limit results for certain mutation rates of particular interest . in the last section [ sec : app - percolation ] , we establish the link to percolation on random recursive trees . appendix [ sec : appendix ] contains some ( standard ) estimates on yule processes , which we use in our analysis . * notation : * we let @xmath41 . if @xmath42 , @xmath43 are two sequences of real numbers , we write @xmath44 if @xmath45 as @xmath11 , and we write @xmath46 or @xmath47 if and only if @xmath48 as @xmath49 . moreover , if @xmath50 and @xmath51 are two positive functions , we say @xmath52 if there exists @xmath53 such that @xmath54 for all @xmath55 , and we write @xmath56 if @xmath57 as @xmath58 . we will use the letters @xmath59 or @xmath60 for small or large generic constants that do not depend on @xmath6 . their values may change from line to line . in this section , we introduce the population system we work with , and present some basic results and heuristics . for background on yule processes , we refer to appendix [ sec : appendix ] , and for more information to chapter @xmath61 of @xcite . let @xmath62 be a standard yule process with unit birth rate per individual and started from one single particle , i.e. @xmath63 . let @xmath64 $ ] . we superpose mutations on @xmath4 as follows : when a new child is born , we declare it to be a clone of its parent with probability @xmath65 , and a mutant with a new genetic type different from all the previous types with probability @xmath1 . we let @xmath66 and write @xmath67 for the sequence of consecutive birth times of individuals which are mutants . more specifically , the ancestor present at time @xmath66 has genetic type @xmath68 , as well as its clone children . the first mutant is born at time @xmath69 and has genetic type @xmath70 . the next mutant appears in the system at time @xmath71 and receives genetic type @xmath61 ; it is a mutant of an individual of either type @xmath68 or of type @xmath70 , and so on . we group individuals of the same genetic type into subpopulations , so that at time @xmath72 , the population system consists of at most @xmath73 many subpopulations . in contrast to @xcite , we will here not be interested in the genealogical structure , but only in the sizes of the subpopulations . for @xmath74 , we let @xmath75 denote the subpopulation process counting the individuals of type @xmath76 , with @xmath77 if @xmath78 . it should be clear that the processes @xmath79 , @xmath74 , form independent yule processes with birth rate @xmath65 per individual and started from one single individual each . the number of subpopulations of different genetic types present at time @xmath80 is denoted by @xmath81 viewed as a process in @xmath72 , @xmath82 is a counting process started from @xmath68 which grows at rate @xmath83 . its predictable compensator is absolutely continuous with density @xmath83 , that is @xmath84 is a martingale . see , e.g. , ( * ? ? ? * theorem 9.15 ) . we next build a process @xmath85 by setting @xmath86 note that @xmath87 . clearly , we can retrieve the total population size @xmath73 at time @xmath72 from @xmath88 , @xmath89 it follows from its construction that the process @xmath88 is markovian with transition rates at time @xmath80 for @xmath90 given by @xmath91 mostly we will stop @xmath88 at the instant when the @xmath6th particle is born , @xmath92 obviously , we have @xmath93 for all @xmath74 and all @xmath64 $ ] . moreover , since @xmath94 is a martingale which converges almost surely to a standard exponential @xmath95 as @xmath72 tends to infinity , see appendix [ sec : appendix ] , @xmath96 in this section , we point at a simple limit theorem in distribution for the size @xmath97 of the subpopulation at time @xmath98 having the same genetic type as the ancestor individual . we obtain the following characterization when @xmath2 . [ prop : ancestralsubp ] for @xmath99 , denote by @xmath100 a geometrically distributed random variable of parameter @xmath34 , and by @xmath101 a standard exponential random variable . then the following holds : 1 . if @xmath102 , then @xmath103{(d ) } 1.$ ] 2 . if @xmath15 for some @xmath10 , then @xmath103{(d ) } \textup{geo}(\e^{-a}).$ ] 3 . if @xmath31 , then @xmath104{(d ) } \textup{exp}(1).$ ] assume @xmath36 , and let @xmath105 . from , we know @xmath106 as @xmath11 . for @xmath107 and @xmath108 , it thus suffices to show that for @xmath109 , @xmath110 where @xmath111 is the distribution function of the stated limit in @xmath107 or @xmath108 , respectively . since @xmath112 is monotone increasing in @xmath72 , we have @xmath113 since @xmath114 has a geometric law with success probability @xmath115 , we obtain @xmath116 it is readily checked that when @xmath109 and @xmath11 , the right side converges for @xmath117 to zero , whereas for @xmath15 , it converges to @xmath118 . writing @xmath119 we see that the same convergence holds for the probability on the left side in . this shows @xmath107 and @xmath108 . the proof of @xmath120 is entirely similar and left to the reader . 1 . the above statements remain true @xmath97 is replaced by any @xmath121 for @xmath74 fixed . 2 . for @xmath122 fixed , it has been shown independently in @xcite and @xcite that @xmath123 converges to a mittag - leffler distributed random variable with parameter @xmath65 . in turn , the latter converges to a standard exponential random variable when @xmath124 . if @xmath39 as @xmath11 , then @xmath125 a.s . , see @xcite . we note that for each @xmath126 $ ] , the mean of @xmath97 can be computed exactly . indeed , we have @xmath127 = \sum_{i=1}^n\p\left(i\in y_1^{(p)}(\tau_n)\right ) = \sum_{i=1}^n\e\left[p^{d(i)}\right],\ ] ] where @xmath128 is the insertion depth of @xmath76 , i.e. , @xmath128 is the height of @xmath76 in the genealogical tree of the yule process stopped at time @xmath98 ( with @xmath129 ) . from @xcite we know that @xmath130 , where @xmath131 are independent bernoulli random variables of parameter @xmath132 . consequently , @xmath133 = 1+p\sum_{i=2}^n\prod_{j=2}^{i-1}\e\left[p^{\xi(j)}\right ] & = 1+p\sum_{i=2}^n\prod_{j=2}^{i-1}\left(\frac{p}{j}+ 1-\frac{1}{j}\right)\\ & = 1+p\sum_{i=1}^{n-1}\frac{1}{i!}\prod_{j=2}^{i}(j-1+p).\end{aligned}\ ] ] clearly , when @xmath134 , the @xmath6 first individuals are all mutants and all subpopulations at time @xmath98 are singletons with high probability . when @xmath135 , standard poisson approximation to the binomial law ( and the fact that the genealogical tree of @xmath136 is not star - like ) shows that the number of subpopulations of size @xmath70 is asymptotically poisson@xmath107-distributed . for general @xmath137 , it is however _ a priori _ not obvious in which regime subpopulations of size @xmath12 will emerge , and if so , how many of them . following the tradition of aldous @xcite , we shall now present an heuristic argument to determine this regime , and will check later on that our informal approach can actually be made rigorous . we are interested in the number of subpopulations of size @xmath138 at time @xmath98 , @xmath139 for @xmath140 and @xmath141 , the birth time @xmath142 of the @xmath76th mutant is close to the birth time @xmath143 of the @xmath76th individual , and by , we have @xmath144 a.s . thus the distribution of @xmath145 should be close to that of a yule process with birth rate @xmath65 evaluated at time @xmath146 , that is to a geometric distribution with parameter @xmath147 . therefore , the bernoulli variable @xmath148 has parameter @xmath149 let us now recall le cam s inequality for the reader s convenience . * inequality of le cam @xcite * _ let @xmath150 be a sequence of integers . for each @xmath55 , let @xmath151 , @xmath152 be independent bernoulli random variables with parameters @xmath153 . set @xmath154 , and let @xmath155=\sum_{i=1}^{a_n } q_{n , i}$ ] . then @xmath156 in particular , if @xmath157 and @xmath158 , then @xmath159{(d ) } \poi(\lambda)$ ] , where @xmath160 is a poisson@xmath100-distributed random variable . _ even though the variables @xmath145 , @xmath74 , which describe the sizes of the subpopulations , are clearly not independent ( obviously , they add up to @xmath6 ) , let us pretend for the purpose of this section that @xmath148 for @xmath161 , form a sequence of independent bernoulli variables . since @xmath162 we infer that for @xmath10 , @xmath163 should be the regime in which the largest subpopulations have size @xmath12 , and more precisely , then @xmath164{(d ) } \poi(\ell!a^{\ell } ) .\ ] ] these informal calculations are of course far from a rigorous proof ; nonetheless we shall show in the forthcoming theorem [ thm : finitesubp ] that the above weak convergence actually holds . for this , we shall first develop a general strategy to analyze the asymptotic behavior of the number of large subpopulations in the next section . we will establish limit laws for the number of subpopulations of size @xmath165 at time @xmath98 , @xmath166 when @xmath7 . the threshold @xmath167 may be fixed or @xmath168 ( depending on the choice of @xmath169 ) . roughly speaking , the key step consists in approximating the population system @xmath88 at time @xmath98 by systems @xmath170 and @xmath171 in which mutants are born at deterministic times , and such that @xmath88 is bounded by @xmath170 from below and by @xmath171 from above . the advantage of working with deterministic times comes from the fact that we can in this way decouple populations sizes from birth times and then deal with independent bernoulli variables . in the new systems , we can compute moments of the quantities corresponding to @xmath172 more easily . provided we construct @xmath170 and @xmath171 such that the expected numbers of subpopulations of a size larger than @xmath167 match asymptotically in the two systems , this will ultimately lead to limit statements for @xmath172 in the original system . in fact , it will be sufficient to control @xmath88 on a set of large probability . in this regard , recall that convergence in distribution of a sequence of random variables @xmath173 to some random variable @xmath174 is equivalent to convergence in distribution of @xmath175 to @xmath174 , provided @xmath176 is a sequence of events with @xmath177 . concerning limit results for @xmath178 , we can therefore restrict ourselves to certain `` good '' events @xmath179 of large probability , which we specify next . [ lem : goodevent ] assume that @xmath36 , and let @xmath180 be a sequence of integers such that @xmath181 and @xmath182 as @xmath11 . there exists an event @xmath179 ( depending on @xmath6 , @xmath3 and @xmath183 ) of probability @xmath184 , on which the following holds true : 1 . @xmath185 , 2 . @xmath186 3 . @xmath187 we call @xmath179 the _ good event _ , and for establishing our limit results for @xmath188 , we will work on @xmath179 . the proof of the lemma uses standard estimates on yule processes and is given in the appendix . note that the event @xmath189 under @xmath107 is precisely the event that the first @xmath190 born individuals ( discounting the ancestor individual ) are all mutants . the choice of @xmath183 will depend on our applications and will be specified later on . roughly speaking , we choose @xmath183 in such a way that with high probability , the first @xmath191 born individuals do not contribute to @xmath188 . define for @xmath74 @xmath192 so that @xmath193 lemma [ lem : goodevent ] immediately implies the following control over birth times on the event @xmath179 . [ cor : goodevent ] on the good event @xmath179 , we have the bounds @xmath194 and @xmath195 in the next two sections , we show how to upper and lower bound @xmath188 on the good event @xmath179 for @xmath196 a sequence of positive integers ( possibly constant ) . for ease of notation , we write @xmath65 instead of @xmath3 , and @xmath167 instead of @xmath197 . we shall always work on the event @xmath179 , with the properties stated in lemma [ lem : goodevent ] . recall the notation . we treat the summands @xmath76 with @xmath198 and @xmath199 separately and first consider the case @xmath199 . from corollary [ cor : goodevent ] , we deduce that for @xmath199 , @xmath200 in the sum , only the terms with @xmath201 do contribute . thus we can restrict ourselves to summands with @xmath202 , where @xmath203 setting @xmath204 , we obtain from and the fact that @xmath205 is monotone increasing in @xmath72 the almost - sure upper bound @xmath206 note that the variables @xmath207 , @xmath202 , are independent of each other , and @xmath207 has the law of the population size of a yule process with birth rate @xmath65 at time @xmath208 when started from a single individual . the summands with @xmath209 we all bound in the same ( rough ) way , by disregarding the individual birth times of the corresponding subpopulations . more specifically , we remark that on @xmath179 , @xmath210 so that , defining @xmath211 and using again monotonicity of @xmath212 , we have almost surely @xmath213 for @xmath209 , the @xmath207 s are independent and identically distributed according to the law of the population size of a yule process with birth rate @xmath65 at time @xmath214 when started from a single individual . moreover , the families @xmath215 and @xmath216 are independent of each other . letting @xmath217 we note that our above estimates yield @xmath218 our lower bound on @xmath219 begins with the trivial estimate @xmath220 that is we will consider only summands with an index @xmath199 . from corollary [ cor : goodevent ] , we see that on the good event @xmath179 , @xmath221 now let @xmath222 defining @xmath223 , we arrive with at the almost - sure lower bound @xmath224 the variables @xmath225 , @xmath226 , are independent , and @xmath225 is distributed as the population size of a yule process with birth rate @xmath65 at time @xmath227 when started from a single individual . with @xmath228 we have shown that @xmath229 note that the @xmath225 s are not independent of the event @xmath179 . by construction , we have @xmath230 almost surely for @xmath231 . the crucial step of our approach is to control the means @xmath232 $ ] and @xmath233 $ ] . in this section , we develop bounds valid for all regimes . the exact asymptotic analysis will then depend on the choices of @xmath169 and @xmath196 and will be postponed to section [ sec : limitresults ] . the behavior of both expectations will be dominated by the following integral . [ lem : integral - gamma ] let @xmath234 , @xmath235 . then @xmath236 where @xmath237 denotes the gamma - function . we let first @xmath238 and then @xmath239 to obtain @xmath240 where the last two identities follow from well - known expressions for the beta- and gamma - functions b and @xmath237 , respectively . from now on , we let for @xmath241 and @xmath234 , @xmath242 the following asymptotics as @xmath11 are immediately derived from stirling s formula . [ lem : iasymp ] when @xmath7 and @xmath168 , then @xmath243 whereas if @xmath7 and @xmath109 is fixed , then @xmath244 in order to evaluate @xmath233 $ ] , recall the definition of @xmath207 for @xmath245 . when evaluated at time @xmath80 , a yule process with birth rate @xmath65 , started from a single individual , follows the geometric law with success probability @xmath246 . for @xmath209 , this implies @xmath247 letting @xmath248 $ ] , we thus have @xmath249 in our applications , we will choose @xmath183 such that @xmath250 as @xmath251 , that is , the first @xmath252 summands are negligible . for the summands with @xmath199 , we get @xmath253=\sum_{i=2k_n}^{n^\ast}\left(1-\exp\left(-p(\ob(n)-\ub(i))\right)\right)^x,\ ] ] and it remains to analyze the sum on the right . we put @xmath254 and obtain @xmath255 in the first step , we used the fact that @xmath256 by , and for the last equality that @xmath257 by definition of @xmath258 , see . with the change of variables @xmath259 and again , the last integral is equal to @xmath260 where the bound on the right holds for @xmath6 sufficiently large , recalling that @xmath181 . for evaluating the integral on the right side , we use lemma [ lem : integral - gamma ] . we have shown that for @xmath6 large enough , @xmath261\leq \e\left[\os^{n , p}(x)\right]\leq \sigma_1^{n , p}(x ) + \left(1 + 10k_n^{-1/3}\right){\rm i}(n , p , x).}\ ] ] we turn to @xmath232 $ ] and write @xmath262= \e\left[\us^{n , p}(x)\right ] - \e\left[\us^{n , p}(x)\1_{(e_n)^c}\right].\ ] ] put @xmath263.$ ] in our applications , @xmath264 will tend to zero . indeed , provided @xmath265 $ ] remains uniformly bounded in @xmath6 , the same holds true for its second moment , see remark [ rem : strategy ] below . we get with cauchy - schwarz @xmath266 as @xmath267 concerning the first term on the right side of , we have @xmath268=\sum_{i=2k_n}^{n_\ast}\left(1-\exp\left(-p(\ub(n)-\ob(i))\right)\right)^x.\ ] ] then , with @xmath269 , a similar calculation as under shows @xmath270 putting @xmath271 , the last integral is bounded from below by @xmath272 now let @xmath273 the term @xmath274 will be negligible in our applications . using again lemma [ lem : integral - gamma ] , we have shown that @xmath275\geq \left(1 - 10k_n^{-1/3}\right){\rm i}(n , p , x)-\sigma_3^{n , p}(x),\ ] ] and therefore @xmath276\geq \e\left[\us^{n , p}(x)\1_{e_n}\right]\geq \left(1 - 10k_n^{-1/3}\right){\rm i}(n , p , x ) -\left(\sigma_2^{n , p}(x)+\sigma_3^{n , p}(x)\right).}\ ] ] for the rest of this text , we refer to the quantities @xmath277 and @xmath264 , @xmath274 as the error terms of @xmath278 $ ] and @xmath279 $ ] , respectively , whereas the term @xmath280 involving the gamma - function ( which is the same for both expectations ) is referred to as the main term . here we outline our general program how to obtain non - degenerate ( poisson ) limit laws for @xmath172 when @xmath7 and @xmath281 may depend on @xmath6 as well . we will control @xmath172 on the good event @xmath179 in terms of the lower and upper bounds @xmath282 and @xmath283 . our strategy is based on the following general observation . [ lem : upperconv ] let @xmath173 , @xmath284 be sequences of @xmath285-valued uniformly integrable random variables with @xmath286 almost surely for all @xmath55 . assume @xmath287{(d ) } v$ ] for some random variable @xmath288 . then , if @xmath289\rightarrow\e[v]$ ] , we have @xmath290{(d ) } v$ ] as @xmath11 . writing @xmath291 , the claim follows if we show that @xmath292 in probability . since @xmath286 almost surely , we have @xmath293= \e[v_n]-\e[u_n]$ ] . uniform integrability and the fact that @xmath294 in distribution implies @xmath295\rightarrow \e[v]$ ] , see , e.g. , ( * ? ? ? * theorem 3.5 ) . since @xmath289\rightarrow \e[v]$ ] by assumption , the proof is complete . we stress that instead of assuming @xmath287{(d ) } v$ ] and @xmath289\rightarrow\e[v]$ ] , one could also _ assume _ @xmath290{(d ) } v$ ] and @xmath295\rightarrow\e[v]$ ] and then _ deduce _ @xmath287{(d ) } v$ ] ( in fact , for positive integrable random variables , this can be proved by fatou s lemma via skorokhod s representation theorem , without assuming uniform integrability ) . however , for us it is more natural to work under the assumption of lemma [ lem : upperconv ] . [ rem : strategy ] uniform integrability will not pose any problem in our setting : if @xmath296 is a sum of indicators of independent events , then @xmath297\leq \e\left[v_n\right ] + \e\left[v_n\right]^2,\ ] ] implying that if @xmath298<\infty$ ] , then @xmath284 has a uniformly bounded second moment as well and is therefore uniformly integrable . we now assume that the sequence @xmath169 is fixed and satisfies @xmath36 . we ask for non - degenerate limits of @xmath188 . recall the definition of @xmath299 . our strategy consists of the following steps . we use the notation from the preceding sections . * find a sequence @xmath300 such that , for a suitable choice of @xmath301 with @xmath302 , @xmath303\geq \lambda\quad\hbox{and}\quad\limsup_{n\rightarrow\infty}\e\left[\os^{n , p_n}(x_n)\right]\leq \lambda\ ] ] for some strictly positive constant @xmath22 . by construction , this implies @xmath304\sim\e\left[n^{n , p_n}(x_n)\1_{e_n}\right]\sim\e\left[\os^{n , p_n}(x_n)\right]\sim \lambda.\ ] ] show that for @xmath301 as under step @xmath68 , with @xmath305 and @xmath306 , @xmath307 le cam s inequality applied to the sum @xmath299 gives @xmath308{(d)}\poi(\lambda)$ ] . apply lemma [ lem : upperconv ] to deduce that @xmath309{(d)}\poi(\lambda)$ ] and hence @xmath310{(d)}\poi(\lambda)$ ] as @xmath11 . note that the constant @xmath34 under step @xmath68 does not depend on the sequence @xmath301 . of course , depending on the point of view , one can also _ fix _ a sequence of thresholds @xmath196 and then _ ask _ for a choice of @xmath169 such that non - degenerate limits of @xmath188 appear . it is also of interest to understand when @xmath311 or @xmath312 in probability . in order to prove such behaviors , we will in the first case apply markov s inequality and then show that the expectation converges to zero , while for the second case , we will make use of the following lemma . [ lem : convtoinfinity ] in the setting of le cam s inequality , cf . section [ sec : poisson - heu ] , assume that @xmath313\rightarrow \infty$ ] as @xmath11 . then @xmath159{(p)}\infty$ ] , i.e. , for each @xmath314 , @xmath315 as @xmath11 . the proof follows from an application of the bienaym - chebycheff inequality . we remark that le cam s inequality can be used to obtain quantitative bounds on the rate of convergence . in our case , since we regard @xmath188 only on the good event @xmath179 , one would have to optimize the choice of this event for establishing good bounds . this will not be our concern here . we will now work out our strategy explained in the last section . even though it can be applied to any choice of @xmath3 such that @xmath316 , we will restrict ourselves to discussing three regimes of particular interest . each of them corresponds to a different limiting behavior of the ancestral subpopulation , as discussed in proposition [ prop : ancestralsubp ] . in section [ sec : poisson - heu ] , we argued heuristically that in the regime @xmath317 there should be a poissonian number of subpopulations of size @xmath318 when @xmath11 . we shall now prove this rigorously , together with the fact that there are no subpopulations of a size strictly larger than @xmath12 , and an unbounded number of subpopulations of size @xmath319 ( or , more generally , of size @xmath13 for each @xmath320 , see corollary [ cor:1 ] ) . we point to theorem 5.4 of @xcite for similar results for percolation on the complete graph , that is , the erds - rnyi random graph model . it will be convenient to use the notation @xmath321 for the number of subpopulations of size equal to @xmath319 . recall that @xmath59 and @xmath60 denote generic constants whose values may change from line to line . [ thm : finitesubp ] fix @xmath8 and @xmath10 , and assume @xmath9 . then , as @xmath11 , @xmath322{(d)}\poi(\ell!a^\ell).\ ] ] moreover , @xmath323{(p ) } 0 $ ] , and @xmath324{(p ) } \infty$ ] . our first two claims follow if we show that @xmath325{(d)}\poi(\ell!a^\ell)$ ] and @xmath323{(p ) } 0 $ ] as @xmath6 tends to infinity . we follow the strategy outlined in section [ sec : strategy ] . we choose @xmath326 and first bound @xmath327 $ ] from above , via . the error term @xmath328 is estimated by @xmath329 for the main term in , we obtain with lemma [ lem : iasymp ] as @xmath11 , @xmath330 the last two display imply @xmath331\leq \ell!a^\ell + o(1).\ ] ] we turn to a lower bound on @xmath332 $ ] , which we will establish via expression . we first show that the error terms @xmath333 and @xmath334 tend to zero . for that purpose , recall that by construction , @xmath335\leq \e\left[\os^{n , p_n}(\ell)\right]\leq c.\ ] ] hence the second moment of @xmath336 is uniformly bounded as well , see remark [ rem : strategy ] , so that with cauchy - schwarz , @xmath337 for the error term @xmath334 in , we note that @xmath338 , whence @xmath339 since the main term of agrees with that of , we get @xmath340\geq \ell!a^\ell - o(1 ) , $ ] and consequently @xmath341\sim \e[n^{n , p_n}(\ell)\1_{e_n } ] \sim \e\left[\os^{n , p_n}(\ell)\right ] \sim \ell!a^\ell\quad\hbox { for } n\rightarrow\infty.\ ] ] step @xmath68 of the strategy is therefore established ( with the constant @xmath34 there given by @xmath342 ) , and so is step @xmath70 , since @xmath343\\ & \leq c\left(1-\exp\left(-p_n(\ob(n)+\ln^2(k_n))\right)\right)^\ell = o(1).\end{aligned}\ ] ] we follow step 3 and obtain @xmath325{(d)}\poi(\ell!a^\ell)$ ] as @xmath6 tends to infinity . for the second part of the theorem , we use markov s inequality to obtain @xmath344+\p(e_n^c)\leq \e\left[\os^{n , p_n}(\ell+1)\right ] + o(1).\ ] ] using again the bound for @xmath345 $ ] , we first estimate the error term @xmath346 and then the main term similarly to above and obtain @xmath345\leq cn^{-1/\ell}$ ] . this proves @xmath323{(p ) } 0 $ ] as @xmath11 . it remains to show that @xmath324{(p)}\infty$ ] , that is for each @xmath347 , @xmath348 as @xmath6 tends to infinity . writing @xmath321 , we have seen that @xmath349 converges in distribution to a poisson random variable , so we may prove the claim for @xmath350 instead of @xmath351 . since @xmath352 almost surely , cf . , we can estimate @xmath353 we analyze the mean of @xmath354 via and obtain with stirling s formula @xmath355 \geq cn^{1/\ell}.\ ] ] an application of lemma [ lem : convtoinfinity ] shows @xmath356{(p ) } \infty$ ] and thus finishes the proof . theorem [ thm : finitesubp ] does not tell us how the number of subpopulations equal to @xmath13 behave when @xmath13 is strictly less than @xmath319 . this can however be deduced from the cases @xmath357 and @xmath358 , as we show next . [ cor:1 ] in the setting of theorem [ thm : finitesubp ] , we have @xmath359{(p ) } \infty$ ] as @xmath11 for each @xmath360 . for @xmath357 , the statement forms already part of the theorem , so we fix @xmath361 with @xmath362 . put @xmath363 . we will show that the number of subpopulations of size @xmath13 that stem from the @xmath364th individual is already unbounded as @xmath11 , see figure 1 . in this regard , we let @xmath365 denote the process that counts the individuals which are descendants of the @xmath364th individual of @xmath4 , i.e. @xmath366 is the number of individuals in the original system at time @xmath367 which have the @xmath364th individual as their common ancestor , no matter whether there are clones or mutants . it should be clear that @xmath368 evolves as a standard yule process started from one individual . next , note that by construction , @xmath369 for @xmath370 a sequence of independent standard exponentials . in particular , @xmath371= ( j/\ell)\ln n + o(1),\quad\textup{var}\left(\tau_n-\tau_{\alpha_n}\right)=o(1),\ ] ] hence @xmath372 in probability as @xmath11 . since @xmath373 converges almost surely to a standard exponential variable @xmath95 as @xmath374 , this implies @xmath375 write @xmath376 for the number of subpopulations of size @xmath377 in the system @xmath88 stopped at time @xmath98 , whose common ancestor is given by the @xmath364th individual , and similarly , define @xmath378 by counting only the subpopulations of size equals @xmath379 that stem from individual @xmath364 . obviously , @xmath380 and @xmath381 . for @xmath314 , we estimate @xmath382 we will show that the first probability tends to @xmath68 for each choice of @xmath383 , while the second can be made as small as we wish provided @xmath383 is sufficiently large . for that purpose , we remark that conditionally on @xmath384 , as a consequence of the dynamics , @xmath376 for @xmath385 has same law as @xmath386 , the number of subpopulations exceeding @xmath379 in @xmath387 . moreover , if @xmath388 with @xmath389 , the variable @xmath386 is stochastically dominated by @xmath390 ( adding individuals can only increase subpopulations ) . now fix @xmath391 . by , we find @xmath392 and @xmath393 such that for all @xmath394 , with @xmath395 and @xmath396 , the event @xmath397\right\}\ ] ] has probability at least @xmath398 . we first look at the second probability in . by our observations from above , we have for @xmath394 , @xmath399 recalling that @xmath400 , we deduce from theorem [ thm : finitesubp ] that for @xmath347 , @xmath401 the right side is smaller than @xmath402 provided @xmath383 is large enough . for the first probability in , we have similarly @xmath403 by theorem [ thm : finitesubp ] , @xmath404 in probability as @xmath6 tends to infinity , hence the probability on the right tends to @xmath68 for each choice of @xmath383 . since @xmath391 was arbitrary , this concludes the proof of the corollary . the next corollary of theorem [ thm : finitesubp ] characterizes the regime where unbounded subpopulations appear in the limit @xmath11 . for the sake of clarity , we write @xmath405 for the law of the system @xmath406 . [ cor:2 ] let @xmath407 then @xmath408 \leftrightarrow \left[p_n\,n^{1/\ell}\rightarrow\infty\textup { for each } \ell\in\n\right].\ ] ] if one of the statements fails , @xmath409 the proof is a direct application of theorem [ thm : finitesubp ] and left to the reader . parts @xmath107 and @xmath108 of proposition [ prop : ancestralsubp ] identify @xmath411 as the regime in which the ancestral subpopulation becomes non - trivial . its size however remains bounded . what are the sizes of the largest subpopulations that do appear ? as a consequence of theorem [ thm : lognsubp2 ] , we will see that if we shift the subpopulation sizes by @xmath412 for some explicit constants @xmath413 , then for any @xmath8 and any @xmath391 , we find @xmath414 such that the @xmath319 largest ( shifted ) sizes are contained in @xmath415 $ ] with probability at least @xmath398 , provided @xmath6 is large enough . while theorem [ thm : lognsubp2 ] holds true whenever @xmath416 , we will first prove theorem [ thm : lognsubp1 ] , which provides a stronger result valid for the case @xmath417 . here we will compute a correction @xmath418 of order one to the above shift , such that the number of subpopulations greater than @xmath419 converges to a poisson limit along all subsequences @xmath420 of @xmath421 , whose fractional part has a limit as @xmath422 tends to infinity . theorem [ thm : lognsubp2 ] then readily follows from adapting some estimates used in the proof of theorem [ thm : lognsubp1 ] . before we give the precise formulation of our results , we need some preparation . for @xmath10 , define @xmath423 on @xmath424 , @xmath425 is a smooth function with @xmath426 as @xmath427 and @xmath428 for @xmath374 . moreover , since on @xmath424 , @xmath429 the function @xmath425 is strictly decreasing , and there is a unique @xmath430 for which @xmath431 . see figure 2 . for @xmath432.,width=377 ] now , for @xmath10 and @xmath22 fixed and @xmath433 the root of @xmath425 , put @xmath434 we use the standard notation @xmath435 to denote the fractional part of a real @xmath436 . as we will see in the following theorem , the barrier @xmath25 defines a precise threshold , in the following sense : whenever @xmath420 is a subsequence of @xmath437 such that @xmath438 exists , then the number of subpopulations exceeding size @xmath439 converges weakly to a poisson@xmath440-distributed random variable with intensity given by @xmath441 we will see in corollary [ cor:3 ] that the restriction to subsequences with converging fractional part is actually necessary . [ thm : lognsubp1 ] assume @xmath442 for some @xmath10 . let @xmath22 , and define @xmath443 as in . let @xmath420 be a subsequence of @xmath437 such that @xmath444 for some @xmath445 . then , with @xmath446 as above , for @xmath447 , @xmath448{(d)}\poi\left(\lambda(a , b,\lambda)\right).\ ] ] we again follow the strategy explained in section [ sec : strategy ] . recall that we require @xmath302 , and we will here choose @xmath449 . putting @xmath450 , the first part of lemma [ lem : iasymp ] and a small calculation show that as @xmath6 tends to infinity , @xmath451 taking the logarithm of the right hand side , we arrive at the expression @xmath452 for @xmath425 as defined under . obviously , @xmath453 , and by taylor s formula , @xmath454 so that @xmath455 taking exponentials , and the last display show that for @xmath422 tending to infinity , @xmath456 we next control the error terms @xmath457 , @xmath458 and @xmath459 . first , recalling that @xmath460 , we have for @xmath6 sufficiently large , @xmath461 in particular , @xmath462\leq \lambda(a , b,\lambda ) + o(1),\ ] ] and , with cauchy - schwarz and remark [ rem : strategy ] , @xmath463 . finally , @xmath464 so that as @xmath447 , @xmath465\sim \e[n^{n(m),p_{n(m)}}(\lfloor y_{n(m)}\rfloor)\1_{e_{n(m ) } } ] & \sim \e\left[\os^{n(m),p_{n(m)}}(\lfloor y_{n(m)}\rfloor)\right]\\ & \sim \lambda(a , b,\lambda).\end{aligned}\ ] ] the condition under step @xmath70 is fulfilled as well , see the estimate for @xmath457 . performing step @xmath61 , it follows that @xmath466 converges to a poisson@xmath440 random variable . theorem [ thm : lognsubp1 ] implies the following remarkable weak convergence along subsequences for the size @xmath467 of the largest subpopulation at time @xmath98 . for @xmath10 , @xmath468 and @xmath433 the root of @xmath425 , put @xmath469 [ cor:3 ] let @xmath10 , @xmath442 and @xmath470 . define @xmath471 as above , and @xmath472 in terms of @xmath471 as in . let @xmath420 be any subsequence of @xmath421 such that @xmath444 for some @xmath445 . then , for @xmath447 , with @xmath473 , @xmath474 the probability on the left hand side is equal to @xmath475 with @xmath476 defined as in . the last asymptotics holds thanks to theorem [ thm : lognsubp1 ] . the claim follows . the limit expression on the right hand side in the above corollary is the distribution function of the gumbel law with location parameter @xmath477 and scale parameter @xmath70 . in view of our heuristics explained in section [ sec : poisson - heu ] and of what is known about the maximum of @xmath6 independent geometrically distributed random variables , it should not come as a surprise that a gumbel distribution appears in the limit for the recentered size of the largest subpopulation . since @xmath478 is a discrete random variable and the gumbel law is a continuous distribution , convergence along the full sequence can not hold . it is instructive to compare our previous two results with theorem 5.10 and corollary 5.11 of @xcite for the erds - rnyi random graph model . from the proof of theorem [ thm : lognsubp1 ] , we easily derive that in the general case @xmath410 , the sizes of the largest subpopulations are concentrated around @xmath479 [ thm : lognsubp2 ] assume @xmath410 for some @xmath10 . define @xmath480 as in , and let @xmath196 be a sequence of positive integers . then the following holds as @xmath11 . 1 . if @xmath481 , then @xmath482 2 . if @xmath483 , then @xmath484 we only have to adapt the estimates given in the proof of theorem [ thm : lognsubp1 ] , so we will only sketch the necessary modifications . we again choose @xmath449 . we treat @xmath107 and @xmath108 together , and in this regard , we note that for @xmath108 , by monotonicity of @xmath485 we can assume that @xmath486 . with @xmath25 everywhere replaced by @xmath197 , we deduce from that if @xmath481 , then @xmath487 , whereas if @xmath483 , then @xmath488 . the error term @xmath489 is seen to be of order @xmath490 with exactly the same argument as in the proof of theorem [ thm : lognsubp1 ] , and so is the error term @xmath491 , using here that @xmath492 . now if @xmath481 , we have by markov s inequality and , @xmath493 + o(1 ) = o(1),\ ] ] and if @xmath483 , then , as in the second part of the proof of theorem [ thm : finitesubp ] , for any @xmath347 , @xmath494 the probability on the right tends to @xmath68 by an appeal to lemma [ lem : convtoinfinity ] . in the regime @xmath496 , the ancestral subpopulation grows like @xmath16 , see part @xmath120 of proposition [ prop : ancestralsubp ] . as we shall prove in the following theorem , when @xmath31 , the sizes of the largest subpopulations are to first order given by @xmath32 . for the case @xmath497 and @xmath22 , we will show that the number of subpopulations greater than @xmath498 is asymptotically poisson@xmath100-distributed . note that as @xmath11 , @xmath499 , cf . figure 3 . as it will become clear from the proof , we require @xmath33 ( and not merely @xmath496 ) only to simplify the calculation ; without this assumption , the exact behavior of @xmath3 has to be taken into account more carefully . [ thm : greatsubp ] assume @xmath500 . then the following holds as @xmath11 . 1 . if @xmath501 , then @xmath502{(p ) } 0 $ ] . if @xmath503 , then @xmath504{(p ) } \infty$ ] . 2 . assume additionally @xmath33 for large @xmath6 . then , with @xmath197 as in , @xmath505{(d)}\poi(\lambda).\ ] ] we work with the choice @xmath506 and follow the steps of the strategy presented in section [ sec : strategy ] . since our proof is similar to the proofs of theorem [ thm : finitesubp ] and theorem [ thm : lognsubp1 ] , we do not provide every detail . we fix a real number @xmath507 and let @xmath508 , @xmath55 . for the main term @xmath509 , lemma [ lem : iasymp ] shows @xmath510 note that since @xmath496 , we have @xmath511 as @xmath11 . by taylor s formula , @xmath512 from the last display , we deduce after a short calculation that the logarithm of the right hand side of behaves asymptotically like @xmath513 in particular , if @xmath501 , then the right hand side diverges to @xmath514 , whereas if @xmath503 , it diverges to @xmath515 . consequently , @xmath516 in the first and @xmath517 in the second case . in order to finish the proof of @xmath107 , it remains to convince ourselves that the error terms @xmath518 and @xmath519 do not contribute when @xmath6 tends to infinity . for the first one , we have @xmath520 since @xmath521 , we see from taking logarithms that @xmath522 , as wanted . the error term @xmath519 is readily seen to be of order @xmath490 as well , and the claims under @xmath107 now follow from the same arguments as in the proof of theorem [ thm : lognsubp2 ] ( or theorem [ thm : finitesubp ] ) . we turn to @xmath108 and fix @xmath22 . if @xmath33 for @xmath6 large , then , with @xmath197 defined as in , @xmath523 performing the above calculation with @xmath25 replaced by @xmath197 and @xmath524 by @xmath525 we deduce from that the logarithm of right side in behaves as @xmath526 , and thus @xmath527 part @xmath108 now follows from the same line of reasoning as in the proof of theorem [ thm : lognsubp1 ] , using that all the error terms @xmath528 , @xmath529 , are of order @xmath490 . in this final section , we make the precise link between the population system @xmath530 defined in section [ sec : yule ] and percolation on random recursive trees . a recursive tree with vertex set @xmath531=\{1,\ldots , n\}$ ] is a tree rooted at @xmath68 with the property that the labels along the unique path from the root to any other vertex form an increasing sequence . a tree chosen uniformly at random among all these @xmath532 recursive trees is called random recursive tree and denoted @xmath37 . the study of bernoulli bond percolation on large random recursive trees can be traced back to the analysis of an algorithm for isolating the root by meir and moon @xcite ; see also drmota _ et al . _ @xcite , iksanov and mhle @xcite and kuba and panholzer @xcite for more recent developments in this direction . it further plays a key role in the construction of the bolthausen - sznitman coalescent by goldschmidt and martin @xcite . we choose @xmath533 and write @xmath534 for the percolation clusters of a bernoulli bond percolation on @xmath37 with parameter @xmath65 . that is , we erase each edge of @xmath37 with probability @xmath1 and independently of each other , and enumerate the connected components according to the increasing order of the label of their root vertex ( i.e. their vertex with the smallest label ) . we use the convention that @xmath535 if there are less than @xmath76 connected components after percolation on @xmath37 . with our ordering , @xmath536 always represents the root cluster containing @xmath68 ; @xmath537 is the cluster rooted at the smallest vertex which does not belong to the root cluster @xmath536 , and so on . we write @xmath538 for the number of vertices of @xmath539 . then the following holds . [ lem : equivrrtyule ] let @xmath126 $ ] . the families @xmath540 and @xmath541 have the same law . we construct percolation on a random recursive tree as a growth process in continuous time as follows . at time @xmath542 , we start from the singleton @xmath543 . given percolation with parameter @xmath65 on a random recursive tree on @xmath544 $ ] , @xmath545 , has been constructed , we equip each vertex @xmath546 $ ] with an independent exponential clock @xmath547 of parameter @xmath68 . at time @xmath548 , we flip a coin with heads probability @xmath65 . if head shows up , we attach a vertex labeled @xmath549 to the vertex with label arg@xmath548 . otherwise , we add vertex @xmath549 to the system without connecting it to any other vertex . it follows from the construction of random recursive trees and the independence of the coin flips that we observe at the instant when the @xmath6th vertex is incorporated a bernoulli bond percolation on @xmath37 with parameter @xmath65 . moreover , if we keep track of the sizes @xmath550 of the growing percolation clusters , where @xmath551 stores the size of the @xmath76th cluster at time @xmath72 and clusters are ordered according to their birth times , we obtain a markov chain with initial state @xmath552 and transition rates at time @xmath80 for @xmath553 given by @xmath554 from the very construction of @xmath88 , it follows that the processes @xmath555 and @xmath556 grow according to the same dynamics , and the statement follows . if we think of the individuals of @xmath88 as being labeled @xmath557 , according to the increasing order of their birth times , there is in fact a one - to - one correspondence between subpopulations and clusters that respects the genealogy , as illustrated by figure 4 . with the above lemma at hand , the results from section [ sec : limitresults ] have a direct interpretation in terms of clusters stemming from a percolation on @xmath37 with parameter @xmath3 . for example , corollary [ cor:2 ] gives a necessary and sufficient condition on the percolation parameter @xmath3 such that in the limit @xmath11 , clusters of unbounded size appear . more generally , the strategy developed in section [ sec : strategy ] gives a concise tool to decide whether for a given percolation sequence @xmath36 and thresholds @xmath197 , there are clusters of a size of order @xmath197 , and if so , how many . indeed , as exemplified in section [ sec : limitresults ] , basically one only has to check the asymptotic behavior of the expression @xmath558 introduced below lemma [ lem : integral - gamma ] . this work completes the study of percolation on random recursive trees initiated in @xcite , where percolation with supercritical parameter @xmath559 , @xmath10 fixed , is studied . in @xcite , non - gaussian fluctuations of the root cluster were proved , and the analysis of percolation clusters was extended in @xcite to all regimes @xmath39 . the works @xcite do additionally contain information on the genealogy of clusters ( and not merely on their sizes ) , a type of question we did not investigate here . as a common feature when @xmath39 , the root cluster containing @xmath68 has size @xmath560 , while the sizes of the next largest clusters are of order @xmath561 . this motivated a sub - classification of regions into weakly supercritical [ @xmath562 when @xmath563 , supercritical [ @xmath564 , @xmath565 fixed ] when @xmath566 , and strongly supercritical [ @xmath567 when @xmath568 . the regime of constant @xmath5 may best be termed `` critical '' , since in this case , the root cluster loses its dominating role ( with respect to the size ) . some results can be found in @xcite and @xcite , although the motivation there is somehow different . for similar reasons , we term the regime @xmath7 considered here `` subcritical '' . the classification into different regimes draws on the terminology which is usually used to describe the erds - rnyi random graph model ( see , e.g. , @xcite ) . we stress that there are other natural families of trees which can be grown according to a probabilistic evolution algorithm , like , e.g. , scale - free random trees or @xmath569-ary random increasing trees . indeed , branching systems with mutations were used in @xcite to study percolation on scale - free random trees when @xmath15 , and in a similar fashion by berzunza in @xcite for @xmath569-ary random increasing trees . in the case of random recursive trees , the underlying population system is particularly simple , so we restricted our discussion of the subcritical regime to these trees , but we certainly expect similar results to hold true for other classes of increasing tree families . in this appendix , we will construct an event @xmath179 with the `` good '' properties specified in lemma [ lem : goodevent ] . we first collect some properties of yule processes . for @xmath126 $ ] , denote by @xmath570 a yule process with birth rate @xmath65 per unit population size . given @xmath571 , write @xmath572 for its law under which @xmath573 , and similarly @xmath574 for its expectation . it is well - known that @xmath575 , @xmath80 , is a square - integrable martingale . under @xmath572 and for @xmath234 , its terminal value @xmath576 is distributed as a sum of @xmath577 standard exponentials and follows thus the gamma@xmath578-law . note that doob s inequality applied to the square - integrable martingale @xmath579 gives we now work in the setting of section [ sec : yule ] . recall the definition of the process @xmath581 counting the number of different genetic types in the population system @xmath88 , starting from @xmath582 . for @xmath74 , @xmath583 denotes the birth time of the @xmath76th individual in the system @xmath88 ( counting both clones and mutants ) , with @xmath4 denoting the underlying standard yule process . the time @xmath584 denotes the first time when an individual of genetic type @xmath76 appears , i.e. , the @xmath584 s are the jump times of @xmath585 . we need the following control over @xmath586 . [ lem : numberoftypes ] let @xmath5 . for @xmath385 , let @xmath587 be the conditional law given @xmath588 ( i.e. the first @xmath190 children are all mutants ) , and let @xmath589 be its expectation . for every @xmath80 , we have @xmath590\leq 4(1-p)k(\e^{t}-1),\ ] ] and @xmath591\leq 324(1-p)k\,\e^{-2t/3}.\ ] ] let @xmath592 denote the natural filtration generated by the process @xmath555 . both processes @xmath586 and @xmath4 are @xmath592-adapted , and @xmath593 is a stopping time . we work now under the probability measure @xmath587 . from the strong markov property and the dynamics of @xmath88 described above , we see that @xmath594 is a martingale with @xmath595 . its quadratic variation is given by @xmath596(s)=t^{(p)}(\tau_k+s)-k$ ] , and its second moment takes the form @xmath597&=\e_{|k}\left[[m^{(p)}](s)\right]=\e_{|k}\left[t^{(p)}(\tau_k+s)\right]-k\\ & = \e_{|k}\left[\int_{\tau_k}^{\tau_k+s}(1-p)z(r)\dt r\right]=\int_{0}^{s}(1-p)\e[z(\tau_k+r)]\dt r= ( 1-p)k(\e^{t}-1).\end{aligned}\ ] ] for the last equality , we used the strong markov property , which entails that @xmath598 is independent of @xmath599 and distributed as a standard yule process started from @xmath379 individuals . an appeal to doob s inequality gives @xmath600 \leq 4 \e_{|k}\left[|m^{(p)}(t)|^2\right]=4(1-p)k(\e^{t}-1).\ ] ] for the second statement , we bound for every @xmath55 @xmath601 by doob s inequality , see the first part , the @xmath602-norm of the right side ( with respect to the conditional measure @xmath587 ) is bounded by @xmath603 thus , summing over @xmath6 and applying the triangle inequality , @xmath604 squaring both sides , the claim follows . * proof of lemma [ lem : goodevent ] : * we assume @xmath36 and fix any sequence of integers @xmath606 satisfying @xmath181 and @xmath182 as @xmath11 . concerning property @xmath107 in lemma [ lem : goodevent ] , we first note that since each new - born child is a mutant with probability @xmath607 , we have @xmath608 moreover , by , we also have @xmath609 , so that the event @xmath610 has probability @xmath611 as @xmath6 tends to infinity . we turn to property @xmath108 . since conditionally on @xmath612 , @xmath613 is a yule process started from @xmath183 individuals , we obtain from and chebycheff s inequality @xmath614 in particular , if we let @xmath615 then @xmath616 . finally , for property @xmath120 we need control over @xmath617 . in this regard , lemma [ lem : numberoftypes ] in combination with chebycheff s inequality shows that conditionally on @xmath618 the event @xmath619 has probability at least @xmath620 . on @xmath621 , we note that for @xmath6 sufficiently large , recalling that @xmath302 , @xmath622 and similarly @xmath623 setting @xmath624 we have constructed an event of probability @xmath611 on which for @xmath6 sufficiently large , @xmath107 , @xmath108 and @xmath120 in the statement of lemma [ lem : goodevent ] are fulfilled .

we calculate the transition form factors in the framework of perturbative qcd to leading power of , being the meson mass . we explain the basic principle by discussing the pion electromagnetic form factor . it is shown that the logarithmic and linear singularities occurring at small momentum fractions of light meson distribution amplitudes do not exist in a self - consistent perturbative analysis , which includes and threshold resummations . epsfig.sty # 1[eq:#1 ] # 1#2#3 _ phys . rev . _ * d#1 * # 2 ( # 3 ) # 1#2#3 _ phys . rev . lett . _ * # 1 * # 2 ( # 3 ) # 1#2#3 _ phys . lett . _ * # 1 * # 2 ( # 3 ) = -10 mm = -6 mm = -6 mm = 10000 * ncku - hep-01 - 04 + dpnu-00 - 13 * 1.cm 0.8 cm * .3 cm t. kurimoto , hsiang - nan li and a.i . sanda * * _ department of physics , toyama university , toyama 930 - 8555 , japan _ * * _ department of physics , national cheng - kung university , + tainan , taiwan 701 , republic of china _ * * _ national center for theoretical sciences , hsinchu , taiwan 300 , republic of china _ * * _ department of physics , nagoya university , nagoya 464 - 8602 , japan _ * 2.0 cm * pacs index : 13.25.hw , 11.10.hi , 12.38.bx , 13.25.ft *

hep-ph0105003

branching ratios of @xmath3 meson two - body nonleptonic decays have been measured by cleoiii , belle and babar collaborations @xcite . cp violations in these modes may be observed in near future . cognizant of this point , we have presented some theoretical anticipations for the @xmath10 @xcite , @xmath11 @xcite , and @xmath12 @xcite decays in the perturbative qcd ( pqcd ) framework . in particular , @xmath13 cp violation is expected in the @xmath10 decays . the @xmath14 transition form factors are the integral part of two - body nonleptonic decay amplitudes . in this paper we shall convince readers that these form factors in the large recoil region of light mesons are calculable in pqcd . this is where our approach starts to differ from other approaches to exclusive @xmath3 meson decays . according to pqcd factorization theorem , a form factor is written as the convolution of a hard amplitude with initial - state and final - state hadron distribution amplitudes @xmath15 , where @xmath16 is the momentum fraction associated with one of the partons . it has been pointed out that perturbative evaluation of the pion form factor suffers nonperturbative enhancement from the end - point region with a momentum fraction @xmath17 @xcite . if this is true , the hard amplitude is characterized by a low scale , such that expansion in terms of a large coupling constant @xmath18 is not reliable . more serious end - point ( logarithmic ) singularities have been observed in the twist-2 ( leading - twist ) contribution to the @xmath19 transition form factor @xcite . the singularities even become linear at twist 3 ( next - to - leading twist ) @xcite . because of these singularities , it was claimed that the @xmath19 form factor is dominated by soft dynamics and not calculable in pqcd @xcite . we shall argue that this conclusion is false . we shall show that at the end points , where the above singularities occur , the double logarithms @xmath20 should be resummed in order to justify perturbative expansion . the result , called threshold resummation @xcite , leads to strong sudakov suppression at @xmath17 @xcite . therefore , the end - point singularities do not exist in a self - consistent pqcd analysis . in this work we shall investigate contributions to the @xmath21 and @xmath22 transition form factors from twist-2 and from two - parton twist-3 distribution amplitudes . in sec . ii we illustrate the pqcd formalism by studying the pion electromagnetic form factor . we review the reasoning why one might conclude that the form factor is not calculable , and explain why these objections are not justified in qcd . in secs . iii and iv we derive the @xmath3 meson transition form factors . it will be shown that the twist-3 contributions , which seem to be proportional to @xmath23 or @xmath24 , do not vanish in the @xmath25 limit . here @xmath26 , @xmath27 , and @xmath2 are the chiral symmetry breaking scale , @xmath28 meson mass , and @xmath3 meson mass , respectively . we record our results of the form factors at large recoil : the @xmath19 form factor @xmath29 and the @xmath22 form factor @xmath30 . meson distribution amplitudes are defined and the sudakov factor from threshold resummation is derived in the appendices . the suggestion that a hadronic form factor is calculable in pqcd was first made in ref . the rough idea is summarized as follows . one expands the bound - state wave function for a pion in terms of fock states containing on - shell partons ( quarks or gluons ) @xcite , @xmath31 define a soft function @xmath32 at a typical hadronic scale @xmath33 as the initial wave function , @xmath34 the wave function @xmath35 can be related to @xmath32 via @xmath36 where @xmath37 is an irreducible kernel and @xmath38 the green function involving only hard loop momenta . the pion electromagnetic form factor @xmath39 is then expressed as a convolution integral , @xmath40 with @xmath41 being the momentum of the initial - state pion , @xmath42 large momentum transfer , and @xmath43 . here we have written the parton momenta associated with the initial state and final state as @xmath44 and @xmath45 , respectively , in the notation @xmath46 , and made explicit the dependence of the two - parton wave function @xmath47 on @xmath41 and @xmath48 . the first term in eq . ( [ fef ] ) contains leading contributions , and ellipses represent those from higher fock states , which are down by powers of @xmath49 in the light - cone gauge and by powers of @xmath18 . the leading diagrams are displayed in fig . [ leading ] . it can be shown that the large momentum transfer @xmath50 flows through the hard amplitude @xmath51 , and that all nonperturbative dynamics goes into wave functions . one can therefore compute @xmath51 perturbatively . however , it was pointed out that the above argument suffers a grave difficulty @xcite : the diagrams in fig . 1 may be infrared divergent , because important contribution to the form factor comes from the region , where the exchanged gluons are soft . pqcd is then not applicable . below we shall examine this difficulty in more details . according to eq . ( [ fef ] ) , the first diagram in fig . [ leading ] gives @xmath52 with the color factor @xmath53 , the number of colors @xmath54 , and the hard amplitude @xmath55^{\gamma\delta } \frac{1}{(k_2-k_1)^2}\left[\gamma^\sigma\frac{\not k_2-\not p_1 } { ( p_1-k_2)^2}\gamma_\mu \right]^{\alpha\beta}\;. \label{perp}\end{aligned}\ ] ] write @xmath56 . if we ignore @xmath57 , it has been shown that the integral in eq . ( [ pff ] ) is dominated by contributions from the end - point regions with @xmath58 , @xmath59 . if the pion wave function does not vanish at @xmath17 , the integral will be even infrared divergent . if we somehow regulate the infrared singularity by an appropriate choice of the wave function , the running coupling constant @xmath60 evaluated at the hard gluon momentum @xmath61 is still too large to make sense out of the perturbative expansion . the above end - point singularity corresponds to the picture of the pion form factor feynman had in mind . in the so - called brick - wall frame the initial - state pion with momentum @xmath62 is struck by a space - like current of momentum @xmath63 , and turns around with momentum @xmath64 as shown in fig . [ feynman ] . feynman pointed out that the major contribution to the form factor comes from the region , where one of the partons carries the full pion momentum . the rest of partons , being all wee partons , do not know in which direction they are moving . the resulting configuration is essentially identical to the initial - state pion except that the momentum of the fast parton is reversed . hence , feynman claimed that the @xmath50 dependence of the pion form factor is related to the probability of finding a single parton carrying all the pion momentum . feynman s picture is consistent with the statement that the form factor is dominated by the singular part of eq . ( [ perp ] ) . because it is singular , we can not compute the form factor . we argue that feynman s picture of the pion form factor is false . consider a qed example . when an electron undergoes hard scattering , it can not help but emit infinitely many photons in the direction of the electron momentum . as a consequence , the elastic scattering cross section @xmath65 at finite angle vanishes at high energy , implying that the probability for the final @xmath66 state being accompanied by no photons diminishes . in other words , the final state must be accompanied by many photons . in the qcd case of the pion form factor , when a quark inside the pion gets hit by a current , the final state will contain many gluons unless the spectator quark is nearby to shield the color charge . when one of the quarks carries all the momentum , the rest of the pion can not shield the color charge of the fast quark , and many gluons will be emitted in arbitrary directions during the hard scattering . thus , the final configuration ending up as a single pion is extremely unlikely . this is so - called sudakov suppression on exclusive processes at kinematic end points . therefore , the contribution from fig . [ feynman ] is negligible , and the end - point singularity does not exist ! in the above argument we have ignored the term @xmath57 in eq . ( [ pff ] ) . small @xmath4 in momentum space corresponds to large transverse distance @xmath67 of two valence quarks . the color charge of the quark , which is struck by the current , is not shielded in this large @xmath67 region , and will emit many gluons . the probability for having a single pion in the final state is then vanishingly small . that is , this configuration can not contribute to the form factor . hence , the momentum space with @xmath68 , where the end - point singularity occurs , is also sudakov suppressed @xcite . the typical behavior of the sudakov factor @xmath69 $ ] , @xmath70 , which is associated with the struck quark , is shown in fig . [ suda2 ] . we observe that the sudakov factor decreases fast at large @xmath67 for @xmath71 ( @xmath72 ) , which corresponds precisely to the end - point region in eq . ( [ perp ] ) . in conclusion , the end - point singularity is absent , and the major contribution to fig . [ leading ] comes from the region with hard gluon exchanges . as derived in the appendix a , a light - cone pion distribution amplitude is written as @xmath73_{\beta\gamma}\phi_\pi(x ) + [ \gamma_5]_{\beta\gamma}m_0\phi^p_\pi(x)\right . \nonumber\\ & & \left.+m_0[\gamma_5(\not n_+\not n_--1)]_{\beta\gamma } \phi^t_\pi(x)\right\}\ ; , \label{fpd}\end{aligned}\ ] ] where @xmath74 is the pion momentum , the light - like vector @xmath75 the coordinate of the @xmath76 quark , and the dimensionless vector @xmath77 parallel to @xmath78 and @xmath79 parallel to @xmath80 . here a four - vector has been expressed in terms of light - cone coordinates , @xmath81 the distribution amplitude @xmath82 is twist-2 , and @xmath83 and @xmath84 proportional to @xmath85 gev , where @xmath86 is the current quark mass of the quark @xmath42 , are twist-3 . the origin of these terms can be simply understood by means of the field - current identity from chiral symmetry , @xmath87 it is easy to observe that twist-3 contributions are suppressed by a power of @xmath88 . the asymptotic behaviors of @xmath82 , @xmath83 and @xmath84 are known to be @xmath89 as the hard amplitude in eq . ( [ perp ] ) is convoluted with these distribution amplitudes , we find that twist-2 contribution is finite , while twist-3 ones are logarithmically divergent without sudakov suppression . the sudakov factor then introduces an effective cut - off to the integral at @xmath90 , and the twist-3 contributions are proportional to @xmath91 . that is , the power counting is not altered by a logarithmic divergence in the factorization formula . as shown later , the power counting for contributions to the @xmath3 meson transition form factors is modified by linear divergences in the factorization formulas . therefore , the different end - point behavior leads to different power counting rules for the pion form factor and for the @xmath3 meson transition form factors . in the @xmath3 meson rest frame , we define the @xmath3 meson momentum @xmath41 and the pion momentum @xmath92 in the light - cone coordinates : @xmath93 with the energy fraction @xmath94 carried by the pion . the spectator momenta @xmath48 on the @xmath3 meson side and @xmath95 on the pion side are parametrized as @xmath96 note that the four components of @xmath48 should be of the same order , @xmath97 , with @xmath98 , @xmath99 being the @xmath67 quark mass . however , since @xmath95 is mainly in the plus direction with @xmath100 , the hard amplitudes will not depend on the plus component @xmath101 as explained below . this is the reason we do not show @xmath101 in eq . ( [ k12 ] ) explicitly . consider the configuration for the semileptonic decay @xmath102 depicted in fig . [ aa ] , which corresponds to soft contribution to the @xmath19 form factor @xmath103 . the @xmath104 quark and the lepton pair fly back to back with energy of @xmath105 . the spectator quark @xmath76 carries a momentum of @xmath97 . if this configuration is responsible for the decay , it is impossible to compute @xmath103 using pqcd . however , applying an argument similar to that used for the pion form factor , we know that the @xmath104 quark recoiling against the lepton pair is bound to emit infinitely many gluons . thus , fig . [ aa ] in fact corresponds to the inclusive decay @xmath106 . the probability that the final state in fig . [ aa ] contains only a single pion is suppressed by the sudakov form factors . a quantitative estimate of sudakov suppression of the soft contribution to @xmath103 in the qcd sum rule formalism will be discussed later . it has been explained that the internal @xmath107 quark involved in the hard amplitude becomes on - shell as the momentum fraction @xmath16 of the @xmath76 quark vanishes @xcite . the contributions to the @xmath21 form factor @xmath103 are then logarithmically divergent at twist 2 and linearly divergent at twist 3 . we argue that as the end - point region is important , the corresponding large double logarithms @xmath20 need to be organized into a jet function @xmath108 as a consequence of threshold resummation @xcite . this jet function vanishes as @xmath109 , and modifies the end - point behavior of meson distribution amplitudes effectively . this modification provides a plausible explanation for the model of the twist-3 pion distribution amplitude proportional to @xmath110 , which was adopted in @xcite . our numerical study shows that the results of the @xmath19 form factor obtained in this work are almost the same as those obtained in @xcite . in the following analysis we shall employ the approximate form , @xmath111^c\ ; , \label{trs}\end{aligned}\ ] ] where the parameter @xmath112 comes from the best fit to the next - to - leading - logarithm threshold resummation in moment space . note that the jet function @xmath113 is normalized to unity . for details of the derivation , refer to the appendix d. similarly , the inclusion of @xmath4 regulates the end - point singularities , and large double logarithms @xmath114 are produced from higher - order corrections . these double logarithms should be also organized to all orders , leading to @xmath4 resummation @xcite . the resultant sudakov form factor , whose explicit expression can be found in our previous works @xcite , controls the magnitude of @xmath115 to be roughly @xmath116 by suppressing the region with @xmath117 . the coupling constant @xmath118 is then small enough to justify the pqcd evaluation of heavy - to - light form factors @xcite . we emphasize that the hard scale for heavy - to - light decays must be @xmath119 in order to define a gauge - invariant @xmath3 meson distribution amplitude @xcite . we shall include the sudakov factor associated with the light spectator quark of the @xmath3 meson . whether this factor is essential will be determined by the @xmath3 meson distribution amplitude . since the @xmath3 meson is dominated by soft dynamics with @xmath120 , the associated sudakov effect is minor compared to that from the energetic pion . with the possible order of magnitude of @xmath121 , a taylor expansion of the hard gluon propagator near the end point , @xmath122 is certainly not appropriate . a more reasonable treatment is to keep @xmath115 in the denominators of internal particle propagators , and to drop @xmath115 in the numerators , which are power - suppressed compared to other @xmath123 terms . under this prescription , the sudakov factor from @xmath4 resummation can be introduced into pqcd factorization theorem without breaking gauge invariance of the hard amplitudes . for the same reason , the terms proportional to @xmath124 in the numerators should be neglected . it is then obvious from eq . ( [ fex ] ) that the hard amplitudes are independent of the component @xmath101 . the @xmath101 dependence of the @xmath3 meson wave function can then be integrated out @xcite , leading to the parametrization in eq . ( [ k12 ] ) . note that the mechanism of threshold and @xmath4 resummations is similar with the former responsible for suppression in the longitudinal direction and the latter for suppression in the transverse direction . as shown below , both twist-2 and twist-3 contributions are well - behaved after including threshold and @xmath4 resummations . hence , the contributions to @xmath103 from fig . [ aa2 ] dominate in the large recoil region . in this configuration the @xmath76 quark gains a large momentum parallel to the @xmath104 quark momentum by exchanging a hard gluon with the @xmath107 or @xmath104 quark . we compute the @xmath19 form factors @xmath125 and @xmath126 defined by the following matrix element , @xmath127 + f_0(q^2)\frac{m_b^2-m_\pi^2}{q^2}q_\mu\;,\end{aligned}\ ] ] where @xmath128 is the lepton - pair momentum . another equivalent definition is @xmath129 in which the form factors @xmath130 and @xmath131 are related to @xmath125 and @xmath126 by @xmath132 the factorization formula for the @xmath19 form factors is written as @xmath133 the pion distribution amplitude @xmath134 has been supplied in eq . ( [ fpd ] ) , and the @xmath3 meson wave function is given by ( see the appendix c ) @xmath135_{\alpha\delta}\;. \label{bwp1}\end{aligned}\ ] ] employing eqs . ( [ fpd ] ) and ( [ bwp1 ] ) , we derive , from eq . ( [ fbpi ] ) , @xmath136 e(t^{(1)})h(x_1,x_2,b_1,b_2)\ ; , \label{f1}\\ f_2&=&16\pi m_b^2c_f\int dx_1dx_2\int b_1db_1 b_2db_2\phi_b(x_1,b_1 ) \nonumber\\ & \times & \biggl\{\left[\phi_\pi(x_2)(1+x_2\eta ) + 2r_\pi\left((\frac{1}{\eta } -x_2 ) \phi_\pi^t(x_2 ) -x_2\phi_\pi^p(x_2 ) \right)\right]e(t^{(1)})h(x_1,x_2,b_1,b_2 ) \nonumber\\ & & + 2r_\pi\phi_\pi^p e(t^{(2})h(x_2,x_1,b_2,b_1)\biggr\}\ ; , \label{fpi}\end{aligned}\ ] ] with the ratio @xmath137 and the evolution factor @xmath138 in the above formulas we have dropped the terms proportional to the momentum fraction @xmath120 as argued before , which are power - suppressed compared to the leading terms such as @xmath139 in the form factor @xmath131 . the explicit expressions of the sudakov exponents @xmath140 and @xmath141 are referred to @xcite . the hard function is written as @xmath142\ ; , \label{dh}\end{aligned}\ ] ] where the factor @xmath113 suppresses the end - point behaviors of the pion distribution amplitudes , especially of the twist-3 ones . the hard scales @xmath143 are defined as @xmath144 it is obvious that if turning off threshold and @xmath4 resummations with @xmath18 fixed , eqs . ( [ f1 ] ) and ( [ fpi ] ) are infrared divergent . we argue that the two - parton twist-3 distribution amplitudes @xmath145 , though proportional to the ratio @xmath23 , need to be taken into account . as stated above , the corresponding convolution integrals for the @xmath19 form factor are linearly divergent without including sudakov effects . these integrals , regulated in some way with an effective cut - off @xmath146 , are proportional to the ratio @xmath147 . combining the two ratios @xmath23 and @xmath147 , the twist-3 contributions are in fact not down by a power of @xmath1 : @xmath148 and should be included in a complete leading - power analysis . we emphasize that the presence of linear divergences modifies the power counting rules , making the difference between the @xmath3 meson transition form factors and the pion form factor . various computing methods have been proposed for the evaluation of the @xmath19 transition form factors @xmath149 in the literature , such as the lattice technique @xcite , light - cone qcd sum rules @xcite , and pqcd @xcite . obviously , lattice calculations become more difficult in the large recoil region of the light meson . however , this region is the one where pqcd is reliable , indicating that the pqcd and lattice approaches complement each other . this complementation will be explicitly exhibited in fig . [ bpi ] below . in light - cone sum rules , dynamics of the @xmath19 form factors have been assumed to be dominated by the large scale of @xmath150 . this is the reason twist expansion into fock states in powers of @xmath151 applies to the pion bound state . if this assumption is valid , pqcd should be also applicable to the @xmath19 form factors . besides , large radiative correction to the @xmath3 meson vertex , which reaches 35% of the full contribution , or about half of the soft ( zeroth - order ) contribution , has been noticed . this @xmath152 correction renders the sum rule for @xmath153 , with @xmath154 being the @xmath3 meson decay constant , quite unstable relative to the variation of input parameters @xcite . to stabilize the sum rule , one considers another sum rule for @xmath154 at the same time , which also receives large radiative correction to the @xmath3 meson vertex . the two large vertex corrections then cancel in the ratio @xmath155 . however , the radiative correction to @xmath154 is then large . a careful look at the light - cone - sum - rule analyses indicates that the soft contribution is more sensitive to the end - point ( @xmath17 ) behavior of the pion distribution amplitude than the @xmath152 correction @xcite . hence , if the end - point behavior of the pion distribution amplitude is modified by the sudakov factor in this work , such that the end - point contribution is not important , perturbative contribution can become dominant . the sudakov effect on the soft contribution to @xmath156 has been investigated in the qcd sum rule formalism @xcite ( without twist expansion for the pion bound state ) . in this analysis , the soft contribution without sudakov suppression was estimated to be between 0.15 ( corresponding to @xmath157 mev ) and 0.22 ( corresponding to @xmath158 mev ) . the soft contribution to @xmath153 obtained in @xcite is consistent with the above range . it was then shown that the sudakov effect decreases the soft contribution by a factor 0.4 - 0.7 , depending on infrared cut - offs for loop corrections to the weak decay vertex . therefore , the soft contribution turns out to be about 0.06 - 0.15 . compared with the lattice results @xmath159 , it is reasonable to conclude that the soft contribution amounts to about 30% , which is consistent with the observation made in @xcite . it is a fair opinion that the estimate of soft contribution is more model - dependent than perturbative one . for example , perturbative contribution is less sensitive to the pion distribution amplitude or to other input parameters such as the borel mass in light - cone sum rules @xcite . in the pqcd approach we calculate the perturbative contribution to @xmath103 , which is more model - independent , and show that the result can more or less saturate the value predicted by lattice technique . for the @xmath3 meson distribution amplitude , we adopt the model @xmath160\ ; , \label{os}\end{aligned}\ ] ] with the shape parameter @xmath161 gev @xcite . the normalization constant @xmath162 is related to the decay constant @xmath163 mev through the relation @xmath164 it is easy to find that eq . ( [ os ] ) has a maximum at @xmath165 . we employ the models for the pion @xcite , @xmath166\ ; , \label{pioa}\\ \phi_\pi^p(x)&=&\frac{f_\pi}{2\sqrt{2n_c } } \left[1 + 0.43c_2^{1/2}(2x-1)+0.09c_4^{1/2}(2x-1)\right]\ ; , \label{piob}\\ \phi_\pi^t(x)&=&\frac{f_\pi}{2\sqrt{2n_c } } ( 1 - 2x ) \left[1 + 0.55(10x^2 - 10x+1)\right]\ ; , \label{pioc}\end{aligned}\ ] ] with the pion decay constant @xmath167 mev . the gegenbauer polynomials are defined by @xmath168 whose coefficients correspond to @xmath169 gev . we first investigate the relative importance of the twist-2 and twist-3 contributions to @xmath170 , and the results are listed in table [ tbl - twist ] . it is observed that the latter are in fact larger than the former , consistent with the argument that the twist-3 contributions are not power - suppressed . the light - cone sum rules also give approximately equal weights to the twist-2 and higher - twist contributions to @xmath125 @xcite . we then compare our results of @xmath170 and @xmath171 for @xmath172 gev@xmath173 with those derived from lattice qcd @xcite and from light - cone sum rules @xcite in fig . [ bpi ] , where lattice results have been extrapolated to the small @xmath174 region . different extrapolation methods cause uncertainty only of about 5% @xcite . the good agreement among these different approaches at large recoil is explicit . the fast rise of the pqcd results at slow recoil indicates that perturbative calculation becomes unreliable gradually . the values of @xmath175 from pqcd for the parameter @xmath176 gev are listed in table [ bpitable ] . the resultant range @xmath177 is in agreement with @xmath178 obtained in @xcite . we shall adopt the same range of @xmath179 in the evaluation of the @xmath22 transition form factors below . we also examine the uncertainty of our predictions from the parametrization of the jet function in eq . ( [ trs ] ) . the values of @xmath170 vary about 15% for the choices of @xmath180 and @xmath181 as shown in table [ rsumtbl - pls ] . the variation for @xmath171 is similar . in a future work we shall incorporate the exact jet function into a convolution integrand in moment space . .contributions to @xmath170 from the twist-2 and two - parton twist-3 pion distribution amplitudes . [ cols="^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] consider the semileptonic decay @xmath182 in the fast recoil region of the @xmath28 meson @xcite . we define the @xmath3 meson momentum @xmath41 as in eq . ( [ pa ] ) , the momentum @xmath92 and the polarization vectors @xmath183 of the @xmath28 meson in light - cone coordinates as @xmath184 with the ratio @xmath185 and the energy fraction @xmath94 carried by the @xmath28 meson . we first keep the @xmath186 dependence of the kinematic variables in eq . ( [ par ] ) , and extract the twist-3 terms proportional to @xmath187 . the parametrization of @xmath92 and @xmath183 is chosen to make this extraction straightforward . the @xmath22 form factors are defined through the following decompositions of hadronic matrix elements , @xmath188 \nonumber\\ & & -a_2(q^2)\frac{\epsilon^*\cdot q}{m_b+m_\rho } \left[p_1^\mu+p_2^\mu-\frac{m_b^2-m_\rho^2}{q^2}q^\mu\right]\;. \label{af}\end{aligned}\ ] ] to calculate the form factors @xmath189 , @xmath190 @xmath191 and @xmath192 , we adopt the following procedures . first , only the transverse polarization vectors @xmath193 are involved in eq . ( [ vf ] ) and associated with the definition of @xmath191 in eq . ( [ af ] ) , through which we evaluate the form factors @xmath189 and @xmath191 , respectively . both the structures associated with @xmath191 and @xmath192 are orthogonal to the lepton pair momentum @xmath42 . contracting eq . ( [ af ] ) with @xmath194 , we have @xmath195 which implies that only the form factor @xmath190 is relevant in two - body nonleptonic decays such as @xmath196 . we calculate @xmath190 from eq . ( [ a0e ] ) using the distribution amplitudes associated with a longitudinally polarized @xmath28 meson . for the longitudinal polarization vector @xmath197 , the structures of @xmath191 and @xmath192 are in fact proportional to each other : @xmath198 = \frac{(\epsilon^*\cdot q)^2(m_b^2-m_\rho^2-q^2 ) } { ( m_b+m_\rho)[(\epsilon^*\cdot q)^2+q^2 ] } \left[\epsilon^{*\mu}-\frac{\epsilon^*\cdot q}{q^2}q^\mu\right]\;,\end{aligned}\ ] ] which can be easily derived via the relation , @xmath199 contracting eq . ( [ af ] ) with @xmath200 , we obtain @xmath201 \gamma_5 u(0)|b(p_1)\rangle \nonumber\\ & & = \frac{2p_2\cdot q}{m_b+m_\rho}\frac{(\epsilon^*\cdot q)^2}{q^2 } \left[a_2-\frac{(m_b+m_\rho)^2}{2p_2\cdot q } \left(1+\frac{q^2}{(\epsilon^*\cdot q)^2}\right)a_1\right]\ ; , \label{ea2}\end{aligned}\ ] ] from which the form factor @xmath192 can be computed . it turns out that @xmath191 and @xmath192 have a simple relation , since the left - hand side of eq . ( [ ea2 ] ) is power - suppressed . we derive the leading - power factorization formulas , @xmath202e(t^{(1)})h(x_1,x_2,b_1,b_2 ) \nonumber \\ & & + r_\rho\left[\phi_\rho^v(x_2)+\phi_\rho^a(x_2)\right ] e(t^{(2)})h(x_2,x_1,b_2,b_1)\biggr\}\ ; , \\ a_0&=&8\pi m_b^2 c_f\int dx_1dx_2\int b_1db_1 b_2db_2\phi_b(x_1,b_1 ) \nonumber\\ & & \times\biggl\{\left[(1+\eta x_2)\phi_\rho(x_2)+r_\rho \left((1 - 2x_2)\phi_\rho^t(x_2)+ \left(\frac{2}{\eta}-1 - 2x_2\right)\phi_\rho^s(x_2)\right)\right ] e(t^{(1)})h(x_1,x_2,b_1,b_2 ) \nonumber \\ & & + 2r_\rho\phi_\rho^s(x_2)e(t^{(2)})h(x_2,x_1,b_2,b_1)\biggr\}\ ; , \\ a_1&=&8\pi m_b^2 c_f\eta\int dx_1dx_2\int b_1db_1 b_2db_2\phi_b(x_1,b_1 ) \nonumber\\ & & \times\biggl\{\left[\phi_\rho^t(x_2)+r_\rho \left(\left(\frac{2}{\eta}+x_2\right)\phi_\rho^v(x_2 ) -x_2\phi_\rho^a(x_2)\right)\right]e(t^{(1)})h(x_1,x_2,b_1,b_2 ) \nonumber \\ & & + r_\rho\left[\phi_\rho^v(x_2)+\phi_\rho^a(x_2)\right ] e(t^{(2)})h(x_2,x_1,b_2,b_1)\biggr\}\ ; , \\ a_2&=&\frac{a_1}{\eta}\ ; , \label{a2f}\end{aligned}\ ] ] with the evolution factor @xmath203 the same as in eq . ( [ evol ] ) . taking the fast recoil limit with @xmath204 and assuming the asymptotic behavior @xmath205 , the above form factors are found to obey the symmetry relations @xcite , @xmath206 where the term @xmath207 , being higher - power , does not appear in eq . ( [ a2f ] ) . note that the form factors , treated as nonperturbative objects , are not calculated in @xcite . instead , the diagrams we have calculated above are regarded as perturbative corrections to the relations in eq . ( [ sr ] ) . we adopt the @xmath28 meson distribution amplitudes given in the appendix b @xcite , @xmath208\ ; , \label{pwr}\\ \phi_{\rho}^t(x)&=&\frac{f^t_{\rho}}{2\sqrt{2n_c } } \left\{3(2x-1)^2 + 0.3(2x-1)^2[5(2x-1)^2 - 3]\right . \nonumber \\ & & \left.+0.21[3 - 30(2x-1)^2 + 35(2x-1)^4]\right\}\ ; , \label{pwt}\\ \phi_{\rho}^s(x ) & = & \frac{3f_\rho^t}{2\sqrt{2n_{c } } } ( 1 - 2x)\left[1 + 0.76(10x^2 - 10x+1)\right]\ ; , \label{pws}\\ \phi_\rho^t(x)&=&\frac{3f_\rho^t}{\sqrt{2n_c } } x(1-x)\left[1 + 0.2c_2^{3/2}(2x-1)\right]\ ; , \label{pwft}\\ \phi_{\rho}^v(x)&=&\frac{f_{\rho}}{2\sqrt{2n_c } } \bigg\{\frac{3}{4}[1+(2x-1)^2]+0.24[3(2x-1)^2 - 1 ] \nonumber \\ & & + 0.12[3 - 30(2x-1)^2 + 35(2x-1)^4]\bigg\}\ ; , \label{pwv}\\ \phi_{\rho}^a(x ) & = & \frac{3f_\rho}{4\sqrt{2n_{c } } } ( 1 - 2x)\left[1 + 0.93(10x^2 - 10x+1)\right]\ ; , \label{pwa}\end{aligned}\ ] ] with the decay constants @xmath209 mev and @xmath210 mev . the @xmath174 dependence of the form factors @xmath189 and @xmath211 with the same @xmath3 meson distribution amplitude in eq . ( [ os ] ) and @xmath212 gev employed , is displayed in fig . our results are consistent with those from light - cone qcd sum rules @xcite at small @xmath174 . it is found that the symmetry relation @xmath213 in eq . ( [ sr ] ) holds very well : @xmath191 is larger than @xmath189 only by 2% in the large recoil region , even after considering the pre - asymptotic forms of @xmath214 and @xmath215 in eqs . ( [ pwv ] ) and ( [ pwa ] ) , respectively . to compare our results with the second symmetry relation , we include next - to - leading power terms in eq . ( [ ea2 ] ) , obtaining @xmath216 e(t^{(1)})h(x_1,x_2,b_1,b_2 ) \nonumber \\ & & + 2r_\rho\phi_\rho^s(x_2 ) e(t^{(2)})h(x_2,x_1,b_2,b_1)\biggr\}\;.\end{aligned}\ ] ] because of the cancellation of the term @xmath217 and the second term in the above expression , the values of @xmath192 only slightly deviate from those in eq . ( [ a2f ] ) . the numerical study shows that @xmath192 is larger than @xmath218 by about 40% , which can be regarded as the estimate of the symmetry breaking effect . in this paper we have presented a complete leading - power and leading - order pqcd evaluation of the @xmath19 , @xmath28 transition form factors in the large recoil region . it has been shown that under sudakov suppression arising from @xmath4 and threshold resummations , the end - point singularities ( logarithmic at twist 2 and linear at twist 3 ) do not exist . the soft contribution to the form factors , being sudakov suppressed , becomes smaller than the perturbative contribution . the physical picture for the mechanism of sudakov suppression has been discussed . we have emphasized that the twist-3 contributions are in fact not power - suppressed in the @xmath219 limit . the treatment of the parton transverse momenta @xmath4 and the light spectator momentum @xmath48 in the @xmath3 meson in the computation of the hard amplitudes has been clearly explained : the hard amplitudes should not be expanded in powers of @xmath115 as the end - point region is important . using the light meson distribution amplitudes derived from qcd sum rules , and choosing an appropriate @xmath3 meson distribution amplitude , we have derived reasonable results for the @xmath19 , @xmath28 form factors , which are in agreement with those from light - cone qcd sum rules and from lattice calculations . our study indicates that in a self - consistent perturbative analysis , the heavy - to - light form factors are calculable . the jet function from threshold resummation needs more thorough exploration . we shall investigate the relevant subjects , such as factorization theorem in moment space , threshold resummation up to next - to - leading logarithms , application to nonleptonic @xmath3 meson decays @xcite , and numerical effects elsewhere . note that if considering only @xmath220 resummation @xcite , twist-3 contributions , though infrared finite , are still too large to give reasonable heavy - to - light transition form factors , because the large double logarithms @xmath20 have not yet been organized . * acknowledgements * we thank s. brodsky and h.y . cheng for helpful discussions . the work was supported in part by grant - in aid for special project research ( physics of cp violation ) , by grant - in aid for scientific research from the ministry of education , science and culture of japan . the work of h.n.l . was supported in part by the national science council of r.o.c . under the grant no . nsc-89 - 2112-m-006 - 033 , and by theory group of slac . the work of t.k was supported in part by grant - in aid for scientific research from the ministry of education , science and culture of japan under the grant no . 11640265 . it has been shown @xcite that the factorization in fermion flow between the pion distribution amplitude and the hard amplitude is achieved by inserting the fierz identity , @xmath221 into the quark and anti - quark lines of the pion , where @xmath222 represents the identity matrix , and @xmath223 is defined by @xmath224/2 $ ] . the insertion of eq . ( [ fe ] ) then leads to various nonlocal matrix elements , @xmath225 each of which is characterized by different twists . the light - like vector @xmath75 is the coordinate of the @xmath76 quark , and @xmath74 the pion momentum . the general expressions of the relevant matrix elements are , quoted from @xcite , @xmath226 where @xmath227 and @xmath228 are the distribution amplitudes of unit normalization , @xmath229 the pion mass , @xmath16 the momentum fraction associated with the @xmath76 quark . it is easy to observe that the contribution from @xmath230 , independent of the pion mass , is twist-2 , and the contribution from @xmath228 is twist-4 because of the factor @xmath231 . the contributions from @xmath232 and @xmath233 , proportional to @xmath234 , are twist-3 . we shall neglect the twist-4 terms and the term @xmath235 in eq . ( [ pt ] ) . it is straightforward to read off the pseudo - vector and pseudo - scalar structures of the pion distribution amplitudes from eqs . ( [ pv ] ) and ( [ ps ] ) . to derive the pseudo - tensor structure from eq . ( [ pt ] ) , we need more effort . using integration by parts , eq . ( [ pt ] ) is rewritten as @xmath236 with the anti - symmetric tensor @xmath237 , @xmath238 . the tensor @xmath237 in eq . ( [ pt3 ] ) contracts to the spin structure @xmath239 in the evaluation of the corresponding hard amplitude . the factor @xmath240 comes from the extra factor @xmath240 associated with the pseudo - tensor structure compared to other structures in eq . ( [ fe ] ) . we have @xmath241 therefore , up to twist-3 , the initial - state @xmath242 meson distribution amplitudes are written as @xmath243_{lj}\phi_\pi(x ) + [ \gamma_5]_{lj}m_0\phi^p_\pi(x)\right . \nonumber\\ & & \left.+m_0[\gamma_5(\not n_-\not n_+-1)]_{lj}\phi^t_\pi(x)\right\}\;,\end{aligned}\ ] ] with @xmath244 for the final - state @xmath242 meson , we consider the adjoints of eqs . ( [ pv ] ) , ( [ ps ] ) and ( [ pt ] ) : @xmath245 it is observed that the pseudo - tensor structure in eq . ( [ pto ] ) acquires an extra minus sign , compared to the other two structures . the pseudo - tensor structure is then given by @xmath246 . therefore , up to twist 3 , we have eq . ( [ fpd ] ) for the final - state @xmath242 meson . note that there is an extra term in the definition of @xmath83 , which contains a differential operator applying to hard amplitudes @xcite . this term , being power - suppressed , is negligible here . the distribution amplitudes @xmath82 and @xmath84 are normalized according to @xmath247 the tensor distribution amplitude is normalized to zero , because of @xmath248 if @xmath233 vanishes at the end points of the momentum fraction . we choose the @xmath28 meson momentum @xmath78 with @xmath249 , which is mainly in the plus direction . the polarization vectors @xmath183 , satisfying @xmath250 , represent one longitudinal polarization vector @xmath197 and two transverse polarization vectors @xmath193 . their explicit expressions in light - cone coordinates have been given in eq . ( [ par ] ) . to arrive at the factorization in fermion flow , we insert the fierz identity into the quark and anti - quark lines of the @xmath28 meson . the spin structures in eq . ( [ fe ] ) lead to the following nonlocal matrix elements , @xmath251 characterized by different twists . the definition of @xmath80 is the same as that for the pion distribution amplitudes in the previous appendix . the general expressions of the above matrix elements are , quoted from @xcite , @xmath252\right\}\ ; , \label{v}\\ \langle \rho^-(p,\epsilon)|{\bar d}(z)\sigma_{\mu\nu}u(0)|0\rangle&= & -i\frac{f^t_{\rho}}{n_c}\left\{(\epsilon_{t\mu}p_{\nu}-\epsilon_{t\nu}p_{\mu } ) \int_0 ^ 1 dx e^{ix p\cdot z}\phi_t(x)\right . \nonumber\\ & & + ( p_{\mu } z_\nu - p_{\nu } z_\mu)\frac{\epsilon\cdot z}{(p\cdot z)^2 } m^2_{\rho}\int_0 ^ 1 dx e^{ix p\cdot z}h^{(t)}_{\parallel}(x ) \nonumber\\ & & \left . + \frac{1}{2}(\epsilon_{t\mu}z_\nu-\epsilon_{t\nu}z_{\mu } ) \frac{m^2_{\rho}}{p\cdot z}\int_0 ^ 1 dx e^{ix p\cdot z}[h_3(x)-\phi_t(x ) ] \right\}\ ; , \label{t}\\ \langle \rho^-(p,\epsilon)|{\bar d}(z)i u(0)|0\rangle&= & -\frac{i}{2n_c}\left(f^t_{\rho}-f_{\rho}\frac{m_u+m_d}{m_{\rho}}\right ) \epsilon\cdot zm^2_{\rho}\int_0 ^ 1 dx e^{ix p\cdot z } h^{(s)}_{\parallel}(x)\ ; , \nonumber\\ & = & \frac{1}{2n_c}\left(f^t_{\rho}-f_{\rho}\frac{m_u+m_d}{m_{\rho}}\right ) \frac{\epsilon\cdot z}{p\cdot z}m^2_{\rho } \int_0 ^ 1 dx e^{ix p\cdot z } \frac{d}{dx}h^{(s)}_{\parallel}(x)\ ; , \label{s}\\ \langle \rho^-(p,\epsilon)|{\bar d}(z)\gamma_5\gamma_\mu u(0)|0\rangle & = & -\frac{1}{4n_c}\left(f_{\rho}-f_{\rho}^t\frac{m_u+m_d}{m_{\rho}}\right ) m_\rho\epsilon_{\mu}^{~\nu\alpha\beta}\epsilon_{t\nu } p_{2\alpha } z_\beta\int_0 ^ 1 dx e^{ix p\cdot z}g^{(a)}_{t}(x)\ ; , \nonumber\\ & = & -\frac{i}{4n_c}\left(f_{\rho}-f_{\rho}^t\frac{m_u+m_d}{m_{\rho}}\right ) \frac{m_\rho}{p\cdot n_-}\epsilon_{\mu\nu\rho\sigma}\epsilon_t^\nu p^\rho n_-^\sigma\int_0 ^ 1 dx e^{ix p\cdot z } \frac{d}{dx}g^{(a)}_{t}(x)\ ; , \label{a}\end{aligned}\ ] ] where @xmath253 and @xmath254 are the decay constants of the @xmath28 meson with longitudinal and transverse polarizations , respectively , and @xmath16 the momentum fraction associated with the @xmath76 quark . we adopt the convention @xmath255 for the levi - civita tensor @xmath256 . the distribution amplitudes @xmath227 , @xmath257 and @xmath258 are normalized to unity . following the similar procedures , we derive the @xmath28 meson distribution amplitudes up to twist 3 , @xmath259_{lj}\phi_\rho(x)\right . \nonumber\\ & & \left.+[\not \epsilon_l\not p]_{lj } \phi_{\rho}^{t}(x ) + m_\rho [ i]_{lj}\phi_\rho^s(x)\right\}\ ; , \label{lpf}\\ \langle\rho^-(p,\epsilon_t)|\bar d(z)_ju(0)_l|0\rangle & = & \frac{1}{\sqrt{2n_c}}\int_0 ^ 1 dx e^{ixp\cdot z } \bigg\{m_\rho[\not \epsilon_t]_{lj}\phi_\rho^v(x)+ [ \not\epsilon_t\not p]_{lj}\phi_\rho^t(x ) \nonumber\\ & & + \frac{m_\rho}{p\cdot n_- } i\epsilon_{\mu\nu\rho\sigma}[\gamma_5\gamma^\mu]_{lj}\epsilon_t^\nu p^\rho n_-^\sigma \phi_\rho^a(x)\bigg\}\ ; , \label{spf}\end{aligned}\ ] ] for longitudinal polarization and transverse polarization , respectively . we have dropped the terms proportional to @xmath260 ( twist-4 ) and the terms @xmath261 in eqs . ( [ s ] ) and ( [ a ] ) . the definitions of the above distribution amplitudes are , @xmath262 according to @xcite , the nonlocal matrix element associated with the @xmath3 meson is written as @xmath263 \right\}_{\delta\alpha}\ ; , \nonumber\\ & = & -\frac{i}{\sqrt{2n_c}}\left\{(\not p_1+m_b)\gamma_5 \left[\phi_b(k_1)-\frac{\not n_+-\not n_-}{\sqrt{2 } } { \bar \phi}_b(k_1)\right]\right\}_{\delta\alpha}\ ; , \label{bwp2}\end{aligned}\ ] ] with the wave functions , @xmath264 because the light meson momenta have been chosen in the plus direction , the hard amplitudes for the heavy - to - light transition form factors are independent of the component @xmath265 as explained in sec . we construct the @xmath3 meson distribution amplitude , @xmath266 with @xmath267 . the two @xmath3 meson distribution amplitudes @xmath268 and @xmath269 are related by the equation of motion @xcite , @xmath270 assuming that @xmath271 vanishes at both ends of the momentum fraction , @xmath17 and @xmath272 , we derive @xmath273 therefore , @xmath274 is normalized to zero . we shall argue that the contribution from the distribution amplitude @xmath274 is negligible compared to that from @xmath275 . consider the reasonable parametrizations , @xmath276\ ; , \nonumber\\ \bar\phi_b(x)&=&\frac{f_b}{2\sqrt{2n_c } } \left[-\frac{\bar\lambda}{2m_b}\delta'\left(x-\frac{\bar\lambda } { m_b}\right)+o\left(\frac{\bar\lambda^2}{m_b^2}\right)\right]\;,\end{aligned}\ ] ] whose moments satisfy eq . ( [ bep ] ) . as shown in sec . iii , the hard amplitudes are approximated by @xmath277 at small @xmath58 . a simple estimation indicates that the contribution from @xmath275 , proportional to @xmath278 , is numerically larger than that from @xmath274 , proportional to a constant . hence , after taking into account eq . ( [ beq ] ) , we consider only a single @xmath3 meson distribution amplitude in this work . in this appendix we supply details of the derivation of the sudakov factor in eq . ( [ trs ] ) . threshold resummation introduces a jet function @xmath108 into the pqcd factorization of the @xmath19 form factors near the end points @xcite , @xmath279 where @xmath280 is an arbitrary real constant larger than all the real parts of poles involved in the integrand . the factor @xmath281 comes from mellin transformation of the initial condition @xmath282 , @xmath283 the sudakov factor @xmath284 in the moment @xmath285 space has been derived explicitly to the accuracy of leading logarithms ( ll ) @xcite , @xmath286\;,\end{aligned}\ ] ] with the anomalous dimension @xmath287 . the contour integral in eq . ( [ mj ] ) leads to @xmath288 which vanishes at @xmath17 , since the integrand is an odd function in @xmath143 , and at @xmath272 due to the factor @xmath289 . in this paper we consider threshold resummation up to next - to - leading logarithms . at this level of accuracy , the anomalous dimension @xmath290 contains two - loop contributions , and the coupling constant @xmath18 becomes running . the sudakov factor @xmath284 is then given by @xmath291\ ; , \label{exa}\end{aligned}\ ] ] with @xmath292\ ; , \label{lk}\ ] ] @xmath293 being the number of quark flavors and @xmath294 a color factor . the anomalous dimension @xmath290 is the same as that for @xmath220 resummation @xcite . it can be shown that @xmath108 still vanishes at the end points @xmath17 and @xmath272 . to simplify the analysis , we propose the parametrization , @xmath111^c\ ; , \label{pst}\end{aligned}\ ] ] whose end - point behavior satisfies the above requirement . in the @xmath295 ( @xmath296 ) limit , _ i.e. _ , without qcd effects , eq . ( [ pst ] ) approaches unity . mellin transformation of @xmath108 gives @xmath297 the variable @xmath298 should be large enough to justify threshold resummation up to the next - to - leading logarithms @xmath299 , and small enough to avoid the divergent running coupling constant @xmath300 in eq . ( [ exa ] ) . performing the best fit of eq . ( [ trf ] ) to @xmath301 for @xmath302 , we determine the parameter @xmath303 . the difference @xmath304 is shown in fig . [ thfit ] for @xmath305 and 0.4 . equation ( [ pst ] ) implies that threshold resummation modifies the end - point behavior of the meson distribution amplitudes , rendering them vanish faster at @xmath17 . cleo coll . , y. kwon _ et al . _ , hep - ex/9908039 . cleo collaboration , hep - ex/0001010 . belle collaboration , talk presented at the 4th international workshop on @xmath3 physics and cp violation , ise , japan , feb . babar collaboration , talk presented at the 4th international workshop on @xmath3 physics and cp violation , ise , japan , feb . , 2001 . keum , h - n . li , and a.i . sanda , phys . b * 504 * , 6 ( 2001 ) ; phys . rev . * d63 * , 054008 ( 2001 ) . l , k. ukai , and m.z . yang , phys . rev . * d63 * , 074009 ( 2001 ) ; c.d . l and m.z . yang , hep - ph/0011238 . chen and h - n . li , phys . d * 63 * , 014003 ( 2001 ) . n. isgur and c.h . llewellyn - smith , nucl . b317 * , 526 ( 1989 ) . szczepaniak , e.m . henley , and s.j . brodsky , phys . b * 243 * , 287 ( 1990 ) . r. akhoury , g. sterman and y.p . yao , phys . d * 50 * , 358 ( 1994 ) . m. beneke and t. feldmann , nucl . phys . * b592 * , 3 ( 2000 ) . a. khodjamirian and r. ruckl , phys . d * 58 * , 054013 ( 1998 ) . g. sterman , phys . b * 179 * , 281 ( 1986 ) ; nucl . b281 * , 310 ( 1987 ) . s. catani and l. trentadue , nucl . phys . * b327 * , 323 ( 1989 ) ; nucl . phys . * b353 * , 183 ( 1991 ) . h - n . li , hep - ph/0102013 . lapage and s.j . brodsky , phys . b * 87 * , 359 ( 1979 ) . g.p . lepage and s.j . brodsky , phys . d * 22 * , 2157 ( 1980 ) . a.v . efremov and a.v . radyushkin , phys . b * 94 * , 245 ( 1980 ) 245 . li and g. sterman , nucl . phys . * b381 * , 129 ( 1992 ) . collins and d.e . soper , nucl . b193 * , 381 ( 1981 ) . j. botts and g. sterman , nucl . * b225 * , 62 ( 1989 ) . li and h.l . yu , phys . lett . * 74 * , 4388 ( 1995 ) ; phys . d * 53 * , 2480 ( 1996 ) . t.w . yeh and h - n . li , phys . d * 56 * , 1615 ( 1997 ) . li , hep - ph/0012140 , to appear in phys . d. s. aoki , talk presented at the international conference on lepton - photon interaction at high energies , 1999 . p. ball j. high energy phys . * 09 * , 005 ( 1998 ) ; z. phys . c * 29 * , 637 ( 1985 ) . m. dahm , r. jakob , and p. kroll , z. phys . c * 68 * , 595 ( 1995 ) . a. khodjamirian _ _ , phys . lett . b * 410 * , 275 ( 1997 ) . a.p . szczepaniak and a.v . radyushkin , phys . d * 59 * , 034017 ( 1999 ) . braun and i.e. filyanov , z phys . c * 48 * , 239 ( 1990 ) ; p. ball , j. high energy phys . * 01 * , 010 ( 1999 ) . et al . _ , b * 486 * , 111 ( 2000 ) . d. becirevic , nucl . . suppl . * 94 * , 337 ( 2001 ) . h - n . li and h.l . yu , phys . b * 353 * , 301 ( 1995 ) . j. charles _ et al . d * 60 * , 014001 ( 1999 ) . p. ball , v.m . braun , y. koike , and k. tanaka , nucl . * b529 * , 323 ( 1998 ) . p. ball and v.m . braun , phys . d * 58 * , 094016 ( 1998 ) . chen , y.y . keum , and h - n . li , in preparation . d. du , d. yang , and g. zhu , hep - ph/0102077 ; hep - ph/0103211 . a.g . grozin and m. neubert , phys . d * 55 * , 272 ( 1997 ) . li , phys . b * 454 * , 328 ( 1999 ) ; chin . * 37 * , 569 ( 1999 ) .

this article is devoted to the analysis of a monte - carlo method to approximate effective coefficients in stochastic homogenization of discrete elliptic equations . we consider the case of independent and identically distributed coefficients , and adopt the point of view of the random walk in a random environment . given some final time , a natural approximation of the homogenized coefficients is given by the empirical average of the final squared positions rescaled by of independent random walks in independent environments . relying on a new quantitative version of kipnis - varadhan s theorem ( which is of independent interest ) , we first give a sharp estimate of the error between the homogenized coefficients and the expectation of the rescaled final position of the random walk in terms of . we then complete the error analysis by quantifying the fluctuations of the empirical average in terms of and , and prove a large - deviation estimate . compared to other numerical strategies , this monte - carlo approach has the advantage to be dimension - independent in terms of convergence rate and computational cost . * keywords : * random walk , random environment , stochastic homogenization , effective coefficients , monte - carlo method , quantitative estimates . * 2010 mathematics subject classification : * 35b27 , 60k37 , 60h25 , 65c05 , 60h35 , 60g50 .

1103.4591

this article is part of a larger program , which consists in devising and quantitatively analyzing numerical methods to approximate effective coefficients in stochastic homogenization of linear elliptic equations . more precisely we tackle here the case of a discrete elliptic equation with independent and identically distributed coefficients ( see however the end of this introduction for more general statistics ) , and present and fully analyze an approximation procedure based on a monte - carlo method . a first possibility to approximate effective coefficients is to directly solve the so - called corrector equation . in this approach , a first step towards the derivation of error estimates is a quantification of the qualitative results proved by knnemann @xcite ( and inspired by papanicolaou and varadhan s treatment of the continuous case @xcite ) and kozlov @xcite . in the stochastic case , such an equation is posed on the whole @xmath3 , and we need to localize it on a bounded domain , say the hypercube @xmath4 of side @xmath5 . as shown in a series of papers by otto and the first author @xcite , and the first author @xcite , there are three contributions to the @xmath6-error in probability between the true homogenized coefficients and its approximation . the dominant error in small dimensions takes the form of a variance : it measures the fact that the approximation of the homogenized coefficients by the average of the energy density of the corrector on a box @xmath4 fluctuates . this error decays at the rate of the central limit theorem @xmath7 in any dimension ( with a logarithmic correction for @xmath8 ) . the second error is the so - called systematic error : it is due to the fact that we have modified the corrector equation by adding a zero - order term of strength @xmath9 ( as standard in the analysis of the well - posedness of the corrector equation ) . the scaling of this error depends on the dimension and saturates at dimension @xmath10 . it is of higher order than the random error up to dimension @xmath11 . the last error is due to the use of boundary conditions on the bounded domain @xmath4 . provided there is a buffer region , this error is exponentially small in the distance to the buffer zone measured in units of @xmath12 . this approach has two main drawbacks . first the numerical method only converges at the central limit theorem scaling in terms of @xmath13 up to dimension @xmath11 , which is somehow disappointing from a conceptual point of view ( although this is already fine in practice ) . second , although the size of the buffer zone is roughly independent of the dimension , its cost with respect to the central limit theorem scaling dramatically increases with the dimension ( recall that in dimension @xmath14 , the clt scaling is @xmath7 , so that in high dimension , we may consider smaller @xmath13 for a given precision , whereas the use of boundary conditions requires @xmath15 in any dimension ) . based on ideas of the second author in @xcite , we have taken advantage of the spectral representation of the homogenized coefficients ( originally introduced by papanicolaou and varadhan to prove their qualitative homogenization result ) in order to devise and analyze new approximation formulas for the homogenized coefficients in @xcite . in particular , this has allowed us to get rid of the restriction on dimension , and exhibit refinements of the numerical method of @xcite which converge at the central limit theorem scaling in any dimension ( thus avoiding the first mentioned drawback ) . unfortunately , the second drawback is inherent to the type of method used : if the corrector equation has to be solved on a bounded domain @xmath4 , boundary conditions need to be imposed on the boundary @xmath16 . since their values are actually also part of the problem , a buffer zone seems mandatory with the notable exception of the periodization method , whose analysis is yet still unclear to us , especially when spatial correlations are introduced in the coefficients . in order to avoid the issue of boundary conditions , we adopt here another point of view on the problem : the random walk in random environment approach . this other point of view on the same homogenization problem has been analyzed in the celebrated paper @xcite by kipnis and varadhan , and then extended by de masi , ferrari , goldstein , and wick @xcite . the strategy of the present paper is to obtain an approximation of the homogenized coefficients by the numerical simulation of this random walk up to some large time . as we did in the case of the approach based on the corrector equation , a first step towards the analysis of this numerical method is to quantify the corresponding qualitative result , namely here kipnis - varadhan s convergence theorem . compared to the deterministic approach based on the approximate corrector equation , the advantage of the present approach is that its convergence rate and computational costs are dimension - independent . as we shall also see , as opposed to the approach based on the corrector equation , the environment only needs to be generated along the trajectory of the random walker , so that much less information has to be stored during the calculation . this may be quite an important feature of the monte carlo method in view of the discussion of ( * ? ? ? * section 4.3 ) . we consider the discrete elliptic operator @xmath17 , where @xmath18 and @xmath19 are the discrete backward divergence and forward gradient , respectively . for all @xmath20 , @xmath21 is the diagonal matrix whose entries are the conductances @xmath22 of the edges @xmath23 starting at @xmath24 , where @xmath25 denotes the canonical basis of @xmath3 . let @xmath26 denote the set of edges of @xmath3 . we call the family of conductances @xmath27 the _ environment_. the environment @xmath28 is random , and we write @xmath29 for its distribution ( with corresponding expectation @xmath30 ) . we make the following assumptions : * the measure @xmath29 is invariant under translations , * the conductances are i. i. d. , * there exists @xmath31 such that @xmath32 almost surely . under these conditions , standard homogenization results ensure that there exists some _ deterministic _ symmetric matrix @xmath33 such that the solution operator of the deterministic continuous differential operator @xmath34 describes the large scale behavior of the solution operator of the random discrete differential operator @xmath17 almost surely ( for this statement , ( h2 ) can in fact be replaced by the weaker assumption that the measure @xmath29 is ergodic with respect to the group of translations , see @xcite ) . the operator @xmath35 is the infinitesimal generator of a stochastic process @xmath36 which can be defined as follows . given an environment @xmath28 , it is the markov process whose jump rate from a site @xmath20 to a neighbouring site @xmath37 is given by @xmath38 . we write @xmath39 for the law of this process starting from @xmath20 . it is proved in @xcite that under the averaged measure @xmath40 , the rescaled process @xmath41 converges in law , as @xmath42 tends to @xmath43 , to a brownian motion whose infinitesimal generator is @xmath34 , or in other words , a brownian motion with covariance matrix @xmath44 ( see also @xcite for prior results ) . we will use this fact to construct computable approximations of @xmath33 . as proved in @xcite , this invariance principle holds as soon as ( h1 ) is true , ( h2 ) is replaced by the ergodicity of the measure @xmath29 , and ( h3 ) by the integrability of the conductances . under the assumptions ( h1-h3 ) , @xcite strengthens this result in another direction , showing that for almost every environment , @xmath41 converges in law under @xmath45 to a brownian motion with covariance matrix @xmath44 . this has been itself extended to environments which do not satisfy the uniform ellipticity condition ( h3 ) , see @xcite . let @xmath46 denote the sequence of consecutive sites visited by the random walk @xmath47 ( note that the `` times '' are different in nature for @xmath48 and @xmath49 ) . this sequence is itself a markov chain that satisfies for any two neighbours @xmath50 : @xmath51 = \frac{\omega_{x , y}}{p_\omega(x)},\ ] ] where @xmath52 . we simply write @xmath53 for @xmath54 . let us introduce a `` tilted '' version of the law @xmath29 on the environments , that we write @xmath55 and define by @xmath56 } \ \d { \mathbb{p}}(\omega).\ ] ] the reason why this measure is natural to consider is that it makes the environment seen from the position of the random walk @xmath57 a stationary process ( see ( [ defenvpart ] ) for a definition of this process ) . interpolating between two integers by a straight line , we can think of @xmath57 as a continuous function on @xmath58 . with this in mind , it is also true that there exists a matrix @xmath59 such that , as @xmath42 tends to @xmath43 , the rescaled process @xmath60 converges in law under @xmath61 to a brownian motion with covariance matrix @xmath62 . moreover , @xmath59 and @xmath33 are related by ( see ( * ? ? ? * theorem 4.5 ( ii ) ) ) : @xmath63 \ { a_\mathrm{hom}^\mathrm{disc}}.\ ] ] given that the numerical simulation of @xmath57 saves some operations compared to the simulation of @xmath64 ( there is no waiting time to compute , and the running time is equal to the number of steps ) , we will focus on approximating @xmath59 . more precisely , we fix once and for all some @xmath65 with @xmath66 , and define @xmath67 , \qquad \sigma^2 = 2 \xi \cdot { a_\mathrm{hom}^\mathrm{disc}}\xi.\ ] ] it follows from results of @xcite ( or ( * ? ? ? * theorem 2.1 ) ) that @xmath68 tends to @xmath69 as @xmath1 tends to infinity . our first contribution is to give a quantitative estimate of this convergence . in particular we shall show that , with i. i. d. coefficients and up to a logarithmic correction in dimension @xmath70 , the difference between @xmath68 and @xmath69 is of order @xmath71 . we now describe a monte - carlo method to approximate @xmath68 . using the definition of the tilted measure ( [ deftdp ] ) , one can see that @xmath72}{t } = \frac{{\mathbb{e}}{\mathbf{e}^\omega}_0[p(\omega ) ( \xi \cdot y(t))^2]}{t { \mathbb{e}}[p]}.\ ] ] assuming that we have easier access to the measure @xmath29 than to the tilted @xmath55 , we prefer to base our monte - carlo procedure on the r. h. s. of the second identity in ( [ enlevetilt ] ) . let @xmath73 be independent random walks evolving in the environments @xmath74 respectively . we write @xmath75 for their joint distribution , all random walks starting from @xmath43 , where @xmath76 stands for @xmath77 . the family of environments @xmath76 is itself random , and we let @xmath78 be the product distribution with marginal @xmath29 . in other words , under @xmath78 , the environments @xmath79 are independent and distributed according to @xmath29 . our computable approximation of @xmath68 is defined by @xmath80 the following step in the analysis is to quantify the random fluctuations of @xmath81 in terms of @xmath2 the number of random walks considered in the empirical average to approximate @xmath68 and @xmath1 . we shall prove a large deviation result which ensures that the @xmath82-probability that the difference between @xmath81 and @xmath68 exceeds @xmath71 is exponentially small in the ratio @xmath83 . the rest of this article is organized as follows . in section [ sec : quantkv ] , which can be read independently of the rest of the paper , we consider a general discrete or continuous - time reversible markov process . kipnis - varadhan s theorem ( and its subsequent development due to @xcite ) gives conditions for additive functionals of this process to satisfy an invariance principle . we show that , under additional conditions written in terms of a spectral measure , the statement can be made quantitative . more precisely , kipnis - varadhan s theorem relies on writing the additive functional under consideration as the sum of a martingale plus a remainder . this remainder , after suitable normalization , is shown to converge to @xmath43 in @xmath6 . under our additional assumptions , we give explicit bounds on the rate of decay . in section [ sec : syserr ] , we make use of this result , in the context of the approximation of homogenized coefficients , to estimate the systematic error @xmath84 . the central achievement of this section is to prove that the relevant spectral measure satisfies the conditions of our quantitative version of kipnis - varadhan s theorem . section [ sec : randfluc ] is dedicated to the estimate of the random fluctuations . these are controlled through large deviations estimates . relying on these results , we give in section [ sec : numtest ] a complete error analysis of the monte - carlo method to approximate the homogenized matrix @xmath59 , which we illustrate by numerical tests . let us quickly discuss the sharpness of these results . if @xmath85 was a periodic matrix ( or even a constant matrix ) the systematic error would also be of order @xmath71 ( without logarithmic correction for @xmath8 ) , and the fluctutations would decay exponentially fast in the ratio @xmath83 as well . this shows that our analysis is optimal ( the additional logarithm seems unavoidable for @xmath8 , as discussed in the introduction of @xcite ) . let us also point out that although the results of this paper are proved under assumptions ( h1)-(h3 ) , the assumption ( h2 ) on the statistics of @xmath28 is only used to obtain the variance estimate of ( * ? ? ? * lemma 2.3 ) . in particular , ( h2 ) can be weakened as follows : * the distribution of @xmath86 may in addition depend on @xmath87 , * independence can be replaced by finite correlation length @xmath88 , that is for all @xmath89 , @xmath90 and @xmath91 are independent if @xmath92 , * independence can be replaced by mixing in the sense of dobrushin and shlosman we refer the reader to work in progress by otto and the first author for this issue @xcite . * notation . * so far we have already introduced the probability measures @xmath45 ( distribution of @xmath57 ) , @xmath75 ( distribution of @xmath93 ) , @xmath29 ( i.i.d . distribution for @xmath27 ) , @xmath55 ( tilted measure defined in ( [ deftdp ] ) ) and @xmath78 ( product distribution of @xmath76 with marginal @xmath29 ) . it will be convenient to define @xmath94 the product distribution of @xmath76 with marginal @xmath55 . for convenience , we write @xmath95 as a short - hand notation for @xmath40 , @xmath96 for @xmath61 , @xmath97 for @xmath98 , and @xmath99 for @xmath100 . the corresponding expectations are written accordingly , replacing `` p '' by `` e '' with the appropriate typography . finally , we write @xmath101 for the euclidian norm of @xmath102 . kipnis - varadhan s theorem @xcite concerns additive functionals of reversible markov processes . it gives conditions for such additive functionals to satisfy an invariance principle . the proof of the result relies on a decomposition of the additive functional as the sum of a martingale term plus a remainder term , the latter being shown to be negligible . in this section , which can be read independently of the rest of the paper , we give conditions that enable to obtain some quantitative bounds on this remainder term . we consider discrete and continuous times simultaneously . let @xmath103 be a markov process defined on some measurable state space @xmath104 ( here , @xmath105 stands either for @xmath106 or for @xmath107 ) . we denote by @xmath108 the distribution of the process started from @xmath109 , and by @xmath110 the associated expectation . we assume that this markov process is reversible and ergodic with respect to some probability measure @xmath111 . we write @xmath112 for the law of the process started from the distribution @xmath111 , and @xmath113 for the associated expectation . to the markov process is naturally associated a semi - group @xmath114 defined , for any @xmath115 , by @xmath116.\ ] ] each @xmath117 is a self - adjoint contraction of @xmath118 . in the continuous - time case , we assume further that the semi - group is strongly continuous , that is to say , that @xmath119 converges to @xmath120 in @xmath118 as @xmath1 tends to @xmath43 , for any @xmath121 . we let @xmath122 be the @xmath118-infinitesimal generator of the semi - group . it is self - adjoint in @xmath118 , and we fix the sign convention so that it is a positive operator ( i.e. , @xmath123 ) . note that in general , one can see using spectral analysis that there exists a projection @xmath124 such that @xmath119 converges to @xmath125 as @xmath1 tends to @xmath43 , @xmath126 . changing @xmath118 to the image of the projection @xmath124 , and @xmath127 for @xmath124 , one recovers a strongly continuous semigroup of contractions , and one can still carry the analysis below replacing @xmath118 by the image of @xmath124 when necessary . in discrete time , we set @xmath128 . again , @xmath122 is a positive self - adjoint operator on @xmath118 . note that we slightly depart from the custom of defining the generator as @xmath129 in order to match more closely the continuous time situation . we denote by @xmath130 the scalar product in @xmath118 . for any function @xmath121 we define the _ spectral measure _ of @xmath122 projected on the function @xmath120 as the measure @xmath131 on @xmath58 that satisfies , for any bounded continuous @xmath132 , the relation @xmath133 the dirichlet form associated to @xmath122 is given by @xmath134 we denote by @xmath135 the completion of the space @xmath136 with respect to this @xmath137 norm , taken modulo functions of zero @xmath137 norm . this turns @xmath138 into a hilbert space , and we let @xmath139 denote its dual . one can identify @xmath139 with the completion of the space @xmath140 with respect to the norm @xmath141 defined by @xmath142 indeed , for all @xmath121 , the linear form @xmath143 has norm @xmath144 , and thus defines an element of @xmath139 ( with norm @xmath144 ) iff @xmath144 is finite . the notion of spectral measure introduced in ( [ defef ] ) for functions of @xmath118 can be extended to elements of @xmath139 . indeed , let @xmath132 be a continuous function such that @xmath145 as @xmath146 . one can check that the map @xmath147 extends to a bounded linear map on @xmath139 . one can then define the spectral measure of @xmath122 projected on the function @xmath120 as the measure @xmath131 such that for any continuous @xmath148 with @xmath145 , ( [ defef ] ) holds . with a slight abuse of notation , for all @xmath149 and @xmath150 , we write @xmath151 for the @xmath152 duality product between @xmath120 and @xmath153 . for any @xmath149 , we define @xmath154 as @xmath155 according to whether we consider the continuous or the discrete time cases . in the continuous case , the meaning of ( [ defzf ] ) is unclear a priori . yet it is proved in ( * ? ? ? * lemma 2.4 ) that for any @xmath156 the map @xmath157 can be extended by continuity to a bounded linear map on @xmath139 , and moreover , that ( [ defzf ] ) coincides with the usual integral as soon as @xmath158 . the following theorem is due to @xcite , building on previous work of @xcite . [ kv ] ( i ) for all @xmath149 , there exists @xmath159 , @xmath160 such that @xmath161 defined in satisfies the identity @xmath162 , where @xmath163 is a square - integrable martingale with stationary increments under @xmath112 ( and the natural filtration ) , and @xmath164 is such that : @xmath165 \xrightarrow[t \to + \infty ] { } 0.\ ] ] as a consequence , @xmath166 converges in law under @xmath112 to a gaussian random variable of variance @xmath167 as @xmath1 goes to infinity , and @xmath168 \xrightarrow[t \to + \infty ] { } \sigma^2(f).\ ] ] ( ii ) if , moreover , @xmath158 and , for some @xmath169 , @xmath170 is in @xmath118 , then the process @xmath171 converges in law under @xmath112 to a brownian motion of variance @xmath172 as @xmath42 goes to @xmath43 . * remarks . * the additional conditions appearing in statement ( ii ) are automatically satisfied in discrete time , due to the fact that @xmath173 in this case . in the continuous - time setting and when @xmath158 , the process @xmath174 is almost surely continuous , and @xmath170 is indeed a well - defined random variable . under some additional information on the spectral measure of @xmath120 , we can estimate the rates of convergence in the limits ( [ e : xitendvers0 ] ) and ( [ e : l2conv ] ) . for any @xmath175 and @xmath176 , we say that the spectral exponents of a function @xmath149 are at least @xmath177 if @xmath178 note that the phrasing is consistent , since if @xmath179 for the lexicographical order , and if the spectral exponents of @xmath120 are at least @xmath177 , then they are at least @xmath180 . in @xcite , it was found more convenient to consider , instead of ( [ specexp ] ) , a condition of the following form : @xmath181 one can easily check that conditions ( [ specexp ] ) and ( [ specexp2 ] ) are equivalent . indeed , on the one hand , one has the obvious inequality @xmath182 which shows that ( [ specexp2 ] ) implies ( [ specexp ] ) . on the other hand , one may perform a kind of integration by parts , use fubini s theorem : @xmath183 and obtain the converse implication by examining separately the integration over @xmath184 in @xmath185 and in @xmath186 . for all @xmath175 and @xmath176 , we set @xmath187 the quantitative version of theorem [ kv ] is as follows . [ quantkv ] if the spectral exponents of @xmath149 are at least @xmath177 , then the decomposition @xmath162 of theorem [ kv ] holds with the additional property that @xmath188 \ = o(\psi_{\gamma , q}(t ) ) \qquad ( t \to + \infty).\ ] ] moreover , @xmath189}{t } = o(\psi_{\gamma , q}(t ) ) \qquad ( t \to + \infty).\ ] ] in the continuous - time setting , the argument for the first estimate is very similar to the one of ( * ? ? * proposition 8.2 ) , and we do not repeat the details here . it is based on the observation that @xmath190 = 2 \int \frac{1-e^{-\lambda t}}{\lambda^2 t } \ \d e_f(\lambda).\ ] ] one needs to take into account the possible logarithmic terms that appear in ( [ specexp2 ] ) and which are not considered in @xcite . some care is also needed because we do not assume that @xmath121 . yet one can easily replace the bound involving the @xmath118 norm of @xmath120 by its @xmath139 norm . the second part of the statement is given by ( * ? ? ? * proposition 8.3 ) . we now turn to the discrete time setting . in this context , identity ( [ normxi2 ] ) should be replaced by @xmath191 = 2 \int \frac{1-(1-\lambda)^t}{\lambda^2 t } \ \d e_f(\lambda).\ ] ] by definition , @xmath192 , where @xmath129 is the semi - group at time @xmath193 . hence the spectrum of @xmath122 is contained in @xmath194 $ ] . one can then follow the same computations as before to prove the first part of theorem [ quantkv ] . somewhat surprisingly , the second part of the statement requires additional attention in the discrete time setting . indeed , in the continuous case , the argument of ( * ? ? ? * proposition 8.3 ) ( which already appears in @xcite ) is that @xmath161 and @xmath195 are orthogonal in @xmath196 , a fact obtained using the invariance under time symmetry . this orthogonality is only approximately valid in the discrete - time setting . indeed , let us recall that @xmath161 is given by ( [ defzf ] ) , while @xmath197 is obtained as the limit in @xmath196 of @xmath198 where @xmath199 . using time symmetry , what we obtain is that @xmath197 is orthogonal to @xmath200 . as a consequence , the cross - product @xmath201 $ ] , which is equal to @xmath43 in the proof of ( * ? ? ? * proposition 8.3 ) , is in the present case equal to @xmath202 $ ] . yet spectral analysis ensures that this term is equal to @xmath203 which is what we need to obtain the second claim of the theorem . we now come back to the analysis of the monte - carlo approximation of the homogenized coefficients within assumptions ( h1)-(h3 ) . the aim of this section is to estimate the difference between @xmath68 and the quantity @xmath69 we wish to approximate ( both being defined in ( [ defsigmat ] ) ) . this difference , that we refer to as the _ systematic error _ after @xcite , is shown to be of order @xmath71 as @xmath1 tends to infinity , up to a logarithmic correction in dimension @xmath70 . [ syserr ] under assumptions ( h1)-(h3 ) , there exists @xmath176 such that , as @xmath1 tends to infinity , @xmath204 theorem [ syserr ] is a discrete - time version of ( * ? ? ? * corollary 2.6 ) . its proof makes use of an auxiliary process that we now introduce . let @xmath205 be the translation group that acts on the set of environments as follows : for any pair of neigbhours @xmath206 , @xmath207 . the _ environment viewed by the particle _ is the process defined by @xmath208 one can check that @xmath209 is a markov chain , whose generator is given by @xmath210 so that @xmath211 = ( i-{\mathcal{l}})f(\omega)$ ] . moreover , the measure @xmath55 defined in ( [ deftdp ] ) is reversible and ergodic for this process ( * ? ? ? * lemma 4.3 ( i ) ) . as a consequence , the operator @xmath212 is ( positive and ) self - adjoint in @xmath213 . the proof of theorem [ syserr ] relies on spectral analysis . for any function @xmath214 , let @xmath131 be the spectral measure of @xmath212 projected on the function @xmath120 . this measure is such that , for any positive continuous function @xmath215 , one has @xmath216 = \int \psi(\lambda ) \ \d e_f(\lambda).\ ] ] for any @xmath175 and @xmath176 , we recall that we say that the spectral exponents of a function @xmath120 are at least @xmath177 if holds . let us define the local drift @xmath217 in direction @xmath218 as @xmath219 = \frac{1}{p(\omega ) } \sum_{|z| = 1 } \omega_{0,z } \ \xi \cdot z.\ ] ] as we shall prove at the end of this section , we have the following bounds on the spectral exponents of @xmath217 . [ p : specexp ] under assumptions ( h1)-(h3 ) , there exists @xmath176 such that the spectral exponents of the function @xmath217 are at least @xmath220 let us see how this result implies theorem [ syserr ] . in order to do so , we also need the following information , that is a consequence of proposition [ p : specexp ] . [ corpoly ] let @xmath221\ ] ] be the image of @xmath217 by the semi - group at time @xmath1 associated with the markov chain @xmath209 . there exists @xmath176 such that @xmath222 = \left| \begin{array}{ll } o\big(t^{-2 } \ { \ln}^q(t)\big ) & \text{if } d = 2,\\ o\big(t^{-(d/2 + 1)}\big ) & \text{if } 3 { \leqslant}d { \leqslant}5 , \\ o\big(t^{-4 } \ { \ln}(t)\big ) & \text{if } d = 6,\\ o\big(t^{-4}\big ) & \text{if } d { \geqslant}7 . \end{array } \right.\ ] ] this result is the discrete - time analog of ( * ? ? ? * corollary 1 ) . it is obtained the same way , noting that @xmath222 = \int ( 1-\lambda)^{2 t } \ \d e_f(\lambda),\ ] ] and that the support of the measure @xmath131 is contained in @xmath194 $ ] . we are now in position to prove theorem [ syserr ] . the proof has the same structure as for the continuous - time case of ( * ? ? ? * proposition 8.4 ) . note that ( * ? ? ? * theorem 2.1 ) ensures that @xmath223 \,=\,\sigma^2.\ ] ] the starting point is the observation that , under @xmath96 , the process defined by @xmath224 is a square - integrable martingale with stationary increments . on the one hand , following ( [ defzf ] ) , we denote by @xmath225 the sum appearing in the r. h. s. of ( [ defnt ] ) . from proposition [ p : specexp ] and theorem [ quantkv ] , we learn that there exist @xmath226 and @xmath176 such that @xmath227 = \left| \begin{array}{ll } o \big ( { \ln}^{q}(t ) \big ) & \text{if } d = 2 , \\ o \big ( 1 \big ) & \text{if } d > 2 . \end{array } \right.\ ] ] on the other hand , since @xmath228 is a martingale with stationary increments , @xmath229 = t { \tilde}{{\mathbb{e}}}_0[(n_1)^2].\ ] ] as in the proof of theorem [ quantkv ] in the discrete time case , we then use that @xmath230 is orthogonal to @xmath231 to turn into @xmath232 = t^{-1}{\tilde}{{\mathbb{e}}}_0[(\xi \cdot y(t))^2 ] + t^{-1}{\tilde}{{\mathbb{e}}}_0[(z_{{\mathfrak}{d}}(t))^2 ] + 2t^{-1 } { \tilde}{{\mathbb{e}}}_0[{\mathfrak}{d}(\omega(t ) ) ( \xi \cdot y(t))].\ ] ] we already control the l. h. s. and the second term of the r. h. s. of ( [ decompnt ] ) . in order to quantify the convergence of @xmath233 $ ] it remains to control the last term . in particular , provided we show that @xmath234 = \left| \begin{array}{ll } o \big ( { \ln}^{q}(t ) \big ) & \text{if } d = 2 , \\ o \big ( 1 \big ) & \text{if } d > 2 , \end{array } \right.\ ] ] , , , and imply first that @xmath235-{\overline}{\sigma}^2 $ ] , and then the desired quantitative estimate . we now turn to and write @xmath236 & = & \sum_{s=0}^{t-1 } { \tilde}{{\mathbb{e}}}_0[{\mathfrak}{d}(\omega(t ) ) ( \xi \cdot ( y(s+1)-y(s ) ) ] \\ & = & \sum_{s=0}^{t-1 } { \tilde}{{\mathbb{e}}}_0[{\mathfrak}{d}_{t - s-1}(\omega(s+1 ) ) ( \xi \cdot ( y(s+1)-y(s))],\end{aligned}\ ] ] where we have used the markov property at time @xmath237 , together with the definition ( [ defdt ] ) of @xmath238 . using cauchy - schwarz inequality and the stationarity of the process @xmath209 under @xmath239 , this sum is bounded by @xmath240}^{1/2}.\ ] ] esimate ( [ controlrest ] ) then follows from corollary [ corpoly ] . this concludes the proof of the theorem . we finally turn to the proof of proposition [ p : specexp ] , which is a discrete - time counterpart of ( * ? ? ? * theorem 5 ) . in ( * theorem 5 ) however , we had proved in addition that the spectral exponents are at least @xmath241 , which is sharper than the exponents of proposition [ p : specexp ] for @xmath242 . in particular for @xmath242 the bounds of ( * ? ? ? * theorem 5 ) follow from results of @xcite , whose adaptation to the discrete time setting is not straightforward . as shown above , the present statement is sufficient to prove the optimal scaling of the systematic error , and we do not investigate further this issue . the proof of proposition [ p : specexp ] is rather involved and one may wonder whether this is worth the effort in terms of the application we have in mind namely theorem [ syserr ] . in order to obtain the optimal convergence rate in theorem [ syserr ] we need the spectral exponents to be larger than @xmath243 . proving that the exponents are at least @xmath243 is rather direct using results of @xcite ( see the first three steps of the proof of proposition [ p : specexp ] ) . yet proving that they are larger than @xmath243 for @xmath244 is as involved as proving proposition [ p : specexp ] itself . this is the reason why we display the complete proof of proposition [ p : specexp ] although the precise values of the spectral exponents are not that important in the context of this paper . the present proof is a direct version of the proof of ( * ? ? ? * theorem 5 ) . in particular , the proof of ( * ? ? ? * theorem 5 ) relies on a nontrivial ( weaker ) estimate of the spectral exponents obtained in @xcite using a covariance estimate . yet if one wants to extend these results to more general statistics of the conductivity function for instance to mixing coefficients in the sense of dobrushin and shlosman one has to give up the covariance estimate ( which we are not able to prove any longer , see @xcite ) . with this in mind we only rely on the variance estimate of ( * ? ? ? * lemma 2.3 ) . our strategy is similar to the strategy used for ( continuous ) elliptic equations in @xcite to prove the estimate of the systematic error . in particular we directly focus on the spectral exponents rather than on some other quantity like the systematic error itself . yet we will go slightly further . starting point is the inequality : @xmath245 which follows from the fact that for @xmath246 , @xmath247 ( here and below @xmath248 and @xmath249 stand respectively for @xmath250 and @xmath251 up to multiplicative constants ) . the variable @xmath252 for @xmath253 large plays the role of @xmath254 in . in what follows we make the standard identification between stationary functions @xmath255 of both the space variable @xmath256 and the environment @xmath28 and their translated versions at 0 @xmath257 depending on the environment only . we define @xmath258 as the unique stationary solution to @xmath259 whose existence and uniqueness follow from lax - milgram s theorem in @xmath260 using the identification between the stationary function @xmath258 and its version defined on the environment only ( see a similar argument of @xcite ) . in particular , with the notation @xmath261 , @xmath262 where @xmath263 is the operator defined in , and the spectral theorem ensures that @xmath264 where @xmath265 is the spectral measure of @xmath263 projected on the drift @xmath266 . we also let @xmath267 be the unique stationary solution to @xmath268 whose existence and uniqueness also follows from lax - milgram s theorem in the probability space as well . this time , @xmath269 and the spectral theorem yields @xmath270 from now on , we shall use the shorthand notation @xmath271 and @xmath272}={\left\langle ( u-{\left\langle u \right\rangle})^2 \right\rangle}$ ] for all @xmath273 . in particular the identity above turns into @xmath274},\ ] ] since @xmath275}\int \psi_t p \ \d \mathbb{p}=\frac{t}{\mathbb{e}[p]}\int \phi_t p \ \d \mathbb{p } = 0 $ ] using equations and . the rest of the proof , which is dedicated to the estimate of @xmath276}$ ] , is divided in six steps . starting point is the application of the variance estimate of ( * ? ? ? * lemma 2.3 ) to @xmath267 , which requires to estimate the susceptibility of @xmath267 with respect to the random coefficients . in view of it is not surprising that we will have to estimate not only the susceptibility of @xmath267 but also of @xmath258 and of some green s function with respect to the random coefficients . in the first step , we establish the susceptibility estimate for the green s function . in step 2 we turn to the susceptibility estimate for the approximate corrector @xmath258 . we then show in step 3 that , relying on @xcite , this implies that the spectral exponents are at least @xmath277 following step is to estimate the susceptibility of @xmath267 . in step 5 we show that this , combined with the suboptimal estimates @xmath278 obtained using , allow us to improve the spectral exponents to @xmath279 in the last step , we quickly argue that in turn these spectral exponents yield the optimal and suboptimal estimates @xmath280 which finally bootstrap to the desired estimates of the spectral exponents , and consequently yield the following optimal estimate of @xmath276}$ ] : @xmath281 } \,\lesssim \ , \left| \begin{array}{rcl } t^2\ln^qt & \mbox { for } & d=2,\\ t^{3/2}&\mbox { for } & d=3,\\ t & \mbox { for } & d=4,\\ \sqrt{t}&\mbox { for } & d=5,\\ \ln t&\mbox { for } & d=6,\\ 1&\mbox { for } & d>6 . \end{array } \right.\ ] ] we do not need step 6 to prove theorem [ syserr ] . yet this step is interesting in itself since it may be the starting point of a subtle `` multi - level '' induction to obtain the sharp spectral exponents in any dimension . _ step _ 1 . susceptility of the green s function . for all @xmath282 we define the green s function @xmath283 with singularity at @xmath37 as the unique solution in @xmath284 to the equation @xmath285 using lax - milgram s theorem . we shall prove for all @xmath286 , @xmath256 , @xmath287 , @xmath288 and @xmath289 where @xmath290 satisfies for some constant @xmath291 ( depending on @xmath292 ) @xmath293 for @xmath244 , and @xmath294 for @xmath8 . we define the elliptic operator @xmath295 as @xmath296 so that takes the form @xmath297 formally differentiating this equation with respect to @xmath90 yields @xmath298 using this identity turns into @xmath299 formally erasing the operator @xmath295 on the l. h. s. then yields the desired identity . to turn this into a rigorous argument , we may proceed as in ( * ? ? ? * proof of lemma 2.5 ) , first consider finite differences instead of derivative w. r. t. @xmath90 , use that @xmath295 is bijective on @xmath284 , and then pass to the limit . we leave the details to the reader and directly turn to . from we infer that @xmath300 using then the uniform pointwise estimate @xmath301 ( see ( * ? ? ? * corollary 2.3 ) , whose proof holds as well in the present case with a non - constant zero order term since @xmath302 ) , the uniform pointwise estimate of the green s function from ( * ? ? ? * lemma 3 ) , and considering this identity as an ode for @xmath303 in function of @xmath90 , we obtain . _ step _ 2 . susceptibility of the approximate corrector @xmath258 . in this step we shall prove that for @xmath304 , @xmath305 and @xmath287 , @xmath306 @xmath307 @xmath308 and for all @xmath309 , @xmath310 as for the green s function , we rewrite the defining equation for @xmath258 as @xmath311 formally differentiating w. r. t. @xmath90 yields @xmath312 which using turns into @xmath313 this ( formally ) shows . to turn this into a rigorous argument , we may use the green representation formula @xmath314 where @xmath315 denotes the sum over all @xmath316 , and proceed as in ( * ? ? ? * proof of lemma 2.4 ) . we now turn to . this estimate follows from , , and the following two facts : @xmath317 starting point to prove is the green representation formula in the form of @xmath318 from which we deduce the claim by a dyadic decomposition of space combined with cauchy - schwarz inequality and ( * ? ? ? * lemma 2.9 ) ( a similar calculation is detailed in ( * ? ? ? * proof of lemma 4 ) ) . for , we first note that implies @xmath319 which seen as an ode w. r. t. @xmath90 yields the claim using the uniform bound @xmath301 of ( * ? ? ? * corollary 2.3 ) and . estimate is a direct consequence of , whereas follows from the leibniz rule combined with , , and . _ step _ 3 . proof of . the estimates of the spectral exponents follow from the more general estimates : for all @xmath320 there exists @xmath321 such that @xmath322 combined with the fact that @xmath323 the proof of is an adaptation of ( * ? ? ? * proof of proposition 2.1 ) which already covers the case of a constant coefficient in the zero order term of @xmath295 , that is for @xmath324 instead of @xmath325 ( no randomness in the zero order term ) . the first step to apply the variance estimate of ( * ? ? ? * lemma 2.3 ) is to show that @xmath258 is measurable with respect to cylindrical topology associated with the random variables . this can be proved exactly as in ( * ? ? ? * lemma 2.6 ) . the estimates , , , and of steps 1 and 2 are like in the auxiliary lemmas of @xcite provided we replace the terms @xmath326 in ( * ? ? ? * lemmas 2.4 & 2.5 ) by @xmath327 . a close look at the proof of ( * ? ? ? * proposition 2.1 ) shows that these terms @xmath326 are either estimated by the green s function itself ( in which case the additional term @xmath328 is of higher order ) , or they are controlled on dyadic annuli by the meyers estimate ( * ? ? ? * lemma 2.9 ) . this lemma shows in particular that there exists @xmath329 such that for all @xmath330 , @xmath331 and @xmath332 , @xmath333 by the properties for @xmath244 and for @xmath8 of the function @xmath334 , it is easy to see that for @xmath244 @xmath335 as well , whereas for @xmath8 @xmath336 hence the proof of ( * ? ? ? * proposition 2.1 ) adapts mutadis mutandis to the present case ( with possibly an additional logarithmic correction for @xmath8 ) , and we have . _ step _ 4 . susceptibility of @xmath267 . in this step we shall prove that for @xmath337 , @xmath305 and @xmath287 , @xmath338 and @xmath339 where @xmath340 starting point is again the green representation formula @xmath341 associated with in the form @xmath342 differentiated w. r. t. @xmath90 it turns into @xmath343 combined with , , and the green representation formula itself , this shows . we now turn to and treat each term of the r. h. s. of separately . we begin with the third line of , appeal to , then bound the gradient of the approximate corrector @xmath344 by the approximate corrector @xmath345 itself , control the green s function @xmath346 by @xmath334 , and use to estimate the supremum in @xmath90 of the gradient of the green s function @xmath347 . this term is thus controlled by the second term of the r. h. s. of . the term in the fourth line of is also estimated by the second term of the r. h. s. of using ( and the uniform bounds @xmath348 ) , whereas the last two terms of are bounded by the first term of the r. h. s. of using . the subtle terms are the first three ones , for which we have to estimate the suprema of @xmath349 , @xmath350 , and @xmath351 w. r. t. @xmath90 . we begin with the following two estimates @xmath352 which considered as a system of two coupled odes show that there exists some @xmath353 such that for all @xmath354 , @xmath355 to prove we consider as an ode on @xmath356 , bound @xmath357 by @xmath358 , and use that the four last terms of the r. h. s. of are bounded by the second term of the r. h. s. of , as discussed above . hence turns into @xmath359 using the uniform bounds @xmath360 , and replacing the gradient of the green s function by the green s function itself in the integral which we then control by @xmath361 , this estimate yields by integrating the ode . we now turn to and infer from that @xmath362 proceeding as from to , this implies , and therefore and . combining the inequality @xmath363 with and yields the last estimate we need : @xmath364 we are finally in position to conclude the proof of : the first three terms of the r. h. s. of are estimated by , , and . replacing then the gradient of the green s function by the green s function itself , and the green s function by @xmath334 , we obtain , as desired . _ step _ 5 . proof of . this is an application of the variance estimate ( * ? ? ? * lemma 2.3 ) on @xmath267 based on . in particular , @xmath365}\,\lesssim\,\sum_{e\in \mathbb{b}}{\left\langle \sup_{\omega_e}\left| \frac{\partial \psi_t(0)}{\partial \omega_e } \right|^2 \right\rangle}.\ ] ] we distinguish the contributions of the two terms of the r. h. s. of in this sum and use the notation @xmath366 the contribution of the first term is estimated as follows : @xmath367 by stationarity of @xmath258 , @xmath267 , and @xmath368 . combined with for @xmath369 , and ( which is proved below ) , this turns into @xmath370 estimate is a consequence of the estimate of the spectral exponents and of the spectral representations @xmath371 since @xmath372 , and by definition of the spectral exponents . we now turn to the second term of the r. h. s. of ( [ eq : step-4 - 2 ] ) , and note that it coincides with the term treated in ( * ? ? ? * step 3 , proof of lemma 5 ) so that we have @xmath373 hence , , , and yield for @xmath244 using and . note that is not optimal for @xmath8 in view of the estimate of @xmath374 in ( which is sharper ) . this comes from our estimate of @xmath375 where we have replaced a gradient of the green s function by the green s function itself and therefore given up some potential better decay ( at least integrated on dyadic annuli ) . in high dimensions however , there is no loss because @xmath334 is then square - integrable itself . anyway , the optimal spectral exponents for @xmath8 have been already obtained in . _ step _ 6 . proof of . we quickly show how allows to get the sharper spectral exponents . this is the same argument as in step 5 , except that can now be boostrapped to using and @xmath376 where @xmath377 has been defined in step 4 . proceeding as in the end of step 5 this yields the spectral exponents , and concludes the proof of the proposition . in this section , we show that the computable quantity @xmath81 defined in ( [ defhata ] ) is a good approximation of @xmath68 , in the sense that its random fluctuations are small as soon as @xmath83 is large . we write @xmath378 for @xmath379 . [ randfluc ] there exists @xmath380 such that , for any @xmath381 , @xmath382 and @xmath1 large enough , @xmath383 { \leqslant}\exp\left ( - \frac{n { \varepsilon}^2}{c t^2}\right).\ ] ] in order to prove this result , we rewrite @xmath81 as @xmath384 where @xmath385},\ ] ] and @xmath386}.\ ] ] both are sums of independent and identically distributed random variables under @xmath97 , thus enabling us to use standard tools from large deviations theory . we start with the following classical but instructing fact . [ gdhatp ] there exists @xmath387 such that , for any @xmath388 and any small enough @xmath382 : @xmath389 { \leqslant}\exp\left ( - n { \varepsilon}^2/c_0 \right).\ ] ] let @xmath390 $ ] . it suffices to show that , for some @xmath391 , @xmath392 { \leqslant}\exp(- n { \varepsilon}^2 / c).\ ] ] chebyshev s inequality implies that , for any @xmath393 , @xmath394 & { \leqslant } & e^{-n { \varepsilon}\lambda } { \mathbb{e}}^\otimes\left[\exp(\lambda({\overline}{p}({\omega}^{(1)})+ \cdots + { \overline}{p}({\omega}^{(n)}))\right ] \\ & { \leqslant } & e^{-n { \varepsilon}\lambda } { \mathbb{e}}\left[\exp(\lambda{\overline}{p}({\omega}))\right]^n.\end{aligned}\ ] ] using a series expansion of the exponential , one can check that there exists @xmath391 such that , for any @xmath395 small enough , @xmath396 { \leqslant}c \lambda^2.\ ] ] as a consequence , for any @xmath395 small enough , @xmath397 { \leqslant}\exp\left ( -n ( \lambda { \varepsilon}- c \lambda^2 ) \right),\ ] ] and for @xmath398 , the latter term becomes @xmath399 . the event @xmath400 can be handled the same way , and we thus obtain ( [ oneside ] ) . what makes the proof of proposition [ gdhatp ] work is the observation ( [ loglapl ] ) that the log - laplace transform of @xmath401 $ ] is quadratic close to the origin . in order to prove the corresponding result for @xmath402 , we will need to control the log - laplace transform of @xmath403 uniformly in @xmath1 . to that end , we use a sharp upper bound on the transition probabilities of the random walk recalled in the following theorem . we refer the reader to @xcite or ( * ? ? ? * theorem 14.12 ) for a proof . [ gauss ] there exists a constant @xmath404 such that , for any environement @xmath28 with conductances in @xmath405 $ ] , any @xmath406 and @xmath20 , @xmath407 { \leqslant}\frac{c_1}{t^{d/2 } } \exp\left(- \frac{|x|^2}{c_1 t } \right).\ ] ] from theorem [ gauss ] we deduce the following result . [ supexp ] let @xmath408 be given by theorem [ gauss ] . for all @xmath409 , one has @xmath410 < + \infty.\ ] ] let @xmath411 . by theorem [ gauss ] , @xmath412 { \leqslant}c_1 { t^{-d/2 } } \sum_{x \in { \mathbb{z}}^d } e^{-\delta |x|^2/t}.\ ] ] if the sum ranges over all @xmath413 , it is easy to bound it by a convergent integral : @xmath414 by symmetry , the estimate carries over to the sum over all @xmath415 . the same argument applies for the sum over all @xmath416 having exactly one component equal to @xmath43 , and so on . the following lemma contains the required uniform control on the log - laplace transform of @xmath403 . [ grandesdev ] there exist @xmath417 and @xmath418 such that , for any @xmath419 and any @xmath406 , @xmath420 { \leqslant}c_2 \lambda^2.\ ] ] it is sufficient to prove that there exists @xmath421 such that , for any @xmath395 small enough and any @xmath1 , @xmath422 { \leqslant}1+c_3 \lambda^2.\ ] ] we use the series expansion of the exponential to rewrite this expectation as @xmath423.\ ] ] the term corresponding to @xmath424 is equal to @xmath193 , whereas the term for @xmath425 vanishes . the remaining sum , for @xmath426 ranging from @xmath70 to infinity , can be controlled using corollary [ supexp ] combined with the bound @xmath427,\ ] ] which follows from the definition of @xmath68 and jensen s inequality . we are now in position to prove theorem [ randfluc ] . starting point is the inequality @xmath428 \\ { \leqslant}{\mathbb{p}}^\otimes_0\left[a_n(t ) - \sigma_t^2 { \geqslant}{\varepsilon}/ t\right ] + { \mathbb{p}}^\otimes_0\left[a_n(t)(\hat{p}_n^{-1 } - 1 ) { \geqslant}{\varepsilon}/ t\right].\end{gathered}\ ] ] we treat both terms of the r. h. s. separately . for the first one , the key observation is that ( recalling the definition of @xmath55 given in ( [ deftdp ] ) ) @xmath429 = { \tilde}{{\mathbb{p}}}^\otimes_0\left[\frac{(\xi \cdot y^{(1)}(t))^2 + \cdots + ( \xi \cdot y^{(n)}(t))^2}{n t } - \sigma_t^2 { \geqslant}{\varepsilon}/ t\right].\ ] ] let @xmath430 . as in the proof of proposition [ gdhatp ] , we bound this term using chebyshev s inequality : @xmath431 \\ & \qquad { \leqslant}{\tilde}{{\mathbb{e}}}^\otimes_0\left [ \exp\left ( \lambda \left(\frac{(\xi \cdot y^{(1)}(t))^2 + \cdots + ( \xi \cdot y^{(n)}(t))^2}{t } - n \sigma_t^2 \right ) \right ) \right ] \ \exp\left ( -\frac{n \lambda { \varepsilon}}{t } \right ) \\ & \qquad { \leqslant}{{\tilde}{{\mathbb{e}}}_0\left [ \exp\left ( \lambda \left(\frac{(\xi \cdot y(t))^2}{t } - \sigma_t^2 \right ) \right ) \right]}^n \ \exp\left ( -\frac{n \lambda { \varepsilon}}{t } \right ) . \end{split}\ ] ] by lemma [ grandesdev ] , the r. h. s. of ( [ comput1 ] ) is bounded by @xmath432 for all @xmath395 small enough . choosing @xmath433 ( which is small enough for @xmath1 large enough ) , we obtain @xmath434 { \leqslant}\exp\left ( - \frac{n { \varepsilon}^2}{4 c_2 t^2 } \right),\ ] ] as needed . we now turn to the second term of the r. h. s. of ( [ decomp2termes ] ) . from inequality ( [ estim1 ] ) , we infer that there exists @xmath435 such that @xmath436 { \leqslant}\exp\left ( - \frac{n { \varepsilon}^2}{4 c_2 t^2 } \right).\ ] ] since @xmath437 is almost surely bounded by a constant , it is enough to evaluate the probability @xmath438,\ ] ] which is controlled by proposition [ gdhatp ] . we have thus obtained the required control of the l. h. s. of ( [ decomp2termes ] ) . the probability of the symmetric event @xmath439\ ] ] can be handled the same way . in this section , we illustrate on a simple two - dimensional example the sharpness of the estimates of the systematic error and of the random fluctuations obtained in theorems [ syserr ] and [ randfluc ] . in the numerical tests , each conductivity of @xmath26 takes the value @xmath440 or @xmath441 with probability @xmath442 . in this simple case , the homogenized matrix is given by dykhne s formula , namely @xmath443 ( see for instance ( * ? ? ? * appendix a ) ) . for the simulation of the random walk , we generate and store the environment along the trajectory of the walk . in particular , this requires to store up to a constant times @xmath1 data . in terms of computational cost , the expansive part of the computations is the generation of the randomness . in particular , to compute one realization of @xmath444 costs approximately the generation of @xmath445 random variables . a natural advantage of the method is its full scalability : the @xmath446 random walks used to calculate a realization of @xmath444 are completely independent . we first test the estimate of the systematic error : up to a logarithmic correction , the convergence is proved to be linear in time . in view of theorem [ randfluc ] , typical fluctuations of @xmath447 are of order no greater than @xmath448 , and thus become negligible when compared with the systematic error as soon as the number @xmath449 of realizations satisfies @xmath450 . we display in table [ tab : syst ] an estimate of the systematic error obtained with @xmath451 realizations . the systematic error is plotted on figure [ fig : syst ] in function of the time in logarithmic scale . the apparent convergence rate ( linear fitting ) is @xmath452 , which is consistent with theorem [ syserr ] , which predicts @xmath453 and a logarithmic correction . for @xmath451 realizations ] we now turn to the random fluctuations of @xmath454 . theorem [ randfluc ] gives us a large deviation estimate which essentially says that the fluctuations of @xmath447 have a gaussian tail , measured in units of @xmath448 . the figures [ fig : histo-1]-[fig : histo-4 ] display the histograms of @xmath455 for @xmath456 and @xmath457 ( with 10000 realizations of @xmath444 in each case ) . as expected , they look gaussian . ] ] ] ] the authors acknowledge the support of inria through the `` action de recherche collaborative '' disco . this work was also supported by ministry of higher education and research , nord - pas de calais regional council and feder through the `` contrat de projets etat region ( cper ) 2007 - 2013 '' . g.c . papanicolaou and s.r.s . boundary value problems with rapidly oscillating random coefficients . in _ random fields , vol . i , ii ( esztergom , 1979 ) _ , volume 27 of _ colloq . jnos bolyai _ , pages 835873 . north - holland , amsterdam , 1981 .

within the frame of macroscopic quantum electrodynamics in causal media , the van der waals interaction between an atomic system and an arbitrary arrangement of dispersing and absorbing dielectric bodies including metals is studied . it is shown that the minimal - coupling scheme and the multipolar - coupling scheme lead to essentially the same formula for the van der waals potential . as an application , the vdw potential of an atom in the presence of a sphere is derived . closed expressions for the long - distance ( retardation ) and short - distance ( non - retardation ) limits are given , and the effect of material absorption is discussed .

quant-ph0303152

the involvement of van der waals ( vdw ) forces in a variety of physicochemical processes and promising potential applications ( such as the construction of atomic - force microscopes @xcite or reflective atom - optical elements @xcite ) have created the need for a very detailed understanding and controlling them . casimir and polder @xcite were the first to study the vdw interaction within the frame of rigorous quantum electrodynamics ( qed ) . investigating the interaction between an atom and a perfectly reflecting semi - infinite ( planar ) body , they found that in the short - distance ( non - retardation ) limit the interaction potential @xmath0 behaves like @xmath1 ( @xmath2 , distance between the atom and the interface ) , whereas in the long - distance ( retardation ) limit it behaves like @xmath3 . in their theory , casimir and polder quantized the ( transverse ) vector potential of the electromagnetic field ( outside the perfectly reflecting body ) in terms of normal modes and coupled them to the atom according to the minimal - coupling hamiltonian , with its coulomb part being determined by means of the method of image charges . they then calculated the ( lowest - order ) change of the ground - state energy of the system arising from this coupling , which is a function of the atomic position and thus plays the role of the potential energy that determines the force acting on the atom . in the earlier experiments @xcite , which studied the deflection of thermal atomic beams by conducting surfaces , the observed signal was extremely low . nevertheless , qualitative trends in agreement with the @xmath1 law were observed . only recent progress in experimental techniques has rendered it possible to detect vdw forces with sufficiently high precision @xcite . in particular , by using atomic passage between two parallel plates , the @xmath3 retarded potential could be verified @xcite . other methods for measuring vdw forces have been based on transmission grating diffraction of molecular beams @xcite , atomic quantum reflection @xcite , evanescent - wave atomic mirror techniques @xcite , and ( indirect ) measurements via spectroscopic means @xcite . proposals have been made on improvements of monitoring the vdw interaction by using atomic interferometry @xcite . since the appearance of casimir s and polder s pioneering article in 1948 there has been a large body of work on the vdw interaction ( see , e.g. , @xcite and references therein ) . roughly speaking , there have been two routes to treat the problem . in the first , which closely follows the ideas of casimir and polder , explicit field quantization is performed by applying standard concepts of qed , such as normal - mode techniques @xcite . in particular , extensions of the theory to one @xcite and two semi - infinite dielectric walls @xcite , thin metallic films @xcite , and cylindrical and spherical dielectric bodies @xcite have been given , and the problem of force fluctuations on short time scales has been studied @xcite . the calculations have typically been based on macroscopic qed , by applying normal - mode decomposition and including in it the bodies by the well - known conditions of continuity at the surfaces of discontinuity . since in such an approach the dependence on frequency of the response to the field of the bodies can not be properly taken into account , material dispersion and absorption are ignored . the problem does of course not occur in microscopic qed , where the bodies are treated on a microscopical level by adopting , e.g. , harmonic - oscillator models ( see , e.g. , @xcite ) . apart from the fact that the calculations are rather involved , the results are model - dependent . to overcome the difficulties mentioned , in the second route , the calculations are based on linear response theory , without ( explicitly ) quantizing the electromagnetic field @xcite . all the relevant entities are expressed in terms of correlation functions which in turn are related , via the fluctuation - dissipation theorem , to response functions . the method has been employed to investigate the vdw energy of an atom near a semi - infinite ( planar ) body made of dielectric @xcite , metallic @xcite , ionic - crystallic @xcite , and birefringent material @xcite . further , the problem of an atom near a a dielectric plate @xcite , a metallic sphere @xcite , and a nanowire @xcite has been considered . since the method borrows from equilibrium statistical mechanics , its applicability is restricted to the vacuum and thermal quantum states . in this article , we give a unified qed approach to the vdw interaction between an atomic system ( such as an atom or a molecule ) and dispersing and absorbing dielectric bodies including metals . starting from the quantized version of the macroscopic maxwell field , with the medium being described in terms of a spatially varying , kramers - kronig consistent ( complex ) permittivity , we derive an expression for the vdw potential that applies to arbitrary body configurations . the formalism can be regarded as being a generalization of the normal - mode formalism of macroscopic qed , so that it can be applied to other than the vacuum and thermal states , and it also allows extensions like the inclusion of dynamical interactions between the atomic system and the medium - assisted electromagnetic field . roughly speaking , the mode expansion is replaced with some source - quantity representation in terms of the green tensor of the macroscopic maxwell equations , in which material dispersion and absorption are automatically included . further , we close the gap between the minimal - coupling scheme and the multipolar - coupling scheme by showing that both approaches lead to equivalent results . to give an application , we consider the vdw potential of an atom in the vicinity of a dispersing and absorbing microsphere . microspheres may be interesting candidates for qed experiments ( see , e.g. , @xcite and references therein ) with atomic beams . the atom - surface distances should be adjusted so that the atoms fly close enough to the surface to facilitate a strong coupling with the microsphere resonances , but not get adsorbed on the microsphere surface because of the attractive vdw force . the article is organized as follows . in section [ sec : basic_eqs ] the formalism is outlined and an expression for the vdw potential is derived . in section [ sec : appl ] the formalism is applied to an atom near a sphere . a summary and some conclusions are given in section [ sec : concl ] . let us consider an atomic system ( such as an atom or a molecule ) interacting with the quantized electromagnetic field in the presence of macroscopic , dispersing and absorbing dielectric bodies . in the nonrelativistic limit , the minimal - coupling hamiltonian in coulomb gauge reads @xcite @xmath4 ^ 2 \nonumber\\ & & + { \textstyle\frac{1}{2}}\int{\,\mathrm{d}}^3r\,\hat{\rho}_a({\bf r } ) \hat{\varphi}_{\rm a}({\bf r } ) + \int{\,\mathrm{d}}^3r\,\hat{\rho}_a({\bf r } ) \hat{\varphi}_{\rm m}({\bf r } ) , \quad\end{aligned}\ ] ] where @xmath5 and @xmath6 are respectively the masses and charges of the particles constituting the atomic system , while @xmath7 and @xmath8 are respectively their coordinates ( relative to the center of mass @xmath9 ) and canonically conjugated momenta . the first term in the hamiltonian describes the combined system of the electromagnetic field plus the macroscopic bodies ( including dissipative systems ) in terms of bosonic vector fields @xmath10 , which satisfy the commutation relations @xmath11 = \delta_{ij}\delta({\bf r}-{\bf r}')\delta(\omega-\omega'),\ ] ] @xmath12 = \left[\hat{f}_i^{\dagger}({\bf r},\omega ) , \hat{f}_j^{\dagger}({\bf r'},\omega')\right ] = 0.\ ] ] the second term is the kinetic energy of the charged particles , while the third term describes their mutual coulomb interaction , where @xmath13 and @xmath14\ ] ] are respectively the scalar potential and the charge density of the atomic system . the last term accounts for the coulomb interaction of the particles with the medium . the vector potential @xmath15 and the scalar potential @xmath16 of the medium - assisted electromagnetic field are given by @xmath17 @xmath18 where the symbols @xmath19 and @xmath20 are used to distinguish transverse and longitudinal vector fields , respectively . in particular , @xmath21 and @xmath22 read @xmath23 where @xmath24 and @xmath25 are the transverse and longitudinal dyadic @xmath26-functions , respectively , and the medium - assisted electric field in the @xmath27-domain , @xmath28 , is expressed in terms of the basic variables @xmath10 as @xmath29&\times\ ; \bm{g}({\bf r},{\bf r}',\omega ) { } \hat{\bf f}({\bf r}',\omega ) \end{split}\ ] ] . here , @xmath30 is the classical green tensor and @xmath31 is the imaginary part of the complex , space- and frequency - dependent ( relative ) permittivity @xmath32 the green tensor , which obeys the inhomogeneous partial differential equation @xmath33 \bm{g}({\bf r},{\bf r}',\omega ) = \bm{\delta}({\bf r}-{\bf r}')\ ] ] together with the boundary condition at infinity , has the following general properties ( see , e.g. , @xcite ) : @xmath34 @xmath35 @xmath36 in this way , all electromagnetic - field quantities can be expressed in terms of the fundamental fields @xmath10 . in particular , the operator for the electric field reads @xmath37 - \bm{\nabla}\hat{\varphi}_{\rm m}({\bf r } ) - \bm{\nabla}\hat{\varphi}_{\rm a}({\bf r } ) \nonumber\\ & = & \hat{\bf e}_{\rm m}({\bf r } ) + \sum_\alpha \frac{q_\alpha[{\bf r}-({\bf r}_{\rm a}+\hat{\bf r}_\alpha ) ] } { 4\pi\varepsilon_0|{\bf r}-({\bf r}_{\rm a}+\hat{\bf r}_\alpha)|^3}\,,\end{aligned}\ ] ] where @xmath38 and the polarization field associated with the dielectric medium is given by @xmath39 where @xmath40 here , @xmath41 is the dielectric susceptibility and @xmath42 is the so - called noise polarization . for the following it is convenient to decompose the hamiltonian ( [ e3 ] ) as @xmath43 where @xmath44 @xmath45 \hspace{-1ex}&=&\hspace{-1ex } \sum_{\alpha } \frac{\hat{{\bf p}}_{\alpha}^2}{2m_{\alpha } } + \sum_{\alpha<\beta } \frac{q_{\alpha}q_{\beta } } { 4\pi\varepsilon_0\left|\hat{{\bf r}}_{\alpha } -\hat{{\bf r}}_{\beta}\right|}\,,\end{aligned}\ ] ] @xmath46 . \quad\end{aligned}\ ] ] obviously , @xmath47 is the hamiltonian of the medium - assisted electromagnetic field , @xmath48 is the hamiltonian of the atomic system with eigenstates @xmath49 and eigenvalues @xmath50 according to @xmath51 and @xmath52 is the interaction energy between them . for a neutral atomic system in the electric - dipole approximation , the latter simplifies to @xmath53 @xmath54 @xmath55 where @xmath56 is the electric dipole operator of the atomic system @xmath57 following the original line of casimir and polder @xcite , we calculate the vdw energy of an atomic system as the position - dependent part of the leading - order correction to the unperturbed ground state energy due to the perturbation according to eq . ( [ e28 ] ) [ together with eqs . ( [ e29 ] ) and ( [ e30 ] ) ] . since the diagonal matrix elements of @xmath58 , eq . ( [ e29 ] ) , are zero , the non - vanishing contribution to the energy correction from first - order perturbation theory is due to @xmath59 , eq . ( [ e30 ] ) . inspection of eq . ( [ e30 ] ) shows that this contribution is of linear order in the fine structure constant . we have thus to include that contribution from second - order perturbation theory which is of the same order in @xmath60 . therefore , we apply first - order perturbation theory for @xmath59 and second - order perturbation theory for @xmath58 . the energy correction to the ground state thus reads @xmath61 where @xmath62 & = & { \langle 0 |}{\langle \{0\ } | } \sum_{\alpha}\frac{q_{\alpha}^2}{2m_{\alpha } } \,\hat{\bf a}^2({{\bf r}_{\rm a } } ) { |\{0\}\rangle}{|0\rangle}\end{aligned}\ ] ] and @xmath63 & & = \frac{1}{\hbar}\sum_n\int_0^{\infty}\!\ ! \frac{{\,\mathrm{d}}\omega}{\omega_n+\omega } \int\!{\,\mathrm{d}}^3{r } \,\bigl |{\langle 0 |}{\langle \{0\ } |}\sum_{\alpha}\frac{q_{\alpha}}{m_{\alpha } } \hat{\bf p}_{\alpha}\hat{\bf a}({{\bf r}_{\rm a } } ) \nonumber\\[1ex]&&\hspace{10ex } -\,\hat{\bf d}\bm{\nabla}\hat{\varphi}_{\rm m}({{\bf r}_{\rm a } } ) { |\{{\bf 1}({\bf r},\omega)\}\rangle}{|n\rangle } \bigr |^2.\end{aligned}\ ] ] here , @xmath64 and @xmath65 are respectively the ground state of the atomic system and the ground state of the medium - assisted electromagnetic field , and @xmath66 denotes single - quantum fock states of the fundamental fields . further , @xmath67 are the transition frequencies between the excited atomic states and the ground state . note that due to the linear dependence of the vector potential and the gradient of the scalar potential on @xmath10 [ and @xmath68 , the nonvanishing matrix elements of @xmath58 in the fock - state basis ( which is defined as the set of eigenvectors of @xmath47 ) are those between fock states that just differ in one quantum . expressing in eq . ( [ e33 ] ) @xmath69 in terms of @xmath70 [ and @xmath71 , on using eqs . ( [ e8 ] ) , ( [ e10 ] ) , and ( [ e11 ] ) , recalling eq . ( [ e35 ] ) , and exploiting the integral relation ( [ e22 ] ) for the contraction of two green tensors , we derive after some lengthy but straightforward calculation @xmath72 where @xmath73 in eq . ( [ e37 ] ) and below , summation over repeated vector indices is understood . using the sum rule @xmath74 ( for a proof , see appendix [ sec : sumrule ] ) , where @xmath75 denote the matrix elements of the electric - dipole operator ( [ e31 ] ) , we may equivalently represent eq . ( [ e37 ] ) in the form of @xmath76 & & = \frac{\mu_0}{\pi } \sum_n \int_0^{\infty}{\,\mathrm{d}}\omega\ , \omega_n { \bf d}_{0n } { \,\mathrm{im}}^{\perp}\bm{g}^{\perp}({{\bf r}_{\rm a}},{{\bf r}_{\rm a}},\omega ) { \bf d}_{n0}. \qquad\end{aligned}\ ] ] in a similar fashion , we find that eq . ( [ e34 ] ) leads to @xmath77 & \times\int{\,\mathrm{d}}^3{s } ' \,{\bf d}_{0n } \bm{\mu}({\bf s},\omega ) { \rm im}\,\bm{g } ( { \bf s},{\bf s}',\omega ) \bm{\mu}({\bf s}',\omega ) { \bf d}_{n0}\biggr ] , \end{split}\ ] ] where , on using the relation ( [ a3 ] ) , the matrix elements of the electric - dipole moment ( [ e42 ] ) have been introduced and the abbreviating notation @xmath78 has been used . we substitute eqs . ( [ e40 ] ) and ( [ e41 ] ) into eq . ( [ e32 ] ) , and obtain , on recalling eq . ( [ e43 ] ) , @xmath79&&\hspace{3ex } -\,\omega_n\omega\bigl [ { \,\mathrm{im}}^\perp\bm{g}^\perp({{\bf r}_{\rm a}},{{\bf r}_{\rm a}},\omega ) + { \,\mathrm{im}}^\perp\bm{g}^\parallel({{\bf r}_{\rm a}},{{\bf r}_{\rm a}},\omega ) \nonumber\\[1ex]&&\hspace{3ex } + \,{\,\mathrm{im}}^\parallel\bm{g}^\perp({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \bigr ] \bigr\ } { \bf d}_{n0}.\end{aligned}\ ] ] let @xmath80 be the ( small ) region of space where the atom is situated and the permittivity can be regarded as being effectively not varying with space , i.e. , if . for , the green tensor can then be given in the form @xmath81 with @xmath82 denoting the ( translationally invariant ) bulk green tensor that corresponds to @xmath83 and @xmath84 being the scattering green tensor that accounts for the spatial variation of the permittivity . in practice , the atom is typically situated in a free - space region , so that @xmath85 is simply the vacuum green tensor . according to the decomposition of the green tensor in eq . ( [ e45 ] ) , the energy correction @xmath86 given by eq . ( [ e44 ] ) consists of two terms , @xmath87 where the @xmath9-independent term @xmath88 , which is related to the bulk green tensor , gives rise to the ( vacuum ) lamb shift , whereas the @xmath9-dependent term @xmath89 , which is related to the scattering green tensor , is just the vdw energy sought : @xmath90&&\quad\times \bigl\{\omega^2 { \,\mathrm{im}}^\parallel\bm{g}^{(1)\parallel}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) -\,\omega_n\omega \nonumber\\[1ex]&&\quad \times\bigl [ { \,\mathrm{im}}^\perp\bm{g}^{(1)\perp}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) + { \,\mathrm{im}}^\perp\bm{g}^{(1)\parallel}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \nonumber\\[1ex]&&\quad + \,{\,\mathrm{im}}^\parallel\bm{g}^{(1)\perp}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \bigr ] \bigr\ } { \bf d}_{n0}.\end{aligned}\ ] ] to further evaluate this expression , it is convenient to express it in terms of the whole scattering green tensor rather than its imaginary part . for this purpose , we write , recall the relation ( [ e20 ] ) , and change the integration variable from @xmath91 to @xmath27 . equation ( [ e47 ] ) then changes to @xmath92 \nonumber\\ & & -\omega^2 { } ^\parallel\bm{g}^{(1)\parallel}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \bigr\ } + \int^0_{-\infty}\frac{{\,\mathrm{d}}\omega}{\omega_n-\omega } \bigl\ { \omega_n\omega \bigl [ { } ^\perp\bm{g}^{(1)\perp}({{\bf r}_{\rm a}},{{\bf r}_{\rm a}},\omega ) + { } ^\perp\bm{g}^{(1)\parallel}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) + { } ^\parallel\bm{g}^{(1)\perp}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \bigr]\nonumber\\ & & + \omega^2 { } ^\parallel\bm{g}^{(1)\parallel}({{\bf r}_{\rm a } } , { { \bf r}_{\rm a}},\omega ) \bigr\ } \biggr ) { \bf d}_{n0}\,.\end{aligned}\ ] ] this equation can be greatly simplified by using contour - integral techniques . note that @xmath93 is an analytic function in the upper complex half plane ( , ) . further , @xmath94 , @xmath95 , and @xmath96 tend to zero as @xmath27 approaches zero , because the asymptote of the green tensor contains no transverse components [ cf . ( [ b4 ] ) ] . finally , the term @xmath97 is also well - behaved for vanishing @xmath27 , as can be seen from eq . ( [ b8 ] ) . consequently , the integrands of the @xmath27-integrals in eq . ( [ e48 ] ) are analytic functions without poles in the whole upper complex half plane , including the real axis . we may therefore apply cauchy s theorem , and replace the integral over the positive ( negative ) real half axis by a contour integral along the positive imaginary half axis ( introducing the purely imaginary coordinate ) and along a quarter circle with infinite radius in the first ( second ) quadrant of the complex frequency plane . since the integrals along the infinitely large quarter circles vanish [ cf . eq . ( [ b3 ] ) ] , we finally arrive at @xmath98 where the identity @xmath99 @xmath100 @xmath101 @xmath102 @xmath103 @xmath102 @xmath104 @xmath102 @xmath105 has been taken into account . introducing the ( lowest - order ) ground - state polarizability tensor @xmath106 of the atomic system ( see , e.g. , @xcite ) , we may represent eq . ( [ e49 ] ) in the equivalent form of @xmath107 it is worth noting that eq . ( [ e51 ] ) directly follows from qed in causal media , without the need of additional assumptions borrowed from other fields . equation ( [ e51 ] ) expresses the vdw potential of an arbitrary atomic system ( such as an atom or a molecule ) in the presence of an arbitrary configuration of dispersing and absorbing macroscopic dielectric bodies in terms of the polarizability tensor of the atomic system in lowest order of perturbation theory and the scattering green tensor of the macroscopic maxwell equations . in particular for an atom , one can make use of the spherical symmetry and reduce eq . ( [ e51 ] ) to @xmath108 where @xmath109 this result agrees with the results inferred from mlassical linear response theory . note that the field susceptibility introduced in refs . @xcite and @xcite differs from the scattering green tensor by a factor of @xmath110 . needless to say , that in the special case of a two - level atom , eq . ( [ e52 ] ) reduces to the result , e.g. , in ref . the derivation of eq . ( [ e52 ] ) shows that it can be regarded as the natural extension of the qed results obtained on the basis of the normal - mode formalism , which ignores material absorption . let us turn to the multipolar - coupling scheme widely used for studying the interaction of electromagnetic fields with atoms and molecules . just as in standard qed , so in the present formalism the multipolar - coupling hamiltonian can be obtained from the minimal - coupling hamiltonian by means of a power zienau transformation , @xmath111 where @xmath112\end{aligned}\ ] ] with @xmath113\ ] ] being the polarization associated with the ( neutral ) atomic system . using @xmath114 from eq . ( [ e3 ] ) , we derive @xcite @xmath115 & & + \ , q_{\alpha}\int_0 ^ 1\!{\rm d}\lambda\,\lambda \hat{\bf r}_{\alpha } \times \hat{\bf b}\left [ { { \bf r}_{\rm a}}\!+\!\lambda \hat{\bf r}_{\alpha } \right ] \bigg\}^2 \nonumber\\[1ex ] & & + \,\frac{1}{2\varepsilon_0 } \!\int\ ! { \rm d}^3{r } \ , \hat{\bf p}_{\rm a}({\bf r } ) \hat{\bf p}_{\rm a}({\bf r } ) - \int\ ! { \rm d}^3{r}\ , \hat{{\bf p}}_{\rm a}({\bf r } ) \hat{\bf e}_{\rm m}({\bf r } ) , \quad \end{aligned}\ ] ] where with @xmath15 from eq . ( [ e8 ] ) [ together with eqs . ( [ e10 ] ) and ( [ e11 ] ) ] , and @xmath116 is defined by eq . ( [ e14 ] ) [ together with eq . ( [ e11 ] ) ] . note that in the multipolar - coupling scheme the operator of the electric field strength is defined according to @xmath117 - \bm{\nabla } \hat{\varphi}_{\rm m}({\bf r } ) - \bm{\nabla } \hat{\varphi}_{\rm a}({\bf r } ) \nonumber\\[1ex ] & = & \hat{\bf e}_{\rm m}({\bf r } ) - \frac{1}{\varepsilon_0}\,\hat{\bf p}_{\rm a}({\bf r}),\end{aligned}\ ] ] i.e. , @xmath118 hence , @xmath119 has the meaning of the displacement field with respect to the polarization of the atomic system . in the electric - dipole approximation , eq . ( [ e57 ] ) simplifies to @xmath120 where @xmath121 and @xmath122 respectively , are the unperturbed hamiltonians of the medium - assisted electromagnetic field and the atomic system , and @xmath123 is the interaction energy between them , where @xmath56 is the atomic dipole operator given by eq . ( [ e31 ] ) . recall that @xmath124 must be thought of a being expressed in terms of the fundamental field variables @xmath125 [ and @xmath126 . comparing eq . ( [ e62 ] ) with eq . ( [ e25 ] ) , we see that , on taking into account the the relationship @xmath127 the atomic hamiltonians @xmath128 and @xmath48 are different from each other , so that the solution of the eigenvalue problem @xmath129 may be different from that defined by eq . ( [ e27 ] ) . keeping in mind this difference , we drop the primes denoting the atomic eigenvalues and eigenstates from here on . in contrast to the interaction energy in the minimal - coupling scheme , eq . ( [ e28 ] ) , the interaction energy in the multipolar - coupling scheme , eq . ( [ e63 ] ) , is linear in @xmath125 and @xmath130 . as a consequence of the latter , there is no first - order correction to the ground state energy . we thus have @xmath131 where @xmath132 [ cf . the first line in eq . ( [ e34 ] ) with @xmath133 instead of @xmath58 ] . in complete analogy to the derivation of eq . ( [ e41 ] ) , we find that @xmath135 to further evaluate the @xmath9-dependent part @xmath136 @xmath100 @xmath137 of @xmath138 , which results from the scattering part of the green tensor , @xmath93 , and gives the vdw energy , we again write , use the relation ( [ e20 ] ) , and change the integration variable from @xmath91 to @xmath27 . after some algebra we arrive at @xmath139 { \bf d}_{n0}\ , . \qquad\end{aligned}\ ] ] as we already know , the integrands of the two frequency integrals appearing in eq . ( [ e68 ] ) are analytic functions in the upper half of the complex frequency plane , including the real axis [ cf . ( [ b8 ] ) ] . we therefore can apply contour integral techniques in a similar way as in the derivation of eq . ( [ e49 ] ) from eq . ( [ e48 ] ) . it is not difficult to see that the result reads @xmath140 which has exactly the same form as the minimal - coupling result ( [ e49 ] ) , so that it can also be given in the form of eq . ( [ e51 ] ) . recall that the values of @xmath141 and @xmath142 obtained in the minimal - coupling scheme may be different from those obtained in the multipolar - coupling scheme , because of the somewhat different eigenvalue equations ( [ e27 ] ) and ( [ e27a ] ) . let us apply the theory to an atom near a dispersing and absorbing dielectric ( micro-)sphere surrounded by vacuum . the material of the sphere of radius @xmath143 is assumed to be homogeneous and isotropic , having a permittivity @xmath144 . the coordinate system is chosen such that its origin lies at the center of the sphere . the scattering green tensor can be given by @xcite @xmath145&&\times\ ; \frac{(n - m)!}{(n+m ) ! } \biggl[b^m_n\sum_{p=-1,1}{\bf m}_{nm , p}({{\bf r}_{\rm a } } ) \otimes{\bf m}_{nm , p}({{\bf r}_{\rm a } } ) \nonumber\\[1ex]&&\hspace{4ex } + \,b^n_n\sum_{p=-1,1}{\bf n}_{nm , p}({{\bf r}_{\rm a } } ) \otimes{\bf n}_{nm , p}({{\bf r}_{\rm a}})\biggr],\end{aligned}\ ] ] where @xmath146 and @xmath147 are even and odd spherical wave vector functions , which can be expressed in terms of spherical hankel functions of the first kind , @xmath148 , and associated legendre functions , @xmath149 , in the following way : @xmath150&&\times\ ; \begin{array}{c}\sin\\ \cos \end{array } ( m\phi){\bf e}_{\theta } -h_n^{(1)}(k_0 r_{\rm a})\ , \frac{{\,\mathrm{d}}p_n^m(\cos\theta)}{{\,\mathrm{d}}\theta}\ , \begin{array}{c}\cos\\ \sin \end{array } ( m\phi){\bf e}_{\phi } , \nonumber\\ & & \\ \label{e72 } \lefteqn { { \bf n}_{nm,\pm1}({{\bf r}_{\rm a } } ) = n(n+1)\,\frac{h_n^{(1)}(k_0 r_{\rm a})}{k_0 r_{\rm a}}\,p_n^m(\cos\theta ) } \nonumber\\&&\times\ ; \begin{array}{c } \cos\\ \sin \end{array}(m\phi){\bf e}_r + \frac{1}{k_0 r_{\rm a } } \frac{{\,\mathrm{d}}[r_{\rm a}h_n^{(1)}(k_0 r_{\rm a})]}{{\,\mathrm{d}}r_{\rm a } } \biggl[\frac{{\,\mathrm{d}}p_n^m(\cos\theta)}{{\,\mathrm{d}}\theta } \nonumber\\&&\times\ ; \begin{array}{c } \cos\\ \sin \end{array } ( m\phi){\bf e}_{\theta } \mp\frac{m}{\sin\theta}\,p_n^m(\cos\theta ) \begin{array}{c } \sin\\ \cos \end{array } ( m\phi){\bf e}_{\phi}\biggr].\end{aligned}\ ] ] here , is the vacuum wave number , and @xmath151 , @xmath152 , @xmath153 , are the mutually orthogonal unit vectors pointing in the directions of @xmath154 , @xmath155 , and @xmath156 , respectively . the coefficients @xmath157 and @xmath158 in eq . ( [ e70 ] ) read @xmath159 & & = - \frac { \bigl [ z_1j_n(z_1)\bigr ] ' j_n(z_0 ) - \bigl [ z_0j_n(z_0)\bigr ] ' j_n(z_1 ) } { \bigl [ z_1j_n(z_1)\bigr ] ' h_n^{(1)}(z_0 ) - \bigl[z_0 h_n^{(1)}(z_0)\bigr ] ' j_n(z_1 ) } \ , , \quad\end{aligned}\ ] ] @xmath160 & & = - \frac { \varepsilon(iu ) j_n(z_1 ) \bigl[z_0 j_n(z_0)\bigr ] ' - j_n(z_0 ) \bigl[z_1 j_n(z_1)\bigr ] ' } { \varepsilon(iu ) j_n(z_1 ) \bigl [ z_0 h_n^{(1)}(z_0)\bigr ] ' -h_n^{(1)}(z_0)\bigl[z_1 j_n(z_1)\bigr ] ' } \ , , \quad \nonumber\\&&\end{aligned}\ ] ] where and , with @xmath161 @xmath100 @xmath162 being the wave number inside the sphere , and @xmath163 is the spherical bessel function of the first kind . the primes indicate differentiations with respect to @xmath164 or @xmath165 , respectively . the coefficients @xmath157 represent contributions from transverse electric ( te ) waves reflected at the surface of the sphere , while the coefficients @xmath158 represent those from transverse magnetic ( tm ) waves . substituting the trace of @xmath166 from eq . ( [ e70 ] ) [ together with eqs . ( [ e71 ] ) and ( [ e72 ] ) ] into eq . ( [ e52 ] ) yields the vdw energy sought . the sums over @xmath167 can then easily be performed using the orthogonality of the unit vectors @xmath151 , @xmath152 , and @xmath168 , and the sum over @xmath169 can be performed with the aid of the summation formulas in appendix [ appc ] . so after a lengthy , but straightforward calculation we arrive at the following result : @xmath170 ^ 2 + n(n+1)b^n_n \left[\frac{h^{(1)}_n(k_0 r_{\rm a } ) } { k_0 r_{\rm a}}\right]^2 \nonumber\\&&\hspace{4ex } + \,b^n_n \left[\frac{1}{k_0 r_{\rm a } } \frac{{\,\mathrm{d}}[r_{\rm a } h^{(1)}_n(k_0 r_{\rm a } ) ] } { { \,\mathrm{d}}r_{\rm a}}\right]^2 \biggr\}\biggr).\end{aligned}\ ] ] note that the vdw potential does not depend on the angle variables of the atomic position , but only on the distance of the atom from the center of the sphere , as can be anticipated from the symmetry of the system . recall that the terms proportional to @xmath171 represent the contributions from the te ( tm ) waves . equation ( [ e75 ] ) applies to an arbitrary dielectric sphere . in particular , when material absorption is omitted , then the result in ref . @xcite can be recovered . a detailed analysis of eq . ( [ e75 ] ) requires numerical computation . here , however , we would like to focus our attention on two interesting limiting cases , where the atom is very far from or very close to the sphere . let us first consider the limit of the atom being far away from the sphere , @xmath172 in this case , eq . ( [ e75 ] ) reduces to @xmath173&&\hspace{10ex}\times\ , \left[2\left(1+z\right)^2+\left(1+z+z^2\right)^2\right]e^{-2z } \qquad\end{aligned}\ ] ] ( see appendix [ appd ] ) , where it turns out that the te waves do not contribute . since the inequalities and are valid , the vdw potential is negative , and the resulting force between the atom and the sphere is attractive . as is seen from eq . ( [ e77 ] ) , the main contribution to the integral comes from the region where . therefore , for sufficiently large distances , the contributions from small frequencies dominate , and we can ( approximately ) replace the atomic polarizability and the material permittivity in eq . ( [ e77 ] ) with their static values and , respectively . the integration can then be performed in closed form to yield the asymptotic distance law @xmath174 in this so - called retarded limit . note that in the opposite nonretarded limit , where the contributions of @xmath175 and @xmath144 at all frequencies have to be retained , eq . ( [ e77 ] ) reduces to the result given in ref . @xcite , where a @xmath176 law was found . the ( formal ) limit in eq . ( [ e78 ] ) obviously corresponds to a metallic sphere @xmath177 note that the decrease of the force with the distance is of three powers stronger than in the case of the atom being near a planar body . in particular , if we introduce the static polarizability of the sphere ( see , e.g. , @xcite ) @xmath178 we may rewrite eq . ( [ e78 ] ) as @xmath179 interestingly , eq . ( [ e80 ] ) also applies to the vdw potential between two atoms @xcite , if the ( static ) polarizability of the sphere is replaced with the polarizability of the second atom . let us now proceed to the short - distance limit of the atom being located at a position very close to the sphere , i.e. , @xmath180 ( @xmath181 @xmath182 @xmath183 @xmath184 @xmath185 ) . in this case , from eq . ( [ e75 ] ) it follows that @xmath186 ( see appendix [ appe ] ) . note that again the te waves do not contribute to @xmath136 . as expected , the dependence on distance of the ( attractive ) vdw potential corresponds to that obtained in the case of the atom being near a planar body . in fact , it exactly looks like that derived in ref . @xcite for an atom in the vicinity of a planar , semi - infinite , non - absorbing dielectric . in particular , if the ( model ) assumption @xmath187 @xmath184 @xmath188/[\varepsilon(iu)$ ] @xmath102 @xmath188 $ ] @xmath100 @xmath189 were made for all values of @xmath190 , then eq . ( [ e82 ] ) would lead , on using eq . ( [ e53 ] ) , to the result @xcite @xmath191 to explore the effect of material absorption , we may assume a permittivity of drude - lorentz type , @xmath192 where @xmath193 and @xmath194 are respectively the ( transverse ) resonance frequencies and the associated absorption constants , and the frequencies @xmath195 are proportional to the so - called oscillator strengths . from eq . ( [ e84 ] ) it is seen that in the limit the resulting static permittivity @xmath196 is independent of the absorption parameters . since it is the static permittivity that enters eq . ( [ e78 ] ) , we see that the long - distance asymptote of the vdw potential is not influenced by material absorption . with decreasing distance the range of frequency that must be taken into account increases . thus , the frequency response of the permittivity becomes crucial to the strength of the vdw force . let us consider the short - distance law ( [ e82 ] ) . from eq . ( [ e84 ] ) it follows that @xmath197 ^ 2 } \frac{2u\omega_{l}^2}{(\omega_{l}^2+u^2+u\gamma_l)^2}\ , , \ ] ] that is to say , @xmath198 with regard to eq . ( [ e82 ] ) , we therefore find that , on recalling that , @xmath199 hence , the vdw force monotonically decreases with increasing absorption constants . [ @xmath200 @xmath100 @xmath201 , @xmath202 @xmath100 @xmath203 as a function of the atom - surface distance for various strengths of material absorption . in the calculation , a metallic permittivity of drude type according to eq . ( [ e84 ] ) and a ( degenerate ) single - resonance atomic polarizability are assumed with , @xmath204 @xmath100 @xmath205 , and @xmath206 @xmath100 @xmath207 ( solid line ) , ( dashed line ) , and @xmath206 @xmath100 @xmath189 ( dotted line ) . ] figure [ force ] illustrates the influence of material absorption on the vdw force acting on an atom located near a metallic sphere in the short - distance limit . it can be seen that the effect of material absorption increases with decreasing atom - surface distance . in particular , at a distance of @xmath181 @xmath208 @xmath209 , an increase of the relative absorption parameter from @xmath206 @xmath100 @xmath207 to @xmath206 @xmath100 @xmath189 would reduce the magnitude of the force by nearly thirty percents . within the frame of macroscopic qed , we have derived an expression for the vdw potential of an atomic system near an arbitrary configuration of dispersing and absorbing bodies . it generalizes the results obtained by means of normal - mode expansion and may be regarded as a foundation of the results inferred from linear response theory . we have performed the calculations for both the minimal - coupling scheme and the multipolar - coupling scheme and shown that the results are essentially the same . we have applied the theory to the vdw interaction between an atom and a sphere . from the integral expression , we have derived the correct long - distance law corresponding to the retardation limit and recovered the short - distance law corresponding to the non - retardation limit . in particular , replacing in the long - distance law the polarizability of the sphere with that of an atom just yields the vdw potential between two atoms . on the other hand , for sufficiently small distances of the atom from the sphere the vdw potential approaches the potential observed for an atom near a planar body . it is worth noting that in the long - distance limit it is the static permittivity that enters the vdw potential . hence material absorption has no effect on it . however , with decreasing distance of the atom from the sphere the relevant frequencies extend for a finite ( increasing ) interval and material absorption becomes substantial , thereby diminishing the strength of the force . in this article , we have restricted our attention to ground - state systems and calculated the vdw potential in lowest - order of perturbation theory with respect to the interaction of the atomic system with the medium - assisted electromagnetic field . the theory allows of course extensions in several respects . as a consequence of the lowest - order perturbation theory , the energy denominators that enter the polarizability of the atomic system are the unperturbed ones , without consideration of the level shift and broadening caused by the presence of the bodies . in fact , the polarizability of an atomic system is expected to drastically change when it becomes close to a macroscopic body and the spontaneous decay thus becomes purely radiationless , with the decay rate being proportional to @xmath210 @xcite . since the level broadening is essentially determined by the spontaneous - decay rate , the polarizability becomes distance - dependent an effect that needs careful consideration . since the electromagnetic field in ( linear ) magnetic media can be quantized analogously @xcite , another interesting extension of the theory be the inclusion in it of composite materials characterized by both a complex permittivity and a complex permeability . interestingly , such materials , which have been fabricated recently , are left - handed . last not least the underlying quantization scheme renders it also possible to extend the theory to atoms and molecules in excited states and treat the motion of driven atomic systems . we would like to thank ludwig knll for valuable discussions . this work was supported by the deutsche forschungsgemeinschaft . from the commutation relation @xmath211 = \biggl[\hat{r}_{\alpha , i}\ , , \sum_{\beta } \frac{\hat{{\bf p}}_{\beta}^2}{2m_{\beta}}\biggr ] = \frac{i\hbar}{m_{\alpha}}\,\hat{p}_{\alpha , i}\ ] ] together with the eigenvalue equation ( [ e27 ] ) we find that @xmath212{|n\rangle } \nonumber\\[1ex ] & \!=&\ ! -\frac{im_{\alpha}}{\hbar}{\langle 0 |}\hat{r}_{\alpha , i}\hat{h}_a -\hat{h}_a\hat{r}_{\alpha , i}{|n\rangle } \nonumber\\[1ex ] & \!=&\ ! -\frac{im_{\alpha}}{\hbar}(e_n - e_0 ) { \langle 0 |}\hat{r}_{\alpha , i}{|n\rangle } \nonumber\\[1ex ] & \!=&\ ! -im_{\alpha}\omega_n{\langle 0 |}\hat{r}_{\alpha , i}{|n\rangle}.\end{aligned}\ ] ] thus @xmath213 & \!=&\ ! -i\omega_n{\langle 0 |}\hat{d}_i{|n\rangle}.\end{aligned}\ ] ] using eq . ( [ a3 ] ) , we derive @xmath214&&\hspace{2ex } = \frac{i}{2\hbar}\sum_{\alpha}\frac{q_{\alpha}}{m_{\alpha } } \sum_n\left({\langle 0 |}\hat{p}_{\alpha , i}{|n\rangle}{\langle n | } \hat{d}_j{|0\rangle}\right . \nonumber \\&&\hspace{20ex } \left.- { \langle 0 |}\hat{d}_j{|n\rangle}{\langle n | } \hat{p}_{\alpha , i}{|0\rangle}\right ) \qquad \nonumber\\[1ex]&&\hspace{2ex } = \frac{i}{2\hbar}\sum_{\alpha}\frac{q_{\alpha}}{m_{\alpha } } { \langle 0 |}\big[\hat{p}_{\alpha , i},\hat{d}_j\big]{|0\rangle } \nonumber\\[1ex]&&\hspace{2ex } = \frac{i}{2\hbar}\sum_{\alpha}\frac{q_{\alpha}}{m_{\alpha } } { \langle 0 |}\big[\hat{p}_{\alpha , i } , \sum_{\beta}q_{\beta}\hat{r}_{\beta , j}\big]{|0\rangle } \nonumber\\[1ex]&&\hspace{2ex } = \sum_{\alpha}\frac{q_{\alpha}^2}{2m_{\alpha}}\delta_{ij},\end{aligned}\ ] ] which is just eq . ( [ e39 ] ) . the asymptotic behavior of the green tensor for large frequencies reads @xcite @xmath215 @xmath216 if @xmath217 and @xmath218 lie in a common region of constant permittivity , we can use eq . ( [ e45 ] ) , and subtract the two equations ( [ b1 ] ) and ( [ b2 ] ) to obtain @xmath219 in the low - frequency limit we have @xcite @xmath220 where @xmath221 recalling that as , @xmath222 ( i\omega)^{-1 } & \mbox{for metals } \end{array } \right.\ ] ] [ cf . ( [ e84 ] ) ] , from eqs . ( [ b4 ] ) and ( [ b4 - 1 ] ) we see that @xmath223 needless to say , that eq . ( [ b8 ] ) is also valid for the scattering part of the green tensor , @xmath224 . the legendre polynomials obey the relation @xcite @xmath225 where @xmath226 for and , eq . ( [ c1 ] ) reduces to @xmath227 differentiating eq . ( [ c1 ] ) twice with respect to @xmath228 and putting and afterwards yield @xmath229 finally , subsequent differentiations of eq . ( [ c1 ] ) with respect to @xmath230 and @xmath231 and again putting and afterwards yield @xmath232 ^ 2 = \frac{n(n+1)}{2}\,.\ ] ] in eq . ( [ e75 ] ) , the spherical hankel functions @xmath233 @xmath100 @xmath234 can be written in the form of @xcite @xmath235 with some complex coefficients @xmath236 . from inspection of eqs . ( [ e73 ] ) and ( [ e74 ] ) it is seen that @xmath237 and @xmath238 can be expanded in powers of @xmath190 at , @xmath239 from eqs . ( [ hexpansion ] ) and ( [ abexpansion ] ) it then follows that the integrand of the ( imaginary ) frequency integral in eq . ( [ e75 ] ) is a sum of terms , which are all of the same general structure @xmath240 ( @xmath241 are nonnegative integers ) . for @xmath242 @xmath243 @xmath244 @xmath184 @xmath245 this is a polynomial in @xmath190 times an exponentially decaying function . the only relevant contributions to the frequency integral come from the maximum of @xmath246 at a frequency @xmath247 satisfying @xmath248 thus @xmath249 where we have used the fact that @xmath250 can be regarded as almost constant for the small frequencies considered here . for @xmath251 @xmath184 @xmath245 , @xmath246 is a monotonically decreasing function , and relevant contributions to the frequency integral can only come from regions , where @xmath252 because for larger frequencies the exponentially decaying factor becomes too small . combining eqs . ( [ cc4 ] ) and ( [ cc5 ] ) , it can be said that the relevant contributions to the frequency integral come from regions , where @xmath190 @xmath253 @xmath254 . in these regions we have @xmath255 this means that in the long - distance limit @xmath256 @xmath257 @xmath185 the main contributions to the integral come from regions , where @xmath258 . we may therefore expand the coefficients @xmath259 in eqs . ( [ e73 ] ) and ( [ e74 ] ) in powers of @xmath260 , on exploiting useful relations in ref . @xcite , and retain only the leading terms : @xmath261,\ ] ] @xmath262 ^ 2 } \frac{\varepsilon(iu)-1 } { \varepsilon(iu ) n + n+1}(k_0r)^{2n+1}.\ ] ] in this way we find that the leading term in eq . ( [ e75 ] ) comes from the two terms containing @xmath158 with @xmath263 @xmath100 @xmath189 . keeping only these terms , using @xcite @xmath264 and changing the integration variable according to @xmath190 @xmath265 @xmath266 @xmath100 @xmath267 , we arrive at eq . ( [ e77 ] ) . provided that @xmath268 the spherical bessel and hankel functions appearing in eqs . ( [ e73 ] ) ( [ e75 ] ) can be approximated by @xcite @xmath269 and @xmath270 respectively . as we know from appendix [ appd ] , the main contribution to the frequency integral in eq . ( [ e75 ] ) is from those values satisfying the condition ( [ cc5 ] ) . hence , the condition ( [ d1 ] ) becomes @xmath271 because @xmath2 @xmath272 @xmath273 @xmath274 in the short - distance limit . substituting in eqs . ( [ e73 ] ) ( [ e75 ] ) for @xmath163 and @xmath275 the expressions ( [ jbign ] ) and ( [ ybign ] ) , we derive after some algebra @xmath276 @xmath277 ^ 2 \nonumber\\ & & + ( 2n+1)b^n_n \left[\frac{{\,\mathrm{d}}[r_{\rm a } h^{(1)}_n(k_0r_{\rm a})]}{k_0r_{\rm a}{\,\mathrm{d}}r_{\rm a}}\right]^2 \nonumber\\ & & \simeq -i\frac{1}{(k_0r_{\rm a})^3 } \frac{\varepsilon(iu)-1}{\varepsilon(iu)+1}n(n+1 ) \left(\frac{r}{r_{\rm a}}\right)^{2n+1 } \nonumber\\&&\quad -i\frac{1}{(k_0r_{\rm a})^3 } \frac{\varepsilon(iu)-1}{\varepsilon(iu)+1 } \frac{(2n+1)^2}{4 } \left(\frac{r}{r_{\rm a}}\right)^{2n+1 } \nonumber\\ & & \simeq -2i\frac{1}{(k_0r_{\rm a})^3 } \frac{\varepsilon(iu)-1}{\varepsilon(iu)+1}n(n+1 ) \left(\frac{r}{r_{\rm a}}\right)^{2n+1}.\end{aligned}\ ] ] whereas in the long - distance limit we could neglect all terms but the @xmath263 @xmath100 @xmath189 one ( appendix [ appd ] ) , in the short - distance limit , as can be seen from eq . ( [ d6 ] ) , the parameter @xmath278 being very close to one , we encounter the opposite extreme , where the main contribution comes from those terms corresponding to high orders @xmath263 . the main contribution to the sum over @xmath263 in eq . ( [ e75 ] ) comes from the peak at @xmath279 determined by @xmath280\right|_{n = n_1 } } \nonumber\\[1ex ] & & \approx\left.\frac{{\,\mathrm{d}}}{{\,\mathrm{d}}n}\left(n^2e^{2n \left(\ln r-\ln r_{\rm a}\right)}\right)\right|_{n = n_1 } \nonumber\\[1ex ] & & = 2n_1\left(\frac{r}{r_{\rm a}}\right)^{2n_1}\left[1+n_1 \left(\ln r-\ln r_{\rm a}\right)\right ] = 0,\end{aligned}\ ] ] from which we find @xmath281 because of @xmath282 @xmath283 @xmath189 . since the main contribution to the sum over @xmath263 in eq . ( [ e75 ] ) comes from values of , where the approximate formulas ( [ d5 ] ) and ( [ d6 ] ) are valid [ cf . ( [ d2 ] ) ] . therefore we introduce only a small error if we extrapolate these formulas to the terms with small @xmath263 . then the sum over @xmath263 is equal to the second derivative with respect to @xmath284 of a geometric sum , which can be performed in closed form to yield eq . ( [ e82 ] ) . m. oria , m. chrevrollier , d. bloch , m. fichet , and m. ducloy , europhys . * 14 * , 527 ( 1991 ) ; v. sandoghdar , c. i. sukenik , e. a. hinds , and s. haroche , phys . 68 * , 3432 ( 1992 ) ; m. fichet , f. schuller , d. bloch , and m. ducloy , phys . a * 51 * , 1553 ( 1995 ) ; m. marrocco , m. weidinger , r. t. sang , and h. walther , phys . lett . * 81 * , 5784 ( 1998 ) ; h. failache , s. saltiel , m. fichet , d. bloch , and m. ducloy , _ ibid . _ * 82 * , 5467 ( 1999 ) . m. bostrm and b. e. sernelius , phys . rev . a * 61 * , 052703 ( 2000 ) . note that the formulas derived in ref . @xcite for frequency - independent , real permittivities are used for studying metals , by putting complex permittivities in them , without any proof . a. m. marvin and f. toigo , phys . a * 25 * , 782 ( 1982 ) . in fact an energy formula based on a normal - mode expansion is combined with elements of the linear - response theory .

dark energy models , such as the chameleon , where the acceleration of the expansion of the universe results from the dynamics of a scalar field coupled to matter , suffer from the potential existence of a fifth force . three known mechanisms have been proposed to restore general relativity in the solar system and the laboratory , which are the symmetron / damour - polyakov effect , the vainshtein property and the chameleon screening . here , we propose to probe the existence of chameleons in the laboratory , considering their particle physics consequences . we envisage the resonant and non - resonant production of chameleons in the sun and their back - conversion into x - ray photons in a solar helioscope pipe such as the one used by cast . we find that the resonant contribution is only present for a very small range of chameleon couplings to matter and occurs in extremely narrow magnetised regions in the sun . however , in the non - resonant case , chameleons are created in all available solar magnetic regions and for a very wide range of chameleon couplings to photons and matter . such chameleons could leave the sun and be back - converted into x - ray photons above the photosphere or in the magnetic pipe of a helioscope . a detection of these x - rays would indicate the existence of chameleons . the number of produced chameleons and regenerated x - ray photons varies depending on the strength of the solar magnetic field in the production shell , the matter and photon couplings of chameleons , and the magnetic field length and magnitude inside a dipole magnet . we focus on a template model for the solar magnetic field : a constant magnetic field in a narrow shell surrounding the tachocline . the x - ray photons in a helioscope pipe obtained from back - conversion of the chameleons created inside the sun have a spectrum which is peaked in the sub - kev region , just below the actual sensitivity range of the present axion helioscopes . nevertheless they are detectable by present day magnetic helioscopes like cast and sumico , which were built originally for solar axions . we also propose a chameleon - through - a - wall experiment whereby x - ray photons from a synchroton radiation source could be converted into chameleons inside a dipole magnet , then pass a wall which is opaque to x - rays before being back - converted into x - ray photons in a second magnet downstream . we show that this could provide a direct signature for the existence of chameleon particles .

1110.2583

the acceleration of the expansion rate of the universe is still a deep mystery despite more than ten years of intense activity both experimentally@xcite and theoretically@xcite . if the acceleration of the universe is due to an unknown form of matter , dark energy , it comprises more than seventy percent of the energy budget of the universe . together with dark matter , some ninety five percent of the energy contents of the universe are unknown to us and only detected through indirect cosmological and gravitational astrophysical probes . of course , the situation would be greatly clarified if some imprint of the existence of dark energy could be discovered experimentally . in this sense , looking for dark energy can be put in parallel with the search for dark matter ( see @xcite and references therein ) . for dark matter , the detection methods comprise the identification of potential candidates in the laboratory ( ranging from experiments at the lhc looking for very heavy new particles to light - shining - through - a - wall experiments at the opposite end of the mass scale ) as well as finding signatures in astrophysical environments like the production of axions in the sun for example . even if dark matter particles were found , dedicated efforts would be required to ascertain that these candidates do act as the pervading astrophysical dark matter . hence experimental searches for dark matter in indirect ( by looking for annihilation products of dark matter particles for example ) or direct channels are also under scrutiny . the latter includes scattering experiments mainly focusing on heavy wimp dark matter in the multi-10 gev range and axion dark matter searches in the @xmath0ev mass regime . searches for dark energy clearly lag behind their dark matter counterparts . a particular reason for this is the lack of particle candidates as well as realistic proposals for smoking - gun detections of dark energy . here we propose astrophysical and laboratory experiments to probe a particular class of dark energy models , i.e. the chameleons@xcite , and their distinctive environmentally dependent features . it should be stressed that even with a clear identification of chameleons in the laboratory , this would not provide a direct explanation to dark energy itself . specialised experiments would have to be pursued in order to prove that chameleons correspond to the dark energy of the universe . however , it would clearly be a big step forward in our understanding of the universe if candidates for dark matter and dark energy were identified in the laboratory . if dark energy is a new form of matter , strong candidates could be scalar fields , which have not been detected so far , interacting with both matter and photons . indeed , scalar fields can provide the potential energy which would drive the acceleration of the expansion rate of the universe . unfortunately this implies that the scalar field mass on cosmological scales is of the order of the hubble rate now , i.e. @xmath1 ev . such a tiny mass would lead to large gravitational effects and the presence of a long range fifth force . the cassini bound on the coupling of such scalar fields to matter@xcite , together with other gravitational tests in the laboratory@xcite , lead to unnatural models if a screening mechanism were not at play both in the solar system and the laboratory . fortunately , known screening mechanisms exist such as the thin shell effect for chameleons and the vainshtein mechanism for galileons , involving highly non - linear lagrangians which can only be understood as low energy effective field theories@xcite . the origin of these models at higher energy is unknown . in this paper , we will focus on chameleons . chameleon models can describe the acceleration of the expansion of the universe on large scales while having a density dependent effective mass which explains why they are screened in local experiments . indeed , the rest mass of chameleons is large inside dense bodies implying that the field inside does not extend further than a scale of order @xmath2 where @xmath3 is the mass at a given point in the astrophysical body . hence , the contributions to the value of the exterior field from inside the body are damped by a large yukawa suppression @xmath4 , where @xmath5 is the distance from a point in the body to another one outside . only a tiny shell at the surface of the body , called a thin shell , contributes to the field outside the body as the field is not yukawa damped across this thin layer . therefore the exterior of a body feels the field created only by a small shell and is thus small , i.e. small enough to create a tiny modification of gravity and pass all the gravitational tests on non - cosmological scales . however , chameleons themselves can have detectable effects in the laboratory . indeed , they can induce a small but detectable deviation from the casimir force between two metallic plates @xcite . they can also couple to two photons in a primakoff fashion@xcite and may lead to observable phenomena in optical experiments , if , for example , a laser beam goes across a magnetic region providing a bath of virtual photons@xcite . using the same coupling to two photons , they could be produced in solar magnetic regions in a way akin to the axionic one@xcite . the main difference between axions and chameleons would be the fact that chameleons couple to the photon polarisation perpendicular to a magnetic field ( pseudo - scalars couple to the parallel polarisation ) , and the crucial variation of their effective mass with the pressure - less energy density of the environment . given the pressure - less energy density profile , the mass essentially depends on two parameters : an index @xmath6 which characterises the dark energy potential and the coupling constant to matter @xmath7 . together with the coupling constant to photons @xmath8 , this forms the three parameter family of chameleon models . notice that for chameleons , the mass has no intrinsic meaning like for axions . here the index @xmath6 and the two coupling constants @xmath7 and @xmath8 suffice . if , in a dense magnetised plasma , the chameleon mass matches the plasma frequency of the environment , chameleons can be resonantly produced in the presence of a magnetic field . this could happen in the sun in a narrow shell whose location depends on the effective rest mass of the chameleons . on the other hand , non - resonant production is more widespread and occurs in all magnetised regions inside the sun . once created either resonantly or non - resonantly across large magnetic regions , chameleons essentially escape from the sun and reach earth where they can enter an helioscope and be back - converted into x - ray photons@xcite . the fact that the emitted chameleons have a spectrum in the immediate sub - kev range helps them to overcome any barrier . note that this only holds true as long as their mass in matter is smaller than the chameleon energy . low energy chameleons are reflected by dense matter . the chameleon emitted by the sun eventually reach earth where they can enter magnetic helioscopes like cast@xcite and sumico@xcite and be back - converted into x - ray photons . additionally , a small fraction of the solar chameleons could also be converted into x - ray photons in the photosphere and above , hence providing a possible explanation to the corona puzzle : the corona is hotter above magnetised solar surfaces . this reasoning has been used also for axions or axion - like particles in @xcite and applies to chameleons as well . in fact , as we show in this work , one can distinguish the photon emission coming from chameleons from that coming from axion conversion . in order to estimate the sensitivity of possible experiments , we have applied present constraints mimicking the ones from the cast experiment and typical of helioscopes , i.e. a background noise of 1 event per hour in the 1 - 7 kev region , to draw the sensitivity region in the chameleon parameter space at the 2@xmath9 level . for the chameleon production inside the sun , we have used the conservative assumption that the solar magnetic field is mainly effective in a shell of 0.01 solar radius around the tachocline where a constant magnetic field of 30 t is commonly assumed to exist with this rather realistic width . with these hypotheses , we deduce both the chameleon spectrum and the regenerated photon spectra in a helioscope pipe , the photosphere and the corona . as we find that the photon spectrum is peaked below the main sensitivity region of the present helioscope detectors , these limits will be improved with future upgrades . constraints on the chameleon parameter space have already been given by the chase experiment at fermilab using optical photons@xcite . in particular for low values of the matter coupling , the coupling to photons is bounded from above @xmath10 . with the assumption concerning the production of chameleons in the sun , due to a shell of 0.01 solar radius ( @xmath11 ) and a magnetic field of 30 t , and a maximal chameleon emission of 10% of the solar luminosity@xcite , we obtain an energy loss bound on the photon coupling @xmath12 . this is comparable to the chase bound obtained with optical photons . however , the results in optical cavities involve many reflections of the chameleons on the beamtube walls and thus are limited by the presence of possible multiple scatterings . to avoid these uncertainties , we propose a new x - ray light - shining - through - a - wall experiment where these issues are absent . similarly , the possible detection of x - ray photons springing from the existence of solar chameleons would be a rather indirect way of studying the properties of chameleons . to overcome these obstacles , we suggest a new experiment where a photon flux would be injected in a magnetic pipe@xcite . as a result , chameleons would be produced . if a wall is introduced after the first magnetic pipe , energetic chameleons would go through the wall and would regenerate photons on the other side downstream . using the energy loss bound on the coupling of chameleons to photons , we find that a 5@xmath9 detection of photons would be within reach with an improved source of photons compared to the ones available now : a smoking gun for chameleons in the laboratory . in the first part of this paper , we recall general results on chameleons and their production in the sun . we then calculate the number of back - converted photons in a helioscope pipe . finally we present a chameleon through wall experiment and we conclude . chameleons are candidates to generate the recent acceleration of the expansion of the universe . on large scales , the chameleon field takes a vacuum expectation value whose energy density drives the acceleration . on smaller scales , this field would be responsible for a fifth force . however , a screening mechanism in massive bodies allows the model to evade all the local gravitational tests . chameleon models depend on the shape of the dark energy potential . in the following we will always assume that @xmath13 is a decreasing function . the coupling to matter @xmath7 is also crucial . chameleons have an effective potential which becomes matter density dependent @xmath14 where @xmath15 is the chameleon field and @xmath16 the non - relativistic matter density . as the potential @xmath13 decreases , the effective potential has always a matter dependent minimum . fluctuations around the vacuum expectation values at the minimum can be seen as chameleon particles which couple to matter . the best constraint on @xmath7 comes from the energy levels of neutrons in the terrestrial gravitational field and reads @xmath17@xcite . the density - dependent minimum is such that the mass of the scalar field becomes also density dependent . we will focus on inverse power law models defined by @xmath18 where we have neglected higher inverse powers of the chameleon field . these models describe chameleon properties with no loss of generality . moreover , although precise numerical results depend on @xmath6 , generic properties are mostly independent of @xmath6 . in the following we will focus on small values of @xmath6 , i.e. @xmath19 and @xmath20 when we investigate chameleon resonances in the sun and later only @xmath20 for the back - conversion of chameleons in the photosphere , the corona and helioscopes . a more thorough scanning of the values of @xmath6 should be envisaged when comparing our results to experimental data in helioscopes . we will choose @xmath21 to accommodate the acceleration of expansion rate of the universe on large scales . the potential has a minimum located at @xmath22 where @xmath16 is the total non - relativistic matter density . the chameleon rest mass at the minimum is @xmath23 we will see that in a cylindrical pipe , this rest mass is only valid at large enough coupling to matter @xmath7 . chameleons also couple to photons in a way akin to the axion coupling @xmath24 where @xmath25 is the determinant of the metric @xmath26 and @xmath27 where @xmath28 is the photon field strength . this implies that the effective matter density in the effective potential is @xmath29 where @xmath30 is the magnetic field . for instance , with @xmath31 and @xmath32 , the contribution of a magnetic field @xmath33 t , like the one around the tachocline , to the energy density is approximately equivalent to @xmath34 , i.e. a negligible contribution to @xmath16 when @xmath35 . the chameleon parameter space depends on the discrete index @xmath6 and two continuous parameters @xmath36 , the coupling to matter , and @xmath37 the suppression scale of the coupling to photons . it is also convenient to introduce the photon coupling parameter @xmath38 . we have already mentioned different experimental and observational bounds on the coupling parameters of chameleons . from a theoretical point of view , very few constraints apply to the parameter space of chameleons . the shape of the potential which we have parameterised using the index @xmath6 is not restricted . we will only invoke a naturality criterion and assume that @xmath6 is not a large number although we are not aware of any deep physical reason for this . when it comes to the coupling constants to matter and photons , we will simply impose that the couplings are greater than one , e.g. corresponding to a scale @xmath39 . again , this type of restriction is only due to the fact that chameleon theories must be viewed as effective field theories only valid below a scale which could be smallest value between @xmath40 and @xmath37 . of course , as the high energy description of chameleon theories valid up to very large scales , maybe the planck scale , is not known , we can not be too strict with this type of argument . on the other hand , when the chameleons are conformally coupled to matter fields , e.g. fermions , and that one studies the low energy physics obtained after integrating out the heavy fields , the coupling constants @xmath7 and @xmath8 can be related@xcite @xmath41 where @xmath42 is the fine structure constant . in this case , their relationship depends on the total number of fermions @xmath43 , due to a conformal anomaly , and the number of heavy fermions @xmath44 . this would naturally lead to a hierarchy between @xmath45 when a large number of fermions is present . in practice and in the absence of knowledge about the number of fermions in the theory at high energy and whether chameleons are conformally coupled in this high energy regime , we have chosen to use @xmath7 and @xmath8 as independent parameters . chameleons can be produced during the propagation of photons inside the macroscopic magnetic fields of the sun . these chameleons would then escape from the sun and reach the earth , where they could be detected by helioscopes such as cast . indeed chameleons interact with matter with a coupling constant depending on the fermion mass and the combination @xmath46 . interactions are therefore stronger with protons than with electrons in the solar plasma . the interaction rate with a fermions can be estimated as @xmath47 taking the proton rate as the dominant part and using electroneutrality , we find that the mean free path @xmath48 reads @xmath49 where @xmath16 is the solar density and @xmath50 the proton mass . the mean free path is shorter in dense regions and reads @xmath51 which far exceeds the size of the sun . hence solar chameleons escape through the solar plasma unscathed . the only possible interaction of chameleons is with photons in a magnetised region . the mixing of photons and chameleons depends on @xmath52 where the effective mass is ( see ( [ mass ] ) ) @xmath53 and @xmath54 is the initial frequency of the incoming photons . this depends on the mixing angle which is given by @xmath55 and the plasma frequency which is @xmath56 where @xmath42 is the fine structure constant . for instance , when @xmath57 and @xmath32 , the magnetic contribution to the mass for @xmath58 t is @xmath59 times smaller than the plasma frequency . electro - neutrality implies that , in the sun , @xmath60 , where @xmath50 is the proton mass . when @xmath61 : as is the case in most helioscope pipes and also inside the sun , we have @xmath62 chameleon production in a magnetised region of size @xmath63 is obtained from the transition probability @xmath64 where @xmath65 . for instance with @xmath66 and @xmath32 , the dependence of this conversion probability on the photon energy @xmath54 can be seen on figure 6 where typical values of order @xmath67 are obtained . as a function of the matter coupling constant @xmath7 for @xmath19 and @xmath20 . notice that only a limited range of @xmath7 can lead to a resonance in the sun . nuclear physics constraints always impose @xmath17@xcite.,title="fig:",width=264 ] as a function of the matter coupling constant @xmath7 for @xmath19 and @xmath20 . notice that only a limited range of @xmath7 can lead to a resonance in the sun . nuclear physics constraints always impose @xmath17@xcite.,title="fig:",width=264 ] it is important to notice that chameleon production is only possible when @xmath68 ( see ( [ k ] ) ) for small values of @xmath69 . we will find that for large values of @xmath7 , this condition is hard to meet in the sun . neglecting the magnetic contribution to the mass and @xmath32 , the magnetic contribution to the energy density is @xmath70 times smaller than the matter density close to the tachocline and only around a few percent in the sparsest regions of the sun s atmosphere . ] , we have the following expression for the effective mass ( [ eff ] ) @xmath71 where we have introduced the density dependent frequency @xmath72 for large enough @xmath7 , the effective mass increases like @xmath73 . this implies that low energy chameleons are attenuated in the sun . chameleons only propagate when @xmath74 let us consider the case of chameleon production in a narrow magnetic region , say around the tachocline @xmath75 . as @xmath76 decreases with decreasing @xmath16 , once a chameleon of energy @xmath54 has been created , it can propagate all the way to the outer sun . denoting by @xmath77 the density in the production region and @xmath78 , a bound on the range of allowed couplings is @xmath79 for a given @xmath7 , we then see that energetic chameleons can always satisfy this bound while chameleons of lower energy are attenuated in the sun . on the other hand if we are interested in back - converted photons from chameleons in a magnetic helioscope with an energy range , e.g. @xmath80 , we find that chameleons can only be produced if @xmath7 is smaller than an upper bound @xmath81 . the value of @xmath81 depends on the density in the production region and the index @xmath6 . for instance , for @xmath20 and a production around the tachocline , we find that @xmath82 for the propagation of chameleons in the detectable energy range @xmath83 . this is a very wide range of couplings which is compatible with the bound from the neutron energy levels in the terrestrial gravitational field . we will estimate the production of chameleons in magnetic regions in the sun . we will focus on a rather conservative assumption , where a constant magnetic field is present in a small shell around the tachocline about 0.7 @xmath11 ( the width of the tachocline is around @xmath84 with a magnetic field in the 20 - 50 t range ) . in this region , the production can be of two types . the first one is due to the presence of a resonance where the effective mass vanishes . this leads to the creation of chameleons over a narrow radial strip of extremely small width . the second type is the non - resonant production over the whole width of the magnetic region . once created , chameleons interact very little with matter and can only occasionally be back - converted to photons . if this happens deep inside the sun , this represents a tiny modification of the solar radiative transfer , but it can be safely neglected . if this chameleon to photon back - conversion takes place in the photosphere and above where the photon mean free path is large , the resulting back converted photons would escape from the sun and have a spectrum which could contribute to the sun s soft x - ray luminosity . interestingly , the sun s x - ray luminosity far exceeds the expected one for a black body at t= 5780 k. this excess radiation reflects the puzzling corona problem , first noticed in 1939@xcite . remarkably , the corona is indeed hotter above magnetised regions like sunspots . chameleons leaving the sun arrive unhindered in the earth s atmosphere where they penetrate due to their high energy . they also go through matter and can be back converted into photons in magnetic helioscopes . we will describe the production mechanisms in the next section . let us consider the resonant case when the effective mass ( [ eff ] ) vanishes somewhere inside the sun@xcite . notice that the actual rest mass of the chameleon does not vanish at the resonance . of course , such a resonant effect is only of interest for this work if it occurs in a magnetised region of the sun . neglecting as always the magnetic contribution to the mass we have @xmath85 where we have introduced @xmath86 the resonance happens in the sun when @xmath87 . this depends on @xmath7 and @xmath6 . at the resonance we have @xmath88 and @xmath89 . in the vicinity of the resonance , the mixing angle is given by @xmath90 where @xmath91 is the magnetic field at the resonance , say the assumed @xmath30 field in the tachocline . the width of the resonance @xmath92 is obtained when @xmath93 . to linear order this implies that @xmath94 depending on the derivative @xmath95 at the resonance . as long as @xmath96 the production probability through one resonance is @xmath97 which can be conveniently written as @xmath98 where @xmath99 leading to a resonance as a function of the radial distance of the resonance ( in units of the solar radius ) when @xmath19 and @xmath20 . nuclear physics constraints impose @xmath17@xcite.,title="fig:",width=264 ] leading to a resonance as a function of the radial distance of the resonance ( in units of the solar radius ) when @xmath19 and @xmath20 . nuclear physics constraints impose @xmath17@xcite.,title="fig:",width=264 ] thermal photons in the sun perform a random walk . in one second , photons scatter @xmath100 times on average . moreover , in one second , a photon covers a distance @xmath101 on average . as a result , the number of times that one photon goes through the resonance in one second is @xmath102 . this is the enhancement factor in the resonant case @xmath103 where @xmath104 is the speed of light . the effective probability is then @xmath105 the chameleon flux is @xmath106 where @xmath107 at the tachocline ( @xmath108 ) is the photon flux at the resonance and @xmath109 the photon spectrum @xmath110 where the average number of photons at temperature @xmath111 is @xmath112 in the following we will assume that the photon flux is conserved and write @xmath113 for the photon flux through a sphere of radius @xmath114 . as a function of @xmath115 for @xmath116 and @xmath117 when a magnetic field is assumed to exist in a shell of width @xmath118 around the tachocline . the width scales like @xmath8 ( see ( [ wi ] ) ) . here we have chosen @xmath8 which corresponds to the saturation of the solar energy loss bound by escaping chameleons . notice that the width is of the order of the photon mean free path @xmath119 cm here and much shorter than the width of the shell , i.e. the width of the shell does not influence the resonance . for each value of @xmath7 the location of the resonance can be deduced from figure 2 . ] chameleons can also be created in the solar magnetic field when no resonance is present . in the non - resonant case , the conversion probability in the sun is obtained from ( [ prob ] ) @xmath120 as long as @xmath121 which is always true in the tachocline for a magnetic field of order 30 t and photon energies less than say 2 kev . on the other hand the behaviour of @xmath122 inside the sun is highly dependent on the value of the coupling to matter @xmath7 . for @xmath123 , we can simply write the production probability over a distance @xmath124 as @xmath125 for higher values of @xmath122 , the production probability oscillates and we can write @xmath126 where we have introduced the coherence length @xmath127 another interesting phenomenon is that the production of chameleons is only allowed when @xmath128 . this becomes possible in smaller and smaller regions around the tachocline as @xmath7 increases . eventually , when @xmath7 is large enough , no chameleons can be produced in the sun . let us come back to the production mechanism in the sun when photons go through a magnetic region of size much larger than the photon mean free path . let us consider all the possible photon paths in this region . photons perform a random walk with many changes of direction . let us consider one such photon which covers a radial distance @xmath129 in one second . the number of scatterings along this trajectory is @xmath130 where @xmath124 is the typical distance between two scatterings and @xmath131 the distance @xmath124 is random and distributed according to a poisson law @xmath132 corresponding to the absence of scattering over a radial length @xmath124 . notice that the average value of @xmath124 is given by the photon mean free path @xmath133 . we will see that , when the coherence length of photons @xmath134 is small compared to the photon mean free path , the conversion probability is dominated by paths with scatterings after a length @xmath135 . on the other hand , when @xmath136 is larger than the photon mean free path , the conversion probability is dominated by paths with @xmath137 . consider the radial region of size @xmath138 which contains the region of size @xmath129 which is the radial distance covered by the photon in its random walk in one second and where both the mean free path length and the magnetic field remain unchanged . there are @xmath139 such patches in a single region of size @xmath138 and @xmath140 regions over the entire sun , in principle . for a given length @xmath124 , the probability of conversion in a region of size @xmath138 is @xmath141 where photons perform independent random walks with scatterings following a poissonian distribution @xmath132 . the total conversion probability in the sun is @xmath142 when @xmath143 and @xmath144 as it is the case in reality , the conversion probability becomes @xmath145 where @xmath146 is the rescaled radial distance and @xmath147 with@xmath148 and the effective mass depends on @xmath114 . typically , we will find that this conversion probability can be as large as @xmath149 . using @xmath150 we find that @xmath151 when @xmath136 is much larger than the mean free path , the probability is dominated by paths of order @xmath137 and @xmath152 when the coherence length @xmath136 is smaller than the mean free path , @xmath153 , the conversion probability is given by @xmath154 of course this is only realistic if the coherence length is large enough that the presence of the solar plasma can be neglected . typically , we will find cases where the coherence length can be as low as @xmath155 ( @xmath156= mean free path of the thermal photons ) . in the following we will cut the probability distribution below the scale @xmath157 . when the coherence length drops below the plasma scale @xmath158 , we average over the oscillatory behaviour of the conversion probability over a length of the order of the mean free path and we get @xmath159 the differential flux of emitted chameleons is then @xmath160 this flux of chameleons escape from the sun and reaches the earth where they can be back - converted to photons inside a magnetic helioscope like cast and sumico . a small fraction of these chameleons can also be back - converted in the magnetised photosphere and above . a strong constraint on chameleons arises from the requirement that the energy carried away by chameleons does not affect the life time of the sun . in practice , we impose that chameleons can not carry more than 10 percent of the total luminosity of the sun , estimated to be @xmath161 @xcite , which then would not affect the evolution of the sun over its lifetime . the total luminosity loss in chameleons is given by @xmath162 where the chameleon flux must include both the resonant and non - resonant contributions . in the numerical examples which will be presented , the energy loss bound is satisfied . let us consider a magnetic helioscope like cast at the surface of the earth . some of the incoming chameleons emerging from the sun will be back - converted provided their energy is higher than their mass in the atmosphere and in matter in front of the helioscope entrance . let us first focus on the resonant case . the number of back - converted photons is @xmath163 where d is the earth - sun distance ( 1 au ) , @xmath164 the aperture and the conversion probability @xmath165 simplifies when the coherence length in the helioscope is larger than the magnetic pipe length @xmath166 where @xmath63 is the helioscope pipe length . integrating over the energy , this gives the total number of back - converted photons from solar chameleons entering the helioscope pipe : @xmath167 and @xmath168 with @xmath169 the riemann @xmath170 function and @xmath171 the solar temperature at the resonance , furthermore : @xmath172 where the bounds of the integral depend on the detector energy range . in the following we will take @xmath173 and @xmath174 corresponding to the present cast detectors . level above the red region for a magnetic helioscope like cast in vacuum at a temperature of 1.8k and for the photon coupling @xmath175 as a function of the matter coupling @xmath176 . we have assumed that there is a constant magnetic field of 30 t in a shell of @xmath118 surrounding the tachocline . the precise form of the diagram depends on @xmath6 although the overall shape remains the same . notice that for values of @xmath7 below the resonance values , where the dip occurs , the 2@xmath9 contour is constant . in this work , we have chosen this value as a template for the photon coupling . larger values of the photon couplings would violate the solar energy loss bound due to escaping chameleons with @xmath177 . as such , the solar model we have chosen for the magnetised region , where chameleons are produced , is almost optimal . notice that the value @xmath178 would only lead to interesting physics at the 2@xmath9 level for very large values of the coupling to photons , higher than the chase bound . ] in the non - resonant case ( section iii - b ) , the number of regenerated photons in the helioscope pipe emitted from a non - resonant region is obtained by multiplying the production flux by the conversion probability in the helioscope pipe : @xmath179 when the coherence length in the helioscope pipe is larger than @xmath63 , we find that @xmath180 and upon integrating over the sun radial distance @xmath181 , we find that @xmath182 the overall dependence on the photon coupling is @xmath183 as @xmath184 varies with @xmath185 . the dependence on the matter coupling enters essentially in the effective mass inside the sun and therefore in the coherence length in the sun . . the mass is constant and independent for low values of @xmath186 , and becomes @xmath7 dependent for larger values . the constant value is obtained from ( [ ma ] ) and the @xmath7 dependence from ( [ mass ] ) . ] as a summary , we have found that a resonance can always generate chameleons for values of the matter coupling @xmath7 in a narrow range for a given value of the index @xmath6 . on the other hand , non - resonance production is generic and happens for the large range of allowed values @xmath82 . it should also be noticed that for values of @xmath7 larger than a @xmath6-dependent value , no chameleon can be produced in the sun . this value is typically around @xmath187 for @xmath20 implying that chameleons can always be non - resonantly produces for @xmath20 and @xmath82 . in the following , we will concentrate on a single value of @xmath31 where no resonance is present and analyse the properties of the non - resonant production . and the photon coupling @xmath116 . the magnetised region is such that @xmath188 . notice the parabolic shape at low energy followed by a sinusoidal trend which would saturate at much higher energies ( not shown here ) . ] before leaving the sun , a tiny fraction of the produced chameleons can be back - converted into photons while propagating through the magnetised photosphere and the other layers of the solar atmosphere . we will be interested in three types of conversion processes . the first one would happen in the photosphere below the solar surface . inside the sun , the mean free path of photons is the maximal extension over which a back - conversion can occur . indeed , on larger distances , photons are scattered and can not mix with the incoming chameleonic wave function . in the photosphere , the magnetic field is of order 0.2 t. above the solar surface , we envisage two situations . the first one corresponds to the quiet sun with a small magnetic field of order 1 gauss distributed over large regions satisfying the empirical law@xcite @xmath189 we also consider flares where the magnetic fields are much larger while the size of the magnetic regions is smaller although still following the same empirical law ( [ emp ] ) . the number of back - converted photons is given by @xmath190 when the coherence length ( in the photosphere ) is much larger than the size of the magnetic region . in the photosphere where the density is relatively high compared to the atmosphere , the coherence length is relatively short and one must average out over the distribution of photon excursions between two scatterings . this implies , using ( [ prob ] ) and the averaged value @xmath191 , that the number of back - converted photons is @xmath192 where @xmath193 is reduced by the larger effective mass in denser regions , such as inside the photosphere compared to the corona . indeed , we find that the back - conversion of outstreaming chameleons from the inner sun is largely suppressed in the photosphere where the density is higher than in the upper atmosphere with a much smaller plasma density . as a result we find that the back - conversion of photons from outstreaming chameleons is mainly taking place in magnetised regions above the chromosphere . as the back - converted photons have a spectrum in the soft x - ray band , this leads to an excess of solar x - ray luminosity given by @xmath194 it is shown below that this radiation excess is tiny compared to the total solar luminosity . however , the solar luminosity in the soft x - ray band due to such back - converted photons for the quiet sun or the active sun is of interest as it could explain the unaccounted part of the solar photon spectrum ( solar corona problem ) . , the photon coupling @xmath195 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline for the resonant case ( left ) and for the non resonant case ( right ) with @xmath31 and @xmath32 . , title="fig : " ] , the photon coupling @xmath195 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline for the resonant case ( left ) and for the non resonant case ( right ) with @xmath31 and @xmath32 . , title="fig : " ] it is therefore relevant to give an order of magnitude estimate of the predicted x - ray intensity from back - converted solar chameleons and compare it with observation . since the chameleon effective mass is low in sparse regions , it decreases outside the sun when chameleons propagate in empty space . moreover the coherence length @xmath136 increases with @xmath196 and the conversion probability with @xmath197 . in the weak but extended magnetic fields @xmath198 which is present all the way to the vicinity of the earth and in particular the orbiting x - ray observatories , the chameleon - to - photon conversion is possible . as an illustration , in figure 6 we see that the chameleon - to - photon conversion efficiency is greater than @xmath199 10@xmath200 in a cast - like pipe for bl=90 tm . the estimated integral of the inter planetary magnetic field @xmath201 is about @xmath202 to @xmath203tm and occasionally even larger @xcite . the conversion probability becomes then 100 times to @xmath204 times larger than the one estimated for a cast - like magnet in vacuum ( see figure 6 ) . taking into account that the maximal allowed solar luminosity in any invisible channel is about 10% of the total solar luminosity , then , a rather conservative estimate of the expected excess x - ray observation at 1 au , is about @xmath205 to @xmath206erg / s . surprisingly , this fits observations ( see ref . @xcite and references therein ) . how could the chameleon origin of this excess be established observationally in a non - equivocal manner ? in fact , the following two measurements can be performed : 1 ) the measurement of the solar x - ray ray spectrum in the energy range below @xmath207 2 kev , which has been done in the recent solar minimum @xcite , and 2 ) the two dimensional reconstruction of the quiet sun x - ray spectrum in the same energy range . if the tachocline were the source of the outstreaming chameleons which would back - convert to soft x - rays on their journey to earth through the magnetic field ( only its transverse component matters ) , the derived two dimensional plot would have a characteristic ring - like structure at about 0.7@xmath208 . this would be a clear signature of the creation of chameleons or other similar exotica around the tachocline . further studies , following the suggestions of this work , would allow to localize the involved processes . moreover , once reconstructed , this would provide a new tool for investigating the solar plasma , either in the interplanetary space or deep inside the sun , i.e. near the tachocline some 200000 km deep inside the sun . this would have the potential to become , along with the solar neutrinos and helioseismology , one of the most precise methods to obtain information from the otherwise perfectly shielded inner sun . of course , laboratory measurements as suggested here ( see section iv ) , or , observations with a magnetic helioscope like cast would provide complimentary information . in fact , cast can increase its sensitivity to solar chameleons by a factor of 10 - 100 in solar chameleon signature , once its detectors can perform with a lower energy threshold , i.e. towards a few 100 ev s ( see also @xcite ) . so far we have presented mechanisms for the creation of chameleons in the sun by the primakoff effect and their back conversion at the surface of the magnetised sun or in magnetic helioscopes by the inverse primakoff effect . of course these results depend on the choice of the solar model for the magnetic region where chameleons are produced . it would be extremely important to have a model - independent probe of the existence of chameleons in the x - ray domain . this would also complement the chase experiment which has been carried out in the ev energy range and might suffer from uncertainties related to multiple reflections of chameleons at the beamtube walls . we suggest an experiment where a powerful source of x - rays shines through a , say , cast - like magnetic pipe , while another identical pipe lies downstream on the other side of a x - ray absorbing thick barrier . of course , on the other side of the barrier no photons should appear conventionally . now the possibility of creating energetic chameleons in the first magnet which would go downstream through the thick barrier allows the regeneration of photons on the other side of the barrier , that by the time - reversed process . we consider that both magnetic pipes on each side are identical . in practice we can have @xmath209 at most so far . the spectrum of regenerated photons is @xmath210 where @xmath211 in the helioscope pipe . we have taken the source spectrum to be @xmath212 such that @xmath213 and @xmath164 is the aperture of the pipe . upon integration this spectrum , we obtain the total number of photons per second @xmath214 when this number is higher or at least not much smaller than the background level noise of the detectors in the appropriate energy range , this gives a new way of searching for chameleons , independently of any astrophysical modeling . of course axions with a similar mass and photon coupling as chameleons would also go through the dense barrier and regenerate x - rays with a resulting spectrum very similar to the chameleonic case . so how can one distinguish chameleons from axions ? in fact , one can use one of the fundamental properties of chameleons : the dependence of their mass on the density in the pipe , while qcd - inspired axions have a fixed rest mass . indeed as we have already exemplified in figure 8 , when the density of the gas filling the pipe is increased , the effective mass of the chameleons becomes also larger and the conversion probability oscillates and is suppressed by @xmath215 , hence a quartic signal suppression in the chameleon through wall experiment . as a result , for a high enough pressure , the number of regenerated photons would become essentially zero . this is a distinguishing property of chameleons . indeed , one could envisage to perform the experiment in vacuum and then with an optically transparent gas in the pipe at a relatively high pressure , exploiting the different dependencies of the chameleon and alp production on the gas pressure in the beam pipe . in fact , a cast pressure of @xmath216 1 mbar is transparent to x - rays and should sufficient to realise the identification of chameleons . a similar technique will be applied for the detection of the chameleon force between metallic plates : by comparing the force in vacuum and with a dense material between the plates , one may detect the presence of chameleons . here the same type of technique may give a direct indication of the existence of chameleons in the kev energy range . for a given @xmath6 , the resonance inside the sun depends on @xmath7 . we find that values of @xmath7 which lead to a resonance are quite limited . outside this small interval , no resonance is present . in figure 1 , we have plotted the density at the resonance and in figure 2 the solar radius of the resonance as a function of @xmath7 for @xmath19 and @xmath20 . typically the resonance exists for @xmath217 for @xmath20 . for large values , @xmath218 , no chameleon can be created inside the sun when we focus on a production region of width 0.01 solar radius around the tachocline at @xmath219 . larger values of @xmath7 are already excluded by nuclear physics measurements of the neutron energy levels in the terrestrial gravitational field@xcite . different values of @xmath6 give almost the same type of results . when it comes to the back - conversion of the emitted chameleons in a helioscope , we concentrate on cast - like conditions . we take into account 100 hours of observation with a detection noise of 1 photons per hour in the 1 - 7 kev range when the vacuum in the pipe is such that the pressure is about @xmath220 mbar . we have taken the pipe length to be 9.26 m , the magnetic field 9 t and the diameter of the pipe to be 43 mm . in the sun , we have considered a constant magnetic field of 30 t above the tachocline in a shell of width @xmath118 . in this case , the width of the resonance as a function of @xmath7 is given in fig . it is of the same order of magnitude as the thermal photon mean free path in the tachocline , i.e. around 0.3 cm . when observations are carried out during h hours , the signal to noise ratio is @xmath221 where the noise level of the helioscope is taken to be @xmath222 per hour as a generic example . we have given in fig.4 a sensitivity plot at the 2@xmath9 level for a 100 hour measurement in vacuum . as can be seen , for values of @xmath7 below the resonance range , the sensitivity region is uniform and the matter coupling must be @xmath223 which is to be compared with the chase bound @xmath224 . outside the resonance values , the coherence length in the sun decreases rapidly and the production probability oscillates very fast leading to a diminishing sensitivity . we note that inside a helioscope pipe in vacuum , the rest mass of the chameleon is determined by the radius @xmath225 of the pipe and is density independent@xcite @xmath226 this is valid at low values of @xmath7 . for higher values , the density dependent rest mass becomes the chameleon mass ( [ mass ] ) , see figure 5 . .the values of the density , photon mean free path and magnetic field as a function of the solar altitude from the solar surface . above an altitude of 1000 km , the considered cases correspond to dynamically changing solar flares . [ cols="^,^,^,^",options="header " , ] we have studied the resonant and non - resonant spectra obtained after back - conversion in the helioscope pipe using for the photon coupling @xmath32 and matter couplings ( @xmath31 , @xmath227 ) in the non - resonant and the resonant case respectively . as can be seen in fig . 7 , the number of resonantly back - converted photons is lower than the non - resonant ones . the width of the resonant region in @xmath7 is so narrow ( see figure 3 ) that one may ignore the resonant case and therefore we focus on the non - resonant one which can occur for all values of @xmath82 . an important consequence of the non - resonant production is that the resulting x - ray spectra appear in sub - kev energies with a peak around @xmath228 600 ev . this is below the present sensitivity of all the magnetic helioscope x - ray detectors . with a better sensitivity in a lower energy range , the potential to detect chameleons would be enormously enhanced . this is an exciting and challenging experimental possibility . we have used the present sensitivity levels of magnetised helioscopes to deduce that values of the coupling constant to photons @xmath229 should be within experimental reach . with an improved sensitivity for detectors in the sub - kev region , we expect that lower values of the photon coupling constants should lead to a detectable number of detected x - ray photons . of course , if the coupling to photons @xmath8 is extremely low , it is unlikely that any signal could be detected . the analysis of the range of @xmath8 with sub - kev detectors is left for future work . we have also investigated what happens when the pressure in the helioscope pipe is not vacuum . as shown in figure 8 , the number of photons decreases as the pressure increases . this is due to the fact that the chameleon mass increases with the pressure . when the coherence length is smaller than the effective length of the helioscope , the conversion probability decreases as @xmath215 implying a decrease of the overall amplitude . moreover , rapid oscillations are present for larger pressures . this is the dependence that we suggest to utilise for ( solar ) chameleon identification . we are also interested in the chameleon spectrum on earth . the number of chameleons per second and @xmath230 has been plotted in fig . a crucial test of our model is whether the luminosity carried by the chameleons is below the energy loss bound for any unknown type of escaping solar radiation which is 10 % of the total solar luminosity@xcite . in figure 9 , we have plotted the analogous spectrum of the escaping chameleons which corresponds to @xmath231 10 % of the total solar luminosity . the saturating bound is @xmath232 as can be seen , the solar model with a magnetised region of width @xmath118 at the tachocline and a magnetic field of 30 t is quite realistic . , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius near the tachocline.,title="fig:",width=264 ] , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius near the tachocline.,title="fig:",width=264 ] + , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius near the tachocline.,title="fig:",width=264 ] , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius near the tachocline.,title="fig:",width=264 ] at 1 au as a function of the chameleon energy in kev . here we have chosen the matter coupling to be @xmath234 , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline . on the right , the chameleon brightness in @xmath235 at the surface of the sun as a function of the energy in kev . the estimated total energy loss in chameleons is @xmath236 @xmath237 , i.e. around 10 % of the photon solar luminosity.,title="fig:",width=264 ] at 1 au as a function of the chameleon energy in kev . here we have chosen the matter coupling to be @xmath234 , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline . on the right , the chameleon brightness in @xmath235 at the surface of the sun as a function of the energy in kev . the estimated total energy loss in chameleons is @xmath236 @xmath237 , i.e. around 10 % of the photon solar luminosity.,title="fig:",width=264 ] + at 1 au as a function of the chameleon energy in kev . here we have chosen the matter coupling to be @xmath234 , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline . on the right , the chameleon brightness in @xmath235 at the surface of the sun as a function of the energy in kev . the estimated total energy loss in chameleons is @xmath236 @xmath237 , i.e. around 10 % of the photon solar luminosity.,title="fig:",width=264 ] at 1 au as a function of the chameleon energy in kev . here we have chosen the matter coupling to be @xmath234 , the photon coupling @xmath233 and a shell magnetic field in the sun of 30 t over a width of 0.01 solar radius above the tachocline . on the right , the chameleon brightness in @xmath235 at the surface of the sun as a function of the energy in kev . the estimated total energy loss in chameleons is @xmath236 @xmath237 , i.e. around 10 % of the photon solar luminosity.,title="fig:",width=264 ] the conversion probability in the helioscope pipe for a small mixing angle @xmath238 is @xmath239 where @xmath240 depends on the effective chameleon mass in the helioscope pipe . when @xmath241 , the probability is constant and given by @xmath242 . as @xmath243 increases and as soon as the oscillations due to @xmath244 appear , the probability becomes proportional to @xmath245 which explains why the envelope of the probability distribution is parabolic for very small values of @xmath54 and resembles a sinusoid for larger values before reaching an asymptotic value ( not shown ) . we are also very interested in the number of chameleons which can be back - converted to photons in the magnetised photosphere and above . we have calculated the corresponding photon spectra in the photosphere , for the quiet sun outside the photosphere and for flares . as can be seen in figure 10 , the back - converted photon luminosity in the photosphere is negligible . for flaring conditions at relatively low altitudes , it is given in figures 10 and 11 . in these cases , the coherence length is small and an average over the size of the magnetic regions has been taken . for flaring conditions at altitudes higher than 10000 km , the back - converted x - ray luminosity can be rather large . a very interesting result is obtained for the quiet sun since the back - converted photon spectrum peaks at @xmath207 1 kev which could explain why the solar corona is even hotter above active regions , see figure 12 . ) leaving the sun as a function of the photon energy . the photons are backconverted at -500 km , -100 km , 0 km and + 500 km with magnetic fields 0.2 t in the first three cases and 0.1 t in the last case . in the first three cases , the conversion takes place over one mean free path of order 15 km , 40 km and 50 km while in the last case the conversion takes place over the size of the magnetic region around 100 km . in each case , the coherence length is much smaller than the size of the chameleon to photon interaction region and the spectrum has been averaged out over many scatterings . here we have chosen the matter and photon coupling to be @xmath234 and @xmath32 respectively.,title="fig:",width=264 ] ) leaving the sun as a function of the photon energy . the photons are backconverted at -500 km , -100 km , 0 km and + 500 km with magnetic fields 0.2 t in the first three cases and 0.1 t in the last case . in the first three cases , the conversion takes place over one mean free path of order 15 km , 40 km and 50 km while in the last case the conversion takes place over the size of the magnetic region around 100 km . in each case , the coherence length is much smaller than the size of the chameleon to photon interaction region and the spectrum has been averaged out over many scatterings . here we have chosen the matter and photon coupling to be @xmath234 and @xmath32 respectively.,title="fig:",width=264 ] + ) leaving the sun as a function of the photon energy . the photons are backconverted at -500 km , -100 km , 0 km and + 500 km with magnetic fields 0.2 t in the first three cases and 0.1 t in the last case . in the first three cases , the conversion takes place over one mean free path of order 15 km , 40 km and 50 km while in the last case the conversion takes place over the size of the magnetic region around 100 km . in each case , the coherence length is much smaller than the size of the chameleon to photon interaction region and the spectrum has been averaged out over many scatterings . here we have chosen the matter and photon coupling to be @xmath234 and @xmath32 respectively.,title="fig:",width=264 ] ) leaving the sun as a function of the photon energy . the photons are backconverted at -500 km , -100 km , 0 km and + 500 km with magnetic fields 0.2 t in the first three cases and 0.1 t in the last case . in the first three cases , the conversion takes place over one mean free path of order 15 km , 40 km and 50 km while in the last case the conversion takes place over the size of the magnetic region around 100 km . in each case , the coherence length is much smaller than the size of the chameleon to photon interaction region and the spectrum has been averaged out over many scatterings . here we have chosen the matter and photon coupling to be @xmath234 and @xmath32 respectively.,title="fig:",width=264 ] ) in solar flares leaving the sun as a function of the photon energy in kev . the photons are backconverted at + 1000 km , + 2000 km , + 10000 km and + 80000 km with magnetic fields 100 gauss , 10 gauss , 10 gauss and 5 gauss respectively . at the altitude of 1000 km , the spectrum has been averaged out over many oscillations . in the other two cases , we have represented the spectrum itself . in each case we have used @xmath246 which determines the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] ) in solar flares leaving the sun as a function of the photon energy in kev . the photons are backconverted at + 1000 km , + 2000 km , + 10000 km and + 80000 km with magnetic fields 100 gauss , 10 gauss , 10 gauss and 5 gauss respectively . at the altitude of 1000 km , the spectrum has been averaged out over many oscillations . in the other two cases , we have represented the spectrum itself . in each case we have used @xmath246 which determines the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] + ) in solar flares leaving the sun as a function of the photon energy in kev . the photons are backconverted at + 1000 km , + 2000 km , + 10000 km and + 80000 km with magnetic fields 100 gauss , 10 gauss , 10 gauss and 5 gauss respectively . at the altitude of 1000 km , the spectrum has been averaged out over many oscillations . in the other two cases , we have represented the spectrum itself . in each case we have used @xmath246 which determines the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] ) in solar flares leaving the sun as a function of the photon energy in kev . the photons are backconverted at + 1000 km , + 2000 km , + 10000 km and + 80000 km with magnetic fields 100 gauss , 10 gauss , 10 gauss and 5 gauss respectively . at the altitude of 1000 km , the spectrum has been averaged out over many oscillations . in the other two cases , we have represented the spectrum itself . in each case we have used @xmath246 which determines the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] ( linear scale on the left and logarithmic scale on the right ) . the photons are backconverted at altitudes larger than 5000 km with a magnetic field of 1 gauss and a density of order @xmath247 . we have used @xmath246 to determine the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] ( linear scale on the left and logarithmic scale on the right ) . the photons are backconverted at altitudes larger than 5000 km with a magnetic field of 1 gauss and a density of order @xmath247 . we have used @xmath246 to determine the length of the magnetic region . here we have chosen the matter coupling to be @xmath234 and the photon coupling is @xmath32 . , title="fig:",width=264 ] the spectrum intensity of regenerated photons depends crucially on the source flux in the x - ray band . we have chosen as a template the desy spectrum@xcite with a total number of @xmath248 . an upgraded version with a spectrum of @xmath249 could be assumed for a future dedicated beam line . we will quote results obtained with this flux . the detection of a low flux of regenerated photon depends also on the detector sensitivity . we will assume in this work @xmath250 . with these specifications , we estimate the 2@xmath9 detection efficiency . for that , we refer to the case @xmath31 and @xmath251 as fixed by the solar energy loss bound . the original source spectrum and the regenerated photons are presented in figure 13 with @xmath252 . the number of seconds for a n@xmath9 detection is @xmath253 we find that a 2@xmath9 level signal would take 300 days . an increase of the number of photons originating from the source by one or two orders of magnitude would reduce this number to a few days only . however , if three equivalent cast magnetic pipes ( e.g two lhc bending magnets on each side of the barrier ) were used on both sides of the thick barrier , a 5@xmath9 level would take only 6 days , which is an experimental challenge for a chameleon smoking gun . a distinguishing feature of chameleons is the fact that there would be no signal on top of the detection noise when the pipe is filled with a gas at relatively high pressure . a comparison between the vacuum and pressured cases inside the magnetic pipes up to a few mbar only may help to identify the regenerated photon signal in a much better fashion . we have studied the creation of chameleons inside the sun by the primakoff effect and their back - conversion to soft x - ray photons in the magnetised photosphere , the corona and helioscopes on earth by the inverse primakoff effect . the production of chameleons is strongly dependent on the assumed features of the magnetised regions inside the sun . depending on the magnetic field in the sun and the size of the magnetised zone , it is possible to deduce the energy loss of the sun escaping as chameleons and impose the energy loss bound : no more than 10 % of the total solar luminosity can be carried away by unknown species . provided this bound is satisfied , escaping chameleons reach the surface of the earth and may be back - converted into soft x - ray photons in the pipe of a magnetic helioscope . moreover , a tiny fraction of the escaping chameleons could also be converted into photons inside the photosphere or the corona . we have found that for a rather conservative solar model with a widely accepted magnetised region of size @xmath118 around the tachocline , the non - resonant production of chameleons would lead to a 2@xmath9 detection of x - rays in a cast - like helioscope and an energy loss of the sun due to chameleons just below 10% . however , using detectors sensitive to sub - kev photons , this would enormousy improve the detection significance . for the same chameleon parameters , we have proposed a shining through wall experiment where chameleons produced in one or more cast - like pipes from a powerful x - ray source would go through a thick barrier and regenerate into x - rays downstream of the barrier in a second magnetic field . the observation of this phenomenon would be a chameleon smoking gun with clear possible chameleon identification . in summary we have shown that a viable candidate for dark energy , the chameleon , could be sought for in magnetic helioscopes and purely laboratory based experiments . both approaches could be pursued in the near future . the combination of their results could give new insights into yet unexplained phenomena . we have also seen that chameleons could shed new light on the solar corona problem . of course , nothing guarantees that the coupling of chameleons to photons is large enough to lead to a detection of chameleons . if no signal were obtained , new bounds on this coupling would ensue corresponding to higher and higher limits on the suppression scale of the coupling of photons to a scalar field . as a function of the photon energy in kev ( left ) and the regenerated spectrum on the other side of the barrier ( right ) . the integrated flux is @xmath254 for the source spectrum . the spectrum of regenerated photons presented here is the one predicted to be seen by a chameleon through a wall experiment using two magnetic pipes on both sides of a thick barrier . the coupling to matter is @xmath31 while the coupling to photons @xmath255 implies a corresponding 2@xmath9 result obtained in 300 days with a detection sensitivity of @xmath256 . for magnetic pipes of 3 times the cast pipe length , a 5@xmath9 result would only take 6 days.,title="fig:",width=340 ] as a function of the photon energy in kev ( left ) and the regenerated spectrum on the other side of the barrier ( right ) . the integrated flux is @xmath254 for the source spectrum . the spectrum of regenerated photons presented here is the one predicted to be seen by a chameleon through a wall experiment using two magnetic pipes on both sides of a thick barrier . the coupling to matter is @xmath31 while the coupling to photons @xmath255 implies a corresponding 2@xmath9 result obtained in 300 days with a detection sensitivity of @xmath256 . for magnetic pipes of 3 times the cast pipe length , a 5@xmath9 result would only take 6 days.,title="fig:",width=340 ] s. perlmutter _ et al . _ [ supernova cosmology project collaboration ] , astrophys . j. * 517 * , 565 ( 1999 ) [ arxiv : astro - ph/9812133 ] . e. j. copeland , m. sami and s. tsujikawa , int . j. mod . d * 15 * ( 2006 ) 1753 [ arxiv : hep - th/0603057 ] . j. khoury and a. weltman , phys . rev . * 93 * ( 2004 ) 171104 [ arxiv : astro - ph/0309300 ] . j. khoury and a. weltman , phys . d * 69 * , 044026 ( 2004 ) [ arxiv : astro - ph/0309411 ] . brax , c. van de bruck , a - c davis , j. khoury and a. weltman , phys.rev.d70:123518,2004 [ arxiv : astro - ph/0408415 ] d. f. mota and d. j. shaw , phys . * 97 * ( 2006 ) 151102 [ arxiv : hep - ph/0606204 ] . d. f. mota and d. j. shaw , phys . d * 75 * ( 2007 ) 063501 [ arxiv : hep - ph/0608078 ] . k. zioutas et al . ( cast collaboration ) , phys . 94 , 121301 ( 2005 ) ; s. andriamonje et al . ( cast collaboration ) , jcap 0704 , 010 ( 2007 ) , 0912 , 002 ( 2009 ) , 1003 , 032 ( 2010 ) ; e. arik ( cast collaboration ) , jcap 0902 , 008 ( 2009 ) ; s. aune et al . ( cast collaboration ) phys . ( 2011 ) accepted ( hep - ex/1106.3919 ) . s. moriyama et al . b 434 , 147 ( 1998 ) ; y. inoue et al . b 536 , 18 ( 2002 ) ; y. inoue et al . b 668 , 93 ( 2008 ) . k. zioutas , m. tsagri , y. semertzidis , e. georgiopoulou , a. gardikiotis , t. dafni , astro - ph/1112.4329 , to appear in the proc . 7th patras workshop on axions , wimps and wisps , mykonos , greece , 2011 ( http://axion-wimp.desy.de/ ) eprint : arxiv:1112.4329v1 [ astro-ph.sr ] k. zioutas on behalf of cast and d. anastassopoulos , o. baker , m. betz , p. brax , f. caspers , j. jaeckel , a. lindner , y. semertzidis , n. spiliopoulos , s. troitsky , a. vradis , cast physics proposal , cern - spsc-2011 - 018 ; spsc - sr-085 ( march 2011 ) ( http://cdsweb.cern.ch/record/1340994 ) .

we study a model of fast magnetic reconnection in the presence of weak turbulence proposed by using three - dimensional direct numerical simulations . the model has been already successfully tested in confirming the dependencies of the reconnection speed on the turbulence injection power and the injection scale expressed by a constraint and no observed dependency on ohmic resistivity . in , in order to drive turbulence , we injected velocity fluctuations in fourier space with frequencies concentrated around , as described in alvelius ( 1999 ) . in this paper we extend our previous studies by comparing fast magnetic reconnection under different mechanisms of turbulence injection by introducing a new way of turbulence driving . the new method injects velocity or magnetic eddies with a specified amplitude and scale in random locations directly in real space . we provide exact relations between the eddy parameters and turbulent power and injection scale . we performed simulations with new forcing in order to study turbulent power and injection scale dependencies . the results show no discrepancy between models with two different methods of turbulence driving exposing the same scalings in both cases . this is in agreement with the lazarian & vishniac ( 1999 ) predictions . in addition , we performed a series of models with varying viscosity . although lazarian & vishniac ( 1999 ) do not provide any prediction for this dependence , we report a weak relation between the reconnection speed with viscosity , . magnetic fields are observed in many astrophysical objects and usually play an important or even crucial role in their dynamics ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * e.g. ) . they are a key ingredient of astrophysical processes such as magneto - rotational instability , magnetic dynamo , transport and acceleration of cosmic rays , accretion disks , turbulence , solar phenomena , gamma ray bursts , etc . . magnetic fields are solenoidal and evolve only through changes in the curl of the electric field . in the limit of zero resistivity the topology of the field lines is a constant of motion and the magnetic flux threading any fluid element is constant . generating large scale magnetic fields requires some kind of battery effect , like the biermann battery and generating strong large scale magnetic fields requires a dynamo ( see * ? ? ? * for example ) . in the limit of very small resistivity , which is typical for astrophysical objects , the magnetic flux is `` frozen in '' and magnetic field lines will resist passing through one another or changing their topology . due to the presence of plasma motions , in particular turbulence , this would result in a very complex tangle of field lines in real objects , with negligible large scale magnetic flux . however , observations indicate that the mean and turbulent components of magnetic fields in many astrophysical objects are of similar strengths ( see * ? ? ? * for example ) . this implies the existence of a process which can violate the frozen - in condition on dynamical time scales , i.e. fast magnetic field reconnection . the first analytic model for magnetic reconnection was proposed independently by and . sweet - parker reconnection has the virtue that it relies on a robust and straightforward geometry . two regions with uniform magnetic fields are separated by thin current sheet . the speed of reconnection is given roughly by the resistivity divided by the sheet thickness . however , the plasma in the current sheet is constrained to move along the local field lines , and is ejected from the edge of the current sheet at the alfvn speed , . since the width of the current sheet limits the flux of expelled plasma , the overall reconnection speed is reduced from the alfvn speed by the square root of the lundquist number , , where is the resistivity and is the length of the current sheet . in most astrophysical contexts is very large and the sweet - parker reconnection speed , , is negligible . fast reconnection requires that the dependence on be erased . given the simplicity of the sweet - parker setup , this requires that the simple geometry of the current sheet must be broken . the realization that sweet - parker reconnection is inadequate to explain magnetic reconnection in an astrophysical context was immediately apparent , and gave rise to decades of research on models of fast reconnection ( see * ? ? ? * ; * ? ? ? * for reviews ) . the first proposal was to replace the current sheet with an x - point configuration , so that the `` sheet '' thickness and length are comparable . this is the basis for petschek s model of fast reconnection . however , a dynamically self - consistent x - point requires that the outflow prevent a general collapse into a narrow current sheet . otherwise we would expect that the same bulk forces that brought the magnetic field lines together would lead to sweet - parker reconnection . proposed that slow - mode shocks on either side of the x - point would serve this purpose . moreover , those shocks are responsible for converting most of the magnetic energy into the heat and kinetic energy . the x - point in this model has an overall size which depends on resistivity , but since the magnetic field decrease logarithmically when approaching the current sheet ( due to the assumption of the current - free magnetic field in the inflow region ) , the resulting reconnection speed is some fraction of . numerical simulations with uniform resistivity have showed that in the mhd limit the shocks fade away and the contact region expands into a sheet . the only way to make the petschek configuration stable is by introducing the local nonuniform resistivity . this leaves the possibility that x - point reconnection is stable when the plasma is collisionless . numerical simulations have been encouraging . however , there are several important issues that remain unresolved . first , it is not clear that this kind of fast reconnection persists on scales greater than the ion inertial scale ( see * ? ? ? * ) . several numerical studies have found large scale reconnection speeds which depend on resistivity , i.e. are not fast . second , in many circumstances the magnetic field geometry does not allow the formation of x - point reconnection . for example , a saddle - shaped current sheet can not be spontaneously replaced by an x - point . the energy required to do so is comparable to the magnetic energy liberated by reconnection , and must be available beforehand . finally , the requirement that reconnection occur in a collision less plasma restricts this model to a small fraction of astrophysical applications . for example , while reconnection in stellar coronae might be described in this way , stellar chromospheres are not . more generally estimated that the scale of the reconnection sheet should not exceed about 40 times the electron mean free path . this condition is not satisfied in many environments which one might naively consider to be collisionless , among them the interstellar medium . the conclusion that stellar interiors and atmospheres , accretion disks , and the interstellar medium in general does not allow fast reconnection is drastic and unpalatable . an alternative to the x - point geometry is to consider magnetic fields that are chaotic , even if only weakly so . requiring the plasma to flow along the local magnetic field implies a powerful constraint on reconnection only if the field lines themselves are laminar . ( * ? ? ? * hereinafter lv99 ) proposed a model for fast reconnection which depends on the presence of turbulence , and its production of weakly stochastic field lines ( also briefly described in section [ sec : lv99model ] ) . turbulence is a natural consequence of convection in stars and of the magnetorotational instability in accretion disks ( for a review see * ? ? ? * ) . in addition , it is now generally accepted that the `` big power law in the sky '' indicates the presence of turbulence on scales from tens of parsecs to thousands of kilometers . among other sources , evidence for this comes from studies of atomic hydrogen spectra in molecular clouds and galaxies ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * see also review by , and references therein ) , as well as recent studies of emission lines and faraday rotation ( see * ? ? ? * ; * ? ? ? * ) . lv99 s model uses the properties of turbulence to predict broad outflows from extended current sheets . the diffusivity of magnetic field line trajectories in a turbulent plasma implies that flows can follow local magnetic field lines without being confined to the current sheet . when the turbulent diffusivity is less than the ohmic resistivity , this model reduces to the sweet - parker reconnection model . the first test of the lv99 model using three - dimensional ( 3d ) mhd simulations was performed in . the main predictions of the model were confirmed . in this paper we provide additional numerical evidence of magnetic reconnection in turbulent environments by testing different mechanisms for injecting turbulence . in [ sec : lv99model ] we briefly review the lv99 model of reconnection and its theoretical predictions . in [ sec : setup ] we describe in details the numerical model studied in this paper and new method of turbulence driving . although the initial setup and boundary conditions are similar to our previous studies , and described in details in , we briefly describe them here for completeness , as well . in [ sec : results ] we present an extensive description of new results obtained from studying our numeric model , which we discuss later in [ sec : discussion ] . in [ sec : summary ] we present our main conclusions .

1203.2971

the notion that turbulence can influence reconnection rate is not unprecedented . the ideas in this regard were discussed long before lv99 . however , they fell short of solving the problem . for instance , @xcite considered the effects of turbulence on microscopic resistivity , @xcite proposed that the current diffusivity should be modified to include the diffusion of electrons across the large scale magnetic field due to the small scale field line stochasticity . the consequent modifications to the ohmic resistivity have only a marginal effect on the sweet - parker reconnection speeds . @xcite studied 2d magnetic reconnection in the presence of external turbulence both theoretically and numerically . they pointed out various turbulence mechanisms that would enhance reconnection rates , including multiple x - points as reconnection sites . however , this work did not include the effect of magnetic field wandering , which is at the core of the lv99 . they did not provide analytical predictions of the reconnection speed either . , which is set by ohmic diffusivity . the length of the current sheet is a macroscopic scale @xmath13 . magnetic field lines are assumed to be laminar . _ middle plot _ : reconnection in the presence of stochastic magnetic field lines . the stochasticity introduced by turbulence is weak and the mean field is clear direction . the outflow width is set by the diffusion of the magnetic field lines , which is a macroscopic process , independent of resistivity . _ low plot _ : an individual small scale reconnection region . the reconnection over small patches of magnetic field determines the local reconnection rate . the global reconnection rate is substantially larger as many independent patches reconnect simultaneously . conservatively , the lv99 model assumes that the small scale events happen at a slow sweet - parker rate . following @xcite and @xcite . [ fig : lv99model],title="fig:",scaledwidth=45.0% ] , which is set by ohmic diffusivity . the length of the current sheet is a macroscopic scale @xmath13 . magnetic field lines are assumed to be laminar . _ middle plot _ : reconnection in the presence of stochastic magnetic field lines . the stochasticity introduced by turbulence is weak and the mean field is clear direction . the outflow width is set by the diffusion of the magnetic field lines , which is a macroscopic process , independent of resistivity . _ low plot _ : an individual small scale reconnection region . the reconnection over small patches of magnetic field determines the local reconnection rate . the global reconnection rate is substantially larger as many independent patches reconnect simultaneously . conservatively , the lv99 model assumes that the small scale events happen at a slow sweet - parker rate . following @xcite and @xcite . [ fig : lv99model],title="fig:",scaledwidth=45.0% ] we begin by offering a brief summary of the differences between the sweet - parker model of the laminar reconnection @xcite and the lazarian - vishniac model which accounts for the effects of turbulence @xcite . the latter can be seen as a generalization of the sweet - parker model ( see fig . [ fig : lv99model ] ) in the sense that the two regions of differing magnetic directions are pressed up against one another over a broad contact region . this is a generic configuration , which should arise naturally whenever a magnetic field has a non - trivial configuration , whose energy could be lowered through reconnection . the outflow of plasma and reconnected flux will fluctuate as the turbulence evolves and the field line connections change , but the long term average will reflect the turbulent diffusion of the field lines . consequently , the essential difference between the sweet - parker model and the lv99 model is that the former the outflow is limited by microphysical ohmic diffusivity , while in the lv99 model the large - scale magnetic field wandering determines the thickness of outflow . for extremely weak turbulence , when the range of magnetic field wandering becomes smaller than the width of the sweet - parker layer @xmath14 , the two models are indistinguishable . by weak turbulence , following lv99 , we understand a regime where the correlation length is much greater than the distance by which individual field lines deviate from a straight line . lv99 considered a large scale , well - ordered magnetic field , of the kind that is normally used as a starting point for discussions of reconnection . in the presence of turbulence the field has some small scale ` wandering ' . lv99 suggested that the presence of a random magnetic field component leads to fast reconnection . there are three phenomena mainly responsible for this : * only a small fraction of any magnetic field line is subject to direct ohmic annihilation , therefore the fraction of magnetic energy that goes directly into heating the fluid drops down to zero as the fluid resistivity vanishes , * the presence of turbulence enables many magnetic field lines to enter the reconnection zone simultaneously , * turbulence broadens the width of the ejection thickness allowing for more efficient removal of the reconnected flux . with the ( * ? ? ? * henceforth gs95 ) model of turbulence , lv99 obtained : @xmath15 where @xmath16 and @xmath17 are the energy injection scale and velocity . this expression assumes that energy is injected isotropically at the scale @xmath16 smaller than the length of current sheet @xmath10 , which for subalfvnic turbulence leads to the generation of weakly interacting waves at that scale . the waves transfer energy to modes with larger values of @xmath18 until strong turbulence sets in . it is important to note that the strongly turbulent eddies have a characteristic velocity of @xmath19 . in other words , the reconnection speed is the large eddy strong turbulent velocity times factors which depend on whether the current sheet is smaller or larger than the large eddies ( whose length is approximately the injection scale ) . in this sense the reconnection speed should be fairly insensitive to the exact mechanism for turbulent power injection . the main purpose of this paper is test whether or not this is true for a simple modification of the driving mechanism used in @xcite . it is important to note three features of equation ( [ eq : constraint ] ) . first , and most important , it is independent of resistivity . this is , by definition , fast reconnection . second , we usually expect reconnection to be close to the turbulent eddy speed , the geometric ratios that enter the expression , i.e. the injection scale @xmath16 divided by the length of the reconnection layer @xmath10 , are typically of order unity . reconnection will occur on dynamical time scales . finally , we note that in particular situations when turbulence is extremely weak the reconnection speed can be much slower than the alfvn speed . more recently , equation ( [ eq : constraint ] ) was derived using the ideas based on the well - known concept of richardson diffusion @xcite . from the theoretical perspective this new derivation avoids rather complex considerations of the cascade of reconnection events that were presented in lv99 to justify the model . @xcite also shows that lv99 model is closely connected to the recently developed idea of `` spontaneous stochasticity '' of magnetic fields in turbulent fluids . in general , the situation in reconnection community now is very different from that a decade ago . currently , possibilities of fast reconnection in mhd regime due to instabilities of the reconnection layers are widely discussed @xcite . these ideas can be traced back to the work of @xcite . the instabilities , like tearing instability , open up the reconnection layer enabling a wide outflow . we expect such an outflow to become turbulent for most of astrophysical conditions . in this case , the process can be important for initiating reconnection in the particular situation when the level of preexisting turbulence is initially low to initiate sufficiently fast reconnection . we feel that exploring the ways of initiation of turbulent reconnection is very synergistic to the lv99 ideas , but in the current paper we focus on the case of preexisting turbulence of sufficient level . this is the primary domain for which lv99 provides predictions . given the limited dynamical range of numerical simulations , we can only inject power on scales less than than @xmath10 . the most convenient numerical parameter is not @xmath17 , but the energy injection power @xmath20 . the power in the turbulent cascade is @xmath21 or @xmath22 . the amount of energy injected during one alfvn time unit @xmath23 , which is constant in our models , is @xmath24 . therefore @xmath25 . substituting @xmath26 in equation ( [ eq : constraint ] ) results in @xmath27 which is the prediction we will test here . in what follows we refer to the injection power and scale using @xmath1 and @xmath2 , respectively . we use a high - order shock - capturing godunov - type scheme based on the monotonicity preserving ( mp ) spatial reconstruction ( see * ? ? ? * ; * ? ? ? * e.g. ) and strong stability preserving runge - kutta ( ssprk ) time integration ( see * ? ? ? * and references therein ) to solve isothermal non - ideal mhd equations , @xmath28 & = & { \textbf{\em f } } , \label{eq : momentum } \\ { \frac{\partial { \textbf{\em a}}}{\partial t } } + { \textbf{\em e } } & = & { \textbf{\em g } } , \label{eq : induction}\end{aligned}\ ] ] where @xmath29 and @xmath30 are plasma density and velocity , respectively , @xmath31 is the vector potential , @xmath32 is the electric field , @xmath33 is the magnetic field , @xmath34 is the current density , @xmath35 is the isothermal speed of sound , @xmath9 is the resistivity coefficient , and @xmath36 and @xmath37 represent the turbulence driving terms either in velocity or vector potential . we used a multi - state harten - lax - van leer ( hlld , * ? ? ? * ) approximate riemann solver for solving the isothermal mhd equations . the hlld riemann solver takes into account magnetic fields , and can follow alfvn waves with minimal numerical dissipation . this is particularly important here , because our simulations are in the quasi - incompressible regime , where most of energy is transported by alfvn waves . the @xmath38 is maintained by solving the induction equation ( eq . [ eq : induction ] ) using the field interpolated constrained transport ( ct ) scheme based on a staggered mesh ( e.g. * ? ? ? * ; * ? ? ? our reconnection simulation setup is illustrated in figure [ fig : setup_planes ] , which is a 2d cut through the problem setup , indicating the location of the diffusion region . the top and bottom of the computational domain contain equal and opposite field components in the @xmath39 direction , as well as a shared component @xmath40 ( see the left panel of fig . [ fig : setup_planes ] ) . magnetic field lines enter through the top and bottom and are bent by the inflow @xmath41 as they move into the diffusion region . the diffusion region has a length @xmath42 in the @xmath39 direction and a thickness @xmath43 in the @xmath44 direction ( see the left panel of fig . [ fig : setup_planes ] ) . the box is periodic in the @xmath45 direction and the diffusion region extends through the entire domain . the projection of the magnetic topology on the xz plane shows that the lines in the upper region ( solid lines in the right panel of fig . [ fig : setup_planes ] ) and in the lower region ( dashed lines ) create an angle @xmath46 determined by the strength of the shared component @xmath47 . once the incoming magnetic lines enter the diffusion region , they are reconnected and the product of this process is ejected along x direction with a speed @xmath48 ( the left panel of fig . [ fig : setup_planes ] ) . we begin with a harris current sheet of the form @xmath49 initialized using the magnetic vector potential @xmath50,and a uniform guide field @xmath51 . the initial setup is completed by setting the density profile from the condition of the uniform total ( thermal plus magnetic ) pressure @xmath52 and setting the initial velocity to zero . magnetic reconnection is initiated by a small perturbation of the vector potential @xmath53 to the initial configuration of @xmath54 whose strength is given by the coefficient @xmath55 . in all our simulations the strength of the magnetic field is expressed in terms of the alfvn velocity defined by the antiparallel component of the unperturbed magnetic field . similarly the density is expressed in terms of the unperturbed density @xmath56 and velocities are expressed as fractions of the fiducial alfvn speed . the length of the box in the @xmath39 direction defines the unit of distance and time is measured in units of @xmath57 . in the new set of models we set the initial strength of the antiparallel magnetic field component @xmath58 and the guide field @xmath59 . we performed modeling for two resistivity coefficients @xmath60 and @xmath61 which are expressed in the dimensionless units . the initial perturbation is set to @xmath62 . in order to avoid the complications of strong compressibility we have set the sound speed to 4.0 . our computational box has a grid of 256x512x256 or for higher resolution runs , 512x1024x512 . in dimensionless units its size is @xmath63 and @xmath64 . we double the size in the @xmath44 direction to keep the driven turbulence away from the inflow boundary . there is no physical reason to do this , but driving turbulence near the inflow boundary produces numerical instabilities . as mentioned earlier , we use three different types of boundary conditions , depending on the direction of the boundary : outflow boundary conditions along the @xmath39 direction , inflow boundary conditions along the @xmath44 direction and ( sometimes ) periodic boundary conditions along the @xmath45 direction . the open boundary conditions are the same as those used in our previous modeling . we refer to @xcite for their detailed description . briefly , we use simple `` zero - gradient '' boundary conditions , setting the normal derivatives of the fluid variables ( density and momentum ) to zero . in the hydrodynamic limit this allows waves to leave the box without significant boundary reflections . in turbulence simulations this can lead to a slight drift in the fluid density . there is no requirement that the boundary density is constant , and inflows and outflows can cause a small net gain or loss from the system . fortunately , changes in the total mass are small and only fluctuate around the initial value @xcite . they do not influence our results significantly . in order to incorporate the magnetic field into the open boundary conditions , we set the transverse components of the vector potential @xmath31 using first order extrapolation . the normal derivative of the normal component is set to zero . in this way the normal derivatives of the transverse components of the magnetic field are zero , while the normal component of magnetic field is calculated from the zero - divergence condition @xmath65 . this approach avoids the generation of nonzero magnetic divergence at the boundary . however , it has the drawback that it creates a small jump in the momentum flux across the boundary resulting from the presence of non - zero terms @xmath66 at the x outflow boundary and @xmath67 at the y inflow boundary . we have evaluated the velocity increment these terms produce at each time step . in models with the strongest turbulence these terms were of order of @xmath68 and @xmath69 at the x and y boundaries , respectively . in the presence of strong outflows and inflows , generally of order unity , they are clearly negligible . simulations with explicit resistivity run into problems at the boundaries . in order to avoid a non - continuous resistive term and difficulties with the treatment of the current density @xmath70 we have introduced a zone of decaying resistivity near the boundary . in a thin layer near the boundary , the value of resistivity @xmath71 decays down to a very small value chosen to be close to the numerical resistivity @xmath72 of our code . in our models we adopt the value of @xmath73 . none of this has an effect on the reconnection speeds . the validation of this method was presented in @xcite . in our previous work we drove turbulence using a method described by @xcite , in which the driving term was implemented in the spectral space with discrete fourier components concentrated around a wave vector @xmath74 corresponding to the injection scale @xmath75 . we perturbed a number @xmath76 of discrete fourier components of velocity in a shell extending from @xmath77 to @xmath78 with a gaussian profile of the half width @xmath79 and the peak amplitude @xmath80 at the injection scale . the amplitude of driving is solely determined by its power @xmath1 , the number of driven fourier components and the time step of driving @xmath81 . the randomness in time makes the force neutral in the sense that it does not directly correlate with any of the time scales of the turbulent flow , and it also determines the power input solely by the force - force correlation . on the right hand side of equation ( [ eq : momentum ] ) , the forcing is represented by a function @xmath82 , where @xmath29 is local density and @xmath83 is random acceleration calculated using the method described above . in a similar way we can drive turbulence in the vector potential or magnetic field , which is represented by term @xmath37 on the right hand side of the induction equation ( eq . [ eq : induction ] ) . in the new method of turbulence driving we add individual eddies with random locations of their centers and random orientations , either to velocity or magnetic field , at random moments in time . this guarantees the randomness of new forcing . each eddy is calculated from a kernel function described by a directional vector @xmath84 ( with amplitude @xmath85 ) multiplied by a gaussian function @xmath86 where @xmath87 is the location of the eddy center , and @xmath43 is the eddy width . an actual eddy is generated from such a kernel function by taking its curl , i.e. @xmath88 or @xmath89 in the case of injection in velocity or magnetic field , respectively . for example , if we assume that we inject one eddy in the magnetic field at @xmath90 , and that the perturbing vector potential fluctuation has only the nonzero component , i.e. @xmath91 , the contribution to magnetic field is expressed by @xmath92 \left ( x , y , z \right ) = \frac{|{\textbf{\em a}}|}{\delta^2 } \exp\left ( - \frac{|{\textbf{\em r}}|^2}{2 \delta^2 } \right ) \left [ \begin{array}{c } - y \\ x \\ 0 \end{array } \right ] dt \ .\ ] ] this function describes an eddy injected in the xy plane with the maximum amplitude @xmath93 at the distance @xmath94 and injection scale @xmath95 . we know the energy injected by one eddy , which is latexmath:[$\delta e_{eddy } = \pi^{3/2 } amplitude @xmath85 from the injection power @xmath1 and the injection rate @xmath97 , which is the number of eddies injected in a time unit , @xmath98 these estimates are done for the 3d case . in the 2d case the eddy energy is @xmath99 , and therefore the eddy amplitude can be determined from @xmath100 . in the new method there is no direct treatment of the velocity - force correlation , therefore there is no guarantee that this correlation is zero and the injected power is completely determined by the force - force correlation . a reasonable solution to this problem would be to control the amount of injected energy and modify the amplitude of injected eddies or the injection rate at each time step in order to compensate the differences in the energy injected in the domain . the performed tests show , however , that although the velocity - force correlation is not zero , it is in fact fluctuating in time with a small amplitude , and giving in result a zero net contribution . the new method drives turbulence directly in the real space , in contrast to the previous one , therefore it can be applied locally . we drive turbulence in a subvolume of the domain . the size of the subvolume is determined by two scales , the radius @xmath101 on the xz plane around the center of the domain and the height @xmath102 describing the thickness of the driving region from the midplane . in this way we avoid driving turbulence at the boundary and reduce the influence of driving on the inflow or outflow . all models are evolved without turbulence for several dynamical times in order to allow the system to achieve stationary laminar reconnection . then , at a given time @xmath103 we start driving turbulence by increasing its amplitude to the desired level , until @xmath104 . in this way we let the system to adjust to a new state . from time @xmath104 the turbulence is driven with the full power @xmath1 . in the next sections we measure the reconnection rate using the new method of reconnection rate measure introduced in @xcite and described by a formula @xmath105\ ] ] where @xmath106 is the strength of reconnecting magnetic component , @xmath107 is the electric field , @xmath108 is area element of an xz plane across which we perform integration , @xmath109 is the line element separating two regions of the yz plane defined by the sign of @xmath106 , @xmath110 is the asymptotic absolute value of @xmath106 , and @xmath111 is the width of the box . this method of the reconnection rate measure was derived from the magnetic flux conservation @xmath112 and takes into account all processes contributing to the change of magnetic flux . the electric field @xmath113 can be further divided into an advection term @xmath114 , a shear term @xmath115 , and a resistive term @xmath116 . with this in mind the line integral can be rewritten as @xmath117 this new reconnection measure contains the time derivative of the absolute value of @xmath106 , and a number of boundary terms , such as advection of @xmath106 across the boundary and the boundary integral of the resistive term @xmath118 . the additional terms include all processes contributing the time change of @xmath119 . in particular , they can have non - zero values . c|cccccccccccc name & @xmath47 & @xmath71 [ @xmath120 & @xmath121 [ @xmath120 & @xmath1 & @xmath122 & driving type + pd & 0.1 & 1.0 & 0.0 & 0.1 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 0.2 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 0.5 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 2.0 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 0.1 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 0.2 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 0.5 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 2.0 & 8 & old in @xmath123 + @xmath124 & 0.1 & 1.0 & 0.0 & 0.2 & 8 & new in @xmath125 + @xmath124 & 0.1 & 1.0 & 0.0 & 0.5 & 8 & new in @xmath125 + @xmath124 & 0.1 & 1.0 & 0.0 & 1.0 & 8 & new in @xmath125 + hr & 0.1 & 0.5 & 0.0 & 1.0 & 8 & new in @xmath123 + sd & 0.1 & 1.0 & 0.0 & 1.0 & 5 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 1.0 & 12 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 1.0 & 16 & old in @xmath123 + & 0.1 & 1.0 & 0.0 & 1.0 & 25 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 5 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 8 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 12 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 16 & old in @xmath123 + & 1.0 & 1.0 & 0.0 & 1.0 & 25 & old in @xmath123 + @xmath124 & 0.1 & 1.0 & 0.0 & 1.0 & 8 & new in @xmath125 + @xmath124 & 0.1 & 1.0 & 0.0 & 1.0 & 24 & new in @xmath125 + @xmath124 & 0.1 & 1.0 & 0.0 & 1.0 & 32 & new in @xmath125 + hr & 0.1 & 0.5 & 0.0 & 1.0 & 8 & new in @xmath123 + vd & 0.1 & 1.0 & 0.2 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 0.5 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 1.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 2.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 3.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 4.0 & 1.0 & 8 & old in @xmath123 + & 0.1 & 1.0 & 5.0 & 1.0 & 8 & old in @xmath123 + in table [ tab : models ] we list parameters of all the models presented in this paper including models from @xcite and models with new driving . as in the previous paper we divided them into several groups . in each group we calculated models in order to study the dependence of the reconnection rate on a characteristic parameter of turbulence or resistivity . we have studied the dependence of reconnection on the power of turbulence ( models pd ) , injection scale ( models sd ) , and viscosity ( models vd ) . models with new driving are marker with a right arrow ( @xmath124 ) , and models with new driving and higher resolution are marked with a symbol only the varying parameters are listed in the table , the strength of guide field @xmath47 , the uniform resistivity @xmath71 , the uniform viscosity @xmath121 , the power of turbulence @xmath1 and its injection scale @xmath74 , and the method of turbulence driving . all models presented in this section were calculated with the grid size @xmath126 corresponding to the resolution 256x512x256 , except the model marked with symbol hr , which was simulated with the resolution 512x1024x512 ( @xmath127 ) . , and kinetic and magnetic energies , @xmath128 and @xmath129 , respectively . two dotted vertical lines bound the period of gradually increasing turbulence . the resistivity in this model is set to @xmath130 and the shared component of magnetic field @xmath131 . in this model we inject turbulence in the magnetic field . [ fig : energies],scaledwidth=45.0% ] in figure [ fig : energies ] we present an example of the evolution of total mass , and kinetic and magnetic energies in a model with @xmath132 , @xmath133 , and @xmath134 . we inject turbulence in magnetic field using new forcing method , gradually increasing its strength from @xmath135 to @xmath136 . this period is marked by two dotted vertical lines in figure [ fig : energies ] . we see an increase of kinetic energy during this period due to the injection and saturation after @xmath136 . the kinetic energy preserves constant value during the turbulent stage very well . the magnetic energy increases during this stage , slowly saturating . this increase is attributed to the injection of magnetic eddies . on the contrary , the total mass in the system decays slowly . we emphasize , that since we use open boundary conditions , not perfect conservation of mass and energies is possible in the presence of turbulence . ( black ) for the same model as in fig . [ fig : energies ] . blue line show the evolution of reconnection in a model with the same parameters in which the turbulence were driven using the old method . in this plot we present the measured rates of the sweet - parker reconnection @xmath137 and during the presence of turbulence , @xmath138 . symbol @xmath139 is the time variance . @xmath140 is the estimated uncertainty of the measure . [ fig : rate],scaledwidth=45.0% ] in figure [ fig : rate ] we show the evolution of reconnection rates @xmath0 for two models with the same set of initial conditions , but in the first model we drove turbulence by injecting magnetic eddies using the new method described in this paper ( black line ) , and in the second model we inject velocity fluctuations using the old method described in @xcite ( blue line ) . in this plot we recognize an increase of both rates during the introduction of turbulence . after the transition period between @xmath135 and @xmath136 , when the system adjusts to a new state , both measures coincide and even though they fluctuate , they reach a stationary state characterized by faster reconnection . both types of turbulence bring the reconnection rate to similar level . somewhat higher reconnection rate in the model with new driving could be attributed to the fact that this model was calculated using the 5th order spatial reconstruction and the 3rd order integration in time , in contrast to the old model where we used the second order methods . lower order , especially in the spatial interpolation , introduces additional numerical diffusion decreasing the amplitudes of turbulent fluctuations at scales comparable to the current sheet scale . in figure [ fig : rate ] we also show the way of measuring the reconnection rates , in the sweet - parker and lv99 stages , @xmath137 and @xmath138 , respectively . because the reconnection rates fluctuate in the presence of turbulence we also measure their time variance @xmath139 using the standard deviation . in addition to the time variance of @xmath0 , we measure their errors by splitting the averaging region into two subregions and after averaging the rates @xmath141 and @xmath142 over each subregion ( see fig . [ fig : rate ] ) , we take the absolute value of their difference @xmath143 . this difference corresponds to the error of @xmath0 , i.e. it is different from zero if the rate is not constant in time . in all further analysis and presented plots we use values estimated in this way . these measures correspond exactly to those presented in @xcite . in this section we compare field topologies in two example models run with the same set of parameters , but with different types of driving . both models have been simulated with the uniform resistivity @xmath134 and the resolution 256x512x256 . we injected turbulence with power @xmath144 at the injection scale @xmath145 . the only difference between models is the way we injected turbulence . in the old model we inject velocity fluctuations with random phases in fourier space , and then transform them to real space and shape by a window in order to limit the injection to the specified region near the current sheet . in the new model , we inject magnetic loops with random locations and random orientations in the 3d volume near the current sheet . the way of injecting turbulence is essentially different in both cases . in figure [ fig : top_turb ] we show examples of xy - cuts ( upper row ) and xz - cuts ( lower row ) through the box for the model with old driving . in the left and middle columns we show topologies of the velocity and magnetic field , respectively , with the intensities corresponding to the amplitude of components parallel to the plotted plane . in the right column we show the absolute value of current density with overplotted magnetic vectors . velocity has a very complex and mixed structure near the midplane due to constant injection of fluctuations in this region ( see the left panel in fig . [ fig : top_turb ] ) . the majority of the velocity fluctuations is perpendicular to the mean magnetic field . this is because we are in the nearly incompressible regime of turbulence ( large plasma @xmath146 ) and most of the fluctuations propagate as alfvn waves along the mean magnetic field . slow and fast waves , whose strengths are significantly reduced , are allowed to propagate in directions perpendicular to the mean field as well . as a result , a big fraction of the turbulent kinetic energy leaves the box along magnetic lines . we observe , however , an efficient bending of magnetic lines at the midplane where the field is weaker ( see the upper middle plot in fig . [ fig : top_turb ] ) . this is not result of a driving , but result of reconnection . in general the interface between positively and negatively directed magnetic lines is much more complex than in the case of sweet - parker reconnection . this complexity favors creation of enhanced current density regions , where the local reconnection works faster ( see the right panel of fig . [ fig : top_turb ] ) . since we observe multiple reconnection events happening at the same time , the global reconnection rate should be significantly enhanced . in figure [ fig : top_new ] we show similar examples of xy - cuts ( upper row ) and xz - cuts ( lower row ) through the box but for a model with the new way of driving turbulence . here , a big number of individual eddies is injected in the magnetic field with random locations and random orientations in domain . comparing to plots in figure [ fig : top_turb ] we see differences but also some clear similarities . among the similarities , we note a highly turbulent region near the current sheet seen in all xy - cuts , with the current sheet itself strongly deformed and fragmented into many small scale current sheets ( the right column of figures [ fig : top_turb ] and [ fig : top_new ] ) . we see also some small increase of magnetic field strength near the current sheet ( middle top panels ) resulting from working turbulence in the injection region . among the differences we can list somewhat different distributions of the fragmented current sheets in the new model with clear enhancements in the locations where the magnetic eddies are injected at that moment . these enhancements are clearly seen in the magnetic topology and current density plots ( middle and right columns ) . in order to decrease those strong disturbances of the magnetic lines , we shall reproduce the same model with higher injection rate and reduced amplitudes of individual eddies . another difference is the strength of current density @xmath70 . in the model with old driving we see more volume in which @xmath147 reaches high magnitude and its structure is elongated with the local field . in the model with new driving , the current density with high strength seems to be less correlated with the local field , probably due to the presence of newly injected eddies . in the intermediate strengths , the structure of @xmath147 seems to be better correlated with the local field . we see from this comparison that models with different driving of turbulence demonstrate different topologies of the fields . in the next sections we show , that the averaged reconnection rates do not change significantly , confirming that the way we inject turbulence is of less importance and only its strength and injection scale have influence on @xmath0 . on @xmath1 updated by symbols from new models . blue symbols show models with new driving in which the eddies where injected in magnetic field instead of velocity , as in the previous models ( black symbols ) . the dotted line corresponds to the sweet - parker reconnection rate for models with @xmath148 . a unique red symbol shows the reconnection rates from model with new driving in velocity performed with higher resolution ( 512x1024x512 ) and resistivity coefficient reduced to @xmath149 . error bars represent the time variance of @xmath0 . the size of symbols corresponds to the error of @xmath0 ( the way we calculate errors is described in [ ssec : evolution ] ) . [ fig : pow_dep ] ] models with the new method of turbulence driving are listed in table [ tab : models ] . we run a few models with varying turbulent powers in order to verify if the new driving modifies our previous results . in these models we kept the same parameters as in the previous ones , which allowed to confirm the dependence of the reconnection rate @xmath0 on the power of injected turbulence @xmath1 . figure [ fig : pow_dep ] shows the values of reconnection speed @xmath0 in models with turbulent power @xmath1 varying in the range of values by more than one order of magnitude , from 0.1 to 2.0 , for all previously shown models ( black symbols ) in @xcite and for new models ( blue and red symbols ) in which we drove turbulence using the new method . because the evolution of @xmath0 in new models reaches stationarity after time @xmath150 , we averaged @xmath0 from @xmath150 to @xmath151 in these models . figure [ fig : rate ] shows that the reconnection rates oscillate around their mean values . in figure [ fig : pow_dep ] we plot how the averaged reconnection speed depends on the strength of turbulence . filled symbols represent the averaged reconnection rate in the presence of turbulence . the dotted line corresponds to the reconnection rate during the sweet - parker process , i.e. without turbulence . the error bars show the time variance of @xmath0 . the size of symbols indicates the uncertainty in our estimate of the reconnection speed @xmath152 normalized to the uncertainty in the reconnection speed during the sweet - parker evolution @xmath153 . it is calculated from a formula @xmath154 . if @xmath152 is of the order of @xmath153 their symbols have the same sizes . the reconnection rates for models with new driving , which is described in [ ssec : forcing ] , confirm the theoretical dependence of @xmath0 on the injected power , which scales as @xmath155 . there is no significant difference between models in which turbulence was driven in velocity and in magnetic field . this is in agreement with the lv99 prediction , that the reconnection rate does not depend on the type of turbulence . the reconnection rate @xmath0 in the presence of turbulence depends only on the strength of turbulence and its injection scale @xmath2 , according to equation [ eq : constraint ] , for a fixed magnitude of the antiparallel magnetic field component . in the previous subsection we presented studies on the turbulent power dependence in this subsection we aim to study the injection scale dependence . for this purpose we performed several models with the new way of driving turbulence , as well , in order to verify if they confirm the dependence of the reconnection speed @xmath0 on the scale @xmath2 at which we inject turbulence . the new models are listed in table [ tab : models ] . we inject turbulence at several scales , from @xmath156 to @xmath157 . at the upper end of this range the turbulence barely broadens the sweet - parker current sheet . at the lower end the turbulent eddies are barely contained within the volume in which we excite turbulent motions . in figure [ fig : sca_dep ] we present the reconnection speed dependence on the injection scale . we plot the averaged @xmath0 for old models ( black symbols ) completed by the values from new models with alternative driving ( blue and red symbols ) . from the plot we clearly see a strong dependence of the reconnection rate on the injection scale . the new models very precisely follow the same dependence , confirming again that the type of turbulent driving has no influence on the process of reconnection , and only the power and injection scale of this driving have strong importance . similarly , as in the power dependence plot , the new models have slightly higher reconnection speeds comparing to the old ones . this is due to reduced numerical dissipation of velocity , since in the new models we used higher order methods . dissipation removes energy at small scales . if it is smaller , due to higher order numerical scheme , the turbulent fluctuations reach higher amplitudes at the current sheet scale . this influences the rate of individual reconnection events improving slightly the global reconnection rate @xmath0 . on @xmath2 with additional models in which turbulence was driven in a new way , as described in [ ssec : forcing ] . similarly to fig . [ fig : pow_dep ] , blue symbols show models with perturbed magnetic field , and red symbols correspond to a high resolution model with reduced uniform resistivity in which turbulence was driven in velocity . the dotted line corresponds to the sweet - parker reconnection rate for models with @xmath148 . error bars and the size of symbols have the same meaning as in fig . [ fig : pow_dep ] . [ fig : sca_dep ] ] figure [ fig : sca_dep ] shows a bit weaker scaling with the injection scale than that predicted by lv99 model , i.e. @xmath158 . we see several possible sources for the discrepancy . for instance , the existence of a turbulent inverse cascade can modify the effective @xmath2 . in addition , reconnection can also modify the characteristics of turbulence , such as the power spectrum and anisotropy . we aim to study these problems in the future work . in @xcite we performed studies of the reconnection rate on the resistivity , both the uniform and anomalous ones , and we obtained great agreement with the sweet - parker scaling @xmath159 for the case without turbulence , and no dependence on resistivity in the presence of turbulence , as was predicted in lv99 . in this section we performed additional studies of the reconnection rate dependence on viscosity . the dissipation scale of turbulent cascade is related to the magnitude of viscosity . if the dissipation works at scales larger than the current sheet thickness , the turbulence cascade stops before reaching the current sheet and the global reconnection rate should be reduced . the reconnection will be enhanced still by the broadened ejection region , allowing for more efficient removal of the reconnected magnetic flux . in figure [ fig : vis_dep ] we show reconnection rates for models with varying viscosity coefficient . although there is not prediction for this dependence in the lv99 model , we could test it numerically . in the figure [ fig : vis_dep ] we see a weak dependence @xmath160 . this dependence might be also useful in understanding the reconnection speed differences between models with the same set of parameters but different resolutions , or solved with different orders of the numerical scheme . at low resolutions or low order schemes , the numerical viscosity is expected to be larger , thus we should observe reduced reconnection speeds in those cases . this is confirmed in figures [ fig : pow_dep ] and [ fig : sca_dep ] where we compare old models done with the second order scheme , and new models done with higher order schemes and higher resolutions . in those plots all new models demonstrate slightly higher reconnection rates . on the uniform viscosity coefficient @xmath5 . as explained in the text , the reconnection speed is reduced with increasing value of @xmath5 . the dotted line corresponds to the sweet - parker reconnection rate . error bars and the size of symbols have the same meaning as in fig . [ fig : pow_dep ] . [ fig : vis_dep ] ] the lv99 model was introduced for both collisional and collisionless media and it claimed that the microphysics of collisionless reconnection events does not change the resulting reconnection rates . this point was subjected to further scrutiny in @xcite who provided a thorough investigation of the problem and concluded that for most of astrophysical collisionless plasmas the lv99 model should be applicable , provided that plasma is turbulent . with turbulence being ubiquitous in astrophysical conditions , this hardly constraints the applicability of the lv99 model . the lv99 model of reconnection is applicable to the collisional medium , such as the ism , which is both turbulent and magnetized , and where the hall - mhd reconnection does not work @xcite . for instance , for hall - mhd reconnection to be applicable , it is required that the sweet - parker current sheet @xmath161 width is smaller than the ion inertial length @xmath162 . thus , the `` reconnection criterion for media to be collisionless '' is @xmath163 , which presents a much severe constraint on the possible rate of collisions . as a result magnetic reconnection happens to be mediated by the hall - mhd only if the extend of the contact region @xmath10 ( see fig . [ fig : lv99model ] ) does not exceed @xmath164 cm . magnetic fields in the ism should interact over much larger scales . the lv99 model works in astrophysical environments to which the hall - mhd reconnection is applicable , as well , like solar corona , interplanetary medium , if the level of turbulence is high enough . the reconnection on microscales can happen fast , i.e. in the hall - mhd fashion . this may not change , however , the global reconnection rate . the lv99 model shows that even with relatively slow sweet - parker reconnection at microscales the global reconnection is limited not by ohmic resistivity , but by the rate of magnetic field wondering . we believe that the hall - mhd _ local _ reconnection of magnetic fields is taking place in the interplanetary medium , which is being tested by local _ in - situ _ measurements , while the _ global _ reconnection rates are determined by magnetic field wandering as prescribed in lv99 . in the absence of quantitative model to be tested , simulations aimed at studying reconnection speed have been done in 2d , both for collisional and collisionless regimes . this allowed to achieve higher resolutions ( compared to those contemporary available in 3d ) but substantially constrained magnetic field dynamics . for instance , the closest study to ours was done by @xcite ( see also * ? ? ? * ) . the authors studied 2d magnetic reconnection in the presence of external turbulence . an enhancement of the reconnection rate was reported , but the numerical setup precluded the calculation of a long term average reconnection rate . a more recent study along the approach in @xcite is one in @xcite , where the effects of small scale turbulence on 2d reconnection were studied and no significant effects of turbulence on reconnection were reported for the setup chosen by the authors . later , @xcite redone the modeling of 2d turbulent reconnection following @xcite with much higher resolutions . they used an advanced technique to detect precisely all x - points in the domain and then performed statistical studies confirming the sweet - parker relation for the reconnection rate as a function of x - point geometry . the development of different techniques to study magnetic reconnection is very important . even though their modeling was limited to one type of highly superalfvnic decaying turbulence ( the initial uniform magnetic field was zero ) , they reported reconnection rates with normalized values @xmath165 and confirmed the importance of turbulence for modifying the character of magnetic reconnection and specifies heating and transport as the effect of particular significance , as well as formation of petschek - type `` x - points '' in 2d turbulence . due to the lack of large scale magnetic field configuration , their model represents a specific case , far from generic situation observed in the astrophysical objects where the mean and turbulent components of magnetic fields have comparable strengths . therefore , these studies can not predict the global reconnection rate , as well . moreover , @xcite interpreted successful numerical confirmation of the lv99 model as a result of strong turbulence , although @xcite addressed this problem carefully showing that the amplitudes of velocity fluctuations , both injected and obtained from spectra of developed and stationary turbulence , are fractions of alfvn speed . the fact that our study is in 3d is essential , as the lv99 model is intrinsically three dimensional . the general picture is of tangled field lines with reconnection taking place via a series of `` y - points '' or modified sweet - parker sheets distributed in some fractal way throughout the turbulence . a large scale sweet - parker sheet will be replaced by a more fractured surface , but the current sheets will occupy a vanishingly small fraction of the total volume and the field reversal will remain relatively well localized . the model predicts that the reconnection speed would be approximately equal to the strong turbulent velocity with a modest dependence on the ratio of the eddy length to the current sheet length . there should be no dependence on resistivity . the major results contained in our figures showing the dependence of the reconnection speed on resistivity , input power and input scale agree with the quantitative predictions of the lv99 model . we are not aware of any competing models to compare our simulations with . the major differences from the present study stem from the fact that we test a 3d model of reconnection , as the lv99 depends on effects , like field wandering , that happen only in 3d . in order to show how different 2d and 3d worlds are , we performed similar studies to those presented in @xcite , but limiting the domain to two dimensions ( see * ? ? ? * ) . in @xcite we demonstrated that 2d magnetic reconnection in the presence of turbulence depends on the ohmic resistivity , therefore , it is not fast . also , the dependencies on the turbulent power and injection scales were significantly weaker than in the lv99 . this dependence of 2d reconnection rate on ohmic resistivity in the presence of turbulence , although weaker than the sweet - parker relation @xmath166 , has been independently confirmed by @xcite studies , performed with very different approach . these differences call for deliberation with simple extension of conclusions coming from the 2d modeling to natural for magnetic field fully three dimensional world . reconnection is one of the most fundamental processes involving magnetic fields in conducting fluids or plasmas . therefore , the identification of a robust process responsible for reconnection has many astrophysically important consequences . below we list a few selected implications of the successful validation of the lv99 model . numerical studies on fermi acceleration in turbulent reconnection have a long history ( e.g. * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in the sweet - parker model , it has been shown that particles can accelerate due to the induced electric field in the reconnection zone @xcite . this @xmath167 acceleration process , however , is constrained by the narrow thickness of the acceleration zone which has to be larger than the particle larmor radius and by the strength of the magnetic field . therefore , the efficiency of this process is rather limited . besides , it also does not predict a power - law spectrum , as generally observed for cosmic rays . observations have always been suggestive that magnetic reconnection can happen at a high speed in some circumstances , in spite of the theoretical difficulties in explaining it . for instance , the phenomenon of solar flares suggests that magnetic reconnection should be first slow in order to ensure the accumulation of magnetic flux and then suddenly become fast in order to explain the observed fast release of energy . the lv99 model can naturally explain this and other observational manifestations of magnetic reconnection . consider a particle entrained on a reconnected magnetic field line ( see fig.[fig : lv99model ] ) . this particle may bounce back and forth between magnetic mirrors formed by oppositely directed magnetic fluxes moving towards each other with the velocity @xmath0 . each bounce will increase the energy of a particle in a way consistent with the requirements of the first - order fermi process @xcite . this is in contrast to the second - order fermi acceleration that is frequently discussed in terms of accelerating particles by turbulence generated by reconnection @xcite . the numerical studies of the particle acceleration supporting these ideas have been already started @xcite . an interesting property of this acceleration mechanism is that it is also potentially testable observationally , since the resulting spectrum of accelerated particles is different from that arising from a shock . @xcite used this mechanism of particle acceleration to explain the synchrotron power - law spectrum arising from the flares of the microquasar grs 1915 + 105 . further applications can be found in the solar physics . following @xcite we note that solar flares inspired much of the earlier research on reconnection ( see * ? ? ? * ; * ? ? ? as the plasma involved is substantially rarefied , the restrictive conditions for the collisionless reconnection are satisfied in this particular environment . @xcite stated that bistable hall reconnection can be important in this case . stochastic reconnection provides an alternative explanation . indeed , an important prediction of the lv99 model is related to the _ reconnection instability _ that arises in the situation when the initial structure of the flux prior to reconnection is laminar . reconnection at the sweet - parker rate is negligible . this allows magnetic flux to accumulate . however , when the degree of stochasticity exceeds a threshold value , the reconnection itself should excite more turbulence , creating a positive feedback resulting in a flare ( see * ? ? ? the instability is a generic property of laminar field reconnection in both collisionless and collisional environments . referring to the sun , one may speculate that the difference between gradual and eruptive flares arises from the original state of magnetic field prior to the flare , at least in some specific situations . in the case when the magnetic field is sufficiently turbulent the accumulation of magnetic flux does not happen and the flare is gradual . similarly , the observed spatial spread of energy release during solar flares may be due to the spread of the region of turbulent fields once reconnection is initiated at one place . recent observations demonstrate that gradual flares occur rather in regions with large scale and weak magnetic fields for which alfvn times are large @xcite . in light of that , the difference in alfvn times may explain different time scales in gradual and impulsive flares . further research is necessary for establishing the role of turbulence in changing the time scale of flare evolution . the lv99 model can find its application in the removal of magnetic flux from the star formation regions . @xcite showed that magnetic field is removed from the star forming core cluster faster than it is allowed by the standard ambipolar diffusion scenario @xcite . @xcite proposed a mechanism using efficient reconnection through `` hyper - resistivity '' . @xcite performed numerical studies of such a concept , replacing the `` hyper - resistivity '' with efficient stochastic reconnection . they reported removal of strong anticorrelations of magnetic field through `` reconnection diffusion '' , which can mimic the effect of enhanced ohmic resistivity . lv99 showed that fast reconnection of stochastic magnetic field makes the models of strong mhd turbulence self - consistent , because the critical balance in the gs95 model requires the existence of eddy - type motions perpendicular to the magnetic field . in the absence of reconnection this would result in unresolved knots that should drain energy from the cascade . the estimates in lv99 showed that the rates of reconnection predicted by the model are sufficient to resolve magnetic knots within one period . the lv99 model of magnetic reconnection in the presence of weakly stochastic magnetic fields was proposed by @xcite in @xcite . however , due to a few objective factors it met less enthusiasm in the community than , for example , the x - point collisionless reconnection . we believe that there were three major factors responsible for this . 1 . the collisionless x - point reconnection was initiated and supported by numerical simulations , while lv99 was a theory . its numerical testing became possible only recently . the reconnection subject had a history of failed theories and models , which without direct numerical support were taken with a grain of salt . the acceptance of the idea of astrophysical fluids generically being in turbulent state had much less observational support at that time compared to the present day . by now we have much more evidence which allows to claim that models not taking the pre - existent turbulence has little relevance to astrophysics . the analytical solutions of lv99 were based on the use of gs95 model of turbulence . the gs95 model of turbulence , in fact , was extended lv99 by introducing the concept of local reference frame for turbulent eddies and by extending the gs95 scalings to the subalfvnic case . the gs95 theory was far from being generally accepted at the time of the lv99 publishing . the situation has changed substantially by now . first of all , gs95 was successfully tested numerically @xcite and their ideas have been extended to describing the alfvnic cascade in compressible mhd turbulence ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? the so - called `` big power law in the sky '' indicating the presence of turbulence on scales from tens of parsecs to thousands of kilometers has been extended @xcite , and the observations of gas and synchrotron emission provided extended number of direct turbulence measurements confirming their presence ( see * ? ? ? * ; * ? ? ? finally , the situation has changed with the numerical testing of the lv99 model . the 3d mhd simulations in @xcite supported the predictions in the lv99 paper and our present work goes further with testing this model , by including different types of energy injection . it is worth noting also , that there is some implicit observational evidence in the favor of the lv99 model , like observations of the thick reconnection current outflow regions observed in the solar flares @xcite . @xcite , explaining quasi - periodic pulsations in observed flaring energy releases at an active region above the sunspot , proposed that the wave packets arising from the sunspots can trigger such pulsations . they established a phenomenological relation between oscillations in a sunspot and pulsations in flaring energy releases . this phenomenon can be naturally explained by the lv99 model . in this article we performed additional testing of the lv99 model of fast reconnection under different types of turbulent driving using 3d numerical simulations . we have introduced a new method of driving turbulence by direct injection of the velocity or magnetic eddies with random locations in the domain . we analyzed the dependence of the reconnection speed on the turbulence injection power , on the injection scale , as well as on the viscosity . we found that : * we observe similar changes of the topology of the magnetic field near the interface of oppositely directed magnetic field lines in models with two different turbulence injection mechanisms . these changes include the fragmentation of the current sheet , favoring multiple simultaneous reconnection events , as well as a substantial increase in the thickness of the outflow of reconnected magnetic flux and matter . * the relation between the reconnection rate @xmath0 and turbulent power @xmath1 remains unchanged under two different mechanisms of energy injection and is confirmed by new models to be @xmath168 , in agreement with the lv99 prediction . moreover , the injection in magnetic field produces similar effects on the reconnection as injection in velocity , remaining the dependence unaltered . * the reconnection rate grows with the size of the injected eddies , which can be directly related to the turbulence injection scale . the rate of growth , for the models with old and new driving mechanism , is approximated by @xmath169 scaling which agrees with the previously obtained scaling . somewhat steeper lv99 prediction , @xmath170 , could results from limitations in the dynamic range available for study . * reconnection in the presence of weak turbulence is only weakly sensitive to viscosity @xmath5 . from performed simulations we obtained a dependence @xmath160 for one set of parameters : @xmath171 , @xmath145 , @xmath134 , @xmath59 . the research of gk is supported by fapesp grant no . 2009/50053 - 8 , al is supported by the center for magnetic self - organization in laboratory and astrophysical plasmas , nsf grant ast-08 - 08118 and nasa grant nnx09ah78 g . the work of etv is supported by the national science and engineering research council of canada . the research of kom is supported by the polish ministry of science and higher education through grants : 92/n - astrosim/2008/0 and 3033/b / h03/2008/35 . this research also was supported by the national science foundation project tg - ast080005n through teragrid resources provided by texas advanced computing center ( tacc : http://www.tacc.utexas.edu ) . part of this work was made possible by the facilities of the shared hierarchical academic research computing network ( sharcnet : http://www.sharcnet.ca ) and the galera supercomputer in the academic computer centre in gdask ( task : http://www.task.gda.pl ) . bhattacharjee , a. , ma , z. w. , and wang , x. : recent developments in collisionless reconnection theory : applications to laboratory and astrophysical plasmas , lec . not . phys . , 614 , 351375 , 2003 . bhattacharjee , a. , huang , y .- m . , yang , h. , and rogers , b. : fast reconnection in high - lundquist - number plasmas due to the plasmoid instability , phys . plas . , 16 , 112102 , 2009 . cho , j. and lazarian , a. : compressible magnetohydrodynamic turbulence : mode coupling , scaling relations , anisotropy , viscosity - damped regime and astrophysical implications , mon . not . r. astron . soc . , 345 , 325339 , 2003 . de gouveia dal pino , e. m. and lazarian , a. : the role of violent magnetic reconnection on the production of the large scale superluminal ejections of the microquasar grs 1915 + 105 , arxiv : astro - ph/0307054 , 2003 . de gouveia dal pino , e. m. and lazarian , a. : production of the large scale superluminal ejections of the microquasar grs 1915 + 105 by violent magnetic reconnection , astron . astrophys . , 441 , 845853 , 2005 . gaensler , b. m. , haverkorn , m. , burkhart , b. , newton - mcgee , k. j. , ekers , r. d. , lazarian , a. , mcclure - griffiths , n. m. , robishaw , t. , dickey , j. m. , and green , a. j. : low - mach - number turbulence in interstellar gas revealed by radio polarization gradients , nature , 478 , 214217 , 2011 . goldstein , m. l. , matthaeus , w. h. , and ambrosiano , j. j. : acceleration of charged particles in magnetic reconnection solar flares , the magnetosphere , and solar wind , geophys . lett . , 13 , 205208 , 1986 . kulpa - dybe , k. , otmianowska - mazur , k. , kulesza - ydzik , b. , hanasz , m. , kowal , g. , wltaski , d. , and kowalik , k. : global simulations of the magnetic field evolution in barred galaxies under the influence of the cosmic - ray - driven dynamo , astrophys . j. lett . , 733 , l18 , 2011 . santos - lima , r. , lazarian , a. , de gouveia dal pino , e. m. , and cho , j. : diffusion of magnetic field and removal of magnetic flux from clouds via turbulent reconnection , astrophys . j. , 714 , 442461 , 2010 . servidio , s. , matthaeus , w. h. , shay , m. a. , dmitruk , p. , cassak , p. a. , and wan , m. : statistics of magnetic reconnection in two - dimensional magnetohydrodynamic turbulence , phys . plas . , 17 , 032315 , 2010 . sweet , p. a. : the neutral point theory of solar flares , electromagnetic phenomena in cosmical physics , 6 , 123134 , 1958 . iau symposium 6 , _ electromagnetic phenomena in cosmical physics _ b. lehnert , ( cambridge , uk : cambridge university press ) sych , r. , nakariakov , v. m. , karlicky , m. , and anfinogentov , s. : relationship between wave processes in sunspots and quasi - periodic pulsations in active region flares , astron . astrophys . , 505 , 791799 , 2009 . valle , j. p. : observations of the magnetic fields inside and outside the milky way , starting with globules ( ~ 1 parsec ) , filaments , clouds , superbubbles , spiral arms , galaxies , superclusters , and ending with the cosmological universe s background surface ( at ~ 8 teraparsecs ) , fund . phys . , 19 , 189 , 1997 . valle , j. p. : observations of the magnetic fields inside and outside the solar system : from meteorites ( ~ 10 attoparsecs ) , asteroids , planets , stars , pulsars , masers , to protostellar cloudlets ( < 1 parsec ) , fund . , 19 , 319422 , 1998 .

we present the first double - lined spectroscopic orbital elements for the central binary in the massive triple , . the solutions are based on fits of cross - correlation functions of _ iue _ high dispersion uv spectra and profiles . the orbital elements for the primary agree well with previous results , and in particular , we confirm the apsidal advance with a period of . we also present tomographic reconstructions of the primary and secondary stars spectra that confirm the o9.5 ii classification of the primary and indicate a b0.5 iii type for the secondary . the relative line strengths between the reconstructed spectra suggest magnitude differences of in the uv and at 6678 . the widths of the uv cross - correlation functions are used to estimate the projected rotational velocities , and for the primary and secondary , respectively ( which implies that the secondary rotates faster than the orbital motion ) . we used the spectroscopic results to make a constrained fit of the _ hipparcos _ light curve of this eclipsing binary , and the model fits limit the inclination to the range . the lower limit corresponds to a near roche - filling configuration that has an absolute magnitude which is consistent with the photometrically determined distance to ori ob1b , the orion belt cluster in which resides . the solution results in masses of and , both of which are substantially below the expected masses for stars of their luminosity . the binary may have experienced a mass ratio reversal caused by case a roche lobe overflow , or the system may have suffered extensive mass loss through a binary interaction ( perhaps during a common envelope phase ) in which most of the primary s mass was lost from the system rather than transferred to the secondary . we also made three component reconstructions to search for the presumed stationary spectrum of the close visual companion , . there is no indication of the spectral lines of this tertiary in the uv spectrum , but a broad and shallow feature is apparent in the reconstruction of indicative of an early b - type star . the tertiary may be a rapid rotator ( ) or a spectroscopic binary .

astro-ph0110683

in this series of papers we have applied a version of the doppler tomography algorithm @xcite to reconstruct the component ultraviolet and optical spectra of massive close binary systems , and we have investigated their spectral morphology , rotational velocities , and relative flux contributions . we were successful in three component spectral reconstructions of the triple systems 55 uma @xcite and @xmath11 cir @xcite , and we are continuing to examine the ultraviolet spectra of several ob systems known to be triples from speckle interferometry @xcite . here we turn our attention to the most famous of this group , and one of the brightest o - stars in the sky , @xmath11 orionis . the visual and spectroscopic components of the @xmath11 orionis system ( mintaka , 34 ori , hd 36486 , hr 1852 , , ads 4134 , hei 42 , hip 25930 ) are depicted in fig 1 . the washington double star catalog @xcite lists two distant visual companions : @xmath11 ori b , a 14.0 mag star 33@xmath12 away , and @xmath11 ori c ( hd 36485 ) , a 6.85 mag star at a separation of 53@xmath12 . both of these targets are far outside the spectroscopic apertures used on the _ international ultraviolet explorer satellite ( iue ) _ , and so will not be considered further here . however , the remaining visual companion , @xmath11 ori ab ( hei 42 ab ) , is very close , and its light will contribute to the spectra we discuss below . this companion was discovered by @xcite in 1978 at a separation of @xmath13 , and subsequent speckle interferometry @xcite indicates a linear motion widening by 5.7 mas y@xmath14 . the _ hipparcos _ satellite found this companion at a separation of @xmath15 in 1993 and determined a magnitude difference between the aa and ab components of @xmath16 mag @xcite . the bright central object , @xmath11 ori a ( o9.5 ii , @xcite ) , is a single - lined spectroscopic binary with a period of @xmath17 @xcite . orbital solutions exist which span the last century @xcite , and these are summarized by @xcite who present the most recent solution based upon 44 high dispersion spectra from _ iue_. in addition , @xcite reanalyzed all the published data and found convincing evidence of a regular advance in the longitude of periastron , giving an apsidal period of @xmath18 y. spectroscopic investigations to date have failed to identify clearly the secondary and tertiary ( hei 42 ab ) spectral components in the composite spectrum . the first detection of the secondary was claimed by @xcite who observed its orbital motion in weak extensions of @xmath19 in 6 of their 140 photographic spectra ( with a semiamplitude ratio of @xmath20 ) . @xcite reported evidence of the secondary star s contribution in the @xmath21 profile , and found the antiphase motion amounting to @xmath22 . @xcite estimated a spectral type of b1 for the secondary and a magnitude difference of @xmath23 mag in the optical region . the central binary also exhibits partial eclipses , and @xcite have analyzed both their photometric observations and the available , historical light curves . the analysis is made difficult by the presence of intrinsic light variability in one or more of the components and the presence of third light . nevertheless , @xcite were able to make a reasonable fit of the light curve by assuming a detached ( but near roche - filling ) configuration . their best fit model led to a primary of @xmath24 , @xmath25 , and @xmath26 k , a secondary of @xmath27 , @xmath28 , and @xmath29 k , an optical magnitude difference @xmath30 mag , and an orbital inclination of about @xmath31 . however , @xcite had to rely on heintz s visual report of a third light magnitude difference of @xmath32 mag , implying a much brighter ab component than found by _ hipparcos _ ( @xmath33 mag ) , and this suggests that their corrected eclipse depths are too deep . in this paper we present our own analysis of the 60 available _ iue _ high - dispersion swp spectra of @xmath11 ori a plus a set of spectra of the @xmath0 line . in the next section ( 2 ) , we describe the _ iue _ observations and our cross - correlation methods for radial velocity measurement . the optical spectra and radial velocity measurements are presented in 3 . a new double - lined orbit is given in 4 along with a discussion of the apsidal motion and long term changes in systemic velocity . next ( 5 ) , we use doppler tomography to separate the individual component spectra , which are then examined to obtain mk classifications , projected rotational velocities , and relative flux contributions . we analyze the _ hipparcos _ light curve in 6 , and discuss in 7 the evolutionary implications of the resulting masses and other parameters . there are 60 high - dispersion , short wavelength prime ( swp ) camera spectra of @xmath11 ori a available in the _ iue _ final archive at the space telescope science institute s multi - mission archive . these spectra have a signal - to - noise ratio ( s / n ) of approximately 10 per pixel . they were made with both the large and small apertures . they were retrieved in newsips format @xcite . a log of observations , including the swp image number and the heliocentric time of mid - observation appears in table 1 . once we retrieved the spectra , they were processed using local procedures written in idl that assemble the spectra into an array of dimensions of wavelength and time . during this process , the spectra are smoothed from the nominal 30 km s@xmath14 fwhm resolution of the raw spectra to 40 km s@xmath14 fwhm . the spectra are rectified using a set of relatively line - free zones , and resampled using a uniform log @xmath34 wavelength grid ( in increments equivalent to @xmath35 ) . the major interstellar lines are used to co - align the spectra on a common velocity frame , and then these lines are replaced by straight - line segments . details are given in @xcite . we then cross - correlated the spectra with the spectrum of a narrow - lined reference template star , ae aurigae ( hd 34078 ; o9.5 v ; @xmath36 ; @xmath37 , @xcite ) . the template was produced from an average of 6 swp spectra of ae aur prepared using the same methods that we used to produce the @xmath11 ori matrix of spectra . the cross - correlation was performed over the velocity range @xmath38 to @xmath39 relative to the reference star to produce a matrix of cross - correlation functions ( the composite ccfs ) of dimensions relative velocity and time . the regions around the broad features of ly@xmath40 , @xmath41 , and @xmath42 were set to unit intensity ( featureless continuum ) before cross - correlation to insure that the resulting ccfs have shapes similar to the rotationally broadened photospheric lines . finally the ccfs were rectified by division of a straight line fit made of the extreme ends of the ccfs . this step makes possible the direct comparison of ccfs made from spectra of varying quality ( s / n ratio ) , but we caution that in this representation the ccf central depths may vary depending on spectrum quality . the final ccfs are presented in figure 2 as a function of absolute radial velocity and orbital phase . the orbital motion of the primary star ( aa1 ) is clearly evident in the varying position of the central minimum and accompanying broad wings of the ccfs ( termed `` pedestal '' by @xcite ) . there is no obvious indication of a secondary ( or a stationary tertiary ) component , but we demonstrate below that the shapes of the ccfs are indeed influenced by the moving secondary component . we first attempted to measure radial velocities for the primary component by fitting a parabola through the 5 points ( spanning @xmath43 ) surrounding the ccf minimum . this procedure proved unsatisfactory because the residuals from the fit were rather large ( due to fitting only a small portion of the ccf ) and because inspection of the quadrature phase ccfs showed asymmetrical extensions indicating the presence of a partially blended , weak secondary component . our solution to the fitting problem involved determining an empirical shape for the ccf component of the primary and a constrained gaussian function for the secondary , and these components were then fit in a least - squares model to determine the positions ( and hence radial velocity shifts ) for each component . the following paragraphs describe the details of the fitting procedure . our final measurements are listed in table 1 , and are discussed below ( 4 ) . the ccfs are dominated by the central core and broad wings of the primary component , and our first task was to isolate the functional shape of this component . during orbital phases when the primary is significantly blue - shifted , the short wavelength half of the profile is mainly clear of the blending influence of the secondary ( or a possible stationary third component ) while the long wavelength side is free of blending at the opposite red - shifted phases . we therefore formed two versions of the primary component s profile by using a shift - and - add algorithm and our preliminary orbital solution to obtain the mean profile near the times of each quadrature . these two means were then averaged using window functions of the form @xmath44 ( with @xmath45 ) for the blue - shifted mean and @xmath46 for the red - shifted mean , to obtain a final primary reference ccf which varied smoothly through the line core ( illustrated in fig . 3 ) . the next step was to define a functional shape for the secondary component in the ccf . this was done by fitting the primary reference ccf to the best resolved quadrature phase ccfs over the range from the ccf minimum to the high velocity wing in the direction away from the secondary . this fit was then subtracted away , and the residual ccf due to the secondary alone was fit with an unconstrained gaussian profile . there were no obvious differences between the derived gaussian widths or depths of the fits from the two quadratures , so we averaged all the gaussian fit parameters to arrive at a mean standard deviation width , @xmath47 , and ratio of central depth to primary reference ccf depth of @xmath48 . we adopted these parameters to fix the shape of the secondary ccf component . the final step was to fit the sum of these profiles to each individual ccf . this was accomplished using a non - linear , least - squares fitting of the primary and secondary functional curves based upon three parameters , the primary and secondary radial velocities and a normalization factor for overall ccf depth ( to account for varying quality of the input spectra ) . an example of the detailed fitting procedure is illustrated in figure 3 which shows the model primary and secondary functions and how their sum gives excellent agreement with the observed ccf ( for a phase near primary velocity maximum ) . this plot demonstrates the importance of defining the shape of the primary component wings in order to place correctly the secondary component . the residuals formed after removal of only the primary component in these fits are shown in figure 4 . this plot shows that although the secondary component is weak , it is nevertheless found in most of the ccfs , moves with the expected period , and is antiphase to the primary s orbit . the average ccf variance per pixel in the inner @xmath49 km s@xmath14 of the velocity range in figure 4 , the range that bounds the secondary s doppler shifts , is 1.5% ( or about one sixth of the height of the scale bar illustrated ) based upon differences in ccfs from multiple spectra obtained sequentially . this is @xmath50 of the the secondary s depth in the residual ccfs , so the secondary s ccf contribution is slightly greater than the noise level in individual pixels . the final velocity measurements on the absolute scale are listed in table 1 , and the radial velocity curves are illustrated in figure 5 . these results indicate that the secondary s contribution to the composite ccf is always weak and partially blended with the primary s ccf component . it is therefore important to review the limitations of our methods and the possible systematic errors that may be present in these difficult measurements . the first shortcoming is that there probably exists an additional stationary component in the composite ccfs that is not included in our simple two - component model . the grayscale portion of figure 4 , for example , indicates that there is a very weak , narrow , and stationary feature in the residuals near a radial velocity of @xmath51 which we suggest results from a correlation between unexcised interstellar lines in the spectrum of @xmath11 ori with their counterparts in the ae aur photospheric spectrum ( since this velocity corresponds to the radial velocity of ae aur ) . it is also possible that there is a weak , stationary ccf component near the systemic velocity due to the tertiary ( although there is no evidence of this in fig . 4 nor from tomographic experiments described below ) . any such stationary feature would tend to pull the two - component fits back towards the stationary component , so that the velocity excursions would be underestimated in our measurements . we did a numerical experiment to estimate the size of this error by artificially removing from the composite ccfs a stationary component to represent a possible tertiary contribution . we assumed that the third component had a shape like the primary s reference ccf but was reduced to 10% of the primary s central depth and centered at the systemic velocity . we then repeated the measurement process by forming a primary reference profile from the quadrature phase ccfs , fitting a gaussian to the isolated secondary component , and then fitting all the adjusted ccfs for the positions of primary and secondary . these fits were generally less satisfactory and we do not list the individual velocities here , but the orbital elements based upon this experiment are given in the fourth column of table 3 under the heading `` adjusted uv ccfs '' . a comparison with our nominal set of elements ( third column of table 3 ) shows that the semiamplitudes may be underestimated ( by @xmath52 8% for @xmath53 and 14% for @xmath54 in this particular case ) if such a third component is present . this effect , if fully present in both components , could raise our estimated masses by 43% ( 6 ) . a second area of concern is the parameterization of the secondary component s width and depth . our choices for these parameters appear to be supported by observations of the optical line @xmath0 ( 3 ) , but the results are sensitive to the values of these parameters . if the secondary profile is set to be wider and/or deeper , then the observed ccfs are fit with smaller radial velocity excursions . for example , we made a test with a secondary gaussian component with a mean standard deviation , @xmath55 , and a ratio of central depth to primary reference ccf depth of @xmath56 , and the resulting measurements led to a secondary semiamplitude of @xmath57 which is much smaller than our nominal ( preferred ) value of @xmath58 . finally , the reader should be aware of potential problems resulting from our assumption that the component ratios and functional shapes are constant with orbital phase . at the conjunction phases for example , the partial eclipses will change the relative contributions of the primary and secondary components so that our fixed ccf depth ratio may lead to systematic errors in the velocity placement . furthermore , the photometric light curve indicates that the primary is close to roche filling @xcite , and consequently we expect that the primary s photospheric lines will show subtle width variations , appearing slightly narrower ( wider or asymmetric ) at conjunctions ( quadratures ) . however , a comparison of our primary reference ccf ( which was formed from the quadrature ccfs ) with conjunction phase ccfs indicates only minor differences so we doubt that our radial velocity measurements are badly in error at conjunctions . the binary was also observed in the optical by @xcite in a program to search for incipient h@xmath40 emission from colliding winds . @xcite found that h@xmath40 is sometimes distorted by residual emission effects , but @xmath59 is relatively free of emission problems most of the time . there are 20 spectra in this set which were obtained with the kitt peak national observatory coude feed telescope and the mount stromlo observatory 74-inch telescope @xcite , and these spectra generally have better resolution ( @xmath60 ) and s / n ratio ( @xmath61 pixel@xmath14 ) than the _ iue _ ccfs discussed above . the @xmath59 profiles are illustrated as a function of orbital phase in figure 6 . there appears to be a weak extension present at both quadratures that is probably the signature of the secondary ( although part of the red wing depression is due to a weak @xmath62 feature in the primary s spectrum ) . the @xmath59 profiles were fit in the same way as the uv ccfs by forming a mean quadrature profile for the primary and fitting the residual secondary component with a gaussian with a mean standard deviation , @xmath63 and ratio of central depth to primary depth of @xmath64 ( both of these parameters are identical within errors with those found from the uv ccfs ) . the fits were generally good , and we show in figure 7 the residuals after the removal of the fitted primary component alone . the motion of the secondary component is clearly seen in these residuals . the radial velocities from the two - component fit are listed in table 2 and plotted in figure 8 . note that we zero weighted three sets of measurements in the orbital solutions below that corresponded to profiles which were visibly contaminated with the emission pattern seen in h@xmath40 . we solved for the orbital elements using the non - linear , least - squares fitting program described by @xcite . the fit progressed in several steps . first , we fit the uv ccf radial velocities of the primary component by solving for all six orbital elements . since the radial velocity measurements of the secondary generally have larger errors , we then fixed the values of the period , @xmath65 , epoch of periastron , @xmath66 , eccentricity , @xmath67 , and longitude of periastron , @xmath68 , to those from the primary s solution , and solved only for the secondary systemic radial velocity , @xmath69 , and semiamplitude , @xmath70 . in table 3 , we compare the results from the uv ccf velocities ( column 3 ) with those of @xcite ( column 2 ) . this table also lists the elements resulting from a numerical experiment ( 2 ) which shows the influence of a potentially unaccounted for tertiary component in our fitting of the uv ccfs ( column 4 ) . we have relatively fewer radial velocities from fits of the @xmath59 profile , and so we fixed the period and eccentricity at the values from the uv ccf solution in fitting the @xmath59 data ( column 5 ) . the computed radial velocity curves are plotted in figures 2 and 4 8 . our results ( with the exception of @xmath68 and @xmath69 ; see below ) are generally in good agreement with the single - lined spectroscopic orbital solution found by @xcite , and our formal errors are similar to or smaller than theirs . we give in table 3 the mean sidereal period rather than the anomalistic period given by @xcite . our derived period agrees with theirs within errors , although we find somewhat better agreement with the photometric period of 5.732476 d adopted by @xcite . the largest errors are associated with the secondary semiamplitude , @xmath54 , which unfortunately propagates into a large error in the quantity @xmath71 . the root - mean - square ( rms ) residuals are smaller for the secondary radial velocity measurements from the higher quality @xmath59 profiles . our value of @xmath54 is much smaller than first claimed by @xcite but is slightly larger than that found by @xcite . given the faintness and line blending problems associated with the secondary s lines , this kind of disagreement is not surprising . the greatest difference between our results and prior work is the evident increase in longitude of periastron , @xmath72 , due to apsidal motion . @xcite performed a comprehensive non - linear , least squares analysis of the entire body of 546 historical and new velocities using @xmath73 as a free parameter and fixing the systemic velocity at @xmath74 . they found an apsidal motion of @xmath75 @xmath76 y@xmath14 , giving a period of @xmath18 y , in excellent agreement with an earlier comprehensive result of @xmath77 y from @xcite . we have added our results from the expanded _ iue _ ccf and @xmath0 solutions to the historical derivations of @xmath72 given in @xcite , and the entire set is plotted as a function of time in figure 9 . the error weighted , linear fit is also shown which yields the same rate for the apsidal advance as found by @xcite , @xmath78 y@xmath14 , giving an apsidal period of @xmath79 y. note that apsidal advance in @xmath11 ori may result from both the tidal forces between the two components of the close binary and the influence of the orbiting third body . it is possible that the reflex motion of the central binary in its orbit about the center of mass of the binary - tertiary system might produce measurable , long - term variations in the systemic velocity of the central binary . we show in figure 10 estimates of the systemic velocity from solutions based on data spanning the last century @xcite plus our own estimates from table 3 . there is a suggestion in this plot of a real variation in which the systemic velocity reached a minimum around 1940 and is currently increasing . the _ iue _ solution corresponds to the highest point in this graph , and we caution that the apparent discrepancy may be due in part to systematic differences between radial velocities for the lines in the uv and optical . @xcite demonstrated that velocity progressions exist in hot , luminous stars between lines formed deep in the photosphere and those of lower excitation formed higher in the atmosphere where acceleration into the stellar wind is occurring . many of the strong optical lines ( the balmer series and lines ) probably reflect the outflow as a negative radial velocity offset compared to the higher excitation lines measured in the uv ccfs ( formed at higher temperature deeper in the photosphere ) , and thus , the systemic velocity from the optical studies would be lower than that from our _ iue _ solution . we note that @xcite found no measurable velocity excitation correlation in his optical spectra of @xmath11 ori , but a difference of @xmath35 between the radial velocities of the low and high excitation lines is smaller than his threshold for detection . we used an iterative doppler tomography algorithm @xcite to reconstruct the spectra of the individual component stars in the @xmath11 orionis a system . the algorithm assumes that each observed composite spectrum is a linear combination of spectral components with known radial velocity curves ( table 3 ) and flux ratios that are constant across the spectral range and throughout the orbital cycle . the latter condition is violated by the partial eclipses that occur at conjunctions ( 5 ) , but since the eclipses are shallow and only a minority of our spectra were obtained during eclipse phases , flux constancy is a reasonable assumption . we estimated the uv flux ratio , @xmath80 , from the ratio of the areas of gaussian fits of the uv ccf reference profiles . note that the actual error is probably larger than quoted since we have not accounted for changes in the ccf amplitude related to differences in the match of the ccf template spectrum of ae aur to the slightly different spectral patterns of the primary and secondary . likewise , we estimated the optical band flux ratio as @xmath81 based upon the ratio of the @xmath0 reference component equivalent widths . again , this estimation does not account for differences in equivalent width of @xmath0 between the spectral types of the primary and secondary . we took the _ hipparcos _ magnitude difference for the tertiary to estimate its optical flux ratio as @xmath82 . note that the tertiary is actually the second brightest star of the three in the visual region . we do not know the uv flux contribution of the tertiary , so we produced uv reconstructions assuming both zero flux and the optical flux ratio . the reconstructed spectra in the vicinity of @xmath0 are illustrated in figure 11 . the primary and secondary components both have a strong @xmath0 feature , and @xmath62 is also visible in the red wing of the primary s profile . the tertiary , on the other hand , has a very shallow and broad @xmath0 feature which is shifted redward by @xmath83 compared to the systemic velocity of the central binary . the equivalent width of the profile in the tertiary reconstruction is comparable to those of the primary and secondary , and this suggests that the tertiary has an early b - type classification . the extreme width of the reconstructed tertiary profile indicates that either it is a rapid rotator ( @xmath10 ) or that it is a spectroscopic binary with a radial velocity range similar to the line width in the reconstruction . a portion of reconstructed uv spectrum of the primary appears in figure 12 , and the s / n in the reconstructed spectrum is excellent thanks to the large number of spectra used in the reconstruction . @xcite developed a system of spectral classification in the uv based upon the equivalent widths of a set of prominent photospheric spectral lines that are sensitive to temperature and luminosity . we used this method with the reconstructed primary spectrum to derive a classification of o9.5 iii based on the two component reconstruction or b0 i based on the three component reconstruction ( in which the primary has deeper lines ) using the following criteria : the equivalent widths of @xmath84 , @xmath85 , @xmath86 , @xmath87 , @xmath88 , and the line ratios of @xmath86/ @xmath85 and @xmath89/ @xmath85 . the luminosity class is set by the equivalent width of @xmath90 which is used to assign a class of i , iii , or v. these uv - based criteria are fully consistent with the optical classification of o9.5 ii @xcite . the depths of the lines in the reconstructed spectrum of the primary offer a means to check on our assumed flux contribution of the tertiary . if we assigned too much flux to the tertiary , for example , then the primary s lines will appear too deep in the reconstructed spectrum . we can test this possibility through a direct comparison of the reconstructed spectrum with that of a single star of the same spectral classification and rotational broadening . we selected 19 cep ( hd 209975 ) for this purpose which has a similar classification ( o9.5 ib ) but slightly narrower lines ( @xmath91 ; @xcite ) . we used the reference ccf widths to estimate the projected rotational velocities using the method of @xcite to arrive at @xmath92 and @xmath93 for the primary and secondary , respectively . then we artificially broadened the spectrum of 19 cep to match the line broadening of the primary . the resulting spectrum of 19 cep ( _ dotted line _ ) is compared to that of the primary in figure 12 for the cases where the tertiary contributes no uv flux ( _ thick solid line _ ) and where it contributes the fraction measured in the optical ( _ thin solid line _ ) . taken at face value , this comparison indicates that the tertiary contributes little if any uv flux . the reconstructed secondary spectrum is noisy because of the star s relative faintness , and this reconstruction is also marred by the effects of incomplete removal of the echelle blaze function ( `` ripple '' ) in the reduction software ( fig . furthermore , several features useful for spectral classification in b - type stars were excised during the process of interstellar line removal ( e.g. , @xmath94 , @xmath95 ) . nevertheless , a number of prominent spectral features are discernible and indicative of an early - b spectral type . since the classification scheme of @xcite is valid only for o - type stars , we relied instead on the classification criteria described by @xcite and @xcite . @xcite found that the equivalent width of @xmath96 is luminosity sensitive in early - b stars , and the measured value in the secondary , @xmath97m@xmath98 indicates luminosity class iii which is consistent with the observed strength of other luminosity sensitive lines discussed by @xcite . based upon a comparison of the relative line strengths in the spectra of giants presented in @xcite , the spectral type falls in the range b0b1 using @xmath99 , b1b3 using @xmath84 , b0.5b1 using @xmath86 , b0b0.5 using @xmath90 , and b0.5b1 using @xmath100 . thus , the secondary has a uv spectrum consistent with a b0.5 iii classification . the secondary spectrum is compared in figure 13 to the co - added spectrum of a similar star , @xmath101 per ( b0.5 iv - iii , @xmath102 ; @xcite ) . the reconstructed secondary does exhibit p cygni type profiles of @xmath103 , @xmath41 , and @xmath42 indicative of an earlier or more luminous star , but we caution that the tomographic reconstructions are problematical in the immediate vicinity of these lines which often lack obvious orbital motion @xcite . our b0.5 iii classification is consistent with the optical detections of the secondary s lines in @xmath19 @xcite , @xmath21 @xcite , and in our tomographic reconstruction of @xmath0 ( fig . 11 ) . in spite of the fact that our algorithm has successfully detected tertiary components in the cases of @xmath11 cir @xcite and 55 uma @xcite , the reconstruction of the uv spectrum of the tertiary ( made assuming it was stationary over the time span of the _ iue _ observations ) is essentially featureless , so that its uv spectral features remain undetected . we show in figure 14 a portion of this spectrum plotted with that of the rapid rotator , @xmath104 oph ( o9.5 v , @xmath105 ; @xcite ) . this comparison spectrum demonstrates how rapid rotation causes most features to become blended into a pseudo - continuum , so that a rapidly rotating tertiary would be difficult to detect in the reconstruction . however , this comparison also shows that certain strong features ( such as @xmath90 ) should still be relatively deep in the uv spectra of rapid rotators , and their absence in the tertiary spectrum reinforces the impression derived from the depths of the primary s lines that the tertiary contributes only a minor fraction of the uv flux ( certainly less than it does in the optical ) . such a low uv flux contribution is puzzling given the relative brightness of the tertiary in the optical and the indication from the @xmath0 reconstruction that it is a b - type star . on the other hand , the lack of features in the tertiary spectrum suggests that the radial velocities derived from our two component fits of the uv ccfs should be relatively free of systematic errors due to line blending with the tertiary spectrum ( so that the trial orbital elements in column 4 of table 3 based on a tertiary with lines 10% as deep as the primary s represents an extreme and unlikely case ) . photometry from the _ hipparcos _ satellite @xcite provides a contemporary light curve for the central binary ( fig . 15 ) , and although this light curve suffers from gaps in coverage and additional intrinsic variability @xcite , we will compare it to model orbital variations to obtain preliminary limits on the system inclination , @xmath106 ( and hence the masses ) . we used the light curve synthesis code gensyn @xcite to produce model @xmath107-band differential light curves ( almost identical to differential _ magnitudes for hot stars ) . the code was written for binaries with circular orbits , and we used multiple runs to synthesize the varying orbital separation in an elliptical orbit @xcite . our approach was to make a constrained fit using as much data as possible from our spectroscopic results . the orbital parameters were taken from the uv ccf solutions ( using an interpolated longitude of periastron , @xmath108 , for the average time of observation , by 1991.2 ; see fig . the stellar temperatures and gravities were taken from @xcite for the primary ( @xmath109 k and @xmath110 ) , and for the secondary we used estimates from @xcite for a comparable star , @xmath101 per ( @xmath111 k and @xmath112 ) . we then estimated the physical fluxes and limb darkening coefficients from tables in @xcite and @xcite , respectively . the secondary s polar radius was set at @xmath113 ( where @xmath114 is the assumed primary s polar radius ) based upon our estimated magnitude difference , @xmath115 ( 5 ) , and the calculated surface fluxes . finally , we set the rotational periods of the stars based upon the assumed values of @xmath114 and @xmath106 and the measured projected rotational velocities ( 5 ) . each trial run of gensyn was defined by two model parameters , @xmath114 and @xmath106 , and all the resulting light curves were corrected for the presence of third light using the _ hipparcos _ flux ratio , @xmath116 . the _ hipparcos _ light curve is illustrated in figure 15 as a function of photometric phase based upon the predicted time of minimum light according to the ephemeris from the uv ccfs for the primary ( table 3 ) . we also show two model light curves which provide only marginally satisfactory fits but which correspond to the probable limits of our fitting parameters . the solid line is the model light curve for @xmath117 and @xmath118 which represents the largest size primary consistent with the fitting scheme . the primary in this model has a polar radius only 1% smaller than the critical roche radius at periastron . the roche radius can be somewhat larger than @xmath119 in models with lower inclination ( since the system dimensions are set by @xmath120 ) , but roche - filling models with lower inclination can not produce eclipses as deep as those observed , so @xmath118 represents a reliable lower limit for the inclination . this is fortuitously close to the inclination adopted by @xcite , @xmath121 , since they adopted very different values for the mass ratio and third light correction . this near roche - filling model is characterized by large ellipsoidal variations that shape the light curve outside of the eclipses . these appear to match the observations better in the photometric phase range 0.6 0.9 than in the range 0.2 0.4 . in fact , the observations in the 0.2 0.4 range are better fit by a smaller , less tidally distorted model primary , and the dashed line indicates such a model for @xmath122 and @xmath123 . this smaller primary model does a better job of matching the relative depths of the eclipses , but the duration of the eclipses may be too short especially when compared to the light curve observations of @xcite . thus , the inclination range @xmath124 appears to span the acceptable limits of plausible fits . this range corresponds to primary and secondary masses of @xmath125 and @xmath126 , respectively ( based upon the uv ccf orbital solutions ) . the _ iue _ ccfs and @xmath0 line profiles have provided us with the first clear and consistent measurements of the orbital motion of the secondary star . however , the secondary s resulting semiamplitude is surprisingly small , and consequently , our results indicate a binary with a smaller semimajor axis ( implying lower masses ) than expected . in particular , the mass of the primary is a factor of 3 to 4 below that predicted for a normal single star with its classification and stellar wind properties @xcite . such low mass results are not unprecedented . for additional examples of orbital masses that are significantly lower ( @xmath52 40% ) than theoretical expectations , see the cases of @xcite and @xcite . in addition , @xcite use model atmospheres to find masses for single o stars , and in some cases , they also find masses that are significantly lower than those expected from evolutionary tracks . to attain a primary mass of @xmath127 would require a 51% upward revision in the secondary s semiamplitude , to @xmath128 ( for @xmath118 and @xmath129 ) , placing the secondary s radial velocity range between @xmath130 and @xmath131 . this would appear to be ruled out by the observed quadrature line profiles ( see figs . 4 and 7 ) . our derived value of @xmath54 ( and the small masses implied ) clearly requires additional verification ( through high s / n and high dispersion spectroscopy of metal lines unaffected by emission , for example ) , but taken at face value , our results suggest that the current state of the primary is the result of extensive mass loss caused by an interaction with the secondary . the absolute magnitudes of the components provide a reference point to assess the apparent discrepancy of the components . the three bright belt stars in orion , @xmath11 ori , @xmath101 ori ( b0 ia ) , and @xmath104 ori ( o9.7 ib ) , form the high luminosity end of the ori ob1b subgroup @xcite which has an estimated age between 1.7 my @xcite and 7 my @xcite . the distance to ori ob1b is estimated to be @xmath132 pc based on photometry @xcite or @xmath133 pc based on _ hipparcos _ parallax measurements @xcite . outside of eclipse , the apparent magnitude of @xmath11 ori is @xmath134 ( see fig . 15 and the @xmath135 to @xmath107 transformation in @xcite ) , which for an extinction of @xmath136 mag @xcite , gives the absolute magnitude of the triple as @xmath137 at @xmath138 pc or @xmath139 at @xmath140 pc . the absolute magnitude is also calculated in the light curve model ( 6 ) based on the assumed sizes and @xmath107-band fluxes of the stars . the predicted absolute magnitudes for the triple are @xmath137 for the @xmath117 model ( near roche - filling ) and @xmath141 for the @xmath122 model . thus , we can obtain consistency between these estimates if we adopt the closer distance for ori ob1b and the larger radius primary model for the binary . a distance of 360 pc corresponds to a parallax of 2.78 mas , which falls within one standard deviation of the hipparcos parallax for @xmath11 ori , @xmath142 mas . the component absolute magnitudes are then @xmath143 for the primary , @xmath144 for the secondary , and @xmath145 for the tertiary . the adopted temperatures and radii result in @xmath146 pairs of ( 4.52 , 5.26 ) and ( 4.44 , 4.08 ) for the primary and secondary , respectively . the non - rotating single star evolutionary tracks of @xcite that pass through the observed @xmath146 points correspond to current masses of about @xmath147 and @xmath148 ( and ages of 4 and 3 my ) for the primary and secondary , respectively . evolutionary models that include rotation @xcite indicate that rapid rotators with temperatures and luminosities comparable to the components of @xmath11 ori aa may appear as much as 0.1 dex more luminous than non - rotating stars of the same mass . by applying an artificial reduction in luminosity of this amount , the resulting evolutionary masses would be approximately @xmath149 and 12 for the primary and secondary , respectively . all of these mass estimates are much larger than those found for our @xmath117 light curve model , where we obtained maximum masses of @xmath150 and 5.6 for the primary and secondary , respectively . thus , both stars are extremely overluminous for their orbitally determined masses . we rule out stellar wind losses as the main mass loss mechanism , because the amount of mass to be removed from the primary , @xmath151 , is much larger than the wind can carry in the time available ( the current mass loss rate is estimated to be @xmath152 = -6.3 $ ] ; @xcite ) . the winds are even weaker for stars in the secondary s mass range . we believe that we can accommodate these mass losses within the framework of current models of roche lobe overflow ( rlof ) . the relative youth of the system and the ori ob1b association suggests that binary mass loss occurred while the stars were still in the core hydrogen burning ( chb ) stage ( case a mass transfer ) . if either star had begun extensive h shell burning or core he burning , we would probably find an enrichment in the products of nuclear burning in one or both stars due to mass transfer or extensive mixing , but in fact , the abundances of both he @xcite and n @xcite in the primary appear to be solar - like . there is no compelling evidence of the strong h@xmath40 emission associated with rlof gas streams at the present time @xcite , and so , we tentatively conclude that @xmath11 ori aa is a post - rlof object that has recently completed case a rlof . @xcite present a series of models of case a evolution based on the assumption of conservative mass transfer , and their results shed some light on the possible evolutionary status of @xmath11 ori aa . rlof begins during the slow expansion of the donor during chb , and continues until the mass ratio is reversed and the system separation grows . the mass donor may end up as a very luminous and hot star at the conclusion of rlof even though it has lost most of its mass @xcite , and the mass gainer will then appear as the more luminous star ( and perhaps overluminous for its mass due to accretion induced mixing ; @xcite ) . if this scenario is applied to @xmath11 ori aa , then we must identify the large , o9.5 ii primary as the mass gainer and the smaller , b0.5 iii secondary as the mass donor ( the originally more massive star ) . the main difficulty with this explanation is the extreme overluminosity of the primary star ( the presumed mass gainer ) . we showed above that @xmath153 primary is radiating a luminosity comparable to that expected for a star more than twice as massive , and it remains to be seen if processes like accretion induced mixing could account for such a large overluminosity . an alternative and more speculative explanation can be constructed based upon non - conservative rlof . if we assume that the high luminosity of the primary indicates it began life as a @xmath154 star , then we must conclude that most of its mass loss left the system entirely since the secondary is too small to have accreted any significant amount of mass . in this scenario the system started with a relatively large mass ratio , @xmath155 , and models suggest that such systems may suffer mergers in a common envelope phase @xcite . it is possible that the @xmath11 ori aa system experienced large scale mass loss during a common envelope phase , and the decrease in system mass caused an increase in binary separation , thereby avoiding a merger . the unusual infrared excess of @xmath11 ori @xcite may be a relic of this substantial mass loss , and it would be worthwhile to search for any other evidence of large quantities of ejected gas . finally we return to the mystery of the tertiary star . the tomographic reconstruction of the @xmath0 profiles produced a very broad and shallow feature for the tertiary spectrum ( fig . this result should be treated with some caution , since this feature is subject to distortion by weak emission , but it would appear to indicate that tertiary is a b - type star . the width of the line suggests that it is either a rapid rotator or a spectroscopic binary ( averaged at many velocity displacements to yield a net broadened profile in the reconstruction ) . according to the absolute magnitude calibration of @xcite , a main sequence star with the observed absolute magnitude of @xmath145 has a mass of @xmath52 @xmath156 ( corresponding to an o8.5 v classification ; @xcite ) . on the other hand , if the tertiary were composed of two identical stars of absolute magnitude @xmath157 , the stars would each have a mass of @xmath52 @xmath158 ( type b0.5 v ) . in either case , we would expect the tertiary to produce a significant uv flux contribution , comparable to or larger than the secondary s flux . our failure to detect the tertiary s uv spectral signature remains unexplained . the tertiary is probably gravitationally bound to the central binary since otherwise its angular motion would have made it a striking double to visual observers in the early centuries of telescopic observation . consequently , we encourage continued spectroscopic and high resolution observations to find the reflex orbital motion hinted at in figure 10 . support for this work was provided in part by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . institutional support was provided from the gsu college of arts and sciences and from the research program enhancement fund of the board of regents of the university system of georgia administered through the gsu office of the vice president for research and sponsored programs . we gratefully acknowledge all this support . this research made use of data obtained from the multimission archive at the space telescope science institute ( mast ) , nasa s astrophysics data system abstract service , the simbad astronomical database of the centre de donnes astronomiques de strasbourg , the washington double star catalog and the third catalog of interferometric measurements of binary stars , both maintained at the u.s . naval observatory , and the national institute of standards and technology s atomic spectra database , version 2.0 . we would like to thank an anonymous referee for a careful reading of the paper and some very helpful suggestions . we thank dr . anthony g. a. brown for helpful comments . finally , we made use of an eclipsing binary simulator , nightfall , written by dr . rainer wichmann , and freely available via the internet . bagnuolo , w. g. , jr . , gies , d. r. , hahula , m. e. , wiemker , r. , & wiggs , m. s. 1994 , , 423 , 446 bieging , j. h. , abbott , d. c. , & churchwell , e. b. 1989 , , 340 , 518 blaauw , a. 1991 , in the physics of star formation and early stellar evolution , nato advanced science institutes ( asi ) series c , vol . 342 , ed . c. j. lada & n. d. kylafis ( dordrecht : kluwer ) , 125 brown , a. g. a. 1996 , ph.d . thesis , univ . of leiden brown , a. g. a. , de geus , e. j. , & de zeeuw , p. t. 1994 , , 289 , 101 curtiss , r. h. 1914 , publ . obs . , 1 , 118 de zeeuw , p. t. , hoogerwerf , r. , de bruijne , j. h. j. , brown , a. g. a. , & blaauw , a. 1999 , , 117 , 354 fullerton , a. w. 1990 , ph.d . thesis , univ . of toronto galkina , t. s. 1976 , izvestiia krymskaia astrofiz . , 54 , 128 garhart , m. p. , smith , m. a. , turnrose , b. e. , levay , k. l. , & thompson , r. w. 1997 , international ultraviolet explorer new spectral image processing system information manual , version 2.0 ( greenbelt : nasa ) gies , d. r. 1995 , in iau symposium 163 , wolf - rayet stars : binaries , colliding winds , evolution , ed . k. a. van der hucht & p. m. williams ( dordrecht : kluwer ) , 373 gies , d. r. , & bolton , c. t. 1986 , , 61 , 419 gies , d. r. , et al . 1999 , , 525 , 420 harmanec , p. 1988 , 39 , 329 harmanec , p. 1998 , , 335 , 173 harries , t. j. , hilditch , r. w. , & hill , g. 1998 , , 295 , 386 hartkopf , w. i. , mason , b. d. , barry , d. j. , mcalister , h. a. , bagnuolo , w. g. , jr . , & prieto , c. m. 1993 , , 106 , 352 hartmann , j. 1904 , , 19 , 268 harvey , a. s. , stickland , d. j. , howarth , i. d. , & zuiderwijk , e. j. 1987 , observatory , 107 , 205 heger , a. , & langer , n. 2000 , , 544 , 1016 heintz , w. d. 1980 , , 44 , 111 herrero , a. , corral , l. j. , villamariz , m. r. , & martin , e. l. 1999 , , 348 , 542 herrero , a. , puls , j. , & villamariz , m. r. , 2000 , , 354 , 193 hnatek , a. 1920 , astronomisches nachrichten , 213 , 17 howarth , i. d. , & prinja , r. k. 1989 , , 69 , 527 howarth , i. d. , siebert , k. w. , hussain , g. a. j. , & prinja , r. k. 1997 , , 284 , 265 hutchings , j. b. 1976 , , 203 , 438 jordan , f. c. 1914 , publ . allegheny obs . , 3 , 125 koch , r. h. , & hrivnak , b. j. 1981 , , 248 , 249 kurucz , r. l. 1994 , solar abundance model atmospheres for 0 , 1 , 2 , 4 , 8 km / s , kurucz cd - rom no . 19 ( cambridge , ma : smithsonian astrophysical obs . ) liu , n. , et al . 1997 , , 485 , 350 luyten , w. j. , struve , o. , & morgan , w. w. 1939 , publ . yerkes obs . , 7 , 256 mason , b. d. , gies , d. r. , hartkopf , w. i. , bagnuolo , w. g. , jr . , ten brummelaar , t. , & mcalister , h. a. 1998 , , 115 , 821 massa , d. 1989 , , 224 , 131 mcalister , h. a. , hartkopf , w. i. , hendry , e. m. , campbell , b. g. , & fekel , f. c. 1983 , , 51 , 309 mcalister , h. a. , hartkopf , w. i. , hutter , d. j. , & franz , o. g. 1987 , , 93 , 688 mcalister , h. a. , hartkopf , w. i. , sowell , j. r. , dombrowski , e. g. , & franz , o. g. 1989 , , 97 , 510 mcalister , h. a. , & hendry , e. m. 1982 , , 49 , 267 mcalister , h. a. , mason , b. d. , hartkopf , w. i. , & shara , m. m. 1993 , , 106 , 1639 meynet , g. , & maeder , a. 2000 , , 361 , 101 miczaika , g. r. 1952 , zeitschrift fur astrophysik , 30 , 299 mochnacki , s. w. , & doughty , n. a. 1972 , , 156 , 51 monet , d. g. 1980 , , 237 , 513 morbey , c. l. , & brosterhus , e. b. 1974 , , 86 , 455 natarajan , v. , & rajamohan , r. 1971 , kodaikanal bull . , 208 penny , l. r. 1996 , , 463 , 737 penny , l. r. , gies , d. r. , & bagnuolo , w. g. , jr . 1997 , , 483 , 439 penny , l. r. , gies , d. r. , & bagnuolo , w. g. , jr . 1999 , , 518 , 450 penny , l. r. , et al . 2001 , , 548 , 889 perryman , m. a. c. 1997 , the _ hipparcos _ and _ tycho _ catalogues ( esa sp-1200 ) ( noordwijk : esa ) pismis , p. , haro , g. , & struve , o. 1950 , , 111 , 509 prieur , j .- l . , oblak , e. , lampens , p. , kurpinska - winiarska , m. , aristidi , e. , koechlin , l. , & ruymaekers , g. 2001 , , 367 , 865 reid , a. h. n. , et al . 1993 , , 417 , 320 rountree , j. , & sonneborn , g. 1993 , spectral classification with the international ultraviolet explorer : an atlas of b - type spectra ( nasa rp-1312 ) ( washington , dc : nasa ) runacres , m. c. , & bloome , r. 1996 , , 309 , 544 schaller , g. , schaerer , d. , meynet , g. , & maeder , a. 1992 , , 96 , 269 schnberner , d. , & harmanec , p. 1995 , , 294 , 509 singh , m. 1982 , , 87 , 269 tarasov , a. e. , et al . 1995 , a&as , 110 , 59 thaller , m. l. 1997a , , 487 , 380 thaller , m. l. 1997b , ph.d . dissertation , georgia state univ . vanbeveren , d. , van rensbergen , w. , & de loore , c. 1998 , the brightest binaries ( dordrecht : kluwer ) voels , s. a. , bohannan , b. , abbott , d. c. , & hummer , d. g. 1989 , , 340 , 1073 wade , r. a. , & rucinski , s. m. 1985 , a&as , 60 , 471 walborn , n. r. 1972 , , 77 , 312 walborn , n. r. 1976 , , 205 , 419 wellstein , s. , langer , n. , & braun , h. 2001 , , 369 , 939 worley , c. e. , & douglass , g. g. 1997 , , 125 , 523 43753.473 & 0.173 & -65.0 & 3.9 & 174.6 & -51.0 & swp02436 + 43755.821 & 0.583 & 94.8 & -1.2 & -158.5 & -61.1 & swp02477 + 43833.823 & 0.190 & -64.8 & 3.2 & 155.7 & -68.1 & swp03401 + 43898.073 & 0.398 & 2.8 & -6.2 & 161.3 & 88.3 & swp04018 + 44126.004 & 0.159 & -71.2 & -2.4 & 129.3 & -96.1 & swp06451 + 44146.137 & 0.671 & 113.5 & -4.6 & -163.0 & -22.4 & swp06677 + 44146.156 & 0.674 & 110.5 & -8.0 & -164.2 & -22.8 & swp06678 + 44146.840 & 0.794 & 104.9 & -4.1 & -165.6 & -42.7 & swp06682 + 44146.859 & 0.797 & 103.8 & -4.2 & -145.9 & -25.0 & swp06683 + 44146.878 & 0.800 & 99.8 & -7.2 & -170.6 & -51.7 & swp06684 + 44146.896 & 0.803 & 102.7 & -3.3 & -155.3 & -38.4 & swp06685 + 44499.006 & 0.227 & -60.0 & 2.1 & 156.3 & -55.9 & swp10153 + 44570.950 & 0.777 & 120.5 & 7.1 & -140.3 & -8.8 & swp10690 + 44631.629 & 0.362 & -10.3 & -1.1 & 163.2 & 54.5 & swp11163 + 44631.693 & 0.373 & -7.7 & -4.1 & 149.0 & 51.4 & swp11166 + 44631.711 & 0.377 & -4.0 & -2.0 & 148.6 & 54.1 & swp11167 + 44652.439 & 0.992 & -4.8 & -0.6 & 142.5 & 43.7 & swp13320 + 44652.504 & 0.004 & -12.1 & -0.6 & 134.3 & 21.1 & swp13323 + 44676.498 & 0.189 & -72.1 & -4.0 & 167.0 & -56.9 & swp13471 + 44678.501 & 0.539 & 77.9 & -0.8 & -183.9 & -120.4 & swp13491 + 44679.183 & 0.658 & 116.6 & 0.6 & -161.7 & -25.1 & swp13499 + 45987.070 & 0.811 & 105.2 & 1.7 & -163.6 & -51.5 & swp24169 + 45991.784 & 0.633 & 110.4 & -0.5 & -51.6 & 75.0 & swp24197 + 45991.947 & 0.661 & 114.5 & -2.1 & -103.6 & 34.2 & swp24202 + 45992.811 & 0.812 & 105.8 & 2.9 & -35.1 & 75.9 & swp24208 + 45993.041 & 0.852 & 90.5 & 5.1 & -104.0 & -27.3 & swp24214 + 45993.139 & 0.869 & 84.5 & 8.2 & -110.0 & -51.1 & swp24217 + 45993.831 & 0.990 & -7.2 & -4.6 & 172.4 & 76.6 & swp24230 + 45994.086 & 0.035 & -29.6 & 0.4 & 184.2 & 34.8 & swp24239 + 45994.168 & 0.049 & -38.7 & -1.0 & 146.1 & -18.4 & swp24241 + 46075.734 & 0.278 & -46.3 & 0.6 & 160.1 & -22.4 & swp24878 + 46076.521 & 0.415 & 14.3 & -3.5 & 108.7 & 52.9 & swp24884 + 46505.425 & 0.234 & -58.0 & 2.3 & 168.3 & -40.5 & swp27920 + 46507.309 & 0.563 & 87.4 & -1.2 & -66.3 & 16.6 & swp27931 + 46510.339 & 0.092 & -52.9 & 3.2 & 188.1 & -12.4 & swp27963 + 46511.414 & 0.279 & -42.1 & 4.1 & 200.4 & 19.3 & swp27975 + 46511.551 & 0.303 & -34.7 & 2.1 & 164.8 & 2.2 & swp27979 + 46511.682 & 0.326 & -28.9 & -2.2 & 127.3 & -15.5 & swp27982 + 46511.821 & 0.350 & -18.1 & -3.0 & 121.8 & 1.5 & swp27985 + 46512.345 & 0.442 & 39.0 & 7.3 & -11.2 & -39.7 & swp27990 + 46512.471 & 0.464 & 43.5 & 0.5 & 54.4 & 48.0 & swp27994 + 46512.609 & 0.488 & 61.8 & 6.8 & -69.7 & -52.6 & swp27998 + 46512.743 & 0.511 & 71.0 & 4.7 & -77.1 & -38.0 & swp28001 + 46515.438 & 0.981 & -0.7 & -4.0 & 172.2 & 88.0 & swp28028 + 46866.743 & 0.264 & -50.3 & 1.3 & 162.5 & -29.1 & swp30490 + 46867.370 & 0.374 & -5.7 & -2.2 & 186.5 & 89.0 & swp30496 + 46867.761 & 0.442 & 40.4 & 8.6 & -39.8 & -68.1 & swp30506 + 46868.314 & 0.538 & 82.4 & 4.0 & -4.5 & 58.4 & swp30513 + 46869.480 & 0.742 & 127.1 & 7.7 & -72.3 & 70.9 & swp30524 + 47044.515 & 0.276 & -45.6 & 2.0 & 188.6 & 4.7 & swp31727 + 47045.147 & 0.386 & 1.5 & -1.1 & 132.8 & 47.3 & swp31745 + 47045.640 & 0.472 & 47.2 & 0.1 & -34.3 & -32.7 & swp31760 + 47045.808 & 0.501 & 66.3 & 4.8 & -20.1 & 9.7 & swp31765 + 47046.176 & 0.565 & 87.7 & -1.7 & -116.8 & -32.3 & swp31774 + 47046.570 & 0.634 & 109.2 & -2.0 & -126.8 & 0.3 & swp31783 + 47046.868 & 0.686 & 117.7 & -2.0 & -135.8 & 8.0 & swp31791 + 47047.174 & 0.739 & 121.9 & 2.3 & -116.7 & 26.9 & swp31801 + 48546.812 & 0.342 & -23.5 & -4.3 & 168.8 & 40.6 & swp42747 + 48546.831 & 0.345 & -21.0 & -3.4 & 157.3 & 32.3 & swp42748 + 48546.851 & 0.349 & -20.2 & -4.3 & 173.7 & 52.0 & swp42749 49612.996 & 0.156 & -44.8 & -1.8 & 147.2 & 9.0 & kpno 1994 + 49612.997 & 0.156 & -45.8 & -2.9 & 148.2 & 10.2 & kpno 1994 + 49613.997 & 0.330 & 60.1 & -6.3 & -81.6 & -27.3 & kpno 1994 + 49615.866 & 0.656 & 91.1 & -3.8 & -114.0 & -9.6 & kpno 1994 + 49615.867 & 0.657 & 92.0 & -2.8 & -104.8 & -0.6 & kpno 1994 + 49617.922 & 0.015 & -80.5 & 8.8 & 235.2 & 15.5 & kpno 1994 + 50150.888 & 0.988 & -58.5 & 28.1 & 220.0 & 5.0 & mso 1996 + 50151.885 & 0.162 & -39.0 & 0.4 & 133.9 & 1.9 & mso 1996 + 50152.879 & 0.335 & 60.9 & -8.0 & -89.5 & -30.9 & mso 1996 + 50153.883 & 0.510 & 117.1 & -4.3 & -134.8 & 16.2 & mso 1996 + 50154.876 & 0.683 & 80.0 & -3.0 & -100.5 & -17.0 & mso 1996 + 50529.892 & 0.103 & -82.5 & -12.1 & 161.7 & -24.8 & mso 1997 + 50530.874 & 0.274 & 40.5 & 7.3 & 1.7 & -2.5 & mso 1997 + 50531.860 & 0.446 & 112.9 & 0.7 & -115.6 & 19.2 & mso 1997 + 50532.863 & 0.621 & 119.2 & 11.9 & -102.8 & 23.5 & mso 1997 + 50533.862 & 0.795 & 19.7 & 3.7 & 30.0 & -4.4 & mso 1997 + 50534.882 & 0.973 & -69.6 & 13.9 & 158.0 & -51.6 & mso 1997 + 50535.860 & 0.144 & -36.4 & 13.5 & 181.5 & 31.1 & mso 1997 + 50536.864 & 0.319 & 69.4 & 9.4 & -40.0 & 3.1 & mso 1997 + 50537.861 & 0.493 & 118.2 & -1.9 & -145.7 & 3.0 & mso 1997 + lcccc @xmath65 ( days ) & 5.732424 ( 50 ) & 5.732503 ( 26 ) & 5.732508 ( 28 ) & 5.732503 + @xmath66 ( hjd 2,400,000 ) & 30802.02 ( 9 ) & 46865.23 ( 10 ) & 46865.19 ( 11 ) & 50535.0 ( 4 ) + @xmath67 & 0.087 ( 9 ) & 0.075 ( 6 ) & 0.076 ( 6 ) & 0.075 + @xmath68 ( @xmath76 ) & 50 ( 6 ) & 112 ( 6 ) & 110 ( 7 ) & 173 ( 23 ) + @xmath53 ( km s@xmath14 ) & 97.9 ( 9 ) & 94.9 ( 6 ) & 102.5 ( 10 ) & 105 ( 4 ) + @xmath159 ( km s@xmath14 ) & 20.3 & 28.6 ( 4 ) & 27.5 ( 7 ) & 24 ( 3 ) + rms@xmath160 ( km s@xmath14 ) & & 4.2 & 4.9 & 7.6 + @xmath54 ( km s@xmath14 ) & & 186 ( 9 ) & 212 ( 11 ) & 186 ( 7 ) + @xmath161 ( km s@xmath14 ) & & 35 ( 7 ) & 36 ( 8) & 20 ( 4 ) + rms@xmath162 ( km s@xmath14 ) & & 51 & 59 & 18 + @xmath163 ( @xmath164 ) & & 8.7 ( 14 ) & 12.4 ( 20 ) & 9.3 ( 10 ) + @xmath165 ( @xmath164 ) & & 4.4 ( 3 ) & 6.0 ( 5 ) & 5.3 ( 7 ) + @xmath166 ( @xmath167 ) & & 31.7 ( 11 ) & 35.5 ( 13 ) & 32.9 ( 9 ) +

we comment on old and new results related to the destruction of a random recursive tree ( rrt ) , in which its edges are cut one after the other in a uniform random order . in particular , we study the number of steps needed to isolate or disconnect certain distinguished vertices when the size of the tree tends to infinity . new probabilistic explanations are given in terms of the so - called cut - tree and the tree of component sizes , which both encode different aspects of the destruction process . finally , we establish the connection to bernoulli bond percolation on large rrt s and present recent results on the cluster sizes in the supercritical regime . * key words : * random recursive tree , destruction of graphs , isolation of nodes , disconnection , supercritical percolation , cluster sizes , fluctuations .

1406.2238

imagine that we destroy a connected graph by removing or cutting its edges one after the other , in a uniform random order . the study of such a procedure was initiated by meir and moon in @xcite . they were interested in the number of steps needed to isolate a distinguished vertex in a ( random ) cayley tree , when the edges are removed uniformly at random from the current component containing this vertex . later on , meir and moon @xcite extended their analysis to random recursive trees . the latter form an important family of increasing labeled trees ( see section [ smaintools ] for the definition ) , and it is the goal of this paper to shed light on issues related to the destruction of such trees . mahmoud and smythe @xcite surveyed a multitude of results and applications for random recursive trees . their recursive structure make them particularly amenable to mathematical analysis , from both a combinatorial and probabilistic point of view . we focus on the probabilistic side . our main tools include the fundamental splitting property , a coupling due to iksanov and mhle @xcite and the so - called _ cut - tree _ ( see @xcite ) , which records the key information about the destruction process . the cut - tree allows us to re - prove the results of kuba and panholzer @xcite on the multiple isolation of nodes . moreover , we gain information on the number of steps needed to disconnect a finite family nodes . finally , we relate the destruction of a random recursive tree to bernoulli bond percolation on the same tree . we explain some results concerning the sizes of percolation clusters in the supercritical regime , where the root cluster forms the unique giant cluster . * figure 1 * in this section , we present some basic tools in the study of random recursive trees which will be useful to our purposes . consider a finite and totally ordered set of vertices , say @xmath0 . a tree on @xmath0 is naturally rooted at the smallest element of @xmath0 , and is called increasing if and only if the sequence of vertices along a segment from the root to an arbitrary vertex increases . most of the time we shall take @xmath1 , which induces of course no loss of generality . more precisely , it is convenient to introduce the following notion . for an arbitrary totally ordered set @xmath0 with cardinality @xmath2 , we call the bijective map from @xmath0 to @xmath3 which preserves the order , the canonical relabeling of vertices . plainly the canonical relabeling transforms an increasing tree on @xmath0 into an increasing tree on @xmath4 . such relabelings enable us to focus on the structure of the rooted tree without retaining specifically the elements of @xmath0 . a _ random recursive tree _ ( in short , rrt ) on @xmath5 is a tree picked uniformly at random amongst all the increasing trees on @xmath5 ; it shall be denoted henceforth by @xmath6 . in particular , @xmath6 has @xmath7 edges and size ( i.e. number of vertices ) @xmath8 . the terminology stems from the easy observation that a version of @xmath6 can be constructed by the following simple recursive random algorithm in which vertices are incorporated one after the other . the vertex @xmath9 is naturally connected by an edge to the root @xmath10 , then @xmath11 is connected either to @xmath10 or to @xmath9 with equal probability @xmath12 , and more generally , the parent of the vertex @xmath13 is chosen uniformly at random amongst @xmath14 and independently of the other vertices . this recursive construction is a close relative to the famous chinese restaurant construction of uniform random permutations ( see , for instance , section 3.1 in pitman @xcite ) , and in particular the number of increasing trees of size @xmath15 equals @xmath16 . another useful observation is that this recursive construction can be interpreted in terms of the genealogy of a yule process . recall that a yule process describes the evolution in continuous time of a pure birth process in which each individual gives birth to a child at unit rate and independently of the other individuals . we label individuals in the increasing order of their birth times , the ancestor receiving by convention the label @xmath10 . if we let the process evolve until the population reaches size @xmath15 , then its genealogical tree , that is the tree where individuals are viewed as vertices and edges connect children to their parent , is clearly a rrt . here is an application to percolation on @xmath6 which will be useful later on . [ l2 ] perform a bernoulli bond percolation on @xmath6 with parameter @xmath17 ( i.e. each edge of @xmath6 is deleted with probability @xmath18 , independently of the other edges ) , and let @xmath19 denote the size of the cluster containing the root . then @xmath20 where @xmath21 a.s . is some random variable . we view @xmath6 as the genealogical tree of a standard yule process @xmath22 up to time @xmath23 . it is well - known that the process @xmath24 is a martingale which converges a.s . to some random variable @xmath25 with the exponential distribution , and it follows that @xmath26 in this setting , performing a bernoulli bond percolation can be interpreted as superposing neutral mutations to the genealogical tree , namely each child is a clone of its parent with probability @xmath27 and a mutant with a new genetic type with probability @xmath18 , independently of the other children . neutrality means that the rate of birth does not depend on the genetic type . then the process @xmath28 of the number of individuals with the same genetic type as the ancestor is again a yule process , but now with birth rate @xmath27 . as a consequence @xmath29 where @xmath30 denotes another exponentially distributed random variable . we then observe that @xmath31 which completes the proof . plainly , the recursive construction can also be interpreted in terms of urns , and we conclude this section by exemplifying this connection . specifically , the size of the root cluster @xmath19 in the above lemma can be identified as the number of red balls in the following plya - hoppe urn . start with one red ball which represents the root of the tree . a draw is effected as follows : ( i ) choose a ball at random from the urn , observe its color , and put the ball back to the urn . ( ii ) if its color was red , add a red ball to the urn with probability @xmath27 , and add a black ball to the urn with probability @xmath18 . if its color was black , add another black ball to the urn . then , after @xmath7 draws , the number of red balls is given by @xmath19 , and in this way , lemma [ l2 ] yields a limit theorem for the proportion of red balls . the choice @xmath32 in this urn scheme corresponds to the usual plya urn . here , if one starts with one red ball and @xmath33 black balls , then the number of red balls after @xmath34 draws is distributed as the size of the subtree @xmath35 of a rrt @xmath6 that stems from the vertex @xmath33 . it is well - known from the theory of plya urns that this number follows the beta - binomial distribution with parameters @xmath36 . moreover , @xmath37 where @xmath38 is a beta@xmath39-distributed random variable . we will use this fact several times below . the _ splitting property _ ( also called randomness preserving property ) reveals the fractal nature of rtt s : roughly speaking , if one removes an edge from a rrt , then the two subtrees resulting from the split are in turn , conditionally on their sizes , independent rrt s . this is of course of crucial importance when investigating the destruction of a rrt , as we can then apply iteratively the splitting property when removing the edges uniformly at random and one after the other . we select an edge of @xmath6 uniformly at random and remove it . then @xmath6 splits into two subtrees , say @xmath40 and @xmath41 , where @xmath40 contains the root @xmath10 . we denote by @xmath42 and @xmath43 the pair of increasing trees which then result from the canonical relabelings of the vertices of @xmath40 and @xmath41 , respectively . introduce also an integer - valued variable @xmath44 with distribution @xmath45 [ p1 ] ( meir and moon @xcite ) in the notation above , @xmath46 has the same law as @xmath44 conditioned on @xmath47 , that is @xmath48 further , conditionally on @xmath49 , @xmath42 and @xmath43 are two independent rrt s with respective sizes @xmath50 and @xmath51 . there are @xmath52 configurations @xmath53 given by an increasing tree @xmath54 on @xmath55 and a distinguished edge @xmath56 . we remove the edge @xmath56 and then relabel vertices canonically in each of the resulting subtrees . let us enumerate the configurations that yield a given pair @xmath57 of increasing trees on @xmath58 and @xmath59 , respectively . let @xmath60 denote the extremity of the edge @xmath56 which is the closest to the root @xmath10 in @xmath54 , and @xmath61 the set of vertices which are disconnected from @xmath33 when @xmath56 is removed . since @xmath54 is increasing , all the vertices in @xmath61 must be larger than @xmath33 , and since we want @xmath62 , there are @xmath63 ways of choosing @xmath61 ( note that this is possible if and only if @xmath64 ) . there are a unique increasing tree structure on @xmath61 and a unique increasing tree structure on @xmath65 that yield respectively @xmath66 and @xmath67 after the canonical relabelings . conversely , given @xmath67 , @xmath66 , @xmath68 and @xmath69 with @xmath62 , there is clearly a unique configuration @xmath53 which yields the quadruple @xmath70 . namely , relabeling vertices in @xmath67 and @xmath71 produces two increasing tree structures @xmath72 and @xmath73 on @xmath65 and @xmath61 , respectively . we let @xmath56 denote the edge @xmath74 and then @xmath54 is the increasing tree obtained by connecting @xmath72 and @xmath73 using @xmath56 . it follows from the analysis above that @xmath75 now recall that @xmath76 to conclude that @xmath77 since there are @xmath78 increasing trees with size @xmath50 and @xmath79 increasing trees with size @xmath51 , this yields the claim . * it can be easily checked that the splitting property holds more generally when one removes a fixed edge , that is the edge connecting a given vertex @xmath80 to its parent . of course , the distribution of the sizes of the resulting subtrees then changes ; see the connection to plya urns mentioned in the beginning . the splitting property was used by meir and moon @xcite to investigate the following random algorithm for isolating the root @xmath10 of a rrt . starting from @xmath6 , remove a first edge chosen uniformly at random and discard the subtree which does not contain the root @xmath10 . iterate the procedure with the subtree containing @xmath10 until the root is finally isolated , and denote by @xmath81 the number of steps of this random algorithm . in other words , @xmath82 is the number of random cuts that are needed to isolate @xmath10 in @xmath6 . iksanov and mhle @xcite derived from proposition [ p1 ] a useful coupling involving an increasing random walk with step distribution given by . specifically , let @xmath83 denote a sequence of i.i.d . copies of @xmath44 and set @xmath84 , @xmath85 further , introduce the last time that the random walk @xmath86 remains below the level @xmath7 , @xmath87 [ c1 ] ( iksanov and mhle @xcite ) one can construct on the same probability space a random recursive tree @xmath6 together with the random algorithm of isolation of the root , and a version of the random walk @xmath86 , such that if @xmath88 denotes the nested sequence of the subtrees containing the root induced by the algorithm , then @xmath89 and @xmath90 let us agree for convenience that @xmath91 for every @xmath92 , and first work conditionally on @xmath93 . introduce a sequence @xmath94 of independent pairs of random variables such that for each @xmath13 , @xmath95 has the bernoulli law with parameter @xmath96 and @xmath97 is an independent variable distributed as @xmath44 conditioned on @xmath98 . then define for every @xmath99 @xmath100 and the partial sums @xmath101 . observe that @xmath102 if and only if @xmath103 , and hence , by construction , there is the identity @xmath104 therefore , follows if we show that @xmath105 are ( unconditionally ) i.i.d . copies of @xmath44 . this is essentially a consequence of the splitting property . specifically , for @xmath106 , we have @xmath107 where we used the notation and the result in proposition [ p1 ] , whereas for @xmath108 we have @xmath109 next , consider the conditional law of @xmath110 given @xmath111 and @xmath112 . of course , @xmath113 , and this inequality is in fact an equality whenever @xmath114 . we know from the splitting property that conditionally on its size , say @xmath115 with @xmath116 , @xmath117 is a rrt . therefore proposition [ p1 ] yields again for @xmath118 @xmath119 similarly for @xmath120 @xmath121 this shows that @xmath110 has the same distribution as @xmath44 and is independent of @xmath111 and @xmath112 . iterating this argument , we get that the @xmath122 form a sequence of i.i.d . copies of @xmath44 , which completes the proof . recall the algorithm of isolation of the root which was introduced in the preceding section , and that @xmath82 denotes its number of steps for @xmath6 , i.e. @xmath82 is the number of random cuts that are needed to isolate the root @xmath10 in @xmath6 . meir and moon @xcite used proposition [ p1 ] to investigate the first two moments of @xmath81 and showed that @xmath123 the problem of specifying the fluctuations of @xmath81 was left open until the work by drmota _ et al . _ , who obtained the following remarkable result . [ t1 ] ( drmota , iksanov , mhle and rsler @xcite ) as @xmath124 , @xmath125 converges in distribution to a completely asymmetric cauchy variable @xmath126 with characteristic function @xmath127 in short , the starting point of the proof in @xcite is the identity in distribution @xmath128 where @xmath129 is a random variable with the law of @xmath44 given @xmath47 , and @xmath129 is assumed to be independent of @xmath130 . more precisely , derives immediately from the splitting property ( proposition [ p1 ] ) . _ deduce from a pde for the generating function of the variables @xmath82 , and then singularity analysis provides the key tool for investigating the asymptotic behavior of this generating function and elucidating the asymptotic behavior of @xmath82 . iksanov and mhle @xcite developed an elegant probabilistic argument which explains the unusual rescaling and the cauchy limit law in theorem [ t1 ] . we shall now sketch this argument . * sketch proof of theorem [ t1 ] : * one starts observing that the distribution in belongs to the domain of attraction of a completely asymmetric cauchy variable @xmath126 whose law is determined by , namely @xmath131 then one deduces from that the asymptotic behavior of the last - passage time is given by @xmath132 see proposition 2 in @xcite . this limit theorem resembles of course theorem [ t1 ] , and the relation between the two is explained by the coupling of the algorithm of isolation of the root and the random walk @xmath86 stated in corollary [ c1 ] , as we shall now see . let the algorithm for isolating the root run for @xmath133 steps . then the size of the remaining subtree that contains the root is @xmath134 , and as a consequence , there are the bounds @xmath135 since at most @xmath136 edge removals are needed to isolate the root in any tree of size @xmath137 . on the other hand , specializing a renewal theorem of erickson @xcite for the increasing random walk @xmath86 , one gets that @xmath138 where @xmath139 is a uniform @xmath140 $ ] random variable . in particular @xmath141 thus theorem [ t1 ] follows from . it should be noted that there exists a vertex version of the isolation algorithm , where one chooses a vertex at random and destroys it together with its descending subtree . the algorithm continues until the root is chosen . using an appropriate coupling with @xmath82 , one readily shows that the number of random vertex removals @xmath142 needed to destroy a rrt @xmath6 satisfies @xmath143 in probability . henceforth , we concentrate on cutting edges . * remark . * weak limit theorems for the number of cuts to isolate the root vertex have also been obtained for other tree models , like conditioned galton - watson trees including e.g. uniform cayley trees and random binary trees ( panholzer @xcite and , in greater generality , janson @xcite ) , deterministic complete binary trees ( janson @xcite ) and random split trees ( holmgren @xcite ) . more generally , addario - berry _ et al . _ @xcite and bertoin @xcite found the asymptotic limit distribution for the number of cuts required to isolate a fixed number @xmath144 of vertices picked uniformly at random in a uniform cayley tree . this result was further extended by bertoin and miermont @xcite to conditioned galton - watson trees . we point to the remark after corollary [ c3 ] for more on this . turning back to rrt s , recent generalizations of theorem [ t1 ] were found first by kuba and panholzer @xcite and then by bertoin @xcite , some of which will be discussed in the reminder of this paper . in @xcite , kuba and panholzer considered the situation when one wishes to isolate the _ first _ @xmath137 vertices of a rrt @xmath6 , @xmath145 , where @xmath144 is fixed . in this direction , one modifies the algorithm of isolation of the sole root in an obvious way . a first edge picked uniformly at random in @xmath6 is removed . if one of the two resulting subtrees contains none of the vertices @xmath145 , then it is discarded forever . else , the two subtrees are kept . in both cases , one iterates until each and every vertex @xmath145 has been isolated , and we write @xmath146 for the number of steps of this algorithm . the approach of kuba and panholzer follows analytic methods similar to the original proof of theorem [ t1 ] by drmota _ we point out here that the asymptotic behavior of @xmath146 can also be deduced from theorem [ t1 ] by a probabilistic argument based on the following elementary observation , which enables us to couple the variables @xmath146 for different values of @xmath137 . specifically , we run the usual algorithm of isolation of the root , except that now , at each time when a subtree becomes disconnected from the root , we keep it aside whenever it contains at least one of the vertices @xmath147 , and discard it forever otherwise . once the root @xmath10 of @xmath6 has been isolated , we resume with the subtree containing @xmath9 which was set aside , meaning that we run a further algorithm on that subtree until its root @xmath9 has been isolated , keeping aside the further subtrees disconnected from @xmath9 which contain at least one of the vertices @xmath148 . we then continue with the subtree containing the vertex @xmath11 , and so on until each and every vertex @xmath145 has been isolated . if we write @xmath149 for the number of steps of this algorithm , then it should be plain that @xmath149 has the same law as @xmath146 , and further @xmath150 . we shall now investigate the asymptotic behavior of the increments @xmath151 for @xmath99 fixed . in this direction , suppose that we now remove the edges of @xmath6 one after the other in a uniform random order until the edge connecting the vertex @xmath13 to its parent is removed . let @xmath152 denote the subtree containing @xmath13 that arises at this step . [ l1 ] for each fixed @xmath99 , @xmath153 where @xmath139 is a uniform @xmath140 $ ] random variable . for the moment , let us take lemma [ l1 ] for granted and deduce the following . [ c2 ] we have @xmath154 where @xmath139 is a uniform @xmath140 $ ] random variable . just observe that @xmath155 has the same law as the number of cuts needed to isolate the root @xmath13 of @xmath156 , and recall from an iteration of the splitting property that conditionally on its size , @xmath156 is a rrt . our statement now follows readily from and lemma [ l1 ] . writing @xmath157 , we now see from theorem [ t1 ] and corollary [ c2 ] that for each fixed @xmath144 , there is the weak convergence @xmath158 which is theorem 1 in @xcite . we now proceed to the proof of lemma [ l1 ] . let @xmath159 denote the subtree of @xmath6 that stems from the vertex @xmath13 , and equip each edge @xmath56 of @xmath6 with a uniform @xmath140 $ ] random variable @xmath160 , independently of the other edges . imagine that the edge @xmath56 is removed at time @xmath160 , and for every time @xmath161 , write @xmath162 for the subtree of @xmath159 which contains @xmath13 at time @xmath163 . hence , if we write @xmath164 for @xmath56 the edge connecting @xmath13 to its parent , then @xmath165 . further , since @xmath139 is independent of the other uniform variables , conditionally on @xmath139 and @xmath159 , @xmath166 can be viewed as the cluster that contains the root vertex @xmath13 after a bernoulli bond percolation on @xmath167 with parameter @xmath168 . thus , conditionally on @xmath169 and @xmath170 , @xmath171 has the same law as @xmath172 in the notation of lemma [ l2 ] . from we know that @xmath173 converges in distribution as @xmath124 to a beta variable with parameters @xmath174 , say @xmath175 , which is of course independent of @xmath139 . on the other hand , conditionally on its size , and after the usual canonical relabeling of its vertices , @xmath159 is also a rrt ( see the remark at the end of section [ smaintools ] ) . it then follows from lemma [ l2 ] that @xmath176 which establishes our claim . imagine now that we remove the edges of @xmath6 one after the other and in a uniform random order , no matter whether they belong to the root component or not . we call this the _ destruction process _ of @xmath6 . after @xmath7 steps , no edges are present anymore and all the vertices have been isolated . in particular , the random variable which counts only the number of edge removals from the root component can be identified with @xmath82 from the previous section . the purpose of this section is to introduce and study the asymptotic behavior of two trees which can be naturally associated to this destruction process , namely the tree of component sizes and the cut - tree . furthermore , we give some applications of the cut - tree to the isolation and disconnection of nodes and comment on ordered destruction of a rrt . in this part , we are interested in the sizes of the tree components produced by the destruction process . our analysis will also prove helpful for studying percolation clusters of a rrt in section [ spercolation ] . the component sizes are naturally stored in a tree structure . as our index set , we use the universal tree @xmath177 with the convention @xmath178 and @xmath179 . in particular , an element @xmath180 is a finite sequence of strictly positive integers @xmath181 , and its length @xmath182 represents the `` generation '' of @xmath183 . the @xmath51th child of @xmath183 is given by @xmath184 , @xmath185 . the empty sequence @xmath186 is the root of the tree and has length @xmath187 . if no confusion occurs , we drop the separating commas and write @xmath188 or simply @xmath189 instead of @xmath181 . also , @xmath190 represents the element @xmath183 . we define a tree - indexed process @xmath191 , which encodes the sizes of the tree components stemming from the destruction of @xmath6 . we will directly identify a vertex @xmath183 with its label @xmath192 . following the steps of the destruction process , we build this process dynamically starting from the singleton @xmath193 and ending after @xmath7 steps with the full process @xmath194 more precisely , when the first edge of @xmath6 is removed in the destruction process , @xmath6 splits into two subtrees , say @xmath195 and @xmath196 , where @xmath195 contains the root @xmath10 . we stress that @xmath195 is naturally rooted at @xmath10 and @xmath196 at its smallest vertex . the size @xmath197 is viewed as the first child of @xmath198 and denoted by @xmath199 . now first suppose that the next edge which is removed connects two vertices in @xmath196 . then , @xmath196 splits into two tree components . the size of the component not containing the root of @xmath196 is viewed as the first child of @xmath200 and denoted by @xmath201 . on the other hand , if the second edge which is removed connects two vertices in @xmath195 , then the size of the component not containing @xmath10 is viewed as the second child of @xmath198 and denoted by @xmath202 . it should now be plain how to iterate this construction . after @xmath7 steps , we have in this way defined @xmath15 variables @xmath203 with @xmath204 , and we extend the definition to the full universal tree by letting @xmath205 for all the remaining @xmath180 . we refer to figure @xmath11 for an example . the tree components whose sizes are encoded by the elements @xmath203 with @xmath182 are called the components of generation @xmath33 . to sum up , every time an edge is removed in the destruction process , a tree component @xmath206 splits into two subtrees , and we adjoin the size of the subtree which does not contain the root of @xmath206 as a new child to the vertex representing @xmath206 . note that the root @xmath207 has @xmath82 many nontrivial children , and they represent the sizes of the tree components which were cut from the root one after the other in the algorithm for isolating the root . * figure 2 * + _ right : the corresponding tree of component sizes , with the vertex sets of the tree components . the elements @xmath203 of size @xmath10 are omitted . _ we now interpret @xmath208 as the genealogical tree of a multi - type population model , where the type reflects the size of the tree component ( and thus takes integer values ) . in particular the ancestor @xmath186 has type @xmath15 ; furthermore , a node @xmath183 with @xmath209 corresponds to an empty component and is therefore absent in the population model . we also stress that the type of an individual is always given by the sum of the types of its children plus @xmath9 . as a consequence , types can be recovered from the sole structure of the genealogical tree . more precisely , the type of an individual is simply given by the total size of the subtree of the genealogical tree stemming from that individual . the splitting property of a rrt immediately transfers into a branching property for this population model . [ l3 ] the population model induced by the tree of component sizes @xmath208 is a multi - type galton - watson process starting from one particle of type @xmath15 . the reproduction distribution @xmath210 of an individual of type @xmath99 is given by the law of the sequence of the sizes of the non - root subtrees which are produced in the algorithm for isolating the root of a rrt of size @xmath13 . even though the coupling of iksanov and mhle is not sufficient to fully describe the reproduction law , it nonetheless provides essential information on @xmath210 in terms of a sequence of i.i.d . copies of @xmath44 . as we will see next , extreme value theory for the i.i.d . sequence then enables us to specify asymptotics of the population model when the type @xmath15 of the ancestor goes to infinity . to give a precise statement , we rank the children of each individual in the decreasing order of their types . formally , given the individual indexed by @xmath180 has exactly @xmath137 children of type @xmath211 , we let @xmath212 be the random permutation of @xmath213 which sorts the sequence of types @xmath214 in the decreasing order , i.e. @xmath215 where in the case of ties , children of the same type are ranked uniformly at random . we extend @xmath212 to a bijection @xmath216 by putting @xmath217 for @xmath218 . we then define the global random bijection @xmath219 recursively by setting @xmath220 , @xmath221 , and then , given @xmath222 , @xmath223 , @xmath180 , @xmath185 . note that @xmath224 preserves the parent - child relationship , i.e. children of @xmath183 are mapped into children of @xmath222 . we simply write @xmath225 for the process which is ranked in this way . now , if the sizes of the components of generation @xmath33 are normalized by a factor @xmath226 , we obtain finite - dimensional convergence of @xmath227 towards the genealogical tree of a continuous - state branching process with reproduction measure @xmath228 on @xmath229 . more precisely , the limit object is a tree - indexed process @xmath230 with initial state @xmath231 , whose distribution is characterized by induction on the generations as follows . 1 . @xmath231 almost surely ; 2 . for every @xmath232 conditionally on @xmath233 , the sequences @xmath234 for the vertices @xmath180 at generation @xmath182 are independent , and each sequence @xmath234 is distributed as the family of the atoms of a poisson random measure on @xmath229 with intensity @xmath235 , where the atoms are ranked in the decreasing order of their sizes . [ pbranching ] as @xmath236 , there is the convergence in the sense of finite - dimensional distributions , @xmath237 we only sketch the proof and refer to the forthcoming paper @xcite for details . basically , if @xmath238 is a sequence of of i.i.d . copies of @xmath44 , then for @xmath239 , the number of indices @xmath240 such that @xmath241 is binomially distributed with parameters @xmath33 and @xmath242 . from and theorem 16.16 in @xcite we deduce that for a fixed integer @xmath51 , the @xmath51 largest among @xmath243 , normalized by a factor @xmath244 , converge in distribution to the @xmath51 largest atoms of a poisson random measure on @xmath229 with intensity @xmath228 . since @xmath245 in probability , finite - dimensional convergence of @xmath246 restricted to generations @xmath247 then follows from . lemma [ l3 ] enables us to transport the arguments to the next generations . consider for a while a deterministic setting where @xmath248 is an arbitrary tree on some finite set of vertices @xmath0 . imagine that its edges are removed one after the other in some given order , so at the end of the process , all the vertices of @xmath248 have been disconnected from each other . we shall encode the destruction of @xmath248 by a rooted binary tree , which we call the cut - tree and denote by @xmath249 . the cut - tree has internal nodes given by the non - singleton connected components which arise during the destruction , and leaves which correspond to the singletons and which can thus be identified with the vertices in @xmath0 . more precisely , the root of @xmath249 is given by @xmath0 , and when the first edge of @xmath248 is removed , disconnecting @xmath0 into , say , @xmath250 and @xmath251 , then @xmath250 and @xmath251 are viewed as the two children of @xmath0 and thus connected to @xmath0 by a pair of edges . suppose that the next edge which is removed connects two vertices in @xmath250 , so removing this second edge disconnects @xmath250 into , say @xmath252 and @xmath253 . then @xmath252 and @xmath253 are viewed in turn as the two children of @xmath250 . we iterate in an obvious way , see figure @xmath254 for an example . ( 400,280)(-10,-30 ) ( 80,10 ) ( 40,10 ) ( 40,50 ) ( 120,50 ) ( 80,90 ) ( 160,90 ) ( 40,130 ) ( 120,130 ) ( 80,170 ) ( 40,20)(0,1)20 ( 50,10)(1,0)20 ( 85,18)(1,1)26 ( 115,58)(-1,1)26 ( 75,98)(-1,1)26 ( 85,98)(1,1)26 ( 125,58)(1,1)26 ( 80,100)(0,1)60 ( 80,10 ) ( 0,0)a ( 40,10 ) ( 0,0)h ( 120,50 ) ( 0,0)g ( 40,50 ) ( 0,0)f ( 80,90 ) ( 0,0)i ( 160,90 ) ( 0,0)d ( 40,130 ) ( 0,0)e ( 120,130 ) ( 0,0)c ( 80,170 ) ( 0,0)b ( 103,29)(0,0)1 ( 60,4)(0,0)6 ( 34,30)(0,0)8 ( 100,63)(0,0)2 ( 102,105)(0,0)5 ( 137,65)(0,0)4 ( 55,110)(0,0)3 ( 75,130)(0,0)7 ( 290,00)(60,15)abcdefghi ( 250,50)(20,15)afh ( 350,50)(40,15)bcdegi ( 230,90)(10,15)a ( 260,90)(15,15)fh ( 230,130)(10,15)f ( 260,130)(10,15)h ( 330,90)(30,15)bcei ( 400,90)(15,15)dg ( 300,130)(20,15)bci ( 340,130)(10,15)e ( 370,130)(10,15)g ( 400,130)(10,15)d ( 300,170)(15,15)bi ( 340,170)(10,15)c ( 270,210)(10,15)b ( 310,210)(10,15)i ( 300,16)(-1,1)34 ( 330,16)(1,1)34 ( 260,66)(-1,1)24 ( 265,66)(0,1)24 ( 260,106)(-1,1)24 ( 265,106)(0,1)24 ( 365,66)(-1,1)24 ( 383,66)(1,1)24 ( 335,106)(-1,1)24 ( 345,106)(0,1)24 ( 400,106)(-1,1)24 ( 405,106)(0,1)24 ( 305,146)(0,1)24 ( 315,146)(1,1)24 ( 300,186)(-1,1)24 ( 313,186)(0,1)24 * figure 3 * _ left : tree @xmath248 with vertices labeled a, ... ,i ; edges are enumerated in the order of the cuts . _ _ right : cut - tree @xmath249 on the set of blocks recording the destruction of @xmath248 . _ it should be clear that the number of cuts required to isolate a given vertex @xmath255 in the destruction of @xmath248 ( as previously , we only count the cuts occurring in the component which contains @xmath255 ) corresponds precisely to the height of the leaf @xmath256 in @xmath249 . more generally , the number of cuts required to isolate @xmath33 distinct vertices @xmath257 coincides with the total length of the cut - tree reduced to its root and the @xmath33 leaves @xmath258 minus @xmath259 , where the length is measured as usual by the graph distance on @xmath249 . in short , the cut - tree encapsulates all the information about the numbers of cuts needed to isolate any subset of vertices . we now return to our usual setting , that is @xmath6 is a rrt of size @xmath15 , whose edges are removed in a uniform random order , and we write @xmath260 for the corresponding cut - tree . we point out that the genealogical tree of component sizes which was considered in the previous section can easily be recovered from @xmath261 . specifically , the root @xmath5 of @xmath261 has to be viewed as the ancestor of the population model , its type is of course @xmath15 . then the blocks of @xmath261 which are connected by an edge to the segment from the root @xmath5 to the leaf @xmath262 are the children of the ancestor in the population model , the type of a child being given by the size of the corresponding block . the next generations of the population model are then described similarly by an obvious iteration . the segment of @xmath261 from its root @xmath5 to the leaf @xmath262 is described by the nested sequence , and the coupling of iksanov and mhle stated in corollary [ c1 ] expresses the sequence of the block - sizes along the portion of this segment starting from the root and with length @xmath133 , in terms of the random walk @xmath86 . we shall refer to this portion as the trunk of @xmath261 and denote it by @xmath263 . the connected components of the complement of the trunk , @xmath264 are referred to as the branches of @xmath261 . roughly speaking , it has been shown in @xcite that upon rescaling the graph distance of @xmath261 by a factor @xmath265 , the latter converges to the unit interval . the precise mathematical statement involves the notion of convergence of pointed measured metric spaces in the sense of the gromov - hausdorff - prokhorov distance . [ t2 ] endow @xmath261 with the uniform probability measure on its leaves , and normalize the graph distance by a factor @xmath265 . as @xmath266 , the latter converges in probability in the sense of the pointed gromov - hausdorff - prokhorov distance to the unit interval @xmath140 $ ] equipped with the usual distance and the lebesgue measure , and pointed at @xmath10 . providing the background on the gromov - hausdorff - prokhorov distance needed to explain rigorously the meaning of theorem [ t2 ] would probably drive us too far away from the purpose of this survey , so we shall content ourselves here to give an informal explanation . after the rescaling , each edge of @xmath261 has length @xmath265 , and it follows from that the length @xmath267 of @xmath263 converges in probability to @xmath9 as @xmath124 . because the trunk is merely a segment , if we equip it with the uniform probability measure on its nodes , then we obtain a space close to the unit interval endowed with the lebesgue measure . the heart of the argument of the proof in @xcite is to observe that in turn , @xmath263 is close to @xmath261 when @xmath7 is large , both in the sense of hausdorff and in the sense of prokhorov . first , as @xmath263 is a subset of @xmath261 , the hausdorff distance between @xmath263 and @xmath261 corresponds to the maximal depth of the branches of @xmath261 , and one thus have to verify that all the branches are small ( recall that the graph distance has been rescaled by a factor @xmath265 ) . then , one needs to check that the uniform probability measures , respectively on the set of leaves of @xmath261 and on the nodes of @xmath263 , are also close to each other in the sense of the prokhorov distance between probability measures on a metric space . this is essentially a consequence of the law of large numbers for the random walk defined in , namely @xmath268 see . theorem [ t2 ] enables us to specify the asymptotic behavior of the number of cuts needed to isolate randomly chosen vertices of @xmath6 . for a given integer @xmath144 and for each @xmath269 , let @xmath270 denote a sequence of i.i.d . uniform variables in @xmath4 . we write @xmath271 for the number of random cuts which are needed to isolate @xmath272 . the following corollary , which is taken from @xcite , is a multi - dimensional extension of theorem 3 of kuba and panholzer @xcite . [ c3 ] as @xmath266 , the random vector @xmath273 converges in distribution to @xmath274 where @xmath275 are i.i.d . uniform @xmath140 $ ] random variables . in particular , @xmath276 converges in distribution to a beta@xmath277 variable . recall that @xmath272 are @xmath137 independent uniform vertices of @xmath6 . equivalently , the singletons @xmath278 form a sequence of @xmath137 i.i.d . leaves of @xmath261 distributed according to the uniform law . let also @xmath279 be a sequence of @xmath137 i.i.d . uniform variables on @xmath140 $ ] . denote by @xmath280 the reduction of @xmath261 to the @xmath137 leaves @xmath281 and its root @xmath55 , i.e. @xmath282 is the smallest subtree of @xmath261 which connects these nodes . similarly , write @xmath283 for the reduction of @xmath284 to @xmath275 and the origin @xmath10 . both reduced trees are viewed as combinatorial trees structures with edge lengths , and theorem [ t2 ] entails that @xmath285 converges in distribution to @xmath283 as @xmath124 . in particular , focusing on the lengths of those reduced trees , there is the weak convergence @xmath286 this yields our claim , as plainly @xmath287 for every @xmath288 . * remark . * the nearly trivial proof of this corollary exemplifies the power of theorem [ t2 ] , and one might ask for convergence of the cut - tree for other tree models . in fact , employing the work of haas and miermont @xcite , it has been shown in @xcite that if @xmath289 is a uniform cayley tree of size @xmath7 , then @xmath290 converges weakly in the sense of gromov - hausdorff - prokhorov to the brownian continuum random tree ( crt ) , see aldous @xcite . since the total length of the crt reduced to the root and @xmath137 i.i.d leaves picked according to its mass - measure follows the chi@xmath291-distribution , one readily obtains the statement corresponding to corollary [ c3 ] for uniform cayley trees ( @xcite and also , by different means , @xcite ) . bertoin and miermont @xcite extended the convergence of the cut - tree towards the crt to the full family of critical galton - watson trees with finite variance and conditioned to have size @xmath7 , in the sense of gromov - prokhorov . as a corollary , one obtains a multi - dimensional extension of janson s limit theorem @xcite . very recently , dieuleveut @xcite proved the analog of @xcite for the case of galton - watson trees with offspring distribution belonging to the domain of attraction of a stable law of index @xmath292 . with corollary [ c3 ] at hand , we can also study the number @xmath293 of random cuts which are needed to isolate the @xmath137 _ last _ vertices of @xmath6 , i.e. @xmath294 , where @xmath144 is again a given integer . as kuba and panholzer @xcite proved in their theorem 2 , @xmath293 has the same asymptotic behavior in law as @xmath271 . the following multi - dimensional version was given in @xcite , relying on theorem 2 of @xcite . here we give a self - contained proof of the same statement . [ c4 ] as @xmath266 , the random vector @xmath295 converges in distribution to @xmath274 where @xmath275 are i.i.d . uniform @xmath140 $ ] random variables . for ease of notation , we consider only the case @xmath296 , the general case being similar . the random variable @xmath297 counts the number of random cuts needed to isolate the vertex @xmath7 , which is a leaf of @xmath6 . if we write @xmath255 for the parent of @xmath7 in @xmath6 , then @xmath255 is uniformly distributed on @xmath298 , and it follows that the number @xmath299 of cuts needed to isolate @xmath255 has the same limit behavior in law as @xmath300 . in view of corollary [ c3 ] , it suffices therefore to verify that @xmath301 we now consider the algorithm for isolating the vertex @xmath255 . clearly , the number of steps of this algorithm until the edge @xmath56 joining @xmath255 to @xmath7 is removed is distributed as @xmath302 . in particular , we obtain a natural coupling between @xmath299 and @xmath302 with @xmath303 . denote by @xmath304 $ ] the segment of @xmath6 from the root @xmath10 to the leaf @xmath7 , and write @xmath33 for the outer endpoint of the first edge from @xmath304 $ ] which is to be removed by the isolation algorithm . since @xmath305|\sim \ln n$ ] in probability ( see e.g. theorem 6.17 of @xcite ) , and since the isolation algorithm chooses its edges uniformly at random , the probability that @xmath33 is equal to @xmath7 tends to zero . moreover , with high probability @xmath306|$ ] will still be larger than @xmath307 , say . by conditioning on @xmath33 and repeating the above argument with @xmath308 $ ] in place of @xmath304 $ ] , we see that we can concentrate on the event that before @xmath7 is isolated , at least two edges different from @xmath56 are removed from the segment @xmath304 $ ] . on this event , after the second time an edge from @xmath304 $ ] is removed , the vertices @xmath255 and @xmath7 lie in a tree component which can be interpreted as a tree component of the second generation in the destruction process . as a consequence of proposition [ pbranching ] , the size of this tree component multiplied by factor @xmath244 converges to zero in probability . since the size of the component gives an upper bound on the difference @xmath309 , the claim follows . as another application of the cut - tree , theorem [ t2 ] allows us to determine the number of cuts @xmath310 which are required to _ disconnect _ ( and not necessarily isolate ) @xmath311 vertices in @xmath6 chosen uniformly at random . for ease of description , let us assume that the sequence of vertices @xmath312 is chosen uniformly at random in @xmath313 without replacement . note that in the limit @xmath236 , it makes no difference whether we sample with or without replacement . we run the algorithm for isolating the vertices @xmath312 , with the modification that we discard emerging tree components which contain at most one of these @xmath137 vertices . we stop the algorithm when @xmath312 are totally disconnected from each other , i.e. lie in @xmath137 different tree components . write @xmath314 for the ( random ) number of steps of this algorithm until for the first time , the vertices @xmath312 do no longer belong to the same tree component , further @xmath315 for the number of steps until for the first time , the @xmath137 vertices are spread out over three distinct tree components , and so on , up to @xmath310 , the number of steps until the @xmath137 vertices are totally disconnected . we obtain the following result . [ c5 ] as @xmath266 , the random vector @xmath316 converges in distribution to @xmath317 where @xmath318 denote the first @xmath136 order statistics of an i.i.d . sequence @xmath275 of uniform @xmath140 $ ] random variables . in particular , @xmath319 converges in distribution to a beta@xmath320 random variable , and @xmath321 converges in distribution to a beta@xmath322 law . since the branches of @xmath261 are asymptotically small compared to the trunk ( see e.g. proposition @xmath9 in @xcite ) , with probability tending to @xmath9 as @xmath236 the @xmath137 vertices @xmath312 are cut from the root component one after the other , i.e. in no stage of the disconnection algorithm , a non - root tree component will contain more than one of the @xmath312 . on this event , writing again @xmath282 for the reduction of @xmath261 to the @xmath137 leaves @xmath278 and its root @xmath313 , the variable @xmath323 is given by the length of the path in @xmath280 from the root to the @xmath13th branch point . now , if @xmath324 and @xmath283 are defined as in the proof of corollary [ c3 ] , the distance in @xmath283 from the root @xmath10 to the @xmath13th smallest among @xmath324 is distributed as @xmath325 . together with , this proves the claim . * remark . * with a proof similar to that of corollary [ c4 ] , one sees that the statement of corollary [ c5 ] does also hold if @xmath326 are replaced by the analogous quantities for disconnecting the @xmath137 last vertices @xmath327 . on the other hand , if one is interested in disconnecting the first @xmath137 vertices @xmath328 , and if @xmath329 denote in this case the quantities corresponding to @xmath326 , one first observes the trivial bound @xmath330 where @xmath146 is the number of steps needed to isolate @xmath331 . now , @xmath332 can be identified with the number of steps in the algorithm for isolating the root until for the first time , an edge connecting one of the vertices @xmath333 to its parent is removed . by similar means as in the proof of lemma [ l1 ] , one readily checks that at this time , the root component has a size of order @xmath334 , with @xmath175 having a beta@xmath335-distribution . in particular , we see that @xmath336 in probability , where @xmath82 is the number of steps to isolate the root @xmath10 . but by , also @xmath337 in probability . therefore , the variables @xmath338 have the same limit behavior in law as @xmath82 , that is as @xmath236 , @xmath339 , @xmath340 , converge all to the same completely asymmetric cauchy variable @xmath126 defined by . here , we consider briefly another natural destruction procedure of a rrt , where instead of removing edges in a uniform random order , we remove them deterministically in their _ natural _ order . that is the @xmath13th edge of @xmath6 which is removed is now the one connecting the vertex @xmath13 to its parent , for @xmath341 . we first point at the fact that the number of ordered edge removals which are now needed to isolate the root ( recall that we only take into account edge removals inside the current subtree containing the root ) can be expressed as @xmath342 , where @xmath343 if the parent of vertex @xmath13 in @xmath6 is the root @xmath10 , and @xmath10 otherwise . that is to say that @xmath344 is the degree of the root . further the recursive construction entails the @xmath345 are independent variables , such that each @xmath345 has the bernoulli distribution with parameter @xmath346 . as is well - known , it then follows e.g. from lyapunov s central limit theorem that @xmath347 we refer to kuba and panholzer @xcite for many more results about the degree distributions in random recursive trees . we then turn our attention to the cut - tree described in section [ scuttree ] , which encodes the ordered destruction of @xmath6 . we write @xmath348 for the latter and observe that the recursive construction of @xmath6 implies that in turn , @xmath348 can also be defined by a simple recursive algorithm . specifically , @xmath349 is the elementary complete binary tree with two leaves , @xmath262 and @xmath350 , and root @xmath351 . once @xmath6 and hence @xmath352 have been constructed , @xmath353 is obtained by incorporating the vertex @xmath15 and creating a new edge between @xmath15 and its parent @xmath354 , which is chosen uniformly at random in @xmath355 . note that this new edge is the last one which will be removed in the ordered destruction of @xmath353 . in terms of cut - trees , this means that the leaf @xmath356 of @xmath348 should be replaced by an internal node @xmath357 to which two leaves are attached , namely @xmath356 and @xmath358 . further , any block ( internal node ) @xmath359 of @xmath348 with @xmath360 should be replaced by @xmath361 . the resulting complete binary tree is then distributed as @xmath362 . if we discard labels , this recursive construction of @xmath363 corresponds precisely to the dynamics of the markov chain on complete binary trees described e.g. in mahmoud @xcite for binary search trees ( in short , bst ) . we record this observation in the following proposition . [ p3 ] the combinatorial tree structure of @xmath364 is that of a bst with @xmath15 leaves . bst have been intensively studied in the literature , see drmota @xcite and references therein , and the combination with proposition [ p3 ] yields a number of precise results about the number of ordered cuts which are needed to isolate vertices in @xmath6 . for instance , the so - called _ saturation level _ @xmath365 in a bst is the minimal level of a leaf , and can then be viewed as the smallest number of ordered cuts after which some vertex of @xmath6 has been isolated . similarly , the height @xmath366 is the maximal level of a leaf , and thus corresponds to the maximal number of ordered cuts needed to isolate a vertex in @xmath6 . the asymptotic behaviors of the height and of the saturation level of a large bst are described in theorem 6.47 of drmota @xcite , in particular one has @xmath367 where @xmath368 are the solutions to the equation @xmath369 . in the same vein , the asymptotic results of chauvin _ et al . _ on the profile of large bst can be translated into sharp estimates for the number of vertices of @xmath6 which are isolated after exactly @xmath33 ordered cuts ( see in particular theorem 3.1 in @xcite ) . finally , let us look at component sizes when edges are removed in their natural order . compared to uniform random edge removal , the picture is fairly different . indeed , when removing an edge from @xmath6 picked uniformly at random , the size of the subtree not containing @xmath10 is distributed according to the law of @xmath44 conditioned on @xmath47 . if , in contrast , the first edge to be removed is the edge joining @xmath9 to its parent @xmath10 , then we know from that both originating subtrees are of order @xmath7 . since the splitting property still holds when we remove a fixed edge , the component sizes again inherit a branching structure . in fact , it is an immediate consequence of the definition that the structure of the tree of component sizes corresponding to the ordered destruction on @xmath6 agrees with the structure of @xmath6 and therefore yields the same rrt of size @xmath15 . in section [ sisoroot ] it has become apparent that bernoulli bond percolation on @xmath6 is a tool to study the sizes of tree components which appear in isolation algorithms . here , we take in a certain sense the opposite point of view and obtain results on the sizes of percolation clusters using what we know about the sizes of tree components . throughout this section , we use the term cluster to designate connected components induced by percolation , while we use the terminology tree components for connected components arising from isolation algorithms . more specifically , the algorithm for isolating the root can be interpreted as a dynamical percolation process in which components that do not contain the root are instantaneously frozen . imagine a continuous - time version of the algorithm , where each edge of @xmath6 is equipped with an independent exponential clock of some parameter @xmath370 . when a clock rings , the corresponding edge is removed if and only if it currently belongs to the root component . at time @xmath371 , the root component can naturally be viewed as the root cluster of a bernoulli bond percolation on @xmath6 with parameter @xmath372 . moreover , under this coupling each percolation cluster is contained in some tree component which was generated by the isolation process up to time @xmath373 . in order to discover the percolation clusters inside a non - root tree component @xmath374 , the latter has to be unfrozen , i.e. additional edges from @xmath374 have to be removed . in particular , the percolation cluster containing the root of @xmath374 can again be identified as the root component of an isolation process on @xmath374 , stopped at an appropriate time . these observations lead in @xcite to the study of the asymptotic sizes of the largest and next largest percolation clusters of @xmath6 , when the percolation parameter @xmath375 satisfies @xmath376 this regime corresponds precisely to the supercritical regime , in the sense that the root cluster is the unique giant cluster , and its complement in @xmath6 has a size of order @xmath7 , too . indeed , the height @xmath377 of a vertex @xmath183 picked uniformly at random in a rrt of size @xmath15 satisfies @xmath378 . since the probability that @xmath183 is connected to the root is given by the first moment of @xmath379 , where @xmath380 denotes the size of the root cluster , one obtains @xmath381 a similar argument shows @xmath382 , which proves @xmath383 in @xmath384 . let us now consider the next largest clusters in the regime . we write @xmath385 for the sizes of the non - root percolation clusters of @xmath6 , ranked in the decreasing order . we quote from @xcite the following limit result . [ p4 ] for every fixed integer @xmath386 , @xmath387 converges in distribution as @xmath236 towards @xmath388 where @xmath389 denotes the sequence of the atoms of a poisson random measure on @xmath229 with intensity @xmath390 . the intensity is better understood as the image of the intensity measure @xmath391 on @xmath392 by the map @xmath393 . in fact , from our introductory remarks and proposition [ pbranching ] it should be clear that the first coordinate of an atom @xmath394 stands for the asymptotic ( and normalized ) size of the tree component containing the percolation cluster , while the second encodes the time when the component was separated from the root . instead of providing more details here , let us illustrate an alternative route to prove the proposition , which was taken in @xcite to generalize the results to _ scale - free random trees_. these random graphs form a family of increasing trees indexed by a parameter @xmath395 that grow according to a preferential attachment algorithm , see @xcite . in the boundary case @xmath396 , one obtains a rrt , while in the case @xmath397 , the @xmath13th vertex is added to one of the first @xmath398 vertices with probability proportional to its current degree . in @xcite , the connection of scale - free random trees to the genealogy of yule processes was employed , and it should not come as a surprise that this approach can be adapted to random recursive trees . in fact , the case of rrt s is considerably simpler , since one has not to keep track of the degree of vertices when edges are deleted . let us sketch the main changes . denote by @xmath399 the genealogical tree of a standard yule process @xmath400 at time @xmath163 . similar to section @xmath254 of @xcite , we superpose bernoulli bond percolation with parameter @xmath401 to this construction . namely , if a new vertex is attached to the genealogical tree , we delete the edge connecting this vertex to its parent with probability @xmath18 . we write @xmath402 for the resulting combinatorial structure at time @xmath163 , and @xmath403 for the sequence of the subtrees at time @xmath163 , enumerated in the increasing order of their birth times , where we use the convention that @xmath404 if less than @xmath51 edges have been deleted up to time @xmath163 . in particular , @xmath405 is the subtree containing the root @xmath10 , and @xmath406 . furthermore , if @xmath407 denotes the birth time of the @xmath13th subtree , then the process @xmath408 is a yule process with birth rate @xmath27 per unit population size , started from a single particle of size @xmath9 . by analyzing the birth times as in @xcite , one readily obtains the analogous statements of section @xmath11 and @xmath254 there . this leads to another proof of proposition [ p4 ] . * remark . * as it is shown in the forthcoming paper @xcite , the approach via yule processes can be extended further to all percolation regimes @xmath409 . moreover , if the entire family of cluster sizes is encoded by a tree structure similar to the tree of component sizes , one can specify the finite - dimensional limit of this `` tree of cluster sizes '' . details will be given in @xcite . we finally take a closer look at the size of the root cluster @xmath380 for supercritical percolation with parameter @xmath410 as we have already discussed , @xmath380 satisfies a law of large numbers , but as we will point out here , @xmath380 exhibits non - gaussian fluctuations . this should be seen in sharp contrast to other graph models , were asymptotic normality of the giant cluster has been established , e.g. for the complete graph on @xmath7 vertices and percolation parameter @xmath411 , @xmath412 fixed ( stephanov @xcite , pittel @xcite , barraez _ et al . _ @xcite ) . for rrt s , the fluctuations can be obtained from a recent result of schweinsberg @xcite . among other things , he studied how the number of blocks in the bolthausen - sznitman coalescent changes over time . the bolthausen - sznitman coalescent was introduced in @xcite in the context of spin glasses , and goldschmidt and martin @xcite discovered the following connection to the random cutting of rrt s : equip each edge of a rrt of size @xmath7 on the vertex set @xmath413 with an independent standard exponential clock . if a clock rings , delete the corresponding edge , say @xmath56 , and the whole subtree rooted at the endpoint of @xmath56 most distant from the root @xmath9 . furthermore , replace the label of the vertex of @xmath56 which is closer to the root @xmath9 , say @xmath13 , by the label set consisting of @xmath13 and all the vertex labels of the removed subtree . then the sets of labels form a partition of @xmath413 , which evolves according to the dynamics of the bolthausen - sznitman coalescent started from @xmath7 blocks @xmath414 ( see proposition 2.2 of @xcite for details ) . note that in this framework , the variable @xmath82 counting the number of steps in the algorithm for isolating the root can be interpreted as the number of collision events which take place until there is just one block left . this statement was re - proved in @xcite , with a different approach which does not rely on the bolthausen - sznitman coalescent . instead , three different growth phases of a rrt @xmath6 are considered , and the effect of percolation is studied in each of these phases . this approach makes again use of the coupling of iksanov and mhle and the connection to yule processes , providing an intuitive explanation for the correction terms in the statement . in : m. drmota , p. flajolet , d. gardy , and b. gittenberger ( eds ) . _ mathematics and computer science iii , algorithms , trees , combinatorics and probabilities _ ( vienna 2004 ) , birkhuser , basel ( 2004 ) 241 - 253 .

we present exact diagonalization and momentum average approximation ( ma ) results for the single polaron properties of a one - dimensional two - band model with phonon - modulated hopping . at strong electron - phonon coupling , we find a novel type of sharp transition , where the polaron ground state momentum jumps discontinuously from to . the nature and origin of this transition is investigated and compared to that of the su - schrieffer - heeger ( ssh ) model , where a sharp but smooth transition was previously reported . we argue that such discontinuous transitions are a consequence of the multi - band nature of the model , and are unlikely to be observed in one - band models . we also show that ma describes qualitatively and even quantitatively accurately this polaron and its transition . given its computationally efficient generalization to higher dimensions , ma thus promises to allow for accurate studies of electron - phonon coupling in multi - band models in higher dimensions .

1611.04543

the coupling between carriers and phonons is known or believed to be important for many materials , including cuprates,@xcite manganites , @xcite , nickelates @xcite and bismuth perovskites.@xcite these materials display a variety of interesting phenomena , including , but not limited to , high - temperature superconductivity ( cuprates , babio@xmath2 ) , layered ferromagnetism ( manganites ) and a spin / charge density wave ( nickelates ) . the carrier - phonon coupling leads to the formation of a polaron , a coherent quasi - particle ( qp ) consisting of the charge carrier and the cloud of phonons surrounding it and moving coherently with it . polarons have been studied extensively especially in the holstein model , @xcite the simplest model where local phonons modify the on - site energy of the carrier , but also to generalizations with short - range and long - range couplings of similar origin , such as the breathing - mode ( bm ) model @xcite , the double - well potential model@xcite and the frhlich model @xcite . the other possibility is that the coupling to phonons modulates the carrier s hopping integrals , a scenario described by the ssh model , @xcite which has seen an increased amount of interest in recent years.@xcite this is because the ssh model exhibits a sharp transition in the properties of its polaron , one signature being the change of the polaron ground state ( gs ) momentum from @xmath1 ( at weak coupling ) to a finite value that smoothly evolves toward @xmath3 ( at strong coupling ) . such transitions were shown to be impossible for models where the phonons modulate the on - site energy @xcite.a study of a model which includes both types of carrier - phonon coupling was carried out by herrera _ et . al _ @xcite and found that in addition to the transition observed in the ssh model , a second transition of the gs momentum also takes place . whether other such transitions can occur and what are their characteristics , is currently an open question . efforts to understand polaron physics have , so far , focused almost exclusively on single - band models . it is therefore a natural question to ask whether the polaron properties of multi - band models are similar , or whether they are qualitatively different from those of single band models . in this article , we answer this question based on a study of the single polaron properties of the two - band model depicted in fig . [ fig : model ] , with two different atoms per unit cell , one of which is light and thus supports lattice vibrations ( optical phonons ) . the coupling to this phonon mode modulates the hopping integrals , in direct analogy with the one - band ssh model.(for the vanishing carrier concentration of interest to us , the ssh - model is a one - band model because the peierls dimerization only occurs at half - filling @xcite ) . due to a phonon on site @xmath4 is also shown . ] given this similarity , one may expect our two - band model to behave like the ssh model , and indeed we find a sharp transition at strong carrier - hole coupling , where the polaron gs momentum changes its value . however , unlike in the one - band ssh model where the gs momentum changes continuously with coupling , in our model the gs momentum jumps discontinuously from @xmath5 to @xmath1 . furthermore , this transition leads to an extreme flattening of the polaron band , unlike in the ssh model . we conclude that this transition is qualitatively different from that of the one - band ssh model . we argue that the mechanism for the transition is a competition between the bare carrier hopping which favors a gs momentum @xmath5 , and the phonon - modulated hopping which favors @xmath1 . while some of this physics is similar to that explaining the transition in the ssh model , we find strong indications that the two - band nature of our model plays a vital role . this makes it unlikely that such a discontinuous transition can occur in a one - band model . the physics described in this article is therefore , to our knowledge , fundamentally new and our findings hint at the possibility that there is still more , new polaron physics to be discovered in other multi - band models . the physical inspiration for our model is the perovskite babio@xmath2 , which is known to have strong electron - phonon coupling and to exhibit superconductivity up to surprisingly high temperatures upon hole doping ( @xmath6 in the case of ba@xmath7k@xmath8bio@xmath2 @xcite and @xmath9 for bapb@xmath7bi@xmath10o@xmath2 ) @xcite , widely believed to be due to a phononic glue @xcite . the relevant valence orbitals are the bi 6@xmath11 orbitals and the o 2@xmath12 orbitals . previous work on polarons and bipolarons in this material was carried out by allen _ _ @xcite and is based on the rice - sneddon model @xcite which assumes that the bi atoms undergo charge disproportionation , bi@xmath13bi@xmath14bi@xmath15bi@xmath16 . this scenario has received wide - spread attention @xcite , and as a consequence many model hamiltonians only take into account the bi 6@xmath11 orbitals . polaronic signatures in agreement with this work have also been found experimentally @xcite . a different scenario is provided by a recent study of foyevtsova _ et . _ @xcite and the experimental as well as theoretical work of menushenkov _ et al _ @xcite . _ argue that the bio@xmath17 octahedra undergo a breathing distortion due to strong hybridization between the bi 6@xmath11 and o 2@xmath12 orbitals , with the holes being located primarily on the o. this picture is similar to that proposed recently for the nickelates @xcite and the inspiration for our `` toy model '' . in this picture both the bi 6@xmath11 and o 2@xmath12 orbitals need to be taken into account . since o atoms are much lighter than bi atoms , only optical phonons on the o atoms are considered and are allowed to modulate the hopping integrals . when compared to other perovskites such as the cuprates , manganites and nickelates , babio@xmath2 is appealing because of its comparatively simple electronic structure and absence of magnetic properties . to simplify things even more , instead of considering a model describing such a material at or near half filling , as it is in reality , we investigate the single polaron physics in its almost fully compensated case , _ i.e. _ like in labio@xmath2 with one extra hole . generalizing to one carrier ( polaron ) per unit cell will be left for future work . moreover , as indicated above , we restrict ourselves to study a 1d bio - like chain , instead of treating the full 3d system . there are two practical reasons for these simplifications : ( i ) they make comparison to the ssh model , where polaron results are currently available only for the 1d model , possible ; ( ii ) they allow us to use exact diagonalization ( ed ) to find essentially exact results very efficiently,@xcite which in turn also allows us to probe a wide range of coupling strengths and phonon frequencies to understand the relevant physics . beside using ed to understand the polaron properties of this model , we also develop and validate here two simple versions of the variational momentum average ( ma ) approximation,@xcite which capture the relevant polaronic physics qualitatively and even quantitatively . ma approaches are very useful because their accuracy improves in higher dimensions while maintaining similar computational efficiency . this is in contrast to ed and most other numerical methods that become very costly due to the significant increase of the hilbert space in higher dimensions . this work can therefore be seen as a first step towards a study of the 3d systems . note that apart from their usefulness in treating more complex problems , developing such approximations also leads to a better understanding of the nature of the polaron s cloud . to summarize , the research presented in this article serves two main purposes : ( i ) to reveal surprising , new polaron physics whose nature appears to be tied to the two - band nature of our model , and ( ii ) to serve as a test - ground for approximations which will be useful in solving the 3d many - band problem , and thus help to improve our understanding of the fascinating compound babio@xmath2 , if the scenario of foyevtsova _ et _ turns out to be valid , or of other many - band materials with hopping - modulated carrier - phonon coupling . the remainder of this article is organized as follows : in sec . [ sec : model ] we introduce the model and in sec . [ sec : methods ] we introduce the ed algorithm , ma and a perturbation theory . the results are presented in sec . [ sec : results ] and sec . [ sec : conclusions ] contains our conclusions . we study the single polaron properties of the 1d , two - band model sketched in fig . [ fig : model ] . there are two atoms per unit cell , one hosting valence electrons in an @xmath11-orbital , and the other in a @xmath12-orbital . the latter atom is assumed to be sufficiently light so that it is a good approximation to ignore the motion of the heavier ones . in other words , we assume that the lighter atoms oscillate inside the `` cage '' made of heavier atoms , giving rise to an optical phonon mode that modulates the @xmath11-@xmath12 hopping of the carriers . we consider the limit of a very lightly doped insulator , _ i.e. _ all states are filled except for a single hole present on the chain . the inspiration to study such a model was discussed in the previous section . the kinetic energy of the hole is described by a nearest neighbor tight - binding hamiltonian : @xmath18 here @xmath19 creates a hole on the @xmath11 ( @xmath12 ) orbital of the atoms in the @xmath20 unit cell ( the spin is an irrelevant degree of freedom here , and we do not write it explicitly ) . their fourier transforms are : @xmath21 and @xmath22 , where @xmath23 is the location of the unit cell @xmath24 , and the number of unit cells @xmath25 . the momentum @xmath26 is restricted to the first brillouin zone ( bz ) , @xmath27 . for an undistorted chain ( no phonons ) , and keeping in mind that @xmath28 and @xmath29 are hole operators , for the choice of lobe orientation shown in fig . [ fig : model ] we have @xmath30 , leading to the kinetic energy : @xmath31 the @xmath32 hopping parameter is given by the overlap of the @xmath11 and @xmath12-orbitals when the atoms are in their equilibrium positions . changing its sign corresponds to changing the sign convention for the @xmath12-orbitals , and therefore does not have any physical effect . as a consequence , the results we present below for a hole - doped chain remain identical for an electron - doped chain as well . when phonons are excited , the hopping amplitudes change from this equilibrium value , resulting in the hole - phonon coupling term discussed below . the hole s on - site energy depends on whether it sits on an @xmath11- or a @xmath12-orbital and leads to a charge transfer term : @xmath33 the difference in on - site energies can have either sign , favoring the @xmath12 ( if @xmath34 ) or @xmath11 ( if @xmath35 ) orbitals . the on - site energy for the @xmath11 orbitals is set to zero . the phonons are assumed to be described by a dispersionless einstein mode with energy @xmath36 ( we set @xmath37 ) : @xmath38 where @xmath39 creates a phonon on the @xmath12-orbital of the lighter atom of the @xmath20 unit cell . as discussed , the heavier atoms are taken to be frozen in their equilibrium positions . this model allows for two types of hole - lattice coupling . one comes from the modulation of on - site energies , because when the distance between neighbor atoms changes , so do the corresponding coulomb interactions . we are not aware of a comprehensive study of this type of coupling for this two - band model , although the asymptotic cases with @xmath40 ( the breathing - mode model @xcite ) and @xmath41 ( the double - well model@xcite ) have been studied and have revealed interesting polaronic behavior , although still rather conventional . instead , we focus here on the hole - lattice coupling arising from the fact that changes in the distance between adjacent atoms also modulate the hopping integrals . as we show below , this modulation of the kinetic energy leads to qualitatively new physics of a kind that , so far as we know , has not been revealed before . thus , the hole - phonon coupling that we study arises from the linear expansion of the hopping amplitudes @xmath42 and @xmath43 as a function of the small oxygen displacement @xmath44 . if we choose coordinates such that @xmath45 for a displacement toward the left , to linear order this expansion gives : @xmath46 = t[1+\tilde{\alpha } ( b_i + b_i^\dagger ) ] \\ & t_{i , i-1 } \approx -t[1-\alpha x_{i-1 } ] = -t[1-\tilde{\alpha } ( b_{i-1 } + b_{i-1}^\dagger)]\end{aligned}\ ] ] within this approximation @xmath47 , where after absorbing all the constants into the coupling @xmath48 , we find : @xmath49\end{aligned}\ ] ] we choose @xmath50 . note that because the sign of @xmath48 is controlled by the choice of the coordinate system , a change @xmath51 is equivalent with choosing @xmath45 for displacements toward the right . consequently the polaron properties only depend on the magnitude of @xmath48 , not on its sign . and @xmath52 for the lower band of energy @xmath53 . in both panels @xmath54 , @xmath55 and @xmath56 . ] the hamiltonian studied here is , therefore , @xmath57 before moving on , we briefly comment on its main limitations . keeping only linear terms in the expansion of the hopping integrals is a valid approximation when @xmath48 is sufficiently small , so that the polaron cloud creates rather small local distortions . for large values of @xmath48 , the effects of higher order coupling terms @xmath58 with @xmath59 , need to be considered because now the local distortions become large so the displacements @xmath60 may no longer be assumed to be small . this is true for all models with electron - phonon coupling . for the holstein and double - well potential models , the effect of such non - linear terms was shown to be significant in the strong coupling limit ( as defined by the linear term).@xcite another approximation is to consider only nearest neighbor ( nn ) hopping . coming back to the inspiration for this model , for a bio - like chain one could argue that the bi 6@xmath11 orbitals are quite broad and therefore next nearest neighbor ( nnn ) hopping between them may play a role . furthermore , in 3d models @xmath61 hopping between nn o 2@xmath12 orbitals needs to be included . in 1d , however , this type of hopping is much less relevant because o atoms are separated by at least on bi atom . finally , as already mentioned , hole - phonon coupling ( linear or to higher order ) resulting from the modulation of the on - site energies could also be included in this model . nevertheless , here we study polaron properties for the linear model discussed above , for a wide range of @xmath48 values , to explore its physics and to provide a baseline from which the effect of such additional terms can be gauged . we also note that all such further extensions can be studied with the methods we use in this work . the free - hole dispersion is obtained by diagonalizing @xmath62 . this results in two bands with eigenenergies @xmath63 , shown in fig . [ fig : h0](a ) . the lower band , which will be referred to as @xmath64 , has its minimum at @xmath5 where the hybridization between @xmath11 and @xmath12-orbitals is maximal . in contrast , the hybridization vanishes at @xmath1 . the free - hole weights @xmath65 are defined as @xmath66 and measure the overlap between the low - energy eigenstate @xmath67 and the free hole in a pure @xmath68 or @xmath12-state , respectively , and are given by : @xmath69 note that @xmath70 confirms that for @xmath1 , the low - energy free hole sits only on the @xmath12-orbitals . we are interested in the evolution of this low - energy band as the coupling to phonons is turned on , and in the nature of the resulting quasiparticle the polaron . a lot of insight can be gained from studying the anti - adiabatic limit @xmath71 , where at sufficiently small @xmath48 the phonon cloud of the polaron is very small , _ i.e. _ with at most one phonon present . the energy correction must be of order @xmath72 because @xmath73 changes the phonon number and thus has vanishing average value in the free hole ground state . consider the effects of @xmath73 . for an @xmath11-orbital hole , it allows it to hop onto the adjacent @xmath12-orbital while also emitting a phonon at this @xmath12-orbital . the phonon needs to be reabsorbed , which can occur in two ways : ( i ) by hopping to the next @xmath11-orbital , or ( ii ) by hopping back to the original @xmath11-orbital . the first process results in an effective @xmath11-@xmath11 hopping of amplitude @xmath74 , whereas the second changes the on - site @xmath11 energy by @xmath75 . for a hole starting from a @xmath12-orbital , emission of a phonon requires the hole to hop to an adjacent @xmath11-orbital . from there the phonon can only be reabsorbed if the hole returns to the original @xmath12-orbital . this process gives an additional on - site @xmath12 energy of @xmath76 . thus , the effective hamiltonian is : @xmath77 & t(1-{\mathrm{e}^{-i k a } } ) \\ t(1-{\mathrm{e}^{i k a } } ) & -2 t \lambda - \delta \\ \end{array } \right ) \label{eq : heff_matrix}\end{aligned}\ ] ] where we introduced the dimensionless , effective coupling @xmath78 . the lowest eigenenergy , _ i.e. _ the polaron dispersion , is given by : @xmath79 -\frac{\delta}{2 } \nonumber \\ & -\sqrt{\left [ t \lambda \cos(ka)-\frac{\delta}{2 } \right]^2 + 4t^2 \sin^2(\frac{ka}{2 } ) } \label{eq : eeff}\end{aligned}\ ] ] this gives a good approximation for the polaron band in the limit @xmath80 , as discussed below . the polaron eigenstates can also be obtained by ed . our implementation is a direct extension of the method proposed in ref . , which has already been successfully applied to a variety of polaronic models : the holstein model in various dimensions,@xcite a generalized holstein model with longer range interactions,@xcite the @xmath81-@xmath82 model with hole - phonon coupling , @xcite and the breathing mode model,@xcite to name a few . we briefly review it here . the hilbert space is spanned by the following translationally invariant basis states : @xmath83 here @xmath84 is an index identifying the orbital , such that @xmath85 and @xmath86 . @xmath87 defines specific phonon cloud configurations , and @xmath26 is the total momentum . following bona _ _ , we construct the hilbert space by acting @xmath88 times with the full hamiltonian @xmath89 on the free carrier states @xmath90 and @xmath91 and all the states which are created in this process . this quickly generates a large enough hilbert space to accurately calculate the ground state energy of the polaron with the lanczos technique.@xcite convergence is reached when an increase in the value of @xmath88 no longer produces a change in the eigenenergy . the number of states contained in this hilbert subspace for different values of @xmath88 is listed in table [ tab : nstates ] . .number of states in the hilbert space [ cols="^,^",options="header " , ] ma is an accurate variational method for calculating propagators of single polaron hamiltonians , from which polaron properties such as its energy and quasiparticle weight can be obtained . its simplest version was introduced for the holstein model,@xcite and then it was shown that it can be systematically improved by increasing the size of the variational space , _ i.e. _ which phonon configurations are included.@xcite apart from the holstein model , ma has also been shown to be very accurate for many other lattice polaron models including the bm model,@xcite the ssh model,@xcite and the double - well model.@xcite here we propose two versions of ma : ma@xmath92 for a variational space containing states with a one - site phonon cloud , and ma@xmath93 which also includes states with one additional phonon on a site adjacent to this one - site phonon cloud . ma@xmath93 is a restricted version of ma@xmath94 which allows the additional phonon to be at any distance from the cloud.@xcite it can also be viewed as a simplified version of the two - site cloud version described in ref . for the breathing mode model . the latter can be implemented easily for this model as well but is more cumbersome to generalize to higher dimensions . as we argue below , these simpler versions already suffice for our purposes . the green s functions ( gf ) of interest are : @xmath95 where again @xmath84 identifies the orbitals . @xmath96^{-1}$ ] is the resolvent of @xmath89 and @xmath97 is a small positive imaginary number indicating that we are computing retarded gfs . a complete derivation of the ma solution can be found in appendix a. we first present the ed results and discuss their meaning and implications . we then use them to gauge the accuracy of the ma@xmath92 and ma@xmath93 results . in fig . [ fig : egs_vs_g ] we plot the highest and lowest ed values of the polaron energy , @xmath98 and @xmath99 , vs. the effective coupling @xmath100 for two values of the phonon frequency . results are qualitatively similar for all other tested values of @xmath36 . for small @xmath100 , the polaron gs energy ( at @xmath0 ) is close to the free hole energy , @xmath101 . at @xmath1 , the polaron band lies just below the polaron+phonon continuum , so @xmath102 . with increasing @xmath100 , the polaron band moves to lower energies and its bandwidth narrows considerably , as the polaron becomes heavier . all this is standard polaronic physics . the surprise is that at sufficiently large @xmath100 , @xmath98 and @xmath99 cross , indicating that the gs momentum changes its value . for @xmath103 , in fig . [ fig : egs_vs_g](a ) , the scale of the graph makes it difficult to see the actual crossing , but its occurrence is verified below , in fig . [ fig : dispersion_close_to_gc ] . it is not a priori obvious that the new gs momentum is necessarily at @xmath1 , but we will show below that this is the case . consequently , we define the critical coupling @xmath104 ( or @xmath105 ) as the value for which @xmath106 . and @xmath99 vs. the effective coupling @xmath100 for @xmath103 ( left panel ) , and @xmath107 ( right panel ) , at @xmath108 . convergence was reached for @xmath109 , respectively . ] and @xmath110 for @xmath103 ( left panel ) and @xmath107 ( right panel ) . note that @xmath111 , because the @xmath11 and @xmath12-orbitals do not hybridize at @xmath1 . the vertical , dashed line marks @xmath104 . convergence was reached for @xmath109 , respectively . ] changing @xmath112 results in a shift of the polaron band . the polaron energy must be smaller than that of the free hole , and the latter is shifted downwards ( upwards ) for @xmath34 ( @xmath35 ) . a similar shift is therefore expected at least for small @xmath100 , and it is seen to appear for all @xmath100 . the effect of @xmath112 on @xmath104 is discussed below . before discussing the nature of the transition in detail , we quickly analyze the qp weight , @xmath113 , defined as the overlap between the polaron eigenstate and the @xmath11 and @xmath12 free - hole states , respectively . they are shown in fig . [ fig : zgs_vs_g ] . note that @xmath114 for any finite @xmath100 , because the @xmath11 and @xmath12 orbitals do not hybridize at @xmath1 ( see eq . ( [ eq : heff_matrix ] ) ) and is therefore not shown . this is in stark contrast to the free hole case , where at @xmath1 all the free - hole weight is on the @xmath12-orbital . consequently there is a discontinuous change in the @xmath1 qp weight when the hole - phonon coupling is turned on . for @xmath0 the qp weight falls off rapidly as @xmath100 increases , whereas for @xmath1 its projection on the @xmath11-orbitals first increases and then falls off more slowly . as pointed out above , this initial increase in qp weight occurs because at @xmath1 and for sufficiently small @xmath100 , the polaron band lies just below the polaron+phonon continuum . the vertical dashed lines mark @xmath104 , the value of @xmath100 at which the @xmath0 and @xmath1 polaron energies cross . note that there is no sudden change in the qp weights at @xmath104 . this is a strong hint that the crossover is not due to a change in the nature of the phonon cloud . ( left panel ) and @xmath115 ( right panel ) for @xmath103 . note that @xmath111 , because the @xmath11 and @xmath12-orbitals do not hybridize at @xmath1 . the vertical , dashed line marks @xmath104 . convergence was reached for @xmath116 the dependence of the qp weight on @xmath112 is shown in fig . [ fig : zgs_vs_g_diff_delta ] for @xmath103 . as expected , @xmath34 increases the amount of @xmath12 character and decreases the @xmath11 character ; @xmath117 has the opposite effect . for larger values of the coupling @xmath100 , the change in qp weights due to @xmath112 becomes negligibly small . at these values of @xmath100 the phonon cloud is already quite sizable , as indicated by the small values of the qp weights . together , fig . [ fig : egs_vs_g ] and fig . [ fig : zgs_vs_g_diff_delta ] show that for strong coupling , @xmath112 shifts the energy of the polaron but does not change the nature of its phonon cloud . this , in turn , suggests that @xmath118 strongly favors a specific kind of phonon cloud . in other words , there appears to be only one dominant mechanism that allows the hole to lower its energy via the emission and absorption of phonons . and values of @xmath119 . the curves have been shifted by @xmath98 for better comparison and the curve at @xmath104 is highlighted in red . the green , dashed lines in the rightmost panel are the perturbative results from eq . ( [ eq : eeff ] ) . all curves are for @xmath120 and the cutoff is @xmath121 , respectively . ] the ( shifted ) polaron dispersion @xmath122 vs @xmath26 , close to the critical coupling @xmath105 , is shown in fig . [ fig : dispersion_close_to_gc ] for @xmath123 and @xmath124 . note that close to @xmath105 , the polaron bandwidth is extremely narrow . these results show that the gs momentum does indeed change discontinuously from @xmath125 to @xmath126 , justifying our definition of @xmath105 . for @xmath127 , the prediction of eq . ( [ eq : eeff ] ) ( dashed lines ) agrees quite well with ed and clearly reproduces the transition . as we show now , a closer examination of this perturbative result indeed unveils a lot of the physics governing this transition . when @xmath128 , the polaron dispersion given by eq . ( [ eq : eeff ] ) is not much different from the free hole dispersion @xmath129 , with the ground state at @xmath0 . for @xmath130 , on the other hand , the polaron dispersion can be approximated as : @xmath131 - \frac{\delta}{2},\end{aligned}\ ] ] with the ground state at @xmath1 and a flat band for @xmath132 . consequently there is a critical value , @xmath104 , where the ground state momentum changes . indeed it can be verified that for @xmath133 the @xmath26-dependence in the square root of eq . ( [ eq : eeff ] ) exactly cancels the @xmath134 term and the band becomes completely flat , @xmath135 . for @xmath136 the ground state is at @xmath0 , and for @xmath137 it is at @xmath1 . it is clear that this change of the ground state momentum is due to a competition between the antisymmetric , bare hopping @xmath138 and the symmetric , phonon - modulated hopping , @xmath118 . to our knowledge this is the first report of a change in the ground state momentum caused by a symmetric phonon - modulated hopping ; such changes have been seen before only for models with antisymmetric , phonon - modulated hopping.@xcite however , while for the latter case the ground - state momentum changes _ smoothly _ for @xmath139 , here we observe a _ discontinuous jump _ from the edge to the centre of the bz . note , furthermore , that the hybridization between the two bands which leads to the square root in eq . ( [ eq : eeff ] ) plays an important role for the nature of the transition and appears to be responsible for the flatness of the band close to @xmath104 . although perturbation theory explains many features of the transition , there are some differences between the large @xmath36 case and the cases where @xmath36 is comparable to @xmath81 . ( [ eq : eeff ] ) predicts a completely flat band at @xmath104 . while this is seen for @xmath127 , it is no seen for smaller @xmath36 . instead , here higher order processes leading to phonon - mediated , longer range effective hopping , cause the bandwidth to remain finite at all values of @xmath100 . we can attribute these processes to longer range effective hopping because they lead to a maximum at @xmath140 , halving the bz . their contribution is very small , as indicated by the small bandwidth close to @xmath104 . this leads us to conclude that the phonon - assisted hopping process described in section [ sec : perturbation - theory ] is indeed primarily responsible for most of the mobility of the polaron . to make these arguments more compelling and valid for phonon clouds with more than one phonon , we now analyze the nature of the polaron cloud in more detail . for @xmath1 , the polaron has qp weight on @xmath141 but not @xmath142 . acting with @xmath118 on @xmath143 ( and ignoring normalization factors ) gives @xmath144 , which we therefore expect to be a part of the polaron wavefunction . acting on this again with @xmath118 gives @xmath145 | 0\rangle$ ] . this pattern continues and we find that for @xmath1 the polaron eigenfunction contains states of the type @xmath146 , _ i.e. _ with an odd number of phonons and the hole on the same @xmath12-orbital , and @xmath147 | 0\rangle$ ] , _ i.e. _ a symmetric state with an even number of phonons and the hole on the adjacent @xmath11-orbitals . while we only considered states with a one - site phonon cloud adjacent to the hole , this structure of the phonon cloud is indeed verified by ed , which finds very small weight for other configurations . for @xmath0 , consequently we need to start the construction of the eigenstate from @xmath149 . acting with @xmath118 gives @xmath150 . this pattern generalizes and mixes states of the form @xmath151 , _ i.e. _ with an even number of phonons and the hole on the same @xmath12-orbital , and @xmath152 |0\rangle$ ] , _ i.e. _ an antisymmetric state with an odd number of phonons and the hole on the adjacent @xmath11-orbitals . this also is verified by ed . thus , the polaron cloud structure is very different at @xmath1 and @xmath125 , and this has important consequences . consider the configuration latexmath:[$\sum_j { \mathrm{e}^{ikr_j } } s_{j}^\dagger b_{j}^{\dagger,2n } times with @xmath118 ; among other states , this links to @xmath154 . acting another @xmath155 times with @xmath118 gives ( among many other states ) @xmath156 , _ i.e _ the original state translated by one site . clearly , such processes contribute to the mobility of the polaron . similarly , states like @xmath157 can be moved by first acting @xmath158 times with @xmath118 linking to @xmath159 , and then to @xmath160 after applying @xmath118 another @xmath158 times . for these processes to work , it is crucial that the number of phonons is even ( odd ) if the carrier is on an @xmath161 orbital . this is the case for @xmath1 , but for @xmath0 we found that exactly the opposite is the case . this explains why the phonon - modulated effective hopping , which dominates at large couplings , favours @xmath1 and this eventually becomes the ground - state . note , furthermore , that this discussion also illustrates why an ma version restricted to a one - site phonon cloud is expected to capture well the phenomenology of this model . defined in eq . ( [ eq : nph ] ) for @xmath103 ( left panel ) and @xmath107 ( right panel ) at different values of @xmath100 . other parameters are @xmath0 and @xmath120 . convergence was reached for @xmath162 and @xmath163 , respectively . ] to analyze the spatial extent of the polaron cloud we calculate the phonon distribution @xmath164 where @xmath165 is the polaron - state with momentum @xmath0 and @xmath166 is the distance between the carrier at site @xmath167 and the phonon at site @xmath168 . this means that @xmath169 when the carrier is on a @xmath12-orbital , and @xmath170 when it is on an @xmath11-orbital . the results are shown in fig . [ fig : nph ] for @xmath103 and @xmath107 . we see that the polaron cloud is located in the immediate vicinity of the carrier , _ i.e. _ we are dealing with a very small polaron . if the carrier is on an @xmath11-orbital the majority of phonons are hosted by the two oxygens next to it , while for a carrier on a @xmath12-orbital the three closest oxygen sites contribute . as @xmath100 is increased past the critical coupling ( @xmath171 for @xmath103 and @xmath172 for @xmath107 ) , there is no noticeable change in the nature of the cloud . we also find that for @xmath1 and sufficiently large @xmath100 the phonon cloud is qualitatively similar ( not shown ) to that for @xmath0 . for small @xmath100 this is not true as here the polaron state at @xmath1 is not well - separated from the polaron + phonon continuum . reducing @xmath36 drastically changes the overall number of phonons , but not the spatial extent of the cloud . normalized by the free hole mass @xmath173 for @xmath107 ( left panel ) and @xmath127 ( right panel ) and @xmath120 . the dashed line marks @xmath104 . the values of @xmath88 are 14 and 10 , respectively . ] apart from the small bandwidth , the heavy nature of the polaron also becomes apparent from the value of its inverse effective mass , @xmath174 , shown in fig . [ fig : meff ] for @xmath175 at @xmath107 and @xmath127 . note that we normalize this with the free hole mass . as @xmath176 from below , @xmath177 decreases indicating that the polaron becomes heavier . this is expected because as the discussion above shows , phonon - modulated hopping can not move the polaron cloud at this momentum , and the bare hole hopping is renormalized to smaller values due to the presence of the phonon cloud . at @xmath178 the ground state changes momentum and consequently we now need to follow @xmath179 which increases because here the phonon - modulated hopping is active and increases the mobility of the polaron . for large values of @xmath36 the increase of @xmath179 for @xmath180 is substantial resulting in a much lighter polaron than at @xmath181 . for smaller values of @xmath36 , on the other hand , @xmath179 levels off quite rapidly and the polaron remains heavy . this is to be expected because for smaller @xmath36 , the phonon cloud is quite large at @xmath104 , suppressing the polaron s mobility . for @xmath71 , on the other hand , the average phonon number is small and an increase in @xmath100 results in an increase of the effective hopping between @xmath11-orbitals . this effective hopping can become larger than the direct hopping , resulting in a lighter polaron . a major difference between the two panels of fig . [ fig : meff ] is that for @xmath107 the inverse effective mass at @xmath1 has a minimum at finite @xmath100 . again , this is due to the presence of the polaron+phonon continuum , which for sufficiently small @xmath100 forces the polaron band to flatten out near @xmath1 . indeed , the location of this minimum agrees with the location of the maximum in @xmath182 in fig . [ fig : zgs_vs_g ] . let us now discuss the role of the charge transfer energy @xmath112 . in fig . [ fig : lambda_c ] we show the critical coupling @xmath104 for different values of @xmath112 and @xmath36 . the dashed lines show the perturbative result , eq . ( [ eq : lambda_c ] ) , which for sufficiently large @xmath36 is in good agreement with the ed results . as a function of @xmath36 and @xmath112 . different values of @xmath88 were used to ensure convergence . dashed lines are the perturbative result from eq . ( [ eq : lambda_c ] ) . ] the critical coupling increases with increasing @xmath112 . this is not surprising since a large value of @xmath112 favors the @xmath12 character , whereas the phonon - modulated hopping favors the @xmath11 character . consequently , a negative value of @xmath112 facilitates the transition . however , this effect is significant only for relatively large values of @xmath36 . as @xmath36 decreases the variation of @xmath104 with @xmath112 becomes very small , suggesting that here @xmath112 has little influence on the nature of the polaron cloud . this agrees with the conclusions drawn from fig . [ fig : zgs_vs_g_diff_delta ] . at small values of @xmath36 the value of @xmath104 is also much larger than that predicted by perturbation theory . however , @xmath183 and in fact the transition from @xmath1 to @xmath0 is actually achieved at smaller values of @xmath48 for small @xmath36 ( see inset of fig . [ fig : ma_gc ] ) . for these values of @xmath36 the phonon cloud is quite large and therefore it is to be expected that the bare hopping , favoring @xmath0 , is renormalized substantially and the phonon - modulated hopping , favoring @xmath1 , wins already at smaller values of @xmath48 . and ma@xmath93 . ( a ) @xmath184 for @xmath185 at different values of @xmath100 . ( b ) the absolute value of the difference in % . ( c ) the dispersion @xmath184 at @xmath186 . ( d ) @xmath184 at @xmath187 . in all cases @xmath103 and @xmath120 . ] this concludes our analysis of the ed results . we now briefly compare the ma predictions with the ed results , in order to validate our choice of the variational space . this is useful because ma can be much more easily and efficiently extended to higher dimensions than ed calculations . moreover , prior work@xcite has shown that any version of ma becomes more accurate in higher dimensions , where the bare propagators decay faster with distance outside the free - hole continuum . note also that ma accuracy improves with increasing phonon frequency @xmath36;@xcite this is why we present results for @xmath103 where the phonon cloud is very large at strong couplings , posing a challenging test for this ( and any other ) approximation . in panel ( a ) of fig . [ fig : ma_1 ] we compare the ma results for the polaron energies @xmath188 at different values of @xmath100 , to the ed results . as expected for a variational approach , both ma@xmath92 and ma@xmath93 values are always larger than the ed ones . panel ( b ) of fig . [ fig : ma_1 ] shows the relative difference between ma and ed . at small values of @xmath100 both ma@xmath92 and ma@xmath93 are very accurate , but at larger values of @xmath100 , ma@xmath93 is clearly superior , showing that its additional configurations acquire finite weight . obviously , adding more cloud configurations will further increase accuracy , but it is clear that this rather small set already suffices to capture quantitatively quite accurately the polaron properties . a comparison for the dispersion @xmath184 is shown in panel ( c ) of fig . [ fig : ma_1 ] for the intermediate coupling @xmath186 and in panel ( d ) for the critical coupling @xmath189 . panel ( c ) shows that both ma@xmath92 and ma@xmath93 give the best results for @xmath0 . this is probably due to the fact that at @xmath186 the gs is still at @xmath0 , and a variational method like ma is expected to perform best for the gs . note that panel ( b ) also shows that for @xmath190 the relative error is smaller for @xmath0 than for @xmath1 . the dispersion shown in panel ( d ) is very narrow ( on this scale ) . ma@xmath92 and ma@xmath93 both reproduce the small bandwidth quite well , but shifted to higher energies by 2% and 6% , respectively . however , neither ma@xmath92 nor ma@xmath93 predict the exact value of @xmath189 . instead , ma@xmath92 predicts a smaller value of 1.1 , whereas ma@xmath93 predicts a larger value of 1.45 . this trend is true for all values of @xmath36 , as shown in fig . [ fig : ma_gc ] . here we also see that ma@xmath93 gives very good predictions for @xmath191 . the agreement becomes gradually worse as @xmath192 , where because of the low cost of phonons , the spatial extent of the phonon cloud increases beyond two sites . it also needs to be pointed out that for small values of @xmath36 small differences in the hole - phonon coupling @xmath48 are amplified in the effective coupling @xmath193 . for small @xmath36 we therefore show the values of @xmath105 in the inset of fig . [ fig : ma_gc ] . obtained with ed for @xmath120 , and the predictions of ma@xmath92 and ma@xmath93 . ] and ma@xmath93 . ( a ) the qp weight @xmath194 for @xmath125 for different values of @xmath100 . note that @xmath111 for all @xmath100 . ( b ) @xmath194 for @xmath195 . in all cases @xmath103 and @xmath120 ] a comparison of the qp weights obtained with ma and ed is shown in fig . [ fig : ma_2 ] . panel ( a ) shows @xmath197 ( as pointed out above , @xmath198 for all @xmath100 ) . panel ( b ) shows @xmath199 for @xmath1 and @xmath0 . as expected for a variational approximation , the ma weights are larger than the ed values everywhere . for small @xmath100 both ma@xmath92 and ma@xmath93 perform very well . the largest quantitative disagreement appears for intermediate values of @xmath100 . here , @xmath200 falls off quite rapidly , and this is captured only qualitatively by ma . note that ma@xmath93 performs much better than ma@xmath92 in this @xmath100-range , indicating that a further improvement of ma by adding phonons on adjacent sites is possible . the results improve substantially for @xmath139 . at these values , the qp weight does not change as rapidly anymore . we find that here ma@xmath92 and ma@xmath93 are on equal footing , with ma@xmath92 actually outperforming ma@xmath93 at @xmath201 . this could be related to the fact that ma@xmath93 predicts a value that is too large for @xmath104 , whereas ma@xmath92 predicts a value that is smaller than the exact result . we have investigated the effect of phonon - modulated hopping in a two - band model describing a chain with alternating atoms with @xmath11 and @xmath12 valence orbitals , respectively . as discussed in the introduction , part of the motivation for this work is to study a toy 1d model inspired by the perovskite labio@xmath2 ( or nearly fully compensated babio@xmath2 ) in the scenario where holes are primarily located on the o atoms , to understand the properties of the resulting polaron and to develop a good approximation that can be easily generalized for similar 3d system . the key finding is that at sufficiently large hole - phonon coupling , the polaron undergoes a sharp transition where its gs momentum jumps from @xmath125 to @xmath126 . sharp polaron transitions have been observed previously in the ssh and related models with phonon - modulated hopping , @xcite however there are qualitative differences between our results and those of these studies . in these other models the momentum changes _ smoothly _ as the effective coupling is increased . also , the polaron mass diverges at the transition but remains finite ( and can be surprisingly light ) in the limit of infinite effective coupling . in our model , on the other hand , there is a _ discontinuous _ jump of the momentum from @xmath0 to @xmath1 . to the best of our knowledge , such a discontinuous transition has not been discussed before . these results suggest that there may be more new physics to be discovered in models of carrier - phonon coupling different from that of the much studied one - band models of the holstein and fr " ohlich type . we argued that this transition is due to a competition between the bare hopping @xmath138 and the phonon - modulated hopping @xmath118 : the former favors a @xmath0 ground state , while the latter increases polaron mobility at @xmath1 but not at @xmath0 . this analysis suggests that a transition can only occur when @xmath138 and @xmath118 have different symmetries . this is supported by the findings of zhang _ et . @xcite who report a transition in an ssh - like model for symmetric bare hopping and antisymmetric phonon - modulated hopping , but not for symmetric phonon - modulated hopping . we have used arguments based on perturbation theory , but also more general symmetry arguments , to support the view that the discontinuous nature of this transition is closely tied to the two - band nature of the model . in perturbation theory the hybridization between the @xmath11 and @xmath12 bands leads to the square root dependence in eq . ( [ eq : eeff ] ) , which is the main ingredient needed for this type of discontinuous jump of the gs momentum . we believe that it is highly unlikely that a transition with such a discontinuous jump can be found in a one - band model . this suggests that the polaron properties of multi - band models are qualitatively different from those of one - band models , and that more work needs to be done to exhaustively study all possible types of polaron transitions . whether such transitions also occurs in 2d and 3d models where nn hopping between oxygens is also included is not a priori clear ; we believe this to be the case and we are currently investigating such models . when extended to one - site phonon cloud configurations with an arbitrary number of phonons , the arguments mentioned above allowed us to elucidate the mechanism behind the transition . they also suggested two types of ma variational approximations , where the phonon cloud is restricted to be one - site in extent ( ma@xmath92 ) and one - site cloud with at most one additional phonon on an adjacent site ( ma@xmath93 ) . we find that ma@xmath92 reproduces the gs energy with an accuracy of 6% or better while ma@xmath93 yields even better results with an accuracy of at least 2% . these values were found for @xmath103 where the phonon cloud is already very large , and therefore show that ma performs well even under difficult conditions . of course , the ma accuracy is significantly less for the qp weights at intermediate coupling @xmath100 . this is not surprising , because the part of the eigenstates ignored by the variational calculation does contribute to the wavefunctions normalization ( and therefore serves to lower the qp weight ) even if it may not have much effect on the eigenenergy . it is important to note , however , that both ma approaches reproduce the qualitative behavior correctly , and that adding more variational states will further improve quantitative agreement . ma is also successful in predicting an accurate value for the critical coupling @xmath104 for not too small phonon frequencies . at small @xmath36 , @xmath104 is underestimated by ma@xmath92 and overestimated by ma@xmath93 , so at least they provide bounds for its true value . the major advantage of ma , when compared to ed , is its numerical cost . for ma@xmath92 one merely needs to multiply @xmath202-matrices , whereas for ma@xmath93 the largest matrix size is @xmath203 . in fact one can also store all of the @xmath204 and @xmath205 in one large sparse matrix , which further increases performance . the essential aspect is that in both cases the number of states for a given number @xmath206 of phonons is independent of @xmath206 . this is why generalizing these ma approaches to 3d will be significantly more efficient than for more general ma schemes , like that of ref . . we have shown that in 1d these schemes already work quite well , and since ma improves its accuracy in higher dimension , these generalizations should suffice to accurately and efficiently obtain polaron properties in the multi - band , 3d model . this work is now in progress . we thank george sawatzky for suggesting this problem , and c.p.j . adolphs for sharing his experience with ed . this work was supported by nserc , qmi and the ubc 4yf ( m.m.m . ) . the equations of motion ( eom ) for @xmath207 can be constructed by repeatedly using dyson s identity @xmath208 , where @xmath209 and @xmath210 is the resolvent for @xmath211 . choosing @xmath212 and doing this once yields : @xmath213 where we defined the generalized gfs : @xmath214 these describe the projection onto states with a one - site phonon cloud , and are therefore the only gfs included in ma@xmath92 . note that we suppressed the @xmath215 index in the definition of the generalized gfs since the eom do not couple to gfs with different @xmath215 . for ma@xmath93 , we also include the contribution from the generalized gfs : @xmath216 with one additional phonon either to the left or to the right of the phonon cloud . note that for @xmath217 , @xmath218 . we now apply dyson s identity again , to obtain the eom for the generalized gfs . suppressing the @xmath219 argument for simplicity , we find : @xmath220 @xmath221 @xmath222 where we introduced the free hole real - space gfs : @xmath223 for simplicity its @xmath224 argument was again suppressed in the equations above . in 1d , these free hole real - space gfs can be obtained analytically , see appendix b. the right hand side of these eom only links to generalized gfs with specific values of @xmath225 . let us first consider the case of ma@xmath92 . here we only consider generalized gfs of the type @xmath226 . by inspection of eq . ( [ eq : eom_f ] ) we find that the eom only link to @xmath227 and @xmath228 . furthermore since @xmath229 and @xmath230 and @xmath231 ( see appendix b ) the eom simplify to : @xmath232 we define the column vector @xmath233 and recast the eom in the form @xmath234 where @xmath235 and @xmath236 are @xmath237-matrices which for @xmath238 can be read off directly from the equations above . to get @xmath239 we use @xmath240 and @xmath241 , where @xmath215 is the same as in eq . ( [ eq : eom_g ] ) . the eom can now be solved with the ansatz @xmath242 , @xcite which is justified because for large @xmath206 , @xmath243 goes to zero and therefore @xmath244 must go to zero as well . plugging this ansatz back into eq . ( [ eq : recurrence ] ) yields @xmath245^{-1}\tilde{\alpha}_n$ ] which can be calculated recursively starting with @xmath246 , where @xmath247 is chosen sufficiently large so that its further increase has no effect on the results . once we have calculated @xmath248 linking @xmath249 to @xmath250 , we can rewrite eq . ( [ eq : eom_g ] ) in matrix form : @xmath251 note that this requires identifying the first row of @xmath252 with @xmath253 and the second row with @xmath254 . @xmath255 is the @xmath202-matrix with @xmath256 . the self - energy @xmath257 is defined by the @xmath258-matrix equation @xmath259^{-1}$ ] . in the ma@xmath92 approximation it is therefore given by : @xmath260 we now show how to use the ma@xmath92 eom to rigorously derive the perturbation result of sec . [ sec : perturbation - theory ] . for a cutoff of @xmath261 , we find that @xmath262 and consequently @xmath263 for @xmath264 we have @xmath265 and @xmath266 ( see appendix b ) and we recover the result of eq . ( [ eq : heff_matrix ] ) . the ma@xmath93 case is treated in exactly the same manner , but is slightly more tedious . for @xmath267 phonons we include the following 17 gfs : @xmath268 with @xmath269 ; @xmath270 with @xmath271 ; @xmath272 with @xmath273 ; @xmath274 with @xmath275 ; @xmath276 with @xmath277 ; and @xmath278 with @xmath279 . for @xmath280 , some of these gfs are identical so that their number is reduced to 12 . similarly , for @xmath217 one only needs to keep 7 gfs . the generalized gfs are again arranged in a vector @xmath281 and the eom recast as a recurrence equation : @xmath282 the matrices @xmath204 and @xmath205 can be read off from eqs . ( [ eq : eom_f ] ) , ( [ eq : eom_f(1,n ) ] ) and ( [ eq : eom_f(n,1 ) ] ) . furthermore , eq . ( [ eq : def_f ] ) indicates that @xmath283 which we use to read off @xmath284 . again the eom are solved with the ansatz @xmath285 @xcite , yielding @xmath286^{-1}\alpha_n$ ] . we then rewrite eq . ( [ eq : eom_g ] ) in matrix form : @xmath287 here @xmath288 is a reduced version of @xmath289 . it is a @xmath290-matrix which only contains the rows of @xmath289 linking @xmath291 and @xmath292 to @xmath293 . this is necessary since the other 4 generalized gfs contained in @xmath294 do not appear in eq . ( [ eq : eom_g ] ) . the matrix @xmath88 is a @xmath295-matrix whose only non - zero elements are @xmath296 and @xmath297 . the self - energy @xmath257 in the ma@xmath93 approximation is , then : @xmath298 the real - space gfs of @xmath299 are defined as @xmath300 . this can be rewritten as @xmath301 , _ i.e. _ @xmath302 measures the probability amplitude that a hole injected at site 0 will be removed at site @xmath225 . the eom are obtained using dyson s identity . for @xmath303 , and suppressing the @xmath219-dependence , we find : @xmath304 where we defined the shorthand @xmath305 . for @xmath306 , the general form of the eom is @xmath307 eliminating @xmath308 , the eom for @xmath309 , with @xmath310 , can be recast as @xmath311 = -t^2 g_{n-1}^{p , p } - t^2 g_{n+1}^{p , p}\end{aligned}\ ] ] similarly we find @xmath312 in the time - domain the small imaginary part @xmath313 corresponds to a finite lifetime of the hole . therefore the probability that a hole travels from site 0 to site @xmath314 falls off exponentially as @xmath315 . and we use the ansatz @xmath316 , for @xmath317 . for @xmath318 we need to use @xmath319 . in both cases , plugging the ansatz back into the eom gives : @xmath320 we need to choose the solution which satisfies @xmath321 . using the ansatz in eq . ( [ eq : eom_g0pp ] ) we obtain @xmath322 from this all the other @xmath309 are obtained as @xmath323 . we can apply the same procedure to find analytical expressions for the @xmath308 . however , we need to be careful since both @xmath324 and @xmath325 link to @xmath326 and therefore need to be treated separately . after some algebra we find @xmath327 @xmath328 similarly one finds : @xmath329 @xmath330 note that for @xmath331 , eq . ( [ eq : z ] ) implies that @xmath332 . using this in the expressions for @xmath333 and @xmath334 we find that they scale as @xmath335 . the diagonal real - space gfs , @xmath336 and @xmath337 on the other hand scale as @xmath338 . all real - space gfs with larger values of @xmath339 go to 0 since @xmath340 . 54ifxundefined [ 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/physrevlett.92.146403 [ * * , ( ) ] \doibase 10.1023/a:1007792713484 [ * * , ( ) ] link:\doibase 10.1038/nmat3378 [ * * , ( ) ] http://stacks.iop.org/0953-8984/20/i=4/a=043201 [ * * , ( ) ] link:\doibase 10.1080/00018730210140805 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.106404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.144101 [ * * , ( ) ] link:\doibase 10.1103/physrevb.50.7222 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.121114 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.256402 [ * * , ( ) ] link:\doibase 10.1103/physrevb.60.1633 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.174306 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.104304 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.115114 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.165113 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.125131 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.97.036402 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.165109 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.174305 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.035453 [ * * , ( ) ] link:\doibase 10.1103/physrevb.78.235120 [ * * , ( ) ] link:\doibase 10.1103/physrevb.89.035122 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.085149 [ * * , ( ) ] link:\doibase 10.1080/14786445008521794 [ * * , ( ) ] link:\doibase 10.1080/00018735400101213 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.42.1698 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.266605 [ * * , ( ) ] link:\doibase 10.1063/1.4733986 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.223002 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.63.63 [ * * , ( ) ] link:\doibase 10.1038/332814a0 [ * * , ( ) ] link:\doibase 10.1016/0038 - 1098(75)90327 - 0 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.017001 [ * * , ( ) ] link:\doibase 10.1103/physrevx.3.021011 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.224509 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.115112 [ * * , ( ) ] link:\doibase 10.1103/physrevb.56.3105 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.47.689 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0038-1098(76)90632-3 [ * * , ( ) ] link:\doibase 10.1107/s0567740879002417 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.61.2713 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.174103 [ * * , ( ) ] link:\doibase 10.1016/j.physc.2015.02.012 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.95.176403 [ * * , ( ) ] @noop ( ) , http://stacks.iop.org/0953-8984/12/i=16/a=303 [ * * , ( ) ] link:\doibase 10.1134/1.1410607 [ * * , ( ) ] link:\doibase 10.1103/physrevb.61.11263 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.156402 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.126404 [ * * , ( ) ] http://stacks.iop.org/0295-5075/102/i=4/a=47003 [ * * , ( ) ] link:\doibase 10.1103/physrevb.77.054519 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.66.763 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.245104 [ * * , ( ) ]

we generalize the notions of flippable and simultaneously flippable edges in a triangulation of a set of points in the plane to so - called _ pseudo - simultaneously flippable edges_. such edges are related to the notion of convex decompositions spanned by . we prove a worst - case tight lower bound for the number of pseudo - simultaneously flippable edges in a triangulation in terms of the number of vertices . we use this bound for deriving new upper bounds for the maximal number of crossing - free straight - edge graphs that can be embedded on any fixed set of points in the plane . we obtain new upper bounds for the number of spanning trees and forests as well . specifically , let denote the maximum number of triangulations on a set of points in the plane . then we show ( using the known bound ) that any-element point set admits at most crossing - free straight - edge graphs , spanning trees , and forests . we also obtain upper bounds for the number of crossing - free straight - edge graphs that have , fewer than , or more than edges , for any constant parameter , in terms of and .

1012.0591

a _ crossing - free straight - edge graph _ @xmath9 is an embedding of a planar graph in the plane such that the vertices are mapped to a set @xmath0 of points in the plane and the edges are pairwise non - crossing line segments between pairs of points in @xmath0 . ( segments are allowed to share endpoints . ) by fry s classical result @xcite , such an embedding is always possible . in this paper , we fix a labeled set @xmath0 of points in the plane , and we only consider planar graphs that admit a straight - edge embedding with vertex set @xmath0 . by _ labeled _ we mean that each vertex of the graph has to be mapped to a unique designated point of @xmath0 . analysis of the number of plane embeddings of planar graphs in which the set of vertices is not restricted to a specific embedding , or when the vertices are not labeled , can be found , for example , in @xcite . a _ triangulation _ of a set @xmath0 of @xmath1 points in the plane is a maximal crossing - free straight - edge graph on @xmath0 ( that is , no additional straight edges can be inserted without crossing some of the existing edges ) . triangulations are an important geometric construct which is used in many algorithmic applications , and are also an interesting object of study in discrete and combinatorial geometry ( recent comprehensive surveys can be found in @xcite ) . improving the bound on the maximum number of triangulations that any set of @xmath1 points in the plane can have has been a major research theme during the past 30 years . the initial upper bound @xmath10 of @xcite has been steadily improved in several paper ( e.g. , see @xcite ) , culminating with the current record of @xmath11 due to sharir and sheffer @xcite . other papers have studied lower bounds on the maximal number of triangulations ( e.g. , @xcite ) , and upper or lower bounds on the number of other kinds of planar graphs ( e.g. , @xcite ) . every triangulation of @xmath0 contains the edges of the convex hull of @xmath0 , and the remaining edges of the triangulation decompose the interior of the convex hull into triangular faces . assume that @xmath0 contains @xmath1 points , @xmath12 of which are on the convex hull boundary and the remaining @xmath13 points are interior to the hull ( we use this notation throughout ) . by euler s formula , every triangulation of @xmath0 has @xmath14 edges ( @xmath12 _ hull _ edges , common to all triangulations , and @xmath15 _ interior _ edges , each adjacent to two triangles ) , and @xmath16 bounded triangular faces . can be flipped to the edge @xmath17 . ( b ) the two bold edges are simultaneously flippable . ( c ) interior - disjoint convex quadrilateral and convex pentagon in a triangulation.,title="fig:",scaledwidth=80.0% ] [ [ edge - flips . ] ] edge flips . + + + + + + + + + + + edge flips are simple operations that replace one or several edges of a triangulation with new edges and produce a new triangulation . as we will see in section [ se : apps ] , edge flips are instrumental for counting various classes of subgraphs in triangulations . in the next few paragraphs , we review previous results on edge flips , and propose a new type of edge flip . we say that an interior edge in a triangulation of @xmath0 is _ flippable _ , if its two adjacent triangles form a convex quadrilateral . a flippable edge can be _ flipped _ , that is , removed from the graph of the triangulation and replaced by the other diagonal of the corresponding quadrilateral , thereby obtaining a new triangulation of @xmath0 . an edge flip operation is depicted in figure 1(a ) , where the edge @xmath18 is flipped to the edge @xmath17 . already in 1936 , wagner @xcite has shown that any _ unlabeled abstract _ triangulation @xmath19 ( in this case , two triangulations are considered identical if we can relabel and change the planar embedding of the vertices of the first triangulation , to obtain the second triangulation ) can be transformed into any other triangulation @xmath20 ( with the same number of vertices ) through a series of edge - flips ( here one uses a more abstract notion of an edge flip ) . when we deal with a pair of triangulations over a specific common ( labeled ) set @xmath0 of points in the plane , there always exists such a sequence of @xmath21 flips , and this bound is tight in the worst case ( e.g. , see @xcite ) . moreover , there are algorithms that perform such sequences of flips to obtain some optimal " triangulation ( typically , the delaunay triangulation ; see @xcite for example ) , which , as a by - product , provide an edge - flip sequence between any specified pair of triangulations of @xmath0 . how many flippable edges can a single triangulation have ? given a triangulation @xmath19 , we denote by @xmath22 the number of flippable edges in @xmath19 . hurtado , noy , and urrutia @xcite proved the following lower bound . [ le : fliporiginal ] _ * @xcite * _ for any triangulation @xmath19 over a set of @xmath1 points in the plane , @xmath23 moreover , there are triangulations ( of specific point sets of arbitrarily large size ) for which this bound is tight . flippable edges.,title="fig:",scaledwidth=45.0% ] to obtain a triangulation with exactly @xmath24 flippable edges ( for an even @xmath1 ) , start with a convex polygon with @xmath25 vertices , triangulate it in some arbitrary manner , insert a new point into each of the @xmath26 resulting bounded triangles , and connect each new point @xmath27 to the three hull vertices that form the triangle containing @xmath27 . such a construction is depicted in figure [ fi : lowflip ] . the resulting graph is a triangulation with @xmath1 vertices and exactly @xmath24 flippable edges , namely the chords of the initial triangulation . next , we say that two flippable edges @xmath28 and @xmath29 of a triangulation @xmath19 are _ simultaneously flippable _ if no triangle of @xmath19 is incident to both edges ; equivalently , the quadrilaterals corresponding to @xmath28 and @xmath29 are interior - disjoint . see figure [ fi : intro](b ) for an illustration . notice that flipping an edge @xmath28 can not affect the flippability of any edge simultaneously flippable with @xmath28 . given a triangulation @xmath19 , let @xmath30 denote the size of the largest subset of edges of @xmath19 , such that every pair of edges in the subset is simultaneously flippable . the following lemma , improving upon an earlier weaker bound in @xcite , is taken from souvaine , tth , and winslow @xcite . [ le : sfliporiginal ] _ * @xcite * _ for any triangulation @xmath19 over a set of @xmath1 points in the plane , @xmath31 . galtier et al . @xcite show that this bound is tight in the worst case , by presenting a specific triangulation in which at most @xmath32 edges are simultaneously flippable . [ [ pseudo - simultaneously - flippable - edge - sets . ] ] pseudo - simultaneously flippable edge sets . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a set of simultaneously flippable edges in a triangulation @xmath19 can be considered as the set of diagonals of a collection of interior - disjoint convex quadrilaterals . we consider a more liberal definition of simultaneously flippable edges , by taking , within a fixed triangulation @xmath19 , the diagonals of a set of interior - disjoint convex polygons , each with at least four edges ( so that the boundary edges of these polygons belong to @xmath19 ) . consider such a collection of convex polygons @xmath33 , where @xmath34 has @xmath35 edges , for @xmath36 . we can then retriangulate each @xmath34 independently , to obtain many different triangulations . specifically , each @xmath34 can be triangulated in @xmath37 ways , where @xmath38 is the @xmath39-th _ catalan number _ ( see , e.g. , ( * ? ? ? * section 5.3 ) ) . hence , we can get @xmath40 different triangulations in this way . in particular , if a graph @xmath41 ( namely , all the edges of @xmath9 are edges of @xmath19 ) does not contain any diagonal of any @xmath34 ( it may contain boundary edges though ) then @xmath9 is a subgraph of ( at least ) @xmath42 distinct triangulations . an example is depicted in figure [ fi : intro](c ) , where by flipping " ( or rather , redrawing ) the diagonals of the highlighted quadrilateral and pentagon , we can get @xmath43 different triangulations ( including the one shown ) , and any subgraph of the triangulation that does not contain any of these diagonals is a subgraph of these ten triangulations . we say that a set of interior edges in a triangulation is _ pseudo - simultaneously flippable _ ( _ ps - flippable _ for short ) if after the deletion of these edges every bounded face of the remaining graph is convex , and there are no vertices of degree 0 . notice that all three notions of flippability are defined within a fixed triangulation @xmath19 of @xmath0 ( although each of them gives a recipe for producing many other triangulations ) . .bounds for minimum numbers of the various types of flippable edges in a triangulation of @xmath1 points . all of these bounds are tight in the worst case . [ cols="<,^",options="header " , ] [ [ our - results . ] ] our results . + + + + + + + + + + + + in section [ sec : psflip ] , we derive a lower bound on the size of the largest set of ps - flippable edges in a triangulation , and show that this bound is tight in the worst case . specifically , we have the following result . [ le : psflip ] let @xmath0 be a set of @xmath1 points in the plane , and let @xmath19 be a triangulation of @xmath0 . then @xmath19 contains a set of at least @xmath44 ps - flippable edges . this bound is tight in the worst case . table [ ta : flippable ] summarizes the bounds for the minimum numbers of the various types of flippable edges in a triangulation . . when completing it into a triangulation , the added ( dashed ) diagonals form a set of ps - flippable edges . this is one of the @xmath45 possible completions.,title="fig:",scaledwidth=20.0% ] we also relate ps - flippable edges to _ convex decompositions _ of @xmath0 . these are crossing - free straight - edge graphs on @xmath0 such that ( i ) they include all the hull edges , ( ii ) each of their bounded faces is a convex polygon , and ( iii ) no point of @xmath0 is isolated . see figure [ fi : decomposition ] for an illustration . l | c | c | c | c graph type & lower bound & previous & new upper & in the form + & & upper bound & bound & @xmath46 + & & & & + [ -2ex ] plane graphs & @xmath47 @xcite & @xmath48 @xcite & @xmath49 & @xmath50 + [ 0.5ex ] spanning trees & @xmath51 @xcite & @xmath52 @xcite & @xmath53 & @xmath54 + [ 0.5ex ] forests & @xmath55 @xcite & @xmath56 ) @xcite & @xmath57 & @xmath58 + [ 1ex ] [ [ counting - plane - graphs - new - upper - bounds . ] ] counting plane graphs : new upper bounds . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in section [ se : apps ] , we use lemma [ le : psflip ] to derive several upper bounds on the numbers of planar graphs of various kinds embedded as crossing - free straight - edge graphs on a fixed labeled set @xmath0 . for a set @xmath0 of points in the plane , we denote by @xmath59 the set of all triangulations of @xmath0 , and put @xmath60 . similarly , we denote by @xmath61 the set of all crossing - free straight - edge graphs on @xmath0 , and put @xmath62 . we also let @xmath2 and @xmath63 denote , respectively , the maximum values of @xmath64 and of @xmath65 , over all sets @xmath0 of @xmath1 points in the plane . since a triangulation of @xmath0 has fewer than @xmath66 edges , the trivial upper bound @xmath67 holds for any point set @xmath0 . recently , razen , snoeyink , and welzl @xcite slightly improved the upper bound on the ratio @xmath68 from @xmath69 down to @xmath70 . we give a more significant improvement on the ratio with an upper bound of @xmath71 . combining this bound with the recent bound @xmath72 @xcite , we get @xmath73 . we provide similar improved ratios and absolute bounds for the numbers of crossing - free straight - edge spanning trees and forests ( i.e. , cycle - free graphs ) . table [ ta : upperbounds ] summarizes these results . we also derive similar ratios for the number of crossing - free straight - edge graphs with exactly @xmath74 edges , with at least @xmath74 edges , and with at most @xmath74 edges , for @xmath75 . for the case of crossing - free straight - edge graphs with exactly @xmath74 we obtain the bound , @xmath76 , and @xmath77 , we neglect polynomial factors . ] @xmath78 where @xmath79 . figure [ fi : plot1 ] contains a plot of the base @xmath80 of the exponential factor multiplying @xmath2 in this bound , as a function of @xmath8 . of the exponential factor in the bound on the number of crossing - free straight - edge graphs with @xmath74 edges , as a function of @xmath8 . the maximum is attained at @xmath81 ( see below ) . , title="fig:",scaledwidth=47.0% ] [ [ notation . ] ] notation . + + + + + + + + + here are some additional notations that we use . = 0em for a triangulation @xmath19 and an integer @xmath82 , let @xmath83 denote the number of interior vertices of degree @xmath84 in @xmath19 . given two crossing - free straight - edge graphs @xmath9 and @xmath85 over the same point set @xmath0 , we write @xmath86 to indicate that every edge in @xmath9 is also an edge in @xmath85 . similarly to the case of edges , the _ hull vertices _ , _ interior vertices _ ) of a set @xmath0 of points in the plane are those that are part of the boundary of the convex hull of @xmath0 ( resp . , not part of the convex hull boundary ) . we only consider point sets @xmath0 in general position , that is , no three points in @xmath0 are collinear . for upper bounds on the number of graphs , this involves no loss of generality , because the number of graphs can only increase if collinear points are slightly perturbed into general position . * separable edges . * let @xmath27 be an interior vertex in a convex decomposition @xmath9 of @xmath0 . following the notation in @xcite , we call an edge @xmath28 incident to @xmath27 in @xmath9 _ separable at @xmath27 _ if it can be separated from the other edges incident to @xmath27 by a line through @xmath27 ( see figure [ fi : separable ] , where the separating lines are not drawn ) . equivalently , edge @xmath28 is separable at @xmath27 if the two angles between @xmath28 and its clockwise and counterclockwise neighboring edges ( around @xmath27 ) sum up to more than @xmath87 . following @xcite , we observe the following easy properties , both of which materialize in figure [ fi : separable ] . a. if @xmath27 is an interior vertex of degree @xmath88 in @xmath9 , its three incident edges are separable at @xmath27 , for otherwise @xmath27 would have been a reflex vertex of some face . b. an interior vertex @xmath27 of degree @xmath89 or higher can have at most two incident edges which are separable at @xmath27 ( and if it has two such edges they must be consecutive in the circular order around @xmath27 ) . in this section , we establish the ps - flippability lemma ( lemma [ le : psflip ] from the introduction ) . we restate the lemma for the convenience of the reader . [ [ lemma - lepsflip ] ] lemma [ le : psflip ] + + + + + + + + + + + + + + + + + _ let @xmath0 be a set of @xmath1 points in the plane , and let @xmath19 be a triangulation of @xmath0 . then @xmath19 contains a set of at least @xmath44 ps - flippable edges . this bound is tight in the worst case . _ starting with the proof of the lower bound , we apply the following iterative process to @xmath19 . as long as there exists an interior edge whose removal does not create a non - convex face , we pick such an edge and remove it . when we stop , we have a crossing - free straight - edge graph @xmath9 , all of whose bounded faces are convex ; that is , we have a locally minimal convex decomposition of @xmath0 . note that all @xmath12 original hull edges are still in @xmath9 , and that every interior vertex of @xmath9 has degree at least @xmath88 ( recall the general position assumption ) . note that every edge of @xmath9 is separable at one or both of its endpoints , for we can remove any other edge and the graph will continue to have only convex faces ( see figure [ fi : nonessential ] ) . we denote by @xmath90 the number of edges of @xmath9 , and by @xmath91 the number of its interior edges . recalling properties ( i ) and ( ii ) of separable edges , we have @xmath92 and @xmath93 where @xmath94 is the number of interior vertices of degree @xmath88 in @xmath9 , and @xmath95 is the number of interior vertices @xmath96 of degree at least @xmath89 in @xmath9 with exactly @xmath84 edges separable at @xmath96 . notice that @xmath97 the estimate in ( [ eq : int1 ] ) may be pessimistic , because it doubly counts edges that are separable at both endpoints ( such as the one in figure [ fi : nonessential](c ) ) . to address this possible over - estimation , denote by @xmath98 the number of edges that are separable at both endpoints , to which we refer as _ doubly separable edges _ , and rewrite ( [ eq : int1 ] ) as @xmath99 denoting by @xmath100 the number of bounded faces of @xmath9 , we have , by euler s formula , @xmath101 ( the expression in the parentheses on the left is the number of faces in @xmath9 , and the expression in the parentheses on the right is the number of edges ) , or @xmath102 let @xmath103 , for @xmath104 , denote the number of interior faces of degree @xmath105 in @xmath9 . by doubly counting the number of edges in @xmath9 , and then applying ( [ eq : mfaces ] ) , we get @xmath106 or @xmath107 the number of edges that were removed from @xmath19 is @xmath108 , because a face of @xmath9 of degree @xmath105 must have had @xmath109 diagonals that were edges of @xmath19 . this number is therefore @xmath110 & = & 2n + h - 2 - ( m_{\text{int}}-n+1 ) = 3n+h-3-a+m_{\text{double } } \label{eq : removed}\end{aligned}\ ] ] ( by first applying ( [ eq : sumfk ] ) , then ( [ eq : mfaces ] ) , and finally ( [ eq : mint ] ) ) . of @xmath0 , its corresponding graph @xmath111 , and the reduced form of @xmath111 after removing vertices of degree 3 . the edges that have been added are dashed.,title="fig:",scaledwidth=85.0% ] we next derive a lower bound for the right - hand side of ( [ eq : removed ] ) . for this , we transform @xmath9 into another graph @xmath111 as follows . we first subdivide each doubly separable edge of @xmath9 at its midpoint , say , and add the subdivision point as a new vertex of @xmath9 ( e.g. , see the vertex @xmath39 in figure [ fi : gprime ] ) . we now modify @xmath9 as follows . we take each vertex @xmath96 of degree @xmath88 in @xmath9 and surround it by a triangle , by connecting all pairs of its three neighbors . notice that some of these neighbors may be new subdivision vertices , and that some of the edges of the surrounding triangle may already belong to @xmath9 . for example , see figure [ fi : gprime ] , where the edges @xmath112 are added around the vertex @xmath12 and the edges @xmath113 are added around the vertex @xmath114 . next we take each interior vertex @xmath96 with two separable edges at @xmath96 and complete these two edges into a triangle by connecting their other endpoints , each of which is either an original point or a new subdivision point ; here too the completing edge may already belong to @xmath9 . for example , see the edge @xmath115 in figure [ fi : gprime ] , induced by the two separable edges of the vertex @xmath84 . we then take the resulting graph @xmath111 and remove each vertex of degree @xmath88 and its three incident edges ; see the reduced version of @xmath111 in figure [ fi : gprime ] . a crucial and easily verified property of this transformation is that the newly embedded edges do not cross each other , nor do they cross old edges of @xmath9 . the number @xmath116 of bounded faces of the new graph @xmath111 is at least @xmath117 , which is the number of triangles that we have created , and the number @xmath118 of its interior vertices is @xmath119 . also , @xmath111 still has @xmath12 hull edges . using euler s formula , as in ( [ eq : mfaces ] ) and ( [ eq : sumfk ] ) above , we have @xmath120 . combining the above , we get @xmath121 or @xmath122 hence , the right - hand side of ( [ eq : removed ] ) is at least @xmath123 @xmath124 & = & 2n + \frac12 h -2 - \frac32 v_3 - \frac32 v_{4,2 } - v_{4,1 } \\[0.4em ] & \ge & 2n + \frac12 h - 2 - \frac32 \left ( v_3 + v_{4,2 } + v_{4,1 } + v_{4,0 } \right ) \\[0.4em ] & = & \frac{n+h}{2 } - 2 = \frac{n}{2 } - 2 .\end{aligned}\ ] ] in other words , the number of edges that we have removed from @xmath19 is at least @xmath24 . on the other hand , we always have @xmath125 and @xmath126 . substituting these trivial bounds in ( [ eq : removed ] ) we get at least @xmath127 ps - flippable edges . this completes the proof of the lower bound . it is easily noticed that only flippable edges of @xmath19 could have been removed in the initial pruning stage . hurtado , noy , and urrutia @xcite present two distinct triangulations that contain exactly @xmath24 flippable edges ( one of those is depicted in figure [ fi : lowflip ] ) . these triangulations can not have a set of more than @xmath24 ps - flippable edges . therefore , there are triangulations for which our bound is tight in the worst case . similarly , for point sets in convex position , all @xmath127 interior edges form a set of ps - flippable edges , showing that the other term in the lower bound is also tight in the worst case . * the proof of lemma 1.3 actually yields the slightly better bound @xmath128 that is , for the bound to be tight , every interior vertex @xmath96 of degree @xmath89 or higher must have two incident edges separable at @xmath96 ( note that this condition holds vacuously for the triangulation in figure [ fi : lowflip ] ) . [ [ convex - decompositions . ] ] convex decompositions . + + + + + + + + + + + + + + + + + + + + + + the preceding analysis is also related to the notion of _ convex decompositions _ , as defined in the introduction . urrutia @xcite asked what is the minimum number of faces that can always be achieved in a convex decomposition of any set of @xmath1 points in the plane . hosono @xcite proved that every planar set of @xmath1 points admits a convex decomposition with at most @xmath129 ( bounded ) faces . for every @xmath130 , garca - lopez and nicols @xcite constructed @xmath1-element point sets that do not admit a convex decomposition with fewer than @xmath131 faces . by euler s formula , if a connected crossing - free straight - edge graph has @xmath1 vertices and @xmath28 edges , then it has @xmath132 faces ( including the exterior face ) . it follows that for convex decompositions , minimizing the number of faces is equivalent to minimizing the number of edges . ( for convex decompositions contained in a given triangulation , this is also equivalent to maximizing the number of removed edges , which form a set of ps - flippable edges . ) lemma [ le : psflip ] directly implies the following corollary . ( the bound that it gives is weaker than the bound in @xcite , but it holds for every triangulation . ) let @xmath0 be a set of @xmath1 points in the plane , so that its convex hull has @xmath12 vertices , and let @xmath19 be a triangulation of @xmath0 . then @xmath19 contains a convex decomposition of @xmath0 with at most @xmath133 convex faces and at most @xmath134 edges . moreover , there exist point sets @xmath0 of arbitrarily large size , and triangulations @xmath135 for which these bounds are tight . in this section we apply the ps - flippability lemma ( lemma [ le : psflip ] ) to obtain several improved bounds on the number of crossing - free straight - edge graphs of various kinds on a fixed set of points in the plane . we begin by recalling some observations already made in the introduction . let @xmath0 be a set of @xmath1 points in the plane . every crossing - free straight - edge graph in @xmath61 is contained in at least one triangulation in @xmath59 . additionally , since a triangulation has fewer than @xmath136 edges , every triangulation @xmath135 contains fewer than @xmath137 crossing - free straight - edge graphs . this immediately implies @xmath138 however , this inequality seems rather weak since it potentially counts some crossing - free straight - edge graphs many times . more formally , given a graph @xmath139 contained in @xmath140 distinct triangulations of @xmath0 , we say that @xmath9 has a _ support _ of @xmath140 , and write @xmath141 . thus , every graph @xmath139 will be counted @xmath142 times in the preceding inequality . recently , razen , snoeyink , and welzl @xcite managed to break the @xmath143 barrier by overcoming the above inefficiency . however , they obtained only a slight improvement , with the bound @xmath144 . we now present a more significant improvement , using a much simpler technique that relies on the ps - flippability lemma . [ th : trivspg ] for every set @xmath0 of @xmath1 points in the plane , @xmath12 of which are on the convex hull , @xmath145 8^n \left(3/8\right)^h \cdot { { \mathsf{tr}}(s ) } , & \text{\quad for } h > n/2 . \end{array } \right.\ ] ] the exact value of @xmath65 is easily seen to be @xmath146 because every graph @xmath9 appears @xmath142 times in the sum , and thus contributes a total of @xmath147 to the count . we obtain an upper bound on this sum as follows . consider a graph @xmath139 and a triangulation @xmath135 , such that @xmath41 . by lemma [ le : psflip ] , there is a set @xmath148 of @xmath149 ps - flippable edges in @xmath19 . the case where @xmath1 is odd is handled in the exact same manner , since a constant change in the size of @xmath148 does not affect the asymptotic bounds . ] let @xmath150 denote the set of edges that are in @xmath148 but _ not _ in @xmath9 , and put @xmath151 . removing the edges of @xmath150 from @xmath19 yields a convex decomposition of @xmath0 which still contains @xmath9 and whose non - triangular interior faces have a total of @xmath39 missing diagonals . suppose that there are @xmath90 such faces , with @xmath152 diagonals respectively , where @xmath153 . then these faces can be triangulated in @xmath154 ways , and each of the resulting triangulations contains @xmath9 . we always have @xmath155 , for any @xmath156 , as is easily verified , and so @xmath157 . ( equality occurs when all the non - triangular faces of @xmath158 are quadrilaterals . ) next , we estimate the number of subgraphs @xmath41 for which the set @xmath150 is of size @xmath39 . denote by @xmath159 the set of edges of @xmath19 that are not in @xmath148 , and assume that the convex hull of @xmath0 has @xmath12 vertices . since there are @xmath160 edges in any triangulation of @xmath0 , @xmath161 . to obtain a graph @xmath9 for which @xmath162 , we choose any subset of edges from @xmath159 , and any @xmath39 edges from @xmath148 ( the @xmath39 edges of @xmath148 that will not belong to @xmath9 ) . therefore , the number of such subgraphs is at most @xmath163 . we can thus rewrite ( [ eq : pgexact ] ) to obtain @xmath164 & = & \displaystyle { { \mathsf{tr}}(s ) } \cdot 2^{3n - h - t-3 } \sum_{j=0}^{t } \binom{t}{j}\frac{1}{2^j } \\[1.3em ] & = & \displaystyle { { \mathsf{tr}}(s ) } \cdot 2^{3n - h - t-3 } \cdot ( 3/2)^{t}. \end{array}\ ] ] if @xmath165 , we get @xmath166 . if @xmath167 , we have @xmath168 . to complete the proof , we note that @xmath169 when @xmath170 . for a lower bound on @xmath171 , we consider the _ double chain _ configurations , presented in @xcite ( and depicted in figure [ fi : doublec ] ) . it is shown in @xcite that , when @xmath0 is a double chain configuration , @xmath172 and @xmath173 ( actually , only the lower bound on @xmath63 is given in @xcite ; the upper bound appears in @xcite ) . thus , we have @xmath174 ( for this set @xmath175 , so @xmath12 has no real effect on the asymptotic bound of theorem [ th : trivspg ] ) . for another lower bound , consider the case where @xmath0 is in convex position . in this case we have @xmath176 , and @xmath177 ( see @xcite ) . hence , @xmath178 , whereas the upper bound provided by theorem [ th : trivspg ] is @xmath179 in this case . informally , the ( rather small ) discrepancy between the exact bound in @xcite and our bound in the convex case comes from the fact that when @xmath39 is large , the faces of the resulting convex decomposition are likely to have many edges , which makes @xmath142 substantially larger than @xmath180 . it is an interesting open problem to exploit this observation to improve our upper bound when @xmath12 is large . finally , recall the notations @xmath181 and @xmath182 . combining the bound @xmath3 , obtained in @xcite , with the first bound of theorem [ th : trivspg ] implies @xmath183 ; see table [ ta : upperbounds ] for comparison with earlier bounds . the bound improves significantly as @xmath12 gets larger . [ [ spanning - trees . ] ] spanning trees . + + + + + + + + + + + + + + + for a set @xmath0 of @xmath1 points in the plane , we denote by @xmath184 the set of all crossing - free straight - edge spanning trees of @xmath0 , and put @xmath185 . moreover , we let @xmath186 . buchin and schulz @xcite have recently shown that every crossing - free straight - edge graph contains @xmath187 spanning trees , improving upon the earlier bound of @xmath188 due to rib mor and rote @xcite . we thus get @xmath189 for every set @xmath0 of @xmath1 points in the plane . the bound from @xcite can not be improved much further , since there are triangulations with at least @xmath190 spanning trees @xcite . however , the ratio between @xmath191 and @xmath64 can be improved beyond that bound , by exploiting and overcoming the same inefficiency as in the case of all crossing - free straight - edge graphs ; that is , the fact that some spanning trees may get multiply counted in many triangulations . we now derive such an improved ratio by using ps - flippable edges . the proof goes along the same lines of the proof of theorem [ th : trivspg ] . [ th : strvtr ] for every set @xmath0 of @xmath1 points in the plane , @xmath192 the exact value of @xmath191 is @xmath193 consider a spanning tree @xmath194 and a triangulation @xmath135 , such that @xmath195 . as in theorem [ th : trivspg ] , let @xmath148 be a set of @xmath24 ps - flippable edges in @xmath19 . ( here we do not exploit the alternative bound of @xmath127 on the size of @xmath148 . ) also , let @xmath196 denote the set of edges that are in @xmath148 but _ not _ in @xmath197 , and put @xmath198 . thus , as argued earlier , @xmath199 . next , we estimate the number of spanning trees @xmath195 for which the set @xmath196 is of size @xmath39 . first , there are @xmath200 ways to choose the @xmath39 edges of @xmath148 that @xmath197 does not use . we next contract the @xmath201 edges of @xmath148 that were chosen to be in @xmath197 ( which will result in having some parallel edges , and possibly also loops ) and then remove the remaining edges of @xmath148 . this produces a non - simple graph @xmath9 with @xmath202 vertices and fewer than @xmath203 edges ( recall that by euler s formula , @xmath9 contains at most @xmath204 edges ) . let @xmath205 denote the set of vertices of @xmath9 , and let @xmath206 denote the degree in @xmath9 of a point @xmath207 . as shown in @xcite , the number of spanning trees in a graph @xmath9 ( not necessarily planar or simple ) is at most the product of the vertex degrees in @xmath9 . thus , the number of ways to complete the tree is at most @xmath208 ( where we have used the inequality of means for the first inequality ) . hence , there are fewer than @xmath209 spanning trees @xmath195 with @xmath210 . however , when @xmath39 is large , it is better to use the bound @xmath187 from @xcite instead . is close to @xmath211 the former bound is smaller ( e.g. , it is @xmath212 for @xmath213 ) , but we do not know how to exploit this observation to improve the bound . ] we thus get , for a threshold parameter @xmath214 that we will set in a moment , @xmath215 the terms in the first sum over @xmath39 increase when @xmath216 , so the sum is at most @xmath211 times its last term . using stirling s formula , we get that for @xmath217 , the last term in the first sum is @xmath218 . since this also bounds the second sum , we get @xmath219 as asserted . ( the optimal parameter @xmath220 was computed numerically . ) combining the bound just obtained with @xmath3 @xcite implies @xmath221 . this improves all previous upper bounds , the smallest of which is @xmath222 @xcite . * remark . * it would be interesting to refine the bound in theorem [ th : strvtr ] so that it also depends on @xmath12 , as in theorem [ th : trivspg ] . an extreme situation is when @xmath0 is in convex position ( in which case @xmath223 ) . in this case it is known that @xmath224 and @xmath225 ( see @xcite ) , so the exact ratio is only @xmath226 . this might suggest that when @xmath12 is large the ratio should be considerably smaller , but we have not pursued this in this paper . [ [ forests . ] ] forests . + + + + + + + + for a set @xmath0 of @xmath1 points in the plane , we denote by @xmath227 the set of all crossing - free straight - edge forests ( i.e. , cycle - free graphs ) of @xmath0 , and put @xmath228 . moreover , we let @xmath229 . buchin and schulz @xcite have recently shown that every crossing - free straight - edge graph contains @xmath230 forests ( improving a simple upper bound of @xmath231 observed in @xcite ) . following the approach of @xcite , we combine the bounds for spanning trees ( just established ) and for plane graphs with a bounded number of edges ( established in section [ ssec : restrict ] below ) , to obtain the following result . [ th : frvstr ] for every set @xmath0 of @xmath1 points in the plane , @xmath232 we define a @xmath105-_forest _ to be a forest that has @xmath105 connected components . the number of @xmath105-forests of a set @xmath0 is denoted by @xmath233 . since any spanning tree has @xmath234 edges , every @xmath105-forest has @xmath235 edges . one way to bound @xmath233 is by counting the number of plane graphs with @xmath235 edges . this number is bounded in theorem [ th : trivspgc ] ( from the following subsection ) , where the parameter @xmath8 in that theorem is equal to @xmath236 ; let us denote this bound as @xmath237 . on the other hand , every @xmath105-forest is obtained by deleting @xmath238 edges from a spanning tree . this allows us to bound the number of @xmath105-forests in terms of @xmath191 . using theorem [ th : strvtr ] we get the bound @xmath239 ; denote this bound as @xmath240 . to bound @xmath241 , we evaluate @xmath242 . a numerical calculation shows that the maximum value is obtained for @xmath243 , and the theorem follows since @xmath244 . as in the previous cases , we can combine this with the bound @xmath3 @xcite to obtain @xmath245 . again , this should be compared with the best previous upper bound @xmath246 ) @xcite . consider once again the case where @xmath0 consists of @xmath1 points in convex position . in this case we have @xmath224 and @xmath247 ( see @xcite ) , so the exact ratio is @xmath248 , again suggesting that the ratio should be smaller when @xmath12 is large . in this subsection we derive upper bounds for the number of crossing - free straight - edge graphs on a set @xmath0 of @xmath1 points in the plane , with some constraints on the number of edges . specifically , we bound the number of crossing - free straight - edge graphs with exactly @xmath7 edges , with at most @xmath7 edges , and with at least @xmath7 edges . the first variant has already been used in the preceding subsection for bounding the number of forests . [ [ crossing - free - straight - edge - graphs - with - exactly - edges . ] ] crossing - free straight - edge graphs with exactly edges . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we denote by @xmath249 the set of all crossing - free straight - edge graphs of @xmath0 with exactly @xmath7 edges , and put @xmath250 . the following theorem , whose proof goes along the same lines of the proof of theorem [ th : trivspg ] , gives a bound for @xmath251 [ th : trivspgc ] for every set @xmath0 of @xmath1 points in the plane and @xmath252 , @xmath253 where @xmath254 and @xmath255 see figure [ fi : plot1 ] for a plot of the base @xmath80 as a function of @xmath8 . the exact value of @xmath251 is @xmath256 where @xmath142 , the support of @xmath9 , is defined as in the case of general crossing - free straight - edge graphs treated in section [ ssec : pg ] . we obtain an upper bound on this sum as follows . consider a graph @xmath257 and a triangulation @xmath135 , such that @xmath41 . by lemma [ le : psflip ] , there is a set @xmath148 of @xmath24 ps - flippable edges in @xmath19 . let @xmath150 denote the set of edges that are in @xmath148 but _ not _ in @xmath9 , and put @xmath151 . as in the preceding proofs , we have @xmath258 . next , we estimate the number of subgraphs @xmath41 for which the set @xmath150 is of size @xmath39 . denote by @xmath159 the set of edges of @xmath19 that are not in @xmath148 . as argued above , @xmath259 . to obtain a graph for which @xmath162 , we choose any @xmath39 edges from @xmath148 ( the @xmath39 edges of @xmath148 that will not belong to @xmath9 ) , and any subset of @xmath260 edges from @xmath159 . if @xmath261 , there are no such graphs and we ignore these values of @xmath39 . the number of ways to pick the edges from @xmath159 is at most @xmath262 this implies that @xmath263 & = & \displaystyle { { \mathsf{tr}}(s ) } \cdot \sum_{j=0}^{n/2 } o^*\left(\binom{5n/2}{(c-1/2)n + j } \cdot \binom{n/2}{j}\right ) \cdot \frac{1}{2^j}. \label{eq : atmost}\end{aligned}\ ] ] ( as already noted , when @xmath264 , only the terms for which @xmath265 are taken into account . ) as in the preceding subsection , it suffices to consider only the largest term of the sum . for this , we consider the quotient of the @xmath39-th and @xmath266-st terms ( ignoring the @xmath267 notation , which will not affect the exponential order of growth of the terms ) , which is @xmath268 to simplify matters , we put @xmath269 and @xmath270 . moreover , since we are only looking for an asymptotic bound , and are willing to incur small multiplicative errors within the @xmath267 notation , we may ignore the two @xmath271 terms in the numerator when @xmath1 is sufficiently large ; we omit the routine algebraic justification of this statement . the above quotient then becomes ( approximately ) @xmath272 , which is larger than 1 whenever @xmath273 & = & \frac{n}{2 } \left(\sqrt{(7/2)^2 + 3c + c^2 } - 5/2 - c \right ) = tn , \end{array}\ ] ] with @xmath274 given in ( [ eq : t ] ) . a simple calculation shows that @xmath275 and @xmath276 for @xmath277 . in other words ( and rather unsurprisingly ) , the index @xmath278 attaining the maximum does indeed lie in the range where the two binomial coefficients in the corresponding terms in ( [ eq : atmost ] ) are both well defined ( non - zero ) . now that we have the largest term of the sum in ( [ eq : atmost ] ) , we obtain @xmath279 using stirling s approximation , we have @xmath280 & = & \displaystyle { { \mathsf{tr}}(s ) } \cdot o^*\left ( \left ( \frac{5^{5/2}}{8(c+t-1/2)^{c+t-1/2 } ( 3-c - t)^{3-c - t}(2t)^t(1/2-t)^{1/2-t } } \right)^n \right ) , \end{array}\ ] ] as asserted . [ [ crossing - free - straight - edge - graphs - with - at - most - edges . ] ] crossing - free straight - edge graphs with at most edges . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for a set @xmath0 of @xmath1 points in the plane and a constant @xmath281 , we denote by @xmath282 the set of all crossing - free straight - edge graphs of @xmath0 with at most @xmath7 edges , and put @xmath283 . the bound for @xmath251 in theorem [ th : trivspgc ] helps us to determine the bound for @xmath284 . [ th : trivspgcc ] for every set @xmath0 of @xmath1 points in the plane and @xmath285 , @xmath286 where @xmath80 is defined as in theorem [ th : trivspgc ] . we begin by noticing that @xmath287 let @xmath288 be the natural logarithm of the non - constant part of the denominator of @xmath80 ( the numerator is a constant ) . that is , @xmath289 since each of the terms in the logarithms is positive when @xmath290 , finding a maximum for @xmath80 in this range is equivalent to finding a minimum for @xmath288 . put @xmath291 ( the derivative of @xmath274 as a function of @xmath8 ) . then @xmath292 & + t'\ln(2 t ) + t ' -t'\ln(1/2-t ) -t ' \\[0.5em ] & = ( 1+t')\ln\left(\frac{c+t-1/2}{3-c - t}\right ) + t'\ln\left(\frac{2t}{1/2-t } \right)\end{aligned}\ ] ] for @xmath81 we have @xmath293 and the arguments of both logarithms are 1 , so @xmath294 . easy calculations show that @xmath295 this is easily seen to imply that @xmath296 and @xmath297 for @xmath290 . this implies that @xmath298 is monotone increasing with @xmath8 , and that @xmath274 is decreasing ( because @xmath297 and @xmath296 ) . it follows that @xmath299 is increasing with @xmath8 , implying that @xmath288 attains its minimum at @xmath81 ( since @xmath300 ) . another easy calculation shows that @xmath301 . for example , theorem [ th : trivspgcc ] implies that there are at most @xmath302 crossing - free straight - edge graphs with at most @xmath1 edges , over any set @xmath0 of @xmath1 points in the plane . in particular , this is also an upper bound on the number of crossing - free straight - edge forests on @xmath0 , or of spanning trees , or of spanning cycles . of course , better bounds exist for these three special cases , as demonstrated earlier in this paper for the first two bounds . [ [ crossing - free - straight - edge - graphs - with - at - least - edges . ] ] crossing - free straight - edge graphs with at least edges . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we next bound the number of plane graphs with at least @xmath7 edges . following the notations used above , we denote by @xmath303 the set of all crossing - free straight - edge graphs of @xmath0 with at least @xmath7 edges , and put @xmath304 . [ th : trivspgccc ] for every set @xmath0 of @xmath1 points in the plane and @xmath285 , @xmath305 similar to equation we can bound @xmath306 by @xmath307 the analysis of is symmetric to the one presented in the proof of theorem [ th : trivspgcc ] , and the theorem follows . and a quadrangulation of @xmath205 that contains it.,title="fig:",scaledwidth=60.0% ] as an application , consider the problem of bounding the number of quadrangulations of @xmath0 , namely crossing - free straight - edge connected graphs on the vertex set @xmath0 with no isolated vertices , that include all the hull edges of @xmath308 , and where every bounded face is a quadrilateral . when @xmath12 is odd , no quadrangulations can be embedded over @xmath0 ( e.g. , see @xcite ) . we may thus assume that @xmath12 is even , and create a superset @xmath309 as follows . we take a quadrilateral @xmath310 that contains @xmath0 in its interior , and add the vertices of @xmath310 to @xmath0 . it is easy to see that every quadrangulation of @xmath0 is contained in at least one quadrangulation of @xmath205 ; see figure [ fi : quadrangulations ] for an illustration and the convex hull of @xmath0 . we start by connecting a vertex of @xmath310 to a vertex of the convex hull , and in each step we add a quadrangle by either marching along two edges of the hull or along one edge of the hull and one edge of @xmath310 . this produces the desired quadrangulation . ] . therefore , it suffices to bound the number of quadrangulations of @xmath205 . using euler s formula , we notice that a quadrangulation of @xmath205 has @xmath311 quadrilaterals , and @xmath312 edges ( since @xmath313 ) . therefore , we can use theorem [ th : trivspgcc ] with @xmath314 , which implies a bound of @xmath315 . ( there are actually @xmath316 points and @xmath317 . however , since we are only interested in the exponential part of the bound , the above bound , with the @xmath267 notation , does hold . ) we are not aware of any previous explicit treatment of this problem . in this paper we have introduced the notion of pseudo - simultaneously flippable edges in triangulations , have shown that many such edges always exist , and have used them to obtain several refined bounds on the number of crossing - free straight - edge graphs on a fixed ( labeled ) set of @xmath1 points in the plane . the paper raises several open problems and directions for future research . one such question is whether it is possible to further extend the notion of ps - flippability . for example , one could consider , within a fixed triangulation @xmath19 , the set of diagonals of a collection of pairwise interior - disjoint simple , but not necessarily convex , polygons . the number of such diagonals is likely to be larger than the size of the maximal set of ps - flippable edges , but it not clear how large is the number of triangulations that can be obtained by redrawing diagonals . we are currently working on two extensions to this work . the first extends our techniques to the cases of crossing - free straight - edge perfect matchings and spanning ( hamiltonian ) cycles . this is done within the linear - algebra framework introduced by kasteleyn ( see @xcite ) , and can be found in @xcite . the second work studies charging schemes in which the charge is moved across certain objects belonging to different crossing - free straight - edge graphs over the same point set . this cross - graph charging scheme allows us to obtain bounds that do not depend on the current upper bound for @xmath2 ( and bounds that depend on @xmath2 in a non - linear fashion ) . 99 o. aichholzer , t. hackl , c. huemer , f. hurtado , h. krasser , and b. vogtenhuber , on the number of plane geometric graphs , _ graphs and combinatorics _ * 23*(1 ) ( 2007 ) , 6784 . m. ajtai , v. chvtal , m.m . newborn , and e. szemerdi , crossing - free subgraphs , _ annals discrete math . _ * 12 * ( 1982 ) , 912 . p. bose and f. hurtado , flips in planar graphs , _ comput . theory appl . _ * 42*(1 ) ( 2009 ) , 6080 . p. bose and g. toussaint , no quadrangulation is extremely odd , _ algorithms and computations _ , lecture notes in comput . , vol 1004 , springer - verlag , berlin , 1995 , 372381 . k. buchin , c. knauer , k. kriegel , a. schulz , and r. seidel , on the number of cycles in planar graphs , _ proc . 17th computing and combinatorics conf . _ , lecture notes comput . , vol 4598 , springer - verlag , berlin , 2007 , 97107 . k. buchin and a. schulz , on the number of spanning trees a planar graph can have , _ proc . 18th annual european symposium on algorithms _ , lecture notes comput . , vol 6346 , springer - verlag , berlin , 2010 , 110121 . j. a. de loera , j. rambau , and f. santos , _ triangulations : structures for algorithms and applications _ , springer - verlag , berlin , 2010 . m. o. denny and c. a. sohler , encoding a triangulation as a permutation of its point set , _ proc . 9th canadian conf . on computational geometry _ ( 1997 ) , 3943 . a. dumitrescu , a. schulz , a. sheffer , and cs . d. tth , bounds on the maximum multiplicity of some common geometric graphs , _ proc . . theoret . aspects comput . sci . _ ( 2011 ) , 637648 . i. fry , on straight line representations of planar graphs , _ acta sci . ( szeged ) _ , * 11 * , 229233 , 1948 . flajolet and m. noy , analytic combinatorics of non - crossing configurations , _ discrete math . _ * 204 * ( 1999 ) , 203229 . s. fortune , voronoi diagrams and delaunay triangulations , in _ computing in euclidean geometry _ , d. a. du , f. k. hwang , eds . , world scientific publishing co. , new york , 1992 , 193233 . j. galtier , f. hurtado , m. noy , s. prennes , and j. urrutia , simultaneous edge flipping in triangulations , _ internat . j. comput . appl . _ * 13*(2 ) ( 2003 ) , 113133 . a. garca , m. noy , and j. tejel , lower bounds on the number of crossing - free subgraphs of @xmath318 , _ comput . theory appl . _ * 16*(4 ) ( 2000 ) , 211221 . j. garca - lopez and m. nicols , planar point sets with large minimum convex partitions , _ abstracts 22nd european workshop on computational geometry _ ( 2006 ) , 5154 . o. gimnez and m. noy , asymptotic enumeration and limit laws of planar graphs , _ j. amer . * 22 * ( 2009 ) , 309329 . hjelle and m. dhlen , _ triangulations and applications _ , springer - verlag , berlin , 2009 . k. hosono , on convex decompositions of a planar point set , _ discrete math . _ * 309 * ( 2009 ) , 17141717 . f. hurtado , m. noy , and j. urrutia , flipping edges in triangulations , _ discrete comput . * 22 * ( 1999 ) , 333346 . l. lovsz and m. plummer , _ matching theory _ , north holland , budapest - amsterdam , 1986 . r. c. mullin , on counting rooted triangular maps , _ canad . _ * 7 * ( 1965 ) , 373382 . a. razen , j. snoeyink , and e. welzl , number of crossing - free geometric graphs vs. triangulations , _ electronic notes in discrete math . _ * 31 * ( 2008 ) , 195200 . a. rib mor , _ realizations and counting problems for planar structures : trees and linkages , polytopes and polyominos _ , ph.d . thesis , freie universitt berlin , 2005 . g. rote , the number of spanning trees in a planar graph , _ oberwolfach reports _ * 2 * ( 2005 ) , 969973 . f. santos and r. seidel , a better upper bound on the number of triangulations of a planar point set , _ j. combinat . theory _ , ser . a * 102*(1 ) ( 2003 ) , 186193 . m. sharir and a. sheffer , counting triangulations of planar point sets , _ electr . j. comb . _ , * 18*(1 ) ( 2011 ) , p70 . m. sharir , a. sheffer , and e. welzl , on degrees in random triangulations of point sets , _ j. combinat . theory a _ * 118 * ( 2011 ) , 19791999 . m. sharir , a. sheffer , and e. welzl , counting plane graphs : perfect matchings , spanning cycles , and kasteleyn s technique , in arxiv:1109.5596 . m. sharir and e. welzl , random triangulations of planar point sets , _ proc . 22nd acm symp . on computational geometry _ ( 2006 ) , 273281 . d. l. souvaine , cs . d. tth , and a. winslow , simultaneously flippable edges in triangulations , _ proc . xiv spanish meeting on comput . ( 2011 ) , 137140 . r. p. stanley , _ enumerative combinatorics _ , * 2 * , cambridge university press , cambridge , 1999 . w. t. tutte , a census of planar maps , _ canad . j. math . _ * 15 * ( 1963 ) , 249271 . j. urrutia , open problem session , in _ proc . 10th canadian conference on computational geometry _ , mcgill university , montral , 1998 . k. wagner , bemerkungen zum vierfarbenproblem , _ j. deutsch . math .- verein . _ * 46 * ( 1936 ) , 2632 .

we present results of wide - fieldco ( ) andco ( ) observations toward the aquila rift and serpens molecular cloud complexes ( 25 and ) at an angular resolution of 3.4 ( 0.25 pc ) and at a velocity resolution of 0.079 km s with the velocity coverage of km s 35 km s . we found that theco emission better traces the structures seen in the extinction map and derived the-factor of this region . applying ` scimes ` to theco data cube , we identified 61 clouds and derived their masses , radii , and line widths . the line - width - radius relation of the identified clouds basically follows those of nearby molecular clouds . majority of the identified clouds are close to virial equilibrium although the dispersion is large . by inspecting theco channel maps by eye , we found several arcs which are spatially extended to 0.2 3 degree in length . in the longitude - velocity diagrams ofco , we also found the two spatially - extended components which appear to converge toward serpens south and w40 region . the existence of two components with different velocities and arcs suggests that large - scale expanding bubbles and/or flows play a role in the formation and evolution of the serpens south and w40 cloud .

1702.01501

large - scale molecular line observations are important to unveil how molecular clouds have been formed and evolved because in interstellar space , large - scale dynamical events such as bubbles and supersonic turbulent flows often influence the structure and kinematics of molecular gas where star formation happens ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the present paper , we investigate the cloud structure and kinematics of the aquila rift and serpens molecular cloud complex , on the basis of large - scale multi - co line observations . aquila rift is located in the first galactic quadrant , spanning from 20@xmath12 to 40@xmath12 in galactic longitude and @xmath13 to 10@xmath12 in galactic latitude . it appears in the optical image as a dark lane that divides the bright band of the milky way longitudinally @xcite . the total molecular gas mass is estimated to be about a few @xmath14 on the basis of the co ( @xmath15 ) observations with an angular resolution of about [email protected] @xcite . @xcite identified a number of dark clouds in this region using the digitized sky survey visual extinction data . these previous studies indicate that the region has complex density and velocity structure , suggesting that the dynamical interaction and events might be ongoing and/or have happened in this region @xcite . in fact , @xcite pointed out that several nearby superbubble shells appear to converge toward the aquila rift region . in the southern part of aquila rift , @xcite suggested that the virial mass of the molecular cloud complex is significantly larger than the molecular gas mass and star formation is less active . however , the molecular gas distribution and star formation activity in the northern part above the galactic plane have not been extensively explored so far . here , we focus on the western part of aquila rift , which stretches from @xmath16 to @xmath17 in galactic longitude and @xmath18 to @xmath19 in galactic latitude . the western part of aquila rift contains several active star - forming regions such as serpens main , serpens south , w40 , and mwc 297 . in a well - studied nearby cluster - forming region , serpens main star - forming region @xcite , several young protostars are blowing out of powerful collimated outflows . recently , the _ spitzer _ observations have discovered an extremely - young embedded cluster of low - mass protostars , serpens south , which contains a large number of class 0/i objects @xcite . in fact , @xcite detected a number of co outflows in the central part of serpens south ( see also * ? ? ? * ) . in more evolved star - forming region , w40 hii region , the expanding structure affects the star formation in this region @xcite . recent studies also suggest that cloud - cloud collision may have triggered star formation in serpens main region @xcite and serpens south region @xcite . mwc 297 is an embedded young massive b1.5ve star , one of the closest massive stars @xcite . a molecular outflow is detected toward the mwc 297 region , implying that active star formation is ongoing . in spite of the effort of these previous studies , it remains uncertain why star formation in the western part is much more active than the southern part . in addition , the distance to the aquila rift and serpens cloud complex is somewhat controversial . for the serpens main star - forming region , the distance of @xmath20 260 pc has been used @xcite . this distance determination is based mainly on measurements of the extinction suffered by stars in the direction of serpens . however , recent vlba measurements of young stellar objects , ec95a and ec95b , suggest a larger distance of 415 pc @xcite . for the w40 cloud , the distance is estimated to be @xmath21 pc and has not yet been determined to a satisfactory precision @xcite . the distance to mwc 297 is estimated to be 250 pc @xcite or 450 pc @xcite on the basis of measurements of the extinction of stars , and the uncertainty is a similar degree to other regions in the aquila rift and serpens cloud complexes . serpens south has a local standard of rest ( lsr ) velocity similar to serpens main @xcite . this fact suggests that serpens south may have a distance similar to serpens main . however , no accurate distance measurements have been done so far . the uncertainty of the distance makes it difficult to clarify the cloud dynamical states in the aquila rift and serpens cloud complexes . in the present paper , we compare the cloud physical quantities for both 260 pc and 415 pc . in the present paper , as a first step toward a better understanding of star formation activity in the aquila rift and serpens molecular cloud complexes , we investigate the large - scale molecular cloud structure of the region with an angular resolution of about 3@xmath5 , on the basis of wide field @xmath0co ( @xmath22 ) and @xmath2co ( @xmath22 ) observations . in section [ sec : obs ] , we present the detail of our observations . in section [ sec : results ] , we describe the large - scale co structure in this region . we find several arcs that may have been formed by the large - scale flows . we also derive the @xmath23 factor of this region by comparing co velocity - integrated intensity map and the 2mass extinction map . in section [ sec : clouds ] , we apply ` scimes ` @xcite to the co data cube , and identify the clouds . in the present paper , we call structures identified by ` scimes ` as `` clouds '' . then , we attempt to assess their dynamical states . finally , we briefly summarize the main results in section [ sec : summary ] . we carried out @xmath0co ( @xmath22 ) , @xmath2co ( @xmath22 ) , and c@xmath24o ( @xmath22 ) mapping observations toward the aquila rift and serpens molecular cloud complexes , during the periods from 2012 february to 2012 march and from 2012 december to 2013 march . the observations were done in the on - the - fly ( otf ) mapping mode with a 2sb sis mixer receiver on the 1.85 m radio telescope of osaka prefecture university . the telescope is installed at the nobeyama radio observatory ( nro ) . the image rejection ratio ( irr ) was measured to be 10 db or higher during the observations . at 230 ghz band , the telescope has a beam size of @xmath25 ( hpbw ) and a main beam efficiency of @xmath26 = 0.6 . at the back end , we used a digital fourier transform spectrometer ( dfs ) with 16384 channels that covers the 1 ghz bandwidth , which allows us to obtain the three molecular lines , @xmath0co ( @xmath22 ) , @xmath2co ( @xmath22 ) , and c@xmath24o ( @xmath22 ) , simultaneously . in the present paper , we focus on the @xmath0co ( @xmath22 ) and @xmath2co ( @xmath22 ) data . the frequency resolution was set to 61 khz , which corresponds to the velocity resolution of @xmath20 0.08 km s@xmath7 at 220 ghz . during the observations , the system noise temperatures were about 200 @xmath11 400 k in a single sideband . further description of the telescope is given by @xcite . the observed area was a rectangle area with 40 square degrees whose coordinates of the bottom - left - corner ( blc ) and top - right - corner ( trc ) are @xmath27 ( @xmath17 , @xmath18 ) and ( @xmath16 , @xmath19 ) , respectively . the area was divided into 40 boxes of @xmath28 , each of which was scanned a few times in both galactic longitude and lattitude with a row spacing of @xmath29 at a scan speed of @xmath30 sec per row . the standard chopper wheel method was used to correct the output signals for the atmospheric attenuation and to convert them into the antenna temperatures ( @xmath31 ) . then , we corrected the antenna temperatures for the main beam efficiency to obtain the brightness temperatures @xmath32 (= @xmath33 ) . we applied a convolution technique with a gaussian function to calculate the intensity onto a regular grid with a spacing of 1@xmath5 . the resultant effective angular resolution was @xmath34 . the rms noise levels of the final @xmath0co and @xmath2co maps vary from box to box , and are summarized in appendix a ( tables a1 and a2 ) . the average noise levels are given in the last column of table 1 . to compare with the molecular line data , we downloaded the near - infrared ( nir ) color excess map of @xmath35 shown in figure [ fig:2mass](a ) from http://darkclouds.u-gakugei.ac.jp/2mass/download.html . the map was originally generated by @xcite and @xcite utilizing the 2mass point source catalog , and sampled at the same @xmath29 grid along the galactic coordinates as the co data . the angular resolution , however , varies in the range of @xmath36 , because the map was drawn using the adaptive grid " technique to achieve a constant noise level . see @xcite and @xcite for details . in section [ sec : av ] , using the @xmath35 color excess map , we construct the visual extinction map of the aquila rift and serpens cloud complexes , which is presented in figure [ fig:2mass](c ) . some of the active star - forming regions are designated in figure [ fig:2mass](c ) . figure [ fig:12co integ map ] shows the @xmath0co ( @xmath22 ) velocity integrated intensity map toward the aquila rift and serpens molecular cloud complexes . for comparison , the @xmath0co ( @xmath22 ) channel maps and the longitude - velocity diagram averaged over the galactic latitude range of @xmath37 are shown in figures [ fig:12co channel map ] and [ fig : pv map ] , respectively . @xmath0co , @xmath2co , and c@xmath24o line spectra at several positions are also presented in figure [ fig : spectra ] . the co emission is extended almost over the entire mapped area . the co integrated intensity takes its maximum at the position of @xmath38 , which is in the molecular cloud associated with the serpens cluster a. the @xmath0co emission associated with the serpens cluster b is not so prominent . other co peaks are located at @xmath39 , which corresponds to the w40 hii region , and at the position of @xmath40 . most co emission comes from the velocity range of 0 @xmath11 20 km s@xmath7 , which is presumably associated with the gould belt , an expanding ring of stars in the local spiral arm . a prominent feature of the longitude - velocity and channel maps is the existence of two spatially - extended components with different velocities , @xmath41 km s@xmath7 and 8 km s@xmath7 in the part of galactic longitude smaller than 29@xmath12 . the former component is recognized in the first panel ( @xmath42 km s@xmath7 ) of the velocity channel map in figure [ fig:12co channel map ] . the latter is more spatially - extended in the entire observed area . in the @xmath0co longitude - velocity diagram ( figure [ fig : pv map ] ) , the two components appear to converge at around the position of the serpens south and w40 cloud , which might suggest the two components interact with each other , triggering the active star formation in serrpens south and w40 . however , there is also a possibility that the two components are simply overlapped along the line - of sight . another interesting characteristic is that in the channel maps , some velocity - coherent , large - scale linear structures can be recognized . for example , in both 1@xmath114 km s@xmath7 ( the first panel of figure [ fig:12co channel map ] ) and 9@xmath1113 km s@xmath7 ( the third panel of figure [ fig:12co channel map ] ) panels , a couple of large arcs with similar morphologies are recognized . to emphasize these structures , we show in figures [ fig:12co channel map2]a and [ fig:12co channel map2]b , the @xmath0co intensity maps integrated over the velocity intervals of @xmath43 km s@xmath7 and @xmath44 km s@xmath7 , respectively , where the velocity - coherent linear structures are indicated with dashed curves . assuming a distance of 260 pc , these linear structures are found to be very large as about 6 @xmath11 20 pc in length . hereafter , we call these linear structures arcs . since ism is highly turbulent , such large - scale structures should be created by large - scale dynamical events such as large - scale turbulent flows and/or supernova shocks . these arcs might be created by the superbubbles converging toward the observed area @xcite . for comparison , we show @xmath2co ( @xmath22 ) velocity integrated intensity map toward the aquila rift and serpens molecular cloud complexes in figure [ fig:13co integ map ] . the @xmath2co ( @xmath22 ) channel maps and the longitude - velocity diagram integrated over the galactic latitude range of @xmath37 are indicated in figures [ fig:13co channel map ] and [ fig : pv map2 ] , respectively . comparison between the @xmath2co map and the extinction map indicates that the @xmath2co emission better traces the extinction image than the @xmath0co emission ( figure [ fig:13co integ map]b ) . two components with different velocities seen in the @xmath0co channel maps are also recognized in the @xmath2co channel maps . some arcs identified with @xmath0co can be seen in the @xmath2co map , indicating that densities of those arcs are relatively high . the longitude - velocity diagram shows that only 8 km s@xmath7 component is prominent in the @xmath2co map . here we compare the co data with dust detected in the @xmath35 map . while the co lines are emitted only in dense regions ( e.g. , @xmath45 @xmath46 ) , the color excess map traces the total dust column density along the line of sight including diffuse regions in the background ( or foreground ) unrelated to the co emitting volumes . we therefore should remove the background of the @xmath35 map to compare the two dataset directly . for compact clouds at high latitudes , such removal of the background can be easily done by subtracting a constant or by fitting the background with an exponential function of the galactic latitudes @xmath47 ( e.g. , * ? ? ? * ) . those methods , however , do not work well in the case of the clouds in aquila rift , because the region suffers from great complexity over a large extent . we therefore decided to use the following procedure to remove the background : we first defined the temporary background as the regions without showing apparent small scale - structures in figure [ fig:2mass](a ) , and masked the apparent clouds in the figure by eye - inspection . we then fitted the unmasked pixels by two dimensional ( 2d ) polynomial function of @xmath48 degree , and subtracted it from the original @xmath35 map to convert the residual map to @xmath49 as @xmath50 where the coefficient is calculated using the reddening law found by @xcite . we smoothed the resulting @xmath49 map with @xmath51 pixels ( @xmath52 ) , and redefined the background as the regions with @xmath53 mag in the smoothed map . the background regions defined in this manner is shown in figure[fig:2mass ] ( b ) . to better assess the background , we further fitted the background regions in the figure with a 2d polynomial function with higher orders ( @xmath54 ) , and decided to adopt @xmath55 degree to produce the final @xmath49 map which we use in the following analyses . we show the map in figure [ fig:2mass](c ) . the maximum value found in the map is @xmath56 mag , and the @xmath57 noise level of the map is @xmath58 mag . in order to infer the possible systematic error due to the background removal , we reproduced background maps with different orders of polynomials @xmath59 to derive @xmath49 maps in the same way as described above . we found that the @xmath49 values in the resulting maps vary by @xmath60 mag at most depending on @xmath54 , which we regard as the uncertainty in our @xmath49 map arising from the background determination . to compare the co and @xmath49 data directly , we smoothed both of the data to the common 6@xmath5 ( fwhm ) angular resolution . we excluded some positions from the analyses where the @xmath49 map suffers from a very low angular resolution ( @xmath61 ) . figures [ fig:13co_vs_av ] and [ fig:12co_vs_av ] show the relations of the @xmath2co and @xmath0co integrated intensities vs. @xmath49 , respectively . in the case of the @xmath2co emission , the relation is simple and can be fitted by a linear function , while that of the @xmath0co emission is complex mainly due to the heavy saturation at higher @xmath49 . this saturation is presumably due to the fact that @xmath0co emission becomes optically thick at higher @xmath49 . in order to estimate the total masses of clouds that we identified with the molecular data ( see section [ sec : clouds ] ) , we formulate the dependence of @xmath62 on @xmath49 . for this purpose , we fitted the @xmath62 vs. @xmath49 relation by a linear function , @xmath63 where the coefficient best fitting the relation is @xmath64 the relations of @xmath65 vs. @xmath49 and @xmath62 vs. @xmath49 appear to be better fitted with exponential functions . here , we attempt to fit these relations with exponential functions . similar attempt has been done by @xcite . for the @xmath62 vs. @xmath49 relation , we consider the following exponential function , @xmath66 \ , \label{eq : exp_13co}\ ] ] where @xmath67 and @xmath68 are constants . the coefficients best fitting the relation are obtained as @xmath69 and @xmath70 for the @xmath65 vs. @xmath49 relation , we consider the following exponential function @xmath71 , \label{eq : exp_12co}\ ] ] where @xmath72 and @xmath73 are constants . the coefficients best fitting the relation are derived as @xmath74 the relations expressed by equations ( [ eq : exp_13co ] ) and ( [ eq : exp_12co ] ) are shown in figures [ fig:13co_vs_av ] and [ fig:12co_vs_av ] , respectively . the @xmath23 factor gives a scaling between co luminosity and molecular cloud mass and we use @xmath23 to derive the masses of the clouds identified in the next section . as shown above , @xmath62 is well correlated with @xmath49 . therefore , we use @xmath2co emission to identify the clouds and evaluate their masses . the @xmath10 factor is calculated as @xmath75 where we used equation ( [ eq : linear_13co ] ) and assumed the relation of @xmath76 @xmath77 mag@xmath7 and @xmath78 @xcite . adopting the above @xmath10 factor , the total molecular gas mass in the observed region is estimated to be about @xmath79 under the assumption of @xmath80 pc . the @xmath23-factors are usually estimated by using the @xmath0co ( @xmath15 ) line . for our galaxy , the averaged @xmath81 is evaluated to be @xmath82 @xmath77 k@xmath7 km@xmath7 s with an uncertainty of @xmath83 @xcite . this values vary from region to region . for example , @xcite derived the @xmath84-factor of @xmath85 @xmath77 k@xmath7 km@xmath7 s in the taurus molecular cloud . for l1551 , @xcite measured the @xmath86 @xmath77 k@xmath7 km@xmath7 s. here , we attempt to compare our @xmath84-factor using higher transition with these previous values using the @xmath15 line . we choose the serpens main cloud because this region basically has single component . in constrast , the serpens south+w40 region contains multiple - components . the part including serpens main cloud has @xmath87 k km s@xmath7 and @xmath88 k km s@xmath7 . in the part including serpens main cloud , the mean intensity of @xmath0co @xmath89 and line width are evaluated to be 5.6 k and 3.5 km s@xmath7 , respectively . the mean intensity of @xmath2co ( @xmath22 ) and line width are evaluated to be 2.5 k and 2.5 km s@xmath7 , respectively . adopting the h@xmath90 density of @xmath91 @xmath46 , @xmath92 k , the h@xmath90 column density of @xmath93 @xmath77 ( @xmath94 ) , the co fractional abundance of @xmath95 , and @xmath0co/@xmath2co ratio of 80 , the intensity of @xmath0co ( @xmath15 ) is derived as @xmath20 7 k using ` radex ` @xcite , and thus @xmath96 k km s@xmath7 . thus , we can convert our @xmath97 to @xmath98 @xmath77 k@xmath7 km@xmath7 s , which is about 2.5 times larger than the standard value of @xcite . therefore , our @xmath97 value is consistent with the previous estimation , taking into account a large variation of @xmath84-factor values . as shown in section [ sec : results ] , the correlation between @xmath2co integrated intensity and extinction @xmath49 is reasonably good . therefore , we identify molecular clouds from the @xmath2co data cube and derive their physical quantities such as cloud masses , radii , and line widths . there are several methods proposed to identify clouds . in the following , we apply a new method called ` scimes ` @xcite to our co data cube . ` scimes ` uses the output of ` dendrogram ` @xcite , and identify the structures that are clustering , based on a graph theory . ` dendrogram ` characterizes the hierarchical structure of the isosurfaces for molecular line data cubes and have been used to identify the structures like cores and clumps @xcite . however , it is sometimes difficult to characterize the structures identified only with ` dendrogram ` . recently , @xcite constructed a new method called ` scimes ` ( spectral clustering for interstellar molecular emission segmentation ) to identify clustered structures with similar emission properties as `` clouds '' based on the hierarchical structures identified by ` dendrogram ` . applying ` scimes ` to the @xmath0co ( @xmath15 ) data cube of orion a , @xcite demonstrated that ` scimes ` can identify the cloud structure well . in the following , we apply ` scimes ` to our @xmath2co data cube because @xmath2co trace the cloud structures better than @xmath0co ( see the previous section ) and call a clustered structure identified by ` scimes ` as a `` cloud '' . before applying ` dendrogram ` to our data , we masked pixels whose intensities are about twice the noise level of each observation box . ` dendrogram ` requires three input parameters : ` min_value ` , ` min_delta ` , and ` min_npix ` . the first parameter , ` min_value ` , specifies the minimum value above which ` dendrogram ` identifies the structures . the second parameter , ` min_delta ` , is the minimum height required for a structure identified . the third parameter , ` min_npix ` , is the minimum number of pixels that a structure should contain in order to remain an independent structure . we apply ` dendrogram ` with the following three input parameters of ` min_value ` = 2 @xmath99 , ` min_delta ` = 3 @xmath99 , and ` min_npix ` = 50 , where @xmath99 is the rms noise level of the data . the value of ` min_npix ` = 50 is adopted to estimate the physical quantities with reasonable accuracy . for the ` dendrogram ` analysis , we adopt @xmath100 k. the ` dendrogram ` classifies two types of structures : _ leaves _ and _ branches_. _ branches _ are the structure which split into multiple sub - structures , and _ leaves _ are the structures which do not have any sub - structures . see @xcite for the detail of the ` dendrogram ` analysis . then , we applied ` scimes ` with the results of ` dendrogram ` . we use the element - wise multiplication of the luminosity and the volume matrix for analyzing the clustering of the dendrogram . see @xcite for more details . in total , we identified 61 clouds . we classified all the structures as clouds . in figure [ fig : scime ] , we also plot the clouds identified on the @xmath2co velocity - integrated intensity map . the structure enclosed by each colored curve corresponds to the cloud identified . in table 2 , we list positions and some physical quantities of the clouds identified such as the cloud velocity @xmath101 , the lengths of major and minor axes , and position angles . the positions of a cloud are determined as the intensity - weighted positions of the structure identified in the corresponding directions . major and minor axes of the projected structure onto the galactic longitude - latitude plane , @xmath102 and @xmath103 , are computed from the intensity - weighted second moment in direction of greatest elongation and perpendicular to the major axis in the galactic longitude - latitude plane , respectively . the position angle is the angle of the major axis in degrees counter - clockwise from the longitude axis . the region including the w40 hii region and serpens south embedded cluster is assigned as a single cloud . the serpens main cloud is also identified as a single cloud . in table 3 , we list cloud radii , masses , and line widths of the clouds identified . the cloud radii are computed as @xmath104 , following the definition of @xcite . the cloud velocities are calculated as the intensity - weighted local standard of rest ( lsr ) velocity of the structure identified . the line widths are the intensity - weighted fwhm values . the cloud mass is computed using the relation @xmath105 where @xmath106 is the intensity in the brightness temperature scale at the @xmath107 , @xmath108 , @xmath109th grid in the position - position - velocity space . the spacings of @xmath110 and @xmath111 are the sizes of each grid in the galactic longitude and latitude directions , respectively , and @xmath112 @xmath113 cm . the velocity width of @xmath114 is the velocity difference of adjacent channels and @xmath115 km s@xmath7 . figures [ fig : radius ] , [ fig : mass ] , and [ fig : linewidth ] show the histograms of the radius , mass , and line width of the identified clouds , respectively . the radii of the identified clouds range from 0.3 to 3 pc and its mean value is around @xmath20 1 pc . the cloud masses range from 5 @xmath116 to @xmath117 . the distribution of cloud masses has a peak at around 30 @xmath116 , which indicate that there are many less - massive clouds . here , we assumed @xmath80 pc . the most massive cloud has about @xmath118 and contains the serpens south and w40 region , which are located near the center of the observation box . this cloud has at least two different components with different velocities ( the velocities of the two peaks are 5.9 and 7.6 km s@xmath7 ) , but it is classified as a single cloud ( no . 16 cloud with @xmath119 km s@xmath7 ) . the second most massive cloud is a molecular cloud which contains the serpens main region or serpens cloud core ( serpens cluster a and b in figure [ fig:2mass ] ) a herbig ae be protostellar system ec 95 is located in serpens cloud core whose line - of - sight velocity is measured as @xmath120 km s@xmath7 @xcite . the distance of ec95 is measured from the vlba observations as 415 pc . in table 2 , the serpens cloud core is associated with no . 2 cloud with @xmath121 km s@xmath7 . if ec95 is really associated with no . 2 cloud , the distance of no . 2 cloud may be almost 415 pc . the third most massive cloud is associated with mwc 297 . this has the largest line width . this is identified as no . 9 cloud with @xmath122 km s@xmath7 . the total mass is estimated to be about @xmath123 . the line widths ranges from @xmath20 0.3 to 4 km s@xmath7 and its distribution has a single peak at around 1 km s@xmath7 . the cloud identified have basically supersonic internal motions . line widths of the clouds provide us with information of internal cloud turbulence . in figure [ fig : line - width - radius]a , we present the line width - radius relation of the identified cloud . here , we assumed a diatance of 260 pc . for comparison , we show the line width - radius relations obtained by @xcite , @xcite , @xcite , and @xcite . for the relations of @xcite and @xcite , we modified the original relations presented in the corresponding papers so as to match the definitions of the radii and line widths . see the appendix of @xcite for the details . it is worth noting that @xcite and @xcite used ` dendrograms ` and ` scimes ` to identify the clouds , respectively , and therefore can be directly compared to our result . the two line - width - size relations of @xcite are derived based on the @xmath2co ( @xmath15 ) data of perseus molecular cloud and n@xmath90h@xmath124 ( @xmath15 ) data of the central molecular zone ( cmz ) . the line - width - size relation of @xcite is derived based on the @xmath0co ( @xmath15 ) data of orion a molecular cloud . the cloud line - width of perseus by @xcite is about 1.5 times larger than that of orion a by @xcite at a given radius . our line - width - radius relation is consistent with the @xcite relation , perseus relation by @xcite , and orion a relation by @xcite . if we adopt a distance of 415 pc , the line - widths of the clouds tend to be smaller than those of @xcite s perseus and orion a relations . for comparison , we show the line - width - radius relation of the identified clouds in figure [ fig : line - width - radius]b , where we assumed @xmath125 pc . for both the distances , the clouds in cmz have significantly larger line - widths , which support the results of @xcite . the gravitational boundedness is often evaluated by the virial parameter . in figure [ fig : virial - mass]a , we plot the virial parameter against the cloud mass . here , we define the virial parameter as @xmath126 where @xmath127 is constant and we set @xmath128 corresponding to a centrally - condense sphere with @xmath129 . for uniform sphere , @xmath130 . we did not apply the correction of the thermal line width of the h@xmath90 gas to the line width @xmath131 because its contribution to the total line width should be small as long as the gas temperature stays at around 10 k. majority of the clouds identified has virial parameters close to unity . even for smaller clouds , the virial parameters are close to or less than the unity . this indicates that most of the clouds identified are gravitationally - bounded . it is worth noting that similarly to the line - width - radius relation , the virial - parameter - mass relation is affected by the adopted distance . if we adopt a distance of 415 pc , the clouds tend to become more gravitationally - bounded . in either way , the majority of the clouds are gravitationally - bounded both for @xmath132 pc and 415 pc . this may be the reason why the northern part of aquila rift contains several active star - forming regions . we speculate that interaction of superbubbles , large - scale turbulent flows and/or cloud - cloud collisions may have trigger star formation in this region . accurate distance measurements are crucial to clarify the dynamical states of the clouds in this region . finally , we present the mass - radius relation of the identified clouds in figure [ fig : massradius ] , where we adopt a distance of 260 pc and 415 pc for the panels ( a ) and ( b ) , respectively . according to @xcite , the clouds forming massive stars ( @xmath133 ) have masses larger than @xmath134 , where we rescaled @xcite s relation by a factor of 1.5 to take into account the difference in the adopted dust parameters , following @xcite . in addition , @xcite derived the threshold column density of 1 g @xmath77 for massive star formation . for comparison , we indicate @xcite s relation and @xcite s criteria for massive star formation in figure [ fig : massradius ] with dashed and dashed - dotted lines , respectively . for both panels , most of the identified clouds are distributed below the two lines . most massive clouds are located near @xcite s line . therefore , we suggest that in the observed area , it may be difficult to form massive stars in near future without any external events such as cloud - cloud collision which can increase the local column densities . in fact , there are only a couple of regions associated with hii regions in the observed area , w40 and mwc 297 . the exciting stars of the hii regions are late o - type and b - type stars with masses of @xmath135 . this fact supports the idea that the aquila rift and serpens cloud complexes may not be a massive star - forming region . we summarize the main results of the present paper as follows . * we carried out large - scale @xmath0co ( @xmath22 ) , @xmath2co ( @xmath22 ) , and c@xmath24o ( @xmath22 ) mapping observations toward the aquila rift and serpens region of @xmath136 and @xmath4 at a effective angular resolution of [email protected] ( @xmath137 pc ) and at a velocity resolution of @xmath20 0.08 km s@xmath7 with the velocity coverage of @xmath8 km s@xmath138 @xmath139 35 km s@xmath7 . * from the co channel maps , we found a number of arcs , which extend over 1 @xmath11 7 pc . these structures may have been formed by the superbubbles that converge toward our observed area ( see also * ? ? ? the velocity integrated intensities of @xmath2co are well correlated to the 2mass visual extinction . we derived the @xmath10 factor , @xmath140 by comparing the @xmath2co velocity integrated intensity map with the 2mass extinction map . * there are two distinct components with having different velocities ( @xmath141 km s@xmath7 and 8 km s@xmath7 ) . they appear to converge at the position of the serpens south and w40 cloud . this is consistent with the scenario that the collision between flows and clouds triggered active star formation in this region @xcite . however , there is a possibility that the two components are simply overlapped along the line of sight . * applying ` dendrogram`+`scimes ` to the @xmath2co data cube , we identified 61 clouds . * the line - width - radius relation of the clouds reasonably agrees with those of nearby star - forming regions . however , if we adopt the uniform distance of 415 pc to the region , the agreement of the line - width - radius relations becomes worse . we speculate that the representative distance to this area is @xmath132 pc , or two components with different distances ( @xmath20 260 pc and @xmath142 415 pc ) are overlapped along the line of sight . * the virial - parameter - mass relation shows that the clouds identified are close to virial equilibrium with large dispersion . this may be the reason why the observed area contains several active star forming regions . this characteristic is contrast to that of the southern part of aquila rift , where the virial parameters tend to be larger than unity @xcite . for a larger assumed distance of @xmath143 415 pc , the majority of the clouds have virial parameter smaller than unity . in other words , the clouds appear too gravitationally - bounded . this research made use of two python packages : ` dendrogram ` @xcite and ` scimes ` version 0.2.0 @xcite . the former is a python package to characterize the hierarchical structure of molecular cloud . the latter is a python package to find relevant structures into dendrograms of molecular gas emission using the spectral clustering approach . we made use of astropy version 1.1.2 python package to make several plots @xcite . without these packages , we could not complete the analyses presented in this paper . we would like to express our special appreciation to all the developers of the packages . we also thank to kazuki tokuda who helped data reduction of the 1.85-m data . andr , p. , menshchikov , a. , bontemps , s. , et al . 2010 , , 518 , l102 astropy collaboration , robitaille , t. p. , tollerud , e. j. , et al . 2013 , , 558 , a33 bohlin , r. c. , savage , b. d. , & drake , j. f. 1978 , , 224 , 132 bolatto , a. d. , wolre , m. , & leroy , a. k. 2013 , ara&a , 51 , 207 colombo , d. , rosolowsky , e. , ginsburg , a. , duarte - cabral , a. , & hughes , a. 2015 , , 454 , 2067 dame , t. m. & thaddeus , p. , 1985 , , 297 , 751 dame , t. m. , ungerechts , h. , cohen , r. s. et al . , 1987 , , 322 , 706 dzib , s. , loinard , l. , mioduszewski , a. j. , et al . 2010 , , 718 , 610 dobashi , k. , uehara , h. , kandori , r. , sakurai , t. , kaiden , m. , umemoto , t. , & sato , f. 2005 , , 57 , s1 dobashi , k. 2011 , , 63 , s1 dobashi , k. , marshall d. j. , shimoikura , t. , & bernard , j .- ph . 2013 , , 65 , 31 drew , j. e. , busfield , g. , hoare , m. g. , murdoch , k. a. , nixon , c. a. , oudmaijer , r. d. 1997 , , 286 , 538 duarte - cabral , a. , dobbs , c. l. , peretto , n. , & fuller , g. a. 2011 , , 528 , 50 eiroa , c. , djupvik , a. a. , & casali , m. m. 2008 , in handbook of star forming regions vol . p.693 frisch , p.c . , 1998 , in iau colloq . 166 , _ the local bubble and beyond _ , ed . d. breitschwerdt , m. freyberg & trumper ( berlin : springer ) , p. 269 heyer , m. h. , & brunt , c. m. 2004 , , 615 , l45 hillenbrand , l. a. , strom , s. e. , vrba , f. j. , keene , j. 1992 , , 397 , 613 kawamura , a. , onishi , t. , mizuno , a. et al . 1999 , , 51 , 851 kauffmann , j. , pillai , t. , shetty , r. , myers , p. c. , & goodman , a. a. 2010 , , 716 , 433 kirk , h. , myers , p. c. , bourke , t. l. , et al . 2013 , , 766 , 115 konyves , v. , andr , p. , menshchikov , a. , et al . 2015 , , 584 , 91 krumholz , m. r. , & mckee , c. f. 2008 , , 451 , 28 larson , r. b. 1981 , , 194 , 809 lin , s .- j . , shimajiri , y. , hara , c. , et al . 2016 , , 826 , 193 mcmullin , j.p . , mundy , l. g. , black , g. a. , et al . , 2000 , , 536 , 845 gutermuth , r. a. , bourke , t. l. , allen , l. e. , et al . , 2008 , 673 , l151 maruta , h. , nakamura , f. , nishi , r. , et al . , 2010 , , 714 , 680 maury , a. , p. andr , menshchikov , a. , konyves , v. , & bontemps , s. 2011 , , 535 , 77 nakamura , f. , sugitani , k. , shimajiri , y. et al . 2011 , , 737 , 56 nakamura , f. , sugitani , k. , tanaka , t. , et al . 2014 , , 791 , 23 onishi , t. , nishimura , a. , ota , y. , et al . 2013 , , 65,78 pineda , j. l. , goldsmith , p. f. , chapman , n. , et al . 2010 , , 721 , 686 plunkett , a. l. , arce , h. , corder , s. a. , et al . 2015 , , 803 , 22 prato , l. , rice , e. l. , & dame , t. m. 2008 , in handbook of star forming regions vol . i. reike , g. h. , & lebofsky , m. j. 1985 , , 288 , 618 rosolowsky , e. , pineda , j. e. , kauffmann , j. , & goodman , a. a. 2008 , , 679 , 1338 rodney , s. a. & reipurth , b. 2008 , handbook of star forming regions , volume ii , 693 shetty , r. , beaumont , c. n. , burton , m. g. , et al . 2012 , , 425 , 720 shimoikura , t. , dobashi , k. , nakamura , f. , et al . 2015 , , 806 , 201 sugitani , k. , nakamura , f. , watanabe , m. , et al . 2011 , , 734 , 63 tanaka , t. , nakamura , f. , awazu , y. , et al . 2013 , , 778 , 34 van der tak , f. f. s. , black , j. h. , scholer , f. l. , jansen , d. j. , van dischoeck , e. f. 2007 , , 468 , 627 llllllll @xmath0co & @xmath22 & 230.5380000 & 2.7 & 0.079 & 0.64 @xmath144 0.10 + @xmath2co & @xmath22 & 220.3986765 & 2.7 & 0.083 & 0.62 @xmath144 0.10 + c@xmath24o & @xmath22 & 219.5603568 & 2.7 & 0.083 & 0.62 @xmath144 0.10 [ tab : obs ] llllllllll 1 & 25.963 & 4.113 & 18h24m01.1s & -04d18m04.1s & 3354.1 & 1422.7 & 89.9 & 2184.4 & 3.6 + 2 & 31.118 & 5.203 & 18h29m36.6s & + 00d45m41.5s & 3621.6 & 1125.6 & 156.9 & 2019.0 & 8.0 + 3 & 30.567 & 3.202 & 18h35m43.7s & -00d38m28.9s & 3124.3 & 597.1 & 173.6 & 1365.8 & 9.5 + 4 & 31.822 & 2.936 & 18h38m57.7s & + 00d21m07.3s & 1449.3 & 1330.8 & 124.9 & 1388.8 & 8.7 + 5 & 31.336 & 4.410 & 18h32m49.7s & + 00d35m34.5s & 1552.5 & 771.4 & 55.4 & 1094.4 & 10.3 + 6 & 27.094 & 4.131 & 18h26m02.9s & -03d17m34.7s & 1042.7 & 671.2 & 91.4 & 836.6 & 3.3 + 7 & 31.576 & 3.958 & 18h34m52.5s & + 00d35m59.8s & 562.4 & 325.7 & 157.1 & 428.0 & 6.7 + 8 & 27.136 & 4.763 & 18h23m53.0s & -02d57m49.8s & 5639.3 & 737.4 & 58.3 & 2039.2 & 8.3 + 9 & 26.874 & 3.514 & 18h27m50.1s & -03d46m21.2s & 2390.2 & 1551.0 & 159.1 & 1925.4 & 5.4 + 10 & 27.529 & 5.652 & 18h21m27.0s & -02d12m21.3s & 1576.8 & 762.2 & 173.4 & 1096.3 & 8.0 + 11 & 25.811 & 3.066 & 18h27m27.7s & -04d55m16.2s & 3181.2 & 1132.9 & 158.2 & 1898.4 & 8.7 + 12 & 30.420 & 2.482 & 18h38m01.3s & -01d06m04.8s & 1552.7 & 1294.8 & 165.7 & 1417.9 & 8.1 + 13 & 26.068 & 4.048 & 18h24m26.8s & -04d14m18.0s & 1525.3 & 973.9 & 163.2 & 1218.8 & 7.9 + 14 & 32.384 & 2.794 & 18h40m29.7s & + 00d47m11.8s & 871.1 & 465.8 & 162.8 & 637.0 & 10.0 + 15 & 32.184 & 2.332 & 18h41m46.4s & + 00d23m54.0s & 1068.8 & 506.7 & 165.3 & 735.9 & 9.6 + 16 & 28.949 & 3.766 & 18h30m45.5s & -01d49m06.7s & 2659.8 & 2070.8 & 73.1 & 2346.9 & 7.1 + 17 & 32.213 & 3.347 & 18h38m12.6s & + 00d53m11.1s & 781.8 & 334.8 & 179.6 & 511.7 & 7.9 + 18 & 32.760 & 2.718 & 18h41m27.0s & + 01d05m11.1s & 987.5 & 732.3 & 176.6 & 850.4 & 11.7 + 19 & 28.849 & 5.550 & 18h24m14.1s & -01d05m13.9s & 1225.3 & 673.4 & 115.6 & 908.4 & 8.5 + 20 & 30.417 & 4.672 & 18h30m13.5s & -00d06m09.7s & 725.0 & 295.8 & 155.0 & 463.1 & 7.8 + 21 & 31.291 & 3.482 & 18h36m03.2s & + 00d07m46.1s & 1052.2 & 600.6 & 105.0 & 794.9 & 6.4 + 22 & 25.465 & 1.944 & 18h30m49.0s & -05d44m47.8s & 1704.6 & 508.9 & 165.1 & 931.4 & 10.0 + 23 & 27.484 & 1.851 & 18h34m52.7s & -03d59m52.9s & 1727.6 & 1148.3 & 155.0 & 1408.5 & 8.8 + 24 & 25.368 & 4.271 & 18h22m21.1s & -04d45m12.6s & 672.3 & 613.4 & 84.2 & 642.2 & 6.6 + 25 & 32.894 & 2.189 & 18h43m34.5s & + 00d57m53.5s & 457.4 & 344.4 & 161.3 & 396.9 & 11.8 + 26 & 25.125 & 3.864 & 18h23m20.9s & -05d09m26.1s & 694.5 & 380.0 & 141.3 & 513.7 & 7.2 + 27 & 26.678 & 2.616 & 18h30m40.1s & -04d21m39.0s & 1609.7 & 771.2 & 94.7 & 1114.2 & 5.6 + 28 & 30.888 & 4.145 & 18h32m57.3s & + 00d04m28.2s & 687.7 & 569.2 & 104.1 & 625.6 & 6.6 + 29 & 29.677 & 5.169 & 18h27m06.3s & -00d31m46.1s & 1502.9 & 724.6 & 53.2 & 1043.6 & 8.8 + 30 & 25.092 & 5.561 & 18h17m15.4s & -04d23m42.0s & 507.4 & 382.4 & 117.9 & 440.5 & 3.0 + 31 & 32.822 & 3.532 & 18h38m39.7s & + 01d30m42.6s & 673.5 & 549.8 & 136.9 & 608.5 & 9.3 + 32 & 28.617 & 5.234 & 18h24m56.0s & -01d26m16.6s & 2363.8 & 683.9 & 165.5 & 1271.5 & 5.4 + 33 & 27.167 & 5.120 & 18h22m40.3s & -02d46m15.7s & 1309.4 & 920.8 & 167.0 & 1098.0 & 3.9 + 34 & 32.359 & 2.507 & 18h41m28.2s & + 00d37m59.9s & 436.0 & 174.1 & 159.6 & 275.5 & 8.8 + 35 & 32.715 & 2.742 & 18h41m16.8s & + 01d03m24.7s & 740.4 & 448.8 & 110.5 & 576.5 & 10.0 + 36 & 30.030 & 1.919 & 18h39m18.5s & -01d42m20.9s & 1561.6 & 797.0 & 161.0 & 1115.6 & 7.4 + 37 & 27.910 & 3.273 & 18h30m36.2s & -02d57m59.2s & 1553.8 & 491.9 & 83.0 & 874.3 & 6.3 + 38 & 30.371 & 4.139 & 18h32m02.2s & -00d23m13.0s & 379.0 & 354.9 & 156.0 & 366.7 & 9.3 + 39 & 31.161 & 5.882 & 18h27m16.3s & + 01d06m35.3s & 842.3 & 370.9 & 173.8 & 558.9 & 7.8 + 40 & 32.700 & 2.295 & 18h42m50.7s & + 00d50m23.8s & 1035.9 & 293.5 & 155.2 & 551.4 & 10.1 + 41 & 25.199 & 5.069 & 18h19m12.3s & -04d31m52.3s & 1556.9 & 938.7 & 55.3 & 1208.9 & 6.2 + 42 & 25.861 & 3.192 & 18h27m06.2s & -04d49m06.0s & 849.9 & 355.6 & 164.4 & 549.8 & 2.5 + 43 & 30.493 & 4.323 & 18h31m36.2s & -00d11m40.1s & 1317.9 & 511.5 & 168.7 & 821.0 & 8.0 + 44 & 25.455 & 3.817 & 18h24m07.8s & -04d53m14.2s & 1175.3 & 649.9 & 169.4 & 874.0 & 6.9 + 45 & 25.139 & 5.629 & 18h17m06.3s & -04d19m22.3s & 735.9 & 437.5 & 96.8 & 567.4 & 4.1 + 46 & 25.689 & 4.375 & 18h22m34.7s & -04d25m17.4s & 696.2 & 286.7 & 150.5 & 446.8 & 4.8 + 47 & 32.226 & 2.389 & 18h41m38.8s & + 00d27m42.3s & 588.8 & 429.3 & 136.8 & 502.7 & 5.4 + 48 & 32.035 & 3.662 & 18h36m46.0s & + 00d52m20.1s & 433.2 & 209.2 & 161.3 & 301.0 & 6.1 + 49 & 31.241 & 4.303 & 18h33m02.3s & + 00d27m35.1s & 440.1 & 279.9 & 141.7 & 351.0 & 6.2 + 50 & 32.311 & 3.911 & 18h36m22.9s & + 01d13m51.3s & 412.1 & 187.4 & 173.8 & 277.9 & 6.4 + 51 & 25.350 & 3.620 & 18h24m37.9s & -05d04m18.7s & 512.5 & 275.8 & 108.8 & 376.0 & 7.8 + 52 & 25.741 & 1.885 & 18h31m32.2s & -05d31m42.5s & 408.3 & 312.1 & 60.3 & 357.0 & 9.6 + 53 & 26.925 & 3.050 & 18h29m34.9s & -03d56m29.7s & 630.8 & 470.6 & 164.2 & 544.8 & 9.7 + 54 & 30.558 & 4.922 & 18h29m35.4s & + 00d08m15.6s & 628.2 & 355.6 & 147.8 & 472.6 & 9.9 + 55 & 26.457 & 3.181 & 18h28m14.8s & -04d17m44.6s & 797.4 & 328.5 & 157.4 & 511.8 & 11.7 + 56 & 30.134 & 4.615 & 18h29m54.6s & -00d22m45.5s & 524.2 & 290.3 & 142.6 & 390.1 & 11.2 + 57 & 28.148 & 1.093 & 18h38m48.1s & -03d45m21.8s & 598.2 & 277.3 & 168.3 & 407.3 & 11.8 + 58 & 27.603 & 1.875 & 18h35m00.6s & -03d52m53.2s & 533.4 & 329.1 & 163.9 & 419.0 & 11.7 + 59 & 32.732 & 2.052 & 18h43m46.1s & + 00d45m26.9s & 345.9 & 185.5 & 172.5 & 253.3 & 13.1 + 60 & 32.092 & 3.376 & 18h37m53.2s & + 00d47m31.2s & 320.4 & 247.7 & 124.8 & 281.8 & 13.7 + 61 & 32.667 & 1.181 & 18h46m45.0s & + 00d18m11.7s & 839.8 & 256.9 & 131.8 & 464.5 & 20.2 [ tab : scime1 ] lllll 1 & 2.8 & 552.5 & 1.51 & 1.4 + 2 & 2.5 & 3875.1 & 1.78 & 0.3 + 3 & 1.7 & 584.2 & 2.07 & 1.6 + 4 & 1.8 & 1722.2 & 2.67 & 0.9 + 5 & 1.4 & 277.8 & 1.13 & 0.8 + 6 & 1.1 & 74.6 & 0.91 & 1.5 + 7 & 0.5 & 31.2 & 0.77 & 1.3 + 8 & 2.6 & 1005.1 & 2.01 & 1.3 + 9 & 2.4 & 1984.3 & 3.53 & 1.9 + 10 & 1.4 & 306.7 & 1.31 & 1.0 + 11 & 2.4 & 464.4 & 1.46 & 1.4 + 12 & 1.8 & 1055.2 & 2.27 & 1.1 + 13 & 1.5 & 308.3 & 1.51 & 1.4 + 14 & 0.8 & 56.4 & 0.62 & 0.7 + 15 & 0.9 & 85.7 & 1.16 & 1.8 + 16 & 3.0 & 6009.1 & 2.75 & 0.5 + 17 & 0.6 & 46.8 & 0.66 & 0.8 + 18 & 1.1 & 216.8 & 1.57 & 1.5 + 19 & 1.1 & 72.9 & 0.95 & 1.8 + 20 & 0.6 & 7.4 & 0.30 & 0.9 + 21 & 1.0 & 172.9 & 1.04 & 0.8 + 22 & 1.2 & 106.7 & 0.94 & 1.2 + 23 & 1.8 & 470.8 & 3.03 & 4.3 + 24 & 0.8 & 40.1 & 0.53 & 0.7 + 25 & 0.5 & 42.3 & 0.94 & 1.3 + 26 & 0.6 & 21.7 & 0.86 & 2.8 + 27 & 1.4 & 470.1 & 1.40 & 0.7 + 28 & 0.8 & 84.9 & 1.47 & 2.5 + 29 & 1.3 & 143.9 & 1.23 & 1.7 + 30 & 0.6 & 14.5 & 0.45 & 1.0 + 31 & 0.8 & 49.6 & 0.89 & 1.5 + 32 & 1.6 & 121.6 & 1.22 & 2.5 + 33 & 1.4 & 73.6 & 0.82 & 1.6 + 34 & 0.3 & 9.8 & 0.48 & 1.0 + 35 & 0.7 & 25.4 & 0.45 & 0.7 + 36 & 1.4 & 372.7 & 1.73 & 1.4 + 37 & 1.1 & 49.2 & 0.43 & 0.5 + 38 & 0.5 & 22.0 & 1.03 & 2.8 + 39 & 0.7 & 34.0 & 0.45 & 0.5 + 40 & 0.7 & 10.1 & 0.33 & 1.0 + 41 & 1.5 & 177.5 & 1.11 & 1.3 + 42 & 0.7 & 39.7 & 0.81 & 1.5 + 43 & 1.0 & 72.8 & 0.77 & 1.1 + 44 & 1.1 & 42.6 & 0.50 & 0.8 + 45 & 0.7 & 31.7 & 0.73 & 1.5 + 46 & 0.6 & 12.8 & 0.64 & 2.3 + 47 & 0.6 & 19.1 & 0.63 & 1.7 + 48 & 0.4 & 11.4 & 0.83 & 2.9 + 49 & 0.4 & 6.7 & 0.42 & 1.5 + 50 & 0.4 & 12.3 & 0.47 & 0.8 + 51 & 0.5 & 5.1 & 0.23 & 0.6 + 52 & 0.4 & 10.6 & 0.49 & 1.3 + 53 & 0.7 & 15.9 & 0.45 & 1.1 + 54 & 0.6 & 14.5 & 0.31 & 0.5 + 55 & 0.6 & 98.7 & 1.55 & 2.0 + 56 & 0.5 & 14.4 & 0.76 & 2.4 + 57 & 0.5 & 20.8 & 1.22 & 4.6 + 58 & 0.5 & 6.7 & 0.22 & 0.5 + 59 & 0.3 & 17.5 & 0.79 & 1.4 + 60 & 0.4 & 14.9 & 0.96 & 2.7 + 61 & 0.6 & 154.1 & 1.17 & 0.6 [ tab : scime2 ] of the aquila rift and serpens molecular cloud complex generated by @xcite and @xcite . the fits data were downloaded from their website ( http://darkclouds.u-gakugei.ac.jp/2mass/download.html ) . ( b ) same as ( a ) , but the map is masked for small - scale structures to determine the background . ( c ) extinction map of @xmath49 derived from the background - subtracted @xmath35 map . the names of the regions discussed in the text are designated in the panel . , width=377 ] the rms noise levels of the final @xmath0co , @xmath2co , and c@xmath24o maps varies from box to box , and were summarized in tables a1 , a2 , and a3 , respectively . the average noise levels are summarized in the last column of table 1 . lllllllll @xmath145 & 0.75 & 0.66 & 0.76 & 0.53 & 0.58 & 0.57 & 0.50 & 0.69 + @xmath146 & 0.65 & 0.65 & 0.80 & 0.43 & 0.45 & 0.44 & 0.49 & 0.71 + @xmath147 & 0.87 & 0.69 & 0.57 & 0.35 & 0.28 & 0.44 & 0.63 & 0.80 + @xmath148 & 0.83 & 0.53 & 0.51 & 0.41 & 0.45 & 0.44 & 0.52 & 0.57 + @xmath149 & 0.54 & 0.56 & 0.55 & 0.70 & 0.68 & 0.61 & 0.53 & 0.63 [ tab : noise12co ] lllllllll @xmath145 & 0.72 & 0.64 & 0.73 & 0.52 & 0.56 & 0.55 & 0.49 & 0.67 + @xmath146 & 0.63 & 0.64 & 0.77 & 0.41 & 0.43 & 0.42 & 0.48 & 0.68 + @xmath147 & 0.85 & 0.68 & 0.56 & 0.35 & 0.28 & 0.43 & 0.61 & 0.77 + @xmath148 & 0.81 & 0.52 & 0.50 & 0.39 & 0.42 & 0.41 & 0.56 & 0.56 + @xmath149 & 0.53 & 0.54 & 0.53 & 0.68 & 0.66 & 0.59 & 0.51 & 0.61 + [ tab : noise13co ] lllllllll @xmath145 & 0.73 & 0.64 & 0.73 & 0.52 & 0.56 & 0.55 & 0.49 & 0.67 + @xmath146 & 0.64 & 0.64 & 0.77 & 0.41 & 0.43 & 0.41 & 0.48 & 0.68 + @xmath147 & 0.84 & 0.68 & 0.56 & 0.34 & 0.28 & 0.43 & 0.60 & 0.78 + @xmath148 & 0.81 & 0.52 & 0.50 & 0.39 & 0.42 & 0.41 & 0.55 & 0.55 + @xmath149 & 0.53 & 0.54 & 0.53 & 0.68 & 0.65 & 0.58 & 0.50 & 0.61 [ tab : noisec18o ]

we discuss the spectral properties of a sample of type-2 seyfert galaxies based upon the analysis of data . in this paper we consider the sources for which the x - ray spectra appear to be dominated by the nuclear continuum , transmitted through a large column of absorbing material . we find that both seyfert-2 galaxies and nelgs show iron k line profiles indicative of reprocessing of nuclear x - rays in a face - on accretion disk . such line profiles are also observed in seyfert-1 galaxies . this result is contrary to unification models , which would predict the inner regions of seyfert-2 galaxies to be observed edge - on . this raises some questions as to the orientation of the circumnuclear absorber . if the observed differences between seyfert type-1 and type-2 galaxies , and nelgs are not due to differences in the orientation of the absorbing material , then we suggest that differences in dust composition and grain size , and in the density of the circumnuclear gas could be of primary importance . = = = 1=1=0pt = 2=2=0pt = 2=2=0pt

astro-ph9708231

the classification of an active galactic nucleus ( agn ) depends upon the wavelength at which one observes . historically , optical observations yielded the categories of seyfert types 1 and 2 , classifying most of the sources which are now `` famous '' . seyfert-2 galaxies differ from type-1 in that the former show only narrow emission lines in their optical spectra . it was soon realized that some seyferts showed weak broad components along with the narrow emission lines and consequently the subclasses seyfert-1.5 , 1.8 and 1.9 were introduced to quantify the differences in strength of the broad - line components relative to the narrow lines . narrow emission line galaxies ( nelgs ) are bright and variable x - ray sources , discovered in early x - ray sky surveys ( marshall 1979 ) . the narrow optical emission lines often have weak , broad h@xmath0 and p@xmath1 emission ( ward 1978 , veron 1980 , shuder , 1980 ) making the optical spectra similar to seyfert-1.9 galaxies . thus the nelg classification is indicative that the source was discovered in an x - ray survey , but many nelgs are otherwise indistinguishable from seyfert-1.9 galaxies . optical spectroscopy and spectropolarimetry , infrared spectroscopy , x - ray spectroscopy and temporal studies plus @xmath2-ray spectra are all important in the determination of the fundamental nature of obscured nuclei . unification models for agn postulate that large amounts of dense , molecular material exists between the broad - emission - line region ( blr ) and the narrow - emission - line region ( nlr ) , in some cases within parsecs of the active nucleus ( see antonucci 1993 for a review of unified models for agn ) . the simplest geometry consistent with observations is a torus , and consequently it has been suggested that one of the primary factors in seyfert classification is the orientation of the absorbing torus to our line - of - sight . this hypothesis is consistent with the existence of circumnuclear molecular gas suggested by absorption measurements in a number of wavebands ( e.g. braatz 1993 , greenhill 1996 ) . in unified models , sources observed within the opening angle of the torus correspond to those classified optically as type-1 agn , while sources with lines - of - sight intersecting the torus correspond to type-2 agn . in the latter case , the nuclear light can nevertheless be observed via scattering or transmission . antonucci & miller ( 1985 ) provided compelling support for this model when they detected broad , seyfert-1 type emission lines in the polarized optical spectrum of the seyfert-2 galaxy ngc 1068 , and similar results were later obtained for a number of seyfert-2 galaxies ( miller and goodrich 1990 ; tran , miller and kay 1992 ) . recently , veilleux ( 1997 ) provided further support for the model with their infrared observations of pa@xmath1 , br@xmath2 and br@xmath0 lines in many seyfert 2 galaxies , revealing hidden blrs which were not always detectable in scattered optical light . optically - thick torii would be expected to result in collimation of nuclear continuum radiation , and imaging of optical emission lines has shown preferential elongation of some narrow - line regions along the radio axis of the agn ( haniff , wilson & ward 1988 ) . these regions must be ionized by a more intense radiation field than is directly observed , again , supporting the unified model ( e.g. pogge 1988 , tadhunter and tsvetanov 1989 ) . further support came from x - ray spectra , which showed large absorbing columns and iron k@xmath0 lines of high equivalent width ( ew ) ( e.g. awaki 1991 ) . the _ asca _ satellite ( makishima et al . 1996 ) consists of four co - aligned grazing - incidence x - ray telescopes ( xrts ; serlemitsos et al . the focal - plane instruments are two solid - state imaging spectrometers ( siss ) , each consisting of four ccd chips , providing an effective bandpass @xmath30.410 kev ( burke et al . 1994 ) , and two gas imaging spectrometers ( giss ) at the focus of the other two xrts , providing coverage over @xmath30.810 kev ( ohashi et al . 1996 and references therein ) . the sources presented here were systematically analyzed in the same way as the broad - line seyfert galaxies presented in nandra 1997 ( hereafter n97 ) . the analysis method is also described in paper i. in a previous paper we presented the basic data - analysis results from a sample of _ asca _ observations of type-2 seyfert galaxies ( turner 1997a , hereafter paper i ) , i.e. those having predominantly narrow optical emission lines . the original sample of 26 observations of 25 narrow - line agn comprised 17 seyfert-2 galaxies and 8 nelgs drawn from the _ asca _ public archive . in paper i we found the 0.6 - 10 kev spectra of a sample of seyfert-2 galaxies and nelgs to be complex , often containing a heavily - absorbed continuum component , a soft x - ray component and numerous x - ray emission lines . we found the 6 - 7 kev regime to be dominated by line flux from gas with ionization - state @xmath4 fe xvi . several sites are expected to produce significant x - ray line emission in agn including the line - of - sight absorber , optically - thick material out of the line - of - sight ( both the putative accretion disk and other larger - scale systems such as the torus ) , ionized ( scattering ) gas and regions of starburst emission . the absence of a strong 6.4 kev iron line component in starburst galaxies ( ptak 1997 ) , indicates that the presence of such a line is likely to be an indication of nuclear activity . iron k@xmath0 yields the strongest observed x - ray emission line in seyfert galaxies , and thus provides an important probe of conditions in the reprocessing material . a discussion of sources dominated by scattered and compton - reflected x - rays was presented in a second paper ( turner 1997b , hereafter paper ii ) . fig . 1 ( repeated from paper ii , for ease of reference ) shows the equivalent width of the iron k - shell line plotted against neutral x - ray absorbing column , @xmath5 , for the sample sources . equivalent widths were measured against the continuum component dominating the 6 - 8 kev range , and based upon the fit to a narrow gaussian profile ( see paper i for details ) . the dot - dashed line in fig 1 denotes the equivalent width of iron k@xmath0 predicted to be produced by transmission through a uniform shell of neutral material ( with solar abundances subtending 4@xmath6 to a continuum source of photon index @xmath7 , leahy & creighton 1993 ) , where the photon flux @xmath8 . the dashed line shows the equivalent width predicted via compton - reflection from optically - thick material , as a function of , assuming that only the power - law is absorbed . in this case the reflection is assumed to be produced from the accretion disk with an equivalent width @xmath9 ev ( n97 ) , as typically observed in seyfert-1 galaxies . coincidently , 230 ev represents both the maximum equivalent width observed when iron k@xmath0 was parameterized as a narrow gaussian line , and the mean equivalent width assuming a relativistic line profile ( n97 ) . sources can lie significantly above both of these model lines if the direct continuum is hidden but the reprocessed emission is observed , as the line equivalent width is then measured against a suppressed continuum . consideration of the iron k@xmath0 , ` [ ` oiii ` ] ` @xmath10 line and x - ray variability together suggested that ngc 1068 , ngc 4945 , ngc 2992 , mrk 3 , mrk 463e and mrk 273 are dominated by reprocessed x - rays ( paper ii ) . these sources were denoted `` group c '' ( marked with squares on fig . 1 ) . sources lying on the `` leahy and creighton line '' were denoted `` group a '' ( marked as circles on fig . 1 ) . thus group a is composed of ngc 1808 , ngc 4507 , ngc 5252 , ngc 6240 , eso 103-g35 , ic 5063 , ngc 7172 and ngc 7582 . ( table 1 shows the group designation of the sources based upon the original sample ) . sources with iron k@xmath0 equivalent widths lying between that line and the 230 ev line are consistent with seyfert-1 spectra transmitted through a high absorbing column and are denoted `` group b '' ( marked as stars on fig . 1 ) . group b is composed of ngc 526a , ngc 2110 , mcg-5 - 23 - 16 and ngc 7314 , which are all nelgs . this classification implies we see the nuclear component directly in group b , and this is supported by the observation of rapid x - ray variability in their flux ( paper i and hayashi 1996 ) and rapid variability of the iron line profile in ngc 7314 ( yaqoob 1996 ) . this division into groups a , b and c leaves mcg-01 - 01 - 043 , ngc 4968 , ngc 6251 , e 0449 , ngc 5135 and ngc 1667 unclassified , i.e. those with the lowest signal - to - noise ratio in the data . as we will demonstrate , this crude classification of sources yields a useful insight into the relative importance of the regions contributing to the x - ray spectra in several different cases . the distribution of sample sources is shown in table 2 . to summarize , in paper ii we showed that in group - c sources the iron k@xmath0 complex contains significant contributions from neutral and high - ionization species of iron , thus compton - reflection , hot gas and starburst emission all could make significant contributions to the observed x - ray spectra . mrk 3 appeared to be the only source which had little contamination by starburst activity and in this case the _ asca _ spectrum below 3 kev is dominated by gas with an x - ray ionization parameter @xmath11 ( as defined by netzer 1996 ) and effective column density @xmath12 . this material is more highly ionized than the zone of material comprising the warm absorber seen in seyfert-1 galaxies ( george 1997 , hereafter g97 ) , but may contain a contribution from shock - heated gas associated with the jet . in this paper the x - ray properties of sources in groups a and b are examined in the context of unified models for agn . paper i suggested the underlying continuum has a power - law with a slope @xmath13 in the 0.6 10 kev band . however , this result might not be meaningful for sources where we used a power - law to parameterize a spectrum dominated by scattered or reflected components ( i.e. group c ) . taking groups a and b and considering the continuum component which dominates the 2 - 10 kev band we obtained weighted mean slopes @xmath14 and @xmath15 respectively , and a dispersion of 0.11 for group a , 0.31 for group b. it is difficult to make a strict comparison between our results and the indices derived for seyfert-1 galaxies . a slope @xmath16 was obtained when the latter where fit to a model which included the effects of compton - reflection ( e.g. g97 ) . type-2 seyferts generally have complex soft x - ray spectra , and it is difficult to constrain the soft component , primary power - law and the effects of compton reflection , using data . however , we can compare the power - law parameterization of the 2 - 10 kev data with a similar parameterization of type-1 seyferts . group a slopes are consistent with the @xmath17 slopes obtained when proportional - counter spectra of seyfert-1 galaxies were modelled with a simple absorbed power - law ( mushotzky 1984 , turner & pounds 1989 ) over the @xmath18 kev range . thus group a appear consistent with the spectra of seyfert-1 galaxies when a comparison is made between similar fits . in group b , ngc 526a and ngc 7314 show a flattening of the spectra to high energies ( e.g. table 12 of paper i ) while the indices inferred for the primary continuum are @xmath19 and @xmath20 respectively . mcg-5 - 23 - 16 and ngc 2110 appear to be dominated by @xmath21 and 1.6 components . the @xmath22 slope is the only one which lies outside of the range of slopes observed in the seyfert-1 sample ( g97 ) . however , the slope implied in the case of ngc 7314 depends strongly upon whether a hard component is allowed . we conclude that groups a and b both show spectra which are generally consistent with the distribution of primary continuum slopes observed in seyfert-1 galaxies . comparing historical x - ray observations of various sources , it seems that some sources show significant index variability , as noted by weaver ( 1997 ) for mcg-5 - 23 - 16 . as discussed in the introduction , most of the sources in the original sample ( of 26 observations ) show a significant iron k@xmath0 line and the mean rest - energy of the line is @xmath23 kev ( paper i ) . the equivalent widths from the narrow gaussian model are shown in fig . 1 . consideration of the ( weighted mean ) equivalent widths in the narrow gaussian model to the line yields @xmath24 and @xmath25 ev for groups a and b respectively . however , the narrow gaussian model ( with energy fixed at 6.4 kev ) could miss some of the flux from a broadened line . the broad gaussian model yields ews @xmath26 ev and @xmath27 ev for a and b. the line energies and widths are @xmath28 kev , @xmath29 kev , @xmath30 kev , @xmath31 kev . to investigate further , we constructed average profiles for each group . papers i and ii showed the average sis+gis iron k@xmath0 line profiles for the individual sources . however , the sis data have superior energy resolution to the gis data , and so examination of the averaged sis0+sis1 data can give the clearest insight into the intrinsic line profiles . obviously , consideration of the sis data alone results in a decrease of signal - to - noise ratio when the data are considered on a source - by - source basis . however , the sis data can provide the most useful information when the data / model ratios for different sources are co - added . the energy - scale of each sis spectrum was redshift - corrected to the rest - frame of the source . the average sis data / model ratio was then calculated ( based upon the best - fit continuum model ) combining sources within each group . the result of this summation is shown for each group in fig . the dotted gaussian profile represents the sis instrument response for a narrow line observed at a rest - energy of 6.4 kev . the profile constructed from the seyfert-1 sample ( n97 ) is also shown for comparison . the group c profile shows line flux at and above 6.4 kev ( a result which is confirmed by consideration of the fits to the sis@xmath32gis data for each individual source ) probably due to the presence of unresolved iron k@xmath0 lines from a range of ionization - states ( as discussed in paper ii ) . interestingly , the group c profile also shows a red wing . while this is small compared to the peak of this line , it is of the same amplitude ( @xmath33 ) as that observed in seyfert-1 galaxies , and group a and b. we return to this point later . the group b profile is not significantly distinguishable from that observed for seyfert-1 galaxies ( n97 ) . the line peak falls at @xmath34 kev , and the line core is significantly broadened towards lower energies . a very broad wing is also evident , with the broad line component peaking at @xmath35 kev , with width @xmath36 kev . a broad iron line with a red - wing is predicted if the line arises close to the black - hole and is subject to gravitational redshift . the iron k@xmath0 emission observed in seyfert-1 galaxies has been successfully modeled with profiles expected for a line originating in an accretion disk around a black - hole ( e.g. tanaka 1995 , n97 ) , hereafter denoted a disk - line . unless the disk is observed face - on , kinematic ( doppler ) effects cause the blue flux of the line to be enhanced relative to the red - wing , contrary to what is observed . hence the requirement for a face - on geometry when the red wing is the dominant feature observed . in the seyfert-1 case , the iron k@xmath0 line profile was interpreted as originating in a relativistic accretion disk inclined close to face - on ( the mean inclination of the seyfert-1 sample was @xmath37 , for a fit where the emissivity was fixed as a function of radius ; n97 , table 5 ) . the similarity in line profiles suggests a comparable orientation for the inner regions of group b sources , which are all nelgs . the group a profile has a core which is consistent with a narrow gaussian line , centered at 6.4 kev . the equivalent width of the core is @xmath24 ev which could be produced by transmission of the power - law continuum through columns of @xmath38 in the line - of - sight ( fig . 1 ) . however , this profile also has excess flux in the wings of the line , reminiscent of the red - wing observed in group b and in seyfert-1 galaxies . in the light of the group b result , we removed the nelgs from the group a profile ( half the group ) . we found that the seyfert-2 galaxies alone also yield a similar profile with a broad red wing . our finding that group a and b have similar profiles is an interesting and surprising result as unification models predict the inner regions of type-2 seyferts to be observed edge - on . we attempted to fit the iron k@xmath0 line in individual sources , using the disk - line model profile of fabian ( 1989 ) . that model assumes a schwarzschild geometry and computes the line profile but not the line strength . in that model the line emissivity is parameterized by a power - law as a function of radius , @xmath39 . as line and continuum parameters can be correlated , and as the fit depends significantly on the assumptions made , we adopted the same assumptions as those used to fit a large sample of seyfert-1 galaxies ( n97 ) , to allow a meaningful comparison between the two sets of results . in each case we assumed the line originates from a region between 6 and 1000 r@xmath40 , where the gravitational radius , @xmath41 , and @xmath42 is the mass of the black hole . the line emissivity function was assumed to be r@xmath43 . we allow the inclination of the disk relative to the observer to be a free parameter ( defined such that @xmath44 is a face - on disk ) , along with the normalization of the line . we assume the line energy @xmath45 kev in the rest - frame of the host galaxy . the disk - line fits were performed assuming the best continuum models from paper i with all continuum parameters allowed to be free . the continuum models in paper i were derived fitting the 0.6 - 5.0 plus 7 - 10 kev data , before consideration of the iron k regime was included in the model . in each case we also considered the results when an additional ( narrow ) line component was allowed , to represent the iron line produced in the line - of - sight absorber . the seyfert-2 galaxies in group a did not generally yield useful constraints on the inclination of any accretion disk system . this is due to the low signal - to - noise ratio in those sources , and the allowance of a relatively strong narrow - line component in the fits ( up to several hundred ev , which could be produced in the line - of - sight absorber ) . however , fitting a disk - line profile to the bright sources from group b yielded some interesting results . the rest energy of the narrow line component was fixed at 6.4 kev , and the equivalent width was constrained to be between 10 50 ev . this range embraces the values expected from the leahy & creighton ( 1993 ) simulations , but allows a wide range about the predicted value , to account for differences in spectrum for each source , and possible geometric effects . errors quoted are 1 @xmath46 . _ ngc 526a _ the preferred continuum model for ngc 526a is a double power - law with @xmath47 , @xmath48 ( paper i ) , flattening to @xmath49 . adding a disk - line profile to the model yields an improvement @xmath50 over the fit to a narrow gaussian line , with an inclination angle @xmath51 and ew@xmath52 ev . if one assumes instead a single power - law continuum partially - covered by the absorber , the fit yields the same inclination and consistent ew@xmath53 ev for the line . addition of a narrow gaussian component ( ew @xmath54 ev ) did not significantly improve the fit . _ ngc 2110 _ the best - fit continuum from paper i was a @xmath55 continuum partially - covered by ionized gas with @xmath56 , @xmath57 . the disk - line yielded an inclination angle @xmath58 , with ew@xmath59 ev and an improvement @xmath60 over the fit where the iron k@xmath0 line was modelled with a narrow gaussian component . addition of a narrow gaussian component ( @xmath61 ev ) did not significantly affect this fit . _ mcg-5 - 23 - 16 _ the nominal best - fit from paper i was @xmath62 , @xmath63 , steepening to @xmath64 at @xmath65 kev . a disk - line fit to mcg-5 - 23 - 16 yielded a reduction @xmath66 ( f=60 ) compared to the fit where the 5 - 7 kev band is modelled with a series of narrow lines ( paper i ; table 15 ) . the line ew was @xmath67 ev and the disk inclination was @xmath68 . addition of a narrow gaussian component ( @xmath61 ev ) did not significantly affect this fit . weaver ( 1997 ) found a somewhat higher inclination angle in their analysis of mcg-5 - 23 - 16 , but all disk - line results depend strongly on the assumptions made in the fit , which were different in this analysis versus weaver ( 1997 ) . the @xmath69 solution is consistent with the observation of some line flux blueward of the peak ( in addition to the red wing ) . _ ngc 7314 _ the nominal best - fit to ngc 7314 was @xmath70 , @xmath71 , @xmath72 . the addition of a disk - line component yields a good description of the iron k@xmath0 line profile improving the fit by @xmath73 , compared to the narrow gaussian parameterization . the fit yields @xmath74 , @xmath75 ev . addition of a narrow gaussian component ( @xmath54 ev ) did not significantly affect this result . however , fits to this source are very sensitive to the underlying continuum and absorption . fitting the disk - line on top of a single power - law continuum , partially covered by a column @xmath76 produces a solution with @xmath77 , @xmath78 ev . the addition of a narrow gaussian component does not significantly affect this solution . the former fit is nominally statistically preferred ( by @xmath79 ) but obviously the conclusions drawn depend strongly on the approach taken in the spectral fitting , this is the origin of differences in the results obtained here versus yaqoob ( 1996 ) . we now consider possible explanations for the ubiquity of the red wing on the iron k@xmath0 profiles of seyfert galaxies . given the correlation between line and continuum parameters , an inadequate modelling of the continuum and absorption properties might be misleading us about the profile of the iron k@xmath0 line . this is more likely to produce a significant effect in type-2 seyferts than in type-1 , since the former show relatively large spectral modification due to attenuation in the middle of the bandpass . however , it would be a strange coincidence for such a problem to so closely mimic the profile observed in seyfert-1 galaxies . consequently we consider this possibility unlikely to be the cause of the similarity in observed profiles . another possibility is that the type-2 objects have another mechanism for producing a red - wing on the iron k@xmath0 line . the existence of a compton - shoulder has been suggested for sources such as ngc 1068 ( iwasawa 1997 ) . however , the compton shoulder can not easily produce a such a strong wing extending down to such low energies as observed in our profiles . despite the systematic effects observed , we note some variety in the profiles of individual sources in the seyfert type-1 and type-2 samples . we also note that the dark earth and coma cluster do not show these characteristic line profiles in their spectra ( n97 ) . these facts suggest that the observed profiles are not an artifact of any outstanding uncertainties with the calibration . we also consider the possibility that the iron k@xmath0 profile is not a good indicator of the orientation of the inner regions of the agn . if strong relativistic effects bend the light produced close to the central nucleus , then the iron k@xmath0 profiles could be distorted , and our fits to a schwarzchild profile could be misleading as to the orientation of the central source . in particular , the kerr metric allows contributions to line emission from a disk extending inwards to within a few gravitational radii , hence yielding more severely skewed line profiles than produced around a non - rotating black hole ( e.g. laor 1991 ) . however , these line profiles also have a distinct dependance on inclination angle , and as the line profiles of the type-1 and type-2 seyferts are so similar , the preferred orientation of the innermost reprocessor must be the same . in the discussion we investigate the consequences of this conclusion in the context of the unified model . observations at different wavelengths provide probes to different depths into the agn . assuming the circumnuclear absorbing material has a galactic dust - to - gas ratio , then a v band extinction of a@xmath801 mag . corresponds to a column of neutral hydrogen @xmath81 . an extinction a@xmath82 allows the broad line region to be observed . a@xmath83 5 - 8 hides the broad optical lines , still these agn can be identified by direct observation of the paschen series of lines in the infrared regime . extinction in excess of a@xmath84 still allow observations of pa@xmath1 ( which probe material up to a@xmath85 ) , br@xmath2 ( up to a@xmath86 ) and br@xmath0 ( up to a@xmath87 ; see vielleux 1997 ) . when the extinction approaches a@xmath88 then only x - ray and @xmath2-ray observations are able to provide a _ direct _ view of the nucleus , and as extinction increases beyond this level , then even the 10 kev photons are hidden , the material is optically - thick to compton scattering and the central source is only detectable via scattering by material out of the line - of - sight . thus the classification assigned to a source may depend on the wavelength at which it is observed . observations in some bandpasses might not show any signature of an important component of the system . this has lead to some historical confusion as to the nature of some sources . as evident in fig . 1 , the column densities for group b sources are clustered around @xmath89 while group a span @xmath90 . as discussed in paper ii , the column densities derived for the scattered x - ray sources ( group c ) generally do not represent the line - of - sight absorption to the nucleus . group c sources were treated in detail in paper ii and that class is not discussed further in this paper . of the sources considered in this paper , only ngc 7172 is consistent with a single power - law absorbed by neutral material . the spectra of all the other sources in groups a and b show a strong soft x - ray flux in excess of that predicted by extrapolation of the hard x - ray power - law , absorbed by neutral material . a `` soft excess '' can arise when there are unattenuated lines - of - sight to the nucleus , when the absorber is ionized or when a soft emission component is present . of the group a sources ; ngc 1808 , ngc 4507 , ngc 6240 , eso 103-g35 and ngc 7582 have an intrinsic 0.5 2.0 kev luminosity in the soft component ( alone ) @xmath91 , when a correction is made only for the galactic line - of - sight column . ( we note this yields a lower 0.5 2.0 kev luminosity than those listed in tables 8 - 13 of paper i , which were corrected for total absorption . ) consideration of the maximum starburst contribution to the x - ray flux ( paper i ) reveals that only ngc 1808 is consistent with all of that soft flux originating from starburst regions ( tacconi - garman , sternberg & eckart 1996 ) , however , soft x - ray luminosities @xmath91 could be accounted for by the summed contributions of starburst emission , hot gas and binary stars in the host galaxy ( although these can not generally account for the observed hard x - ray luminosities ) . no significant variability is evident in the soft flux or absorbing column in ngc 4507 , ngc 1808 or ngc 6240 ( for which few x - ray observations are available ) , consistent with an active nucleus covered by a uniform absorber with an extended soft x - ray emission component providing the spectral complexity . variations in x - ray absorption have been claimed for eso 103-g35 ( warwick 1988 ) on the basis of two observations which revealed a drop in column @xmath92 in 90 days . the timescale for variability indicated that absorber to be consistent with clouds at a radius coincident with the blr . the observation reveals an absorbing column consistent with the most heavily absorbed spectrum , as does analysis of a later observation ( 26 september 1995 ) , raising the question as to whether a contaminating source could have affected the apparent measurement of a lower column at one of the epochs . the absorption of ngc 7582 also appears lower ( @xmath93 ) than that observed during the observation ( turner & pounds 1989 ) . the nearby bl lac pks2316 - 423 , may have contaminated the spectrum obtained with the collimated medium energy ( me ) instrument . however we would expect any such contamination to have made the column measurement attributed to ngc 7582 to appear lower than the measurement , rather than higher . the nearby clusters abell 1111s and sersic 159 - 03 also posed potential sources of contamination for the observations , and could have yielded an erroneous column measurement . examination of the archives shows that the source and background observations had the me proportional counter offset and with a roll angle designed to minimize the contamination from sersic 159 - 03 . however , given the large number of contaminating sources in the me field , any claim of column variability in ngc 7582 requires confirmation using further observations of the source . in the cases of ngc 526a , ngc 7314 , mcg-5 - 23 - 16 and ngc 2110 in group b , plus ngc 5252 and ic 5063 from group a , the soft emission exceeds that expected from stellar origins by at least an order of magnitude . these luminous soft components could originate from scattered nuclear flux , or leakage of the primary continuum through unattenuated line - of - sight . the history of the soft x - ray flux could , in principle , distinguish between these two . variations in absorbing column could indicate that the absorber had a variable ionization - state , covering fraction or column . analysis of a _ rosat _ hri observation of ngc 5252 ( made in 1995 july ) showed the soft flux to be a factor of several lower than that observed by ( 1994 january ) . as the spectrum suggests the 0.5 - 2.0 kev flux is dominated by scattered or unattenuated nuclear emission this variability favors partial - covering of the absorber or a sub - parsec size for the scattering region . our analysis of the _ rosat _ pspc observation of ic 5063 showed the soft flux to be consistent with that observed by allowing no constraint in that case . x - ray column variability has been claimed for ngc 2110 on timescales of years ( hayashi 1996 ) based upon x - ray observations covering heao-1 ( 1978 ) to ( 1994 ) , although we note that it is difficult to make a conclusive comparison based upon different spectral fits performed across a variety of bandpasses , and this result needs confirmation . ngc 526a showed strong spectral changes , consistent with column variability , on a timescale of years across five observations ( turner & pounds 1989 ) . historic x - ray observations of ngc 7314 and mcg-5 - 23 - 16 show column measurements consistent with no variation . a power - law source partially - covered by neutral material would be expected to show a leaking fraction with the same spectrum as that observed in the hard x - ray regime . simple electron scattering of the hard continuum is also expected to yield the same spectrum in scattered x - rays , as the primary continuum . however , many datasets marginally prefer the double power - law model ( paper i ) with some sources showing a steep soft x - ray component , and others a hardening of the spectrum to higher energies . this indicates some spectral curvature which is not accounted for by the aforementioned models . this could be due to curvature in the underlying continuum . however , partial - covering by ionized material could also yield a difference in soft and hard spectra . this is the preferred model for seyfert 1 galaxies ( reynolds 1997 , g97 ) and for ngc 2110 and ngc 5252 from the type-2 sample . thus we consider the model in more depth in the discussion . in summary , there is only tentative evidence for column variability in the sample sources . the the spectral shapes can generally be described either by a continuum attenuated by a patchy or ionized absorber , or by the presence of a second emission component . x - ray monitoring of one or more well - chosen type-2 seyferts will help clarify the picture . both of our subsamples of type-2 seyferts show an iron k@xmath0 profile which contains a significant red - wing at the same level as that observed in the seyfert-1 sample . this skewed profile implies that type-2 seyferts contain an iron k@xmath0 line component from a reprocessor close to the central black hole . furthermore this reprocessor is implied to be orientated preferably face - on to our line - of - sight . this result is contrary to unification models , which predict the inner regions of type-2 seyfert galaxies are observed preferentially edge - on . this is all particularly interesting in the light of the fact that the host galaxies of type-2 seyferts are also observed preferentially face - on . specifically , kirhakos & steiner ( 1990 ) find the distribution of inclinations to be indistinguishable between type-1 and type-2 hosts when they consider a sample of 288 seyfert galaxies . furthermore , both seyfert type-1 and type-2 samples are observed preferentially face - on compared to a large control sample of normal galaxies . even ngc 1068 and mrk 3 , which appear to be observed in scattered and reflected x - rays , are decidedly face - on ( with b / a=0.89 and 0.72 , respectively ) . however , maiolino & reike ( 1995 ) made a study where the `` intermediate '' seyfert 1.8 1.9 galaxies were treated separately to the classical seyfert 2 galaxies . they found that both seyfert 1 and seyfert 2 galaxies are preferentially face - on , but the `` intermediate '' seyferts are preferentially edge - on . this result inferred that the absorption in the seyfert 1.8 1.9 galaxies might often be due to material in the plane of the host galaxy rather than circumnuclear material . nelgs are `` intermediate '' seyferts on the basis of their optical properties , however , as the mean iron k@xmath0 profiles clearly indicate a preference for a face - on orientation of the inner regions , there can not be many edge - on sources in our sample or the mean iron k@xmath0 profiles would not be so similar to the mean profile for type-1 seyferts . if the putative accretion disk is totally decoupled from the orientation of the host galaxy , then we should observe a variety of inclinations in the iron k@xmath0 profiles , resulting in a mean profile of @xmath94 for the sample . thus the preference for a face - on orientation of the nuclear regions and the host might suggest a link between the angular momentum of the two . this might be expected if the accretion process ultimately feeds on material lying in the plane of the galaxy . if this is true , then it is difficult to understand why the putative obscuring torus , which ( if it is present ) must lie between these two systems , would be decoupled from this preferred orientation . this problem might be avoided if the absorber is a distribution of clouds , and the type-1 or type-2 nature of a source is simply determined by whether or not a cloud lay across the observers line - of - sight . differences in the distribution and composition of the clouds could then account for differences in observed properties . several seyfert galaxies vary back and forth between type-1 and type-2 properties on timescales of years . in at least some sources the observed behaviour has been attributed to variations in the reddening , consistent with clouds of dust moving across the line - of - sight ( e.g. goodrich 1989 ) . the timescales and repetition of this type of behaviour argues against both evolutionary models and orientation of large - scale structures of the absorbing material to explain the difference between seyfert classes , although the `` orientation effect '' of a cloud being in the line of sight is consistent with the observations . in the context of cloud models , another possibility is that all classes of seyfert - type agn show a similar mean iron k@xmath0 profile because the line is produced by reprocessing in a spherically symmetric distribution of material . spherical accretion of clouds of material could produce a line showing the strong gravitational effects observed , but with the same profile observed from each viewing direction . the construction of a mean profile for a sample of edge - on seyferts would test this theory . the intermediate seyfert galaxies might provide an edge - on sample . alternatively , kirhakos & steiner ( 1990 ) find that iras galaxies are observed preferentially edge - on . if the orientation of the host galaxy is associated with that of the inner accretion disk / reprocessing region then iras galaxies should show an iron k@xmath0 profile indicative of edge - on systems , although such a line will be broad , and possibly difficult to detect . we note that the edge - on galaxy ngc 2992 ( b / a=0.27 ) shows an iron k@xmath0 line profile with a blue wing ( paper i , paper 2 ) . weaver ( 1996 ) suggest that the large equivalent width of the iron k@xmath0 line in ngc 2992 is due to a lag of several years between the direct and reprocessed components . the data are consistent with a narrow component of high equivalent width , as suggested by weaver ( 1996 ) , plus a disk - line component from an edge - on accretion disk , which would support a connection between the orientation of the host galaxy and the inner regime . however , the data are also consistent with contributions to the iron k@xmath0 line profile from highly ionized species of iron . a search for variability could distinguish between these possibilities , as the occurrence of rapid variability in the blue wing would favor the disk - line scenario . the existence of `` ionization cones '' in a dozen or so seyfert-2 galaxies ( e.g. pogge 1988 , tadhunter and tsvetanov 1989 ) poses an interesting problem if we assume the inner regions are viewed with a face - on orientation . such cones should only be seen in systems observed at a different angle to the escape direction of the radiation . however , many ionization cones are one - sided . this would be expected if we observe close to the direction of the collimated radiation , as one side of the cone will escape in a direction almost directly away from us . in this case the cone will be foreshortened and the back - side will be heavily absorbed by any intervening galactic material . we note that one of the best examples of a two - sided cone is in ngc 5252 , in which the host galaxy is observed at an intermediate orientation ( b / a=0.37 ) . the fact that observations of the iron k@xmath0 line profiles imply a near face - on geometry for the inner regions of both seyfert types 1 and 2 necessitates a discussion of ways in which factors other than the orientation of the absorbing material can explain the observed differences between the two classes . the obscuration of the nlr is similar in optically selected seyfert-1 and seyfert-2 galaxies , indicating that the obscuration responsible for the differences between these classes is located within the nlr , or out of the line - of - sight to the nlr ( lawrence & elvis 1982 , de zotti & gaskell 1985 ) . in nelgs the x - ray absorbing columns are @xmath95 ( ngc 7314 to eso 103-g35 ) while the optical lines imply an attenuation @xmath96 ( assuming a galactic gas - to - dust ratio ) . this implies that the x - ray absorber lies within the blr , or is highly ionized , otherwise it would produce stronger suppression of the blr than that which is observed . in seyfert-2 galaxies x - ray column densities are @xmath97 or greater . extinction of the broad lines occurs for columns in excess of @xmath98 . on this basis , the x - ray absorption in seyfert-2 galaxies can only be said to exceed that required to fully suppress the broad optical lines , yielding no constraint on the location of the absorber . granato , danese and franceschini ( 1997 ) compare the infrared broad - band spectra and spectroscopy results with x - ray spectra of a sample of type-2 seyferts ( smith & done 1996 ) . for convenience we define the infrared extinction as @xmath99 and the x - ray extinction as @xmath100 where @xmath101 , @xmath102 is the observed flux , and @xmath103 is the intrinsic flux . granato find the infrared extinction , @xmath99 is approximately consistent with the x - ray extinction , @xmath100 , for columns @xmath104 , while @xmath105 for @xmath106 , and @xmath107 for @xmath108 . thus , their finding indicates that where there is a large x - ray absorption , the material likely sits predominantly within the dust sublimation radius . evidence for an ionized absorber covering ngc 2110 and ngc 5252 makes it interesting to examine the general results from such a model , particularly in the light of the strong evidence for ionized absorbers in most seyfert-1 galaxies ( e.g. reynolds 1997 , g97 ) . to examine the unification of type-1 and type-2 seyfert galaxies we examined the distribution of column , ionization parameter and luminosity for both types . we took the seyfert-1 results from g97 . those sources were analyzed in the same way as our sample of type-2 seyferts , allowing a direct comparison to be made . 3 shows @xmath109 and @xmath5 , the ionization - state and column of ionized material versus ( unabsorbed ) x - ray luminosity adapted from a figure in g97 ( group b sources are marked with stars , group a with circles , seyfert-1 galaxies with open squares ) . this comparison demonstrates that while the mean column appears to be higher for type-2 , some seyfert-1 galaxies are also consistent with having large columns of circumnuclear material ( which is so highly ionized that it does not provide much opacity below @xmath110 kev ) . examination of fig . 3 shows that there is a large overlap in intrinsic ( absorption - corrected ) luminosity of the type-2 and type-1 sample sources , although the mean intrinsic luminosity of the type-2 seyferts is lower . for gas of the same density and distance from the ionizing source , then the ionization parameter is , by definition , proportional to the ionizing luminosity . the seyfert 1 sample appears consistent with @xmath111 , and hence the circumnuclear gas is consistent with similar density and location across the type-1 sample . however , type-2 seyferts are clearly inconsistent with that picture , most notably due to several sources absorbed by material of very low ionization - state . we also note that overlaying three quasars on this plot ( marked with solid squares ; from george 1997b ) indicates that there is no simple proportionality between @xmath109 and @xmath112 . this indicates that the absorbing material in type-2 seyferts exists under significantly different conditions than that in type-1 galaxies . differences in gas density , geometry and location between individual objects could cause the observed effects . type-1 seyferts do not show such pronounced suppression of the broad optical lines as type-2 sources with the same nuclear ( x - ray ) luminosity . this indicates systematic differences in the location of the dust sublimation radius between classes of agn . reverberation studies have indicated blr sizes @xmath113pc ( peterson 1993 ; netzer 1990 ) , although the blr probably extends over an order of magnitude in radius ( korista 1995 ) . for bolometric luminosities @xmath114 ( appropriate for our sample sources ) we estimate the blr to be @xmath115 cm ( @xmath3 a few light days ) . barvainis ( 1987 ) shows that the dust sublimation radius , @xmath116 pc , where @xmath117 is the uv luminosity in units of @xmath118 and @xmath119 is the grain sublimation temperature in units of 1500 k. order - of - magnitude estimates show the dust sublimation radius to lie close to or just outside of the blr . barvainis also notes that the highest sublimation temperature occurs for dust composed of graphite grains , and is @xmath120 k. dust containing silicate grains would have a lower temperature and hence a larger sublimation radius . evidently , different mixtures of dust grains will allow the sublimation radius to change by up to a factors of a few . the size of the dust grains yields an even bigger effect . laor & draine ( 1993 ) show that graphite grains of size @xmath121 can exist almost an order of magnitude closer to the nucleus than grains of size @xmath122 . thus differences in the distribution of grain sizes could allow significant differences in the location of @xmath123 relative to the blr in different sources , @xmath123 will be closer to the nuclei in those sources with the largest dust grains . however , as also shown by laor & draine ( 1993 ) some distributions of grain size yield dust which is ineffective at reddening , i.e. the effective extinction can have a flat dependance over a large frequency range . the data require that for seyfert-2 galaxies , the blr is deeply situated well within @xmath123 , for nelgs the blr is probably very close to @xmath123 because the broad lines suffer some obscuration . thus differences in dust content and in the distribution of grain sizes could be the key to unification of type-1 and type-2 seyfert galaxies . in fact konigl & kartje ( 1994 ) suggested a dusty wind model to unify agn , with partially - ionized gas in the inner regions of the wind accounting for the bulk of x - ray absorption seen in face - on sources . spectral variability in the optical and x - ray regimes could be driven by changes in the nuclear flux , which will move the effective radii of ionization and sublimation . differences in the distribution of circumnuclear material and in the total mass of gas surrounding the nuclei may also play a role . a model where clouds of optically thick material cause the x - ray obscuration can explain many observed properties ( guilbert & rees 1988 ) . the location , size and column density of the clouds can determine whether or not column variability or a leaking fraction of nuclear flux are observed . compton - reflection can occur from the surface of the blobs ( nandra & george 1994 ) while scattering of nuclear radiation can occur from a region of hot gas , so long as there are some unattenuated lines of sight by which to view this . collimation of radiation ( to produce ionization cones ) could also occur , if the blobs are distributed around a preferred plane . interestingly , maiolino ( 1995 ) find differences in infrared colors indicating that the host galaxies of type-2 nuclei have significantly higher levels of starburst activity than type-1 nuclei . this starburst activity occurs close to the nucleus , and as starburst regions contain large amounts of dusty gas , perhaps the presence of these starburst regions results in a different gas and dust content in type-2 seyfert galaxies . thus we concur with the suggestion of maiolino that consideration of the starburst regions seems likely to provide an important component of unification models . differences in dust content of the circumnuclear gas might also be related to the apparent preferential occurrence of masers in type-2 seyferts . the relationship between a high x - ray absorbing column and the existence of water masers in type-2 seyferts ( braatz , wilson & henkel 1997 ) indicates that the material in which the maser amplification occurs is related to significant x - ray opacity . _ if _ the molecular gas is in a well - defined plane then masers will be preferentially observed in sources viewed within a few degrees of that plane . the sources in our original sample which are known to have masers are ngc 1068 , ngc 4945 and eso103-g35 ( braatz , wilson & henkel 1997 ) . ngc 1068 and ngc 4945 are so highly absorbed that the x - ray spectra are seen only in reprocessed photons , and the data are consistent with the high degree of opacity expected from an edge - on masing disk . this again raises the interesting question as to the relationship of any preferred planes of material on size scales of the galaxy , the torus / masing disk or clouds , and the nucleus . the fact that masers are only observed in type-2 seyferts might indicate that conditions in those agn result in a layer of shielded masing material . kartje , konigl & elitzur ( 1996 ) propose that maser flux arises from clouds in a magnetized accretion disk and that these clouds are shielded from the nuclear flux by gas and dust . they suggest the inner edge - of the maser is dictated by the dust sublimation radius , where effective shielding begins . examination of the iron k@xmath0 line from a sample of nelgs and seyfert-2 galaxies shows profiles similar to those observed in seyfert-1 galaxies , and indicative of an origin in an accretion disk orientated face - on . this result is contrary to unification models , which predict the inner regions of seyfert-2 galaxies to be observed edge - on . the preference for a face - on orientation of both the accretion disk and host galaxy of type-1 and type-2 seyfert galaxies poses some questions as to the orientation and geometry of gas comprising the circumnuclear absorbing material . if the absorber is composed of clouds then differences in gas density , size of dust grains and dust composition and the distribution of the clouds seem a likely explanation of the observed differences in the x - ray and optical absorption properties of seyfert-1 galaxies , seyfert-2 galaxies and nelgs . these results indicate that further refinement is required for unified models for agn . we are grateful to team for their operation of the satellite and to keith gendreau , tahir yaqoob and ski antonucci for useful comments . this research has made use of the nasa / ipac extragalactic database , which is operated by the jet propulsion laboratory , caltech , under contract with nasa ; of the simbad database , operated at cds , strasbourg , france ; and data obtained through the heasarc on - line service , provided by nasa / gsfc . we acknowledge the financial support of universities space research association ( img , tjt ) and the national research council ( kn ) . antonucci , r. , 1993 , ara&a , 31 , 473 antonucci , r. , miller , j.s . , 1985 , , 297 , 621 awaki , h. , koyama , k. , inoue , h. , halpern , j.p . , 1991 , 43 , 195 barvainis , r. , 1987 , 320 , 537 . braatz , j.a . , wilson , a. , gezari , d.y . , varosi , f. , & beichman , c.a . , 1993 , 409 , l5 braatz , j.a . , wilson , a. , henkel , 1997 , 110 , 321 burke , b.e . , mountain , r.w . , daniels , p.j . , dolat , v.s . , 1994 , ieee trans . 41 , 375 de zotti , g. , gaskell , c.m . , 1985 , a&a , 147 , 1 hayashi , i. , koyama , k. , awaki , h. , yamauchi , s. , ueno , s. , 1996 , 48 , 219 fabian , a.c . , rees , m.j . , stella , l. , white , n.e . 1989 , 238 , 729 george , i.m . , turner , t.j . , netzer , h. , nandra , k. , mushotzky , r.f . , yaqoob , t. , 1997 , submitted ( g97 ) george , i.m . , nandra , k. , netzer , h. , laor , a. , turner , t.j . , mushotzky , r.f . , fiore , f. , 1997b , in prep granato , l.g . , danese , l. , franceschini , a. 1997 , 486 , 147 goodrich , r. , 1989 , 340 , 190 greenhill , l.j . , gwinn , c.r . , antonucci , r. , barvainis , r. , 1996 , 472 , l21 guilbert , p.w . , rees , m.,j . , 1988 , 233 , 475 haniff , c.a . , wilson , a.s . , ward , m.j . , 1988 , , 334 , 104 hayashi , i. , koyama , k. , awaki , h. , yamauchi , s. , ueno , s. , 1996 , 48 , 219 iwasawa , k. , fabian , a.c . & matt , g. , 1997 , 289 , 443 kirhakos , s.d . , steiner , j.e . , 1990 , a&a , 99 , 1435 konigl , a. & kartje , j.f . , 1994 , 434 , 446 konigl , a. & kartje , j.f . , elitzur , m. , 1996 , baas , 188 , 0402 korista _ et al_. , 1995 , 97 , 285 laor , a. , 1991 , 376 , 90 laor , a. , draine , b.t . , 1993 , 402 , 441 lawrence , a. , elvis , m.e . , 1982 , 256 , 410 leahy , d.a . & creighton , j. 1993 , 263 , 314 marshall , f.e . , boldt , e.a . , holt , s.s . , mushotzky , r.f . , rothschild , r.e . , serlemitsos , p.j . , pravdo , s.h . , 1979 , 40 , 657 maiolino , r. , rieke , g.h . , 1995 , 454 , 95 maiolino , r. , ruiz , m. , rieke , g.h . , keller , l.d . , 1995 , 446 , 561 makishima , et al 1996 , , 48 , 171 miller , j.s . , goodrich , r.w . , 1990 , 355 , 456 mushotzky , r.f . , 1984 , adv . space research , 3 , 10 nandra , k. , george , i.m . , 1994 , 267 , 974 nandra , k. , pounds , k.a . , 1994 , , 268 . 405 nandra , k. , george , i.m . , mushotzky , r.f . , turner , t.j . , yaqoob , t. , 1997 ( n97 ) , , 477 , 602 netzer , h. , 1990 , in active galactic nuclei ( saas - fee , advanced course 20 ) , ed . courvoisier , & m. mayor , ( berlin : springer ) , 57 netzer , h. , 1996 , 473 , 781 ohashi , t. , et al . , 1996 , , 48 , 157 peterson , b. , 1993 , 105 , 247 pogge , r.w . , 1988 , , 328 , 519 ptak , a. , serlemitsos , p , yaqoob , t. , mushotzky , r.f . , tsuru , t. 1997 , aj , 113 , 1286 reynolds , c. , 1997 , 286 , 513 serlemitsos , p.j . , et al . , 1995 , , 47 , 105 shuder , j.m . , 1980 , , 240 , 32 smith , d. , done , c. , 1996 , , 280 , 355 tacconi - garman , l.e . , sternberg , a. , eckart , a. , 1996 , 112 , 918 tadhunter , c. , tsvetanov , z. , 1989 , nature , 341 , 422 tanaka , y. , et al . 1995 , nature , 375 , 659 . tran , h.d . , miller , j.s . , kay , l.e . , 1992 , , 397 , 452 turner , t.j . , pounds , k.a . , 1989 , , 240 , 833 turner , t.j . , george , i.m . , nandra , k. & mushotzky , r.m . , 1997a ( paper i ) , in press turner , t.j . , george , i.m . , nandra , k. & mushotzky , r.m . , 1997b ( paper ii ) , in press veilleux , s. , goodrich , r.w . , & hill , g.j . , 1997 , 477 , 631 veron , p. , lindblad , p.o . , zuiderwijk , e. , j. , adam , g. , veron , m.p . , 1980 , a&a , 87 , 245 ward , m.j . , wilson , a.s . , penston , m.v . , elvis , m. , maccacaro , t. , tritton , k.p . , 1978 , , 223 , 788 warwick , r. , pounds , k.a . , turner , t.j . , 1988 , 231 , 1145 weaver , k.a . , nousek , j. , yaqoob , t. , mushotzky , r.f . , makino , f. , otani , c. , 1996a , 458 , 160 weaver , k.a . , , yaqoob , t. , mushotzky , r.f . , nousek , j. , hayashi , i. , koyama , k. , 1997 , 474 , 675 yaqoob , t. , serlemitsos , p.j . , turner , t.j . , george , i.m . , nandra , k. , 1996 , 470 , 27 fig 1 - equivalent width of the narrow 6.4 kev line versus absorbing column ( from paper i ) for the full sample ( excluding ngc 1667 which had a line equivalent width of zero , and ngc 5695 which was not detected ) . the dot - dash line shows the line equivalent width expected from a uniform shell of material encompassing the continuum source ( leahy & creighton 1993 ) . the dashed line shows the predicted equivalent width from reflection as a function of , assuming that only the power - law is absorbed , but that the reflection component remains unchanged . data points are annotated with an abbreviation of the source name . group a sources are marked as circles , group b sources as stars and group c sources as squares . a few of the lowest signal - to - noise datasets have not been classified . fig 2 - the average data / model ratios showing the iron k@xmath0 line emission for each group , and that of the seyfert-1 sample from n97 , for comparison . each dataset was corrected to the rest - frame energy before the objects were combined and so the x - axis shows the rest - frame energy . the dotted gaussian profile represents the sis instrument response for a narrow line observed at a rest - energy of 6.4 kev . fig 3 - the absorption - corrected 210 kev luminosity versus ionization - state and column of the circumnuclear gas , @xmath109 and @xmath5 . the open squares represent the seyfert-1 sample of george ( 1997 ) , the open circles are the group a sources , stars are group b and filled squares represent three quasars ( see 4.2 ) . mcg-01 - 01 - 043 & 1.0 & 00 10 03.5 & -04 42 18&0.0300 & s2 & 3.27@xmath124 ngc 526a ( b)&0.70&01 23 55.1&-35 04 04 & 0.0192&nelg@xmath125&2.33@xmath124ngc 1068 ( c ) & 0.89 & 02 42 40.8&-00 00 47&0.0038&s2@xmath126 & 2.93@xmath127 ngc 1667 & 0.83 & 04 48 37.2 & -06 19 12 & 0.0152 & s2 & 5.46@xmath127e0449 - 184 & & 04 51 38.8 & -18 18 55 & 0.338 & s2 & 3.73@xmath127 ngc 1808 ( a ) & 0.60 & 05 07 42.3 & -37 30 46 & 0.0033 & sb / s2 & 2.42@xmath127 ngc 2110 ( b ) & 0.79 & 05 52 11.4 & -07 27 22 & 0.0076 & nelg & 18.60@xmath124 mkn 3 ( c ) & 0.72 & 06 15 36.3 & 71 02 15 & 0.0135 & s2@xmath125&8.70@xmath127 ngc 2992 ( c ) & 0.27 & 09 45 41.9&-14 19 35&0.0077&nelg@xmath125&5.56@xmath124 mcg-5 - 23 - 16 ( b)&0.45&09 47 40.2&-30 56 54&0.0083&nelg@xmath125 & 8.82@xmath127 ngc 4507 ( a ) & 0.80 & 12 35 36.5 & -39 54 31 & 0.0118 & s2 & 7.04@xmath127 ngc 4945 ( c)&0.13 & 13 05 26.2 & -49 28 16&0.0019 & s2@xmath128 & ngc 4968 & 0.47 & 13 07 06 & -23 40 43 & 0.0100 & s2 & 8.37@xmath127 ngc 5135 & 0.80 & 13 25 44 & -29 50 01 & 0.0137 & s2 & 4.66@xmath127 ngc 5252 ( a)&0.37 & 13 38 15.9 & 04 32 33&0.0230&s1.9&1.97@xmath129 mkn 273 ( c ) & 0.27 & 13 44 42.1 & 55 53 13 & 0.0378 & s2 & 1.02@xmath124 mkn 463e ( c ) & 0.57 & 13 56 02.9 & 18 22 19 & 0.0500 & s2@xmath125&2.08@xmath127 ngc 5695 & 0.73 & 14 37 22.0 & 36 34 04 & 0.0141 & s2 & 1.01@xmath127 ngc 6251 & 0.97 & 16 32 31.9 & 82 32 17 & 0.0230 & s2 & 5.48@xmath127 ngc 6240 - 49 ( a ) & 0.43 & 16 52 59.3 & 02 23 59 & 0.0245&s2 & 5.45@xmath127 eso 103-g35 ( a ) & 0.51 & 18 38 20.2 & -65 25 42 & 0.0133&s2/nelg@xmath128&ic 5063 ( a)&0.67&20 52 02.9 & -57 04 14 & 0.0113 & s2@xmath130 & ngc 7172 ( a)&0.42 & 22 02 02.1 & -31 52 12&0.0086&s2/nelg@xmath125&1.63@xmath124 ngc 7314 ( b)&0.43&22 35 45.7 & -26 03 03 & 0.0047 & s1.9/nelg & 1.45@xmath124 ngc 7582 ( a ) & 0.46&23 18 23.2 & -42 22 11&0.0053&s2/nelg@xmath131&1.47@xmath124

we calculate the lateral casimir force between corrugated parallel plates , described by-function potentials , interacting through a scalar field , using the multiple scattering formalism . the contributions to the casimir energy due to uncorrugated parallel plates is treated as a background from the outset . we derive the leading- and next - to - leading - order contribution to the lateral casimir force for the case when the corrugation amplitudes are small in comparison to corrugation wavelengths . we present explicit results in terms of finite integrals for the case of the dirichlet limit , and exact results for the weak - coupling limit , for the leading- and next - to - leading - orders . the correction due to the next - to - leading contribution is significant . in the weak coupling limit we calculate the lateral casimir force exactly in terms of a single integral which we evaluate numerically . exact results for the case of the weak limit allows us to estimate the error in the perturbative results . we show that the error in the lateral casimir force , in the weak coupling limit , when the next - to - leading order contribution is included is remarkably low when the corrugation amplitudes are small in comparison to corrugation wavelengths . we expect similar conclusions to hold for the dirichlet case . the analogous calculation for the electromagnetic case should reduce the theoretical error sufficiently for comparison with the experiments .

0805.2776

the casimir force , as exhibited between neutral metallic parallel plates , was discovered theoretically in 1948 @xcite . the casimir torque between asymmetric materials was first studied in 1973 @xcite . recently , theoretical study of the lateral casimir force between corrugated parallel plates was pioneered and developed by the mit group in @xcite . in particular , in @xcite , the authors evaluated analytic expressions for the lateral casimir force , to the leading order , between two corrugated parallel plates perturbatively . experimentally , the casimir interaction between corrugated surfaces was explored during the same period of time by roy and mohideen in @xcite . this experiment measured the lateral casimir force between a plate , with small sinusoidal corrugations , and a large sphere with identical corrugations . the motivation was to study the nontrivial boundary dependence in the casimir force . the experimental parameters in our notation are ( see figure [ corru ] ) : @xmath1 , @xmath2 , and @xmath3 , where @xmath4 is the height of the corrugations , @xmath5 is the wavelength of the corrugations , and @xmath6 is the mean distance between the plates . the corresponding dimensionless quantities are : @xmath7 , @xmath8 , and @xmath9 , where @xmath10 is the wavenumber related to the spatial wavelength of the corrugations . experimental data was analyzed based on the theoretical results obtained from the proximity force approximation ( pfa ) , and has been presented in @xcite . the validity of the pfa in the above analysis has been the topic of a recent debate and controversy , see @xcite . theoretical results based on perturbative approximations as done in @xcite do not settle the issue because the error keeping only the leading order may be high . it is generally believed that the next - to - leading - order calculation will be able to throw light on the issue . we carry out this calculation for the case of scalar fields . the analogous calculation for the electromagnetic case should now be straightforward . this paper in principle is an offshoot of @xcite where we shall deal with corrugated cylinders to study non - contact gears . while evaluating the leading order for the case of corrugated cylinders it was noticed that it would be possible to extend the calculation to the next - to - leading order . this led to the study in the present paper . in this installment we present the next - to - leading - order calculation for the case of corrugated parallel plates . the leading order calculation for the corrugated cylinders , which in itself is a significant result , will form the sequel @xcite of this paper . the next - to - leading - order calculation for the corrugated cylinders is in progress . in this section we shall describe the formalism and derive the key formula used for calculating the casimir energy . this has been done in various papers before , ( see @xcite , @xcite , @xcite , and references therein ) . we hope our derivation using schwinger s quantum action principle techniques will be illuminating . in an earlier paper @xcite describing the multiple scattering formalism it was mentioned that the use of the scattering matrix , @xmath11 , was equivalent to using the full green s function , and required the same computational effort . as a justification of this comment we exclusively use the full green s function in this article . let us consider a scalar field , @xmath12 , interacting with a scalar background potential , @xmath13 , described by the lagrangian density @xmath14 in terms of the source function , @xmath15 , corresponding to the scalar field , we write the action for this description to be @xmath16 = \int d^4x \big [ k(x)\phi(x ) + { \cal l}(\phi(x ) ) \big ] . \label{wpk}\ ] ] the vacuum to vacuum persistence amplitude , @xmath17 = \langle 0_+|0_- \rangle^k,\ ] ] which generates the amplitudes for all the physical processes , satisfies schwinger s quantum action principle , @xmath18 = i\,\langle 0_+|\,\delta w[\phi;k ] \,|0_- \rangle^k . \label{qap}\ ] ] our immediate task will be to get a formal solution for the vacuum amplitude , @xmath19 $ ] , in the form @xmath17 = e^{i \,w[k ] } , \label{z = eiw}\ ] ] where @xmath20 $ ] , which is not an action anymore , is dependent only on the source function . note that the action , @xmath21 $ ] in eq . , which satisfied the action principle , was described in terms of both the ( operator ) scalar field and the source function . variation with respect to the source function in the quantum action principle in eq . allows us to write @xmath22 } = \frac{1}{z[k ] } \frac{1}{i } \frac{\delta z[k]}{\delta k(x ) } , \label{eff - field}\ ] ] where the redefined scalar field , on the left of the above expression , is an effective field . this can be used to replace operator fields with functional derivatives with respect to the sources . variation with respect to the scalar field in eq . gives us @xmath23 \ , \frac{1}{i } \frac{\delta z[k]}{\delta k(x ) } = k(x ) z[k ] , \label{ginz = kz}\ ] ] which can be immediately inverted and written in the form , after using eq . , @xmath24 } \frac{1}{i } \frac{\delta z[k]}{\delta k(x ) } = \int d^4 x^\prime g(x , x^\prime ) k(x^\prime ) , \label{zinz = gz}\ ] ] where we defined the inverse of the differential operator , the green s function , as @xmath23 \ , g(x , x^\prime ) = \delta^{(4 ) } ( x - x^\prime ) . \label{green}\ ] ] the solution to eq . is a gaussian in the functional sense , and when written in the desired form in eq . , lets us identify @xmath25 = q[v ] + \frac{1}{2 } \int d^4 x \int d^4 x^\prime k(x ) g(x , x^\prime ) k(x^\prime),\ ] ] where @xmath26 $ ] is a functional of the background potential alone . for the case when the background potential is switched off , described by @xmath27 , we have @xmath28 = \text{exp}(i w_0[k])$ ] , where @xmath29 = q[0 ] + \frac{1}{2 } \int d^4 x \int d^4 x^\prime k(x ) g_0(x , x^\prime ) k(x^\prime ) , \label{free - act}\ ] ] where , @xmath30 is the corresponding green s function in eq . for the case when @xmath27 . now , in the absence of a source function the vacuum should not decay , which amounts to the statement @xmath31 . this implies that @xmath32 = 0 $ ] which when used in eq . lets us conclude that @xmath33 = 0 $ ] . variation with respect to the background potential in eq . yields @xmath18 = - \frac{i}{2 } \int d^4x \ , \delta v(x ) \,\langle 0_+| \,\phi(x ) \phi(x)\ , |0_- \rangle,\ ] ] where we can replace the operator fields with functional derivatives and then write the formal solution as @xmath17 = e^{- \frac{i}{2 } \int d^4x \ , v(x ) \frac{1}{i } \frac{\delta}{\delta k(x ) } \frac{1}{i } \frac{\delta}{\delta k(x ) } } e^{\frac{i}{2 } \int d^4 x \int d^4 x^\prime k(x ) g_0(x , x^\prime ) k(x^\prime)}.\ ] ] using the standard identity , we have @xmath25 = - \frac{i}{2 } \,\text{tr } \,\text{ln } \,g g_0^{-1 } + \frac{1}{2 } \int d^4 x \int d^4 x^\prime k(x ) g(x , x^\prime ) k(x^\prime ) , \label{action}\ ] ] which uses @xmath34 . using eq . in eq . we observe that in the presence of a background the vacuum to vacuum transition amplitude is not unity , instead it evaluates to @xmath35 } = e^{\frac{1}{2 } \,\text{tr } \,\text{ln } \,g g_0^{-1}}. \label{z = iw0}\ ] ] for the case when the process under investigation is time independent , denoting @xmath36 to be the time associated with the physical process , we can evolve the vacuum state using the hamiltonian of the system and thus conclude @xmath37 where we assumed the vacuum to be an eigenstate of the hamiltonian . comparing the two forms , eq . , we thus identify the energy of the vacuum in the presence of a background to be @xmath38 which serves as the central formula for calculating the casimir energy . further , for the case of a time independent situation , making a fourier transformation in the time variable and using translational independence in time , we can formally write @xmath39 + \frac{i}{2 } \int \frac{d\omega}{2\pi } \,\text{tr } \,\text{ln } \big [ - \omega^2 - \nabla^2 \big]\ ] ] where we used eq . to define the fourier transformed green s function in the form @xmath40 $ ] . integrating by parts and throwing away the boundary terms we derive @xmath41 which is an alternate expression for evaluating the casimir energy . this latter form has been used in studies @xcite related to surface divergences which requires the evaluation of the energy density rather than the total energy . we consider two semitransparent , corrugated plates , parallel to the @xmath42 plane , described by the potentials , @xmath43 where @xmath44 , are labels that identify the individual plates , and we define @xmath45 . the functions @xmath46 are designed to describe the corrugations associated with the individual plates . let us define the function @xmath47 which measures the relative corrugations between the two plates . we shall define the corrugations @xmath46 so that the mean of the relative corrugations is @xmath6 . in general we require @xmath48 , which is a restriction on the corrugations . translational invariance is assumed in the @xmath49 direction . the change in energy due to the change in the mean distance between the plates , @xmath6 , leads to the conventional casimir force which points in the direction perpendicular to the surface of the plates , and is expressed as @xmath50 where @xmath51 is the total casimir energy associated with the corrugated plates including the divergent contributions associated with the single plates . the divergent contributions being independent of @xmath6 , do not contribute to the casimir force . when the plates have corrugations on them we expect to have a change in the total energy due to a shift of one of the plates parallel ( lateral ) to the other plate . the force corresponding to the change in the total energy due to this shift acts in the lateral direction and is called the lateral casimir force . the shift is mathematically described by a translation in the @xmath52-coordinate , @xmath53 , which corresponds to a lateral shift of one plate with respect to the other . the lateral force is expressed as @xmath54 where @xmath55 is the measure of the translational shift . we note that there will be no lateral force between the plates if the corrugations are switched off by setting @xmath56 , @xmath57 and @xmath58 . the physical quantities associated with this configuration will thus act as a background , and a reference , and we shall find it convenient to denote them by the superscript @xmath59 to mean zeroth order . the potential for the background will thus be described by @xmath60 which has no @xmath52 dependence . the total casimir energy associated with the background , due to the two uncorrugated plates , will be denoted as @xmath61 , which will include the divergent contributions associated with the single plates . this background energy will be independent of the displacement @xmath55 due to the @xmath52 independence of this configuration . thus , we can conclude that @xmath62 we further note that there will be no lateral force between the plates if either one of the plates have their corrugations switched off by setting @xmath56 , @xmath57 or @xmath58 . the casimir energy associated with this configuration can be written as @xmath63 where @xmath64 is the additional contribution to the energy with respect to the background energy due to the presence of the corrugations on one of the plates . throughout this article we shall use @xmath65 to mean the deviation from the background . for example , we will have @xmath66 . using the argument of @xmath52 independence , or translational symmetry , we can again conclude that @xmath67 in light of the above observations we are led to write the total casimir energy , for the case when both the plates have their corrugations switched on , in the form @xmath68 where @xmath69 is the contribution to the total energy due to the interaction between the corrugations in the plates . only this part of the total energy contributes to the lateral casimir force . thus , we conclude that @xmath70 our central problem will be to evaluate @xmath69 using eq . . using the central formula derived in the multiple scattering formalism in eq to evaluate eq . we have @xmath71 where @xmath72 is the free green s function introduced in eq . . note that @xmath72 cancels in the above expression and the reference is now with respect to the uncorrugated plates . the green s function @xmath73 satisfies eq . with potentials described by eq . , which in symbolic notation reads @xmath74 g = 1,\ ] ] and the corresponding green s function associated with the background satisfies the differential equation , @xmath75 g^{(0 ) } = 1 . \label{g0eqn}\ ] ] the above two equations can be used to deduce @xmath76^{-1}.\ ] ] next , one makes the observation that the above expression can be rewritten in the form @xmath77^{-1 } g_1 { g^{(0)}}^{-1 } , \label{gg0 - 1}\ ] ] where @xmath78 ( @xmath44 ) are the green s functions for the parallel plates when only one of the plates has corrugations on it . the differential equations for @xmath78 s are @xmath79g_i = 1,\ ] ] which together with eq . can be used to deduce @xmath80 using eq . in eq . we we immediately obtain eq . . the last term in eq . , @xmath69 , which is the only term that contributes to the lateral force between the two corrugated plates , can be read out from eq . to be given by the expression @xmath81 . \label{de12}\ ] ] we could have written this down immediately , but we hope it was instructive to go through the steps because it clarifies our notation , which would anyhow have required us to write many of the above equations , and also to emphasize that the configuration due to parallel plates is treated as a background from the outset . formally expanding the logarithm in the above expression and using eq . to expand @xmath82 in terms of @xmath83 we can write @xmath84^m . \label{formal - series}\ ] ] our potentials in eq . can be formally expanded in powers of @xmath85 in the form @xmath86^n}{n ! } \frac{\partial^n}{\partial z^n } v_i^{(0)}(z ) = \lambda_i \big[e^{- h_i(y ) \frac{\partial}{\partial z}}-1\big]\delta ( z - a_i ) , \label{pot - series}\ ] ] so we can further write the series expansion as @xmath87\delta_1 \right\}^{n_1 } \left\ { g^{(0 ) } \big[e^{- h_2 \partial } - 1\big]\delta_2 \right\}^{n_2 } \bigg]^m,\ ] ] where we again use symbolic notation and suppress the variable dependence in @xmath85 , @xmath88 , and the delta functions . in this section we shall derive the leading order contribution to the casimir energy when the corrugation amplitude is small in comparison to the corrugation wavelength . this has been evaluated for the dirichlet scalar case and the electromagnetic case in @xcite . we obtain the result for scalar @xmath0-function potentials as a warm up exercise in preparation for the calculation for the next - to - leading order . in this section we shall illustrate our methodology which will be further used in the higher order calculation . for the particular case when the corrugations can be treated as small perturbations we can approximate the potentials by keeping a few terms in the expansion in eq . where we use the superscripts @xmath89 to represent the @xmath90-th order perturbation in a quantity . thus , to the leading order the interaction energy of the corrugations in eq . takes the form @xmath91 . \label{de12 - 2}\ ] ] we observe that the potentials in the @xmath90-th order are given by derivatives acting on @xmath0-functions in eq . which can be transferred over to the green s functions after integration by parts . the background green s function , which is a solution to eq . , can be written as @xmath92 where @xmath93 , with @xmath94 , where @xmath95 and @xmath96 are the fourier variables corresponding to the space - time coordinates @xmath97 and @xmath98 respectively . the reduced green s function , @xmath99 , satisfies the equation @xmath100 g^{(0)}(z , z^\prime;\kappa ) = \delta ( z - z^\prime ) . \label{g0zzp}\ ] ] in terms of the reduced green s function defined above we can write eq . as @xmath101 where @xmath102 is a large length in the @xmath49 direction and where @xmath103 are the fourier transforms of the functions @xmath46 , which describe the corrugations on the parallel plates , @xmath104 the kernel @xmath105 in eq . is suitably expressed in the form @xmath106 where we have switched to imaginary frequencies by a euclidean rotation , @xmath107 , and defined @xmath108 \bigg|_{\bar{z}=a_1 , z = a_2 } , \label{i2}\ ] ] where @xmath109 . using the reciprocal symmetry in the green s function in eq . we deduce that @xmath110 we evaluate the derivatives in the expression for @xmath111 , using the prescription described in appendix [ ddz - green ] , as @xmath112 , \label{i2-gen}\end{aligned}\ ] ] where @xmath113 s are given by eq . after replacing @xmath114 . for the case of the dirichlet limit ( @xmath115 ) the expression for @xmath111 in eq . takes on the relatively simple form @xmath116 where the subscript @xmath117 stands for dirichlet limit . using the above expression in eq . we have @xmath118 we describe the correspondence of the expression for the interaction energy , in eq . , in the dirichlet limit , to the result in @xcite in appendix [ app - emig ] . we shall digress a little to answer how the above result is related to the dirichlet green s function . we start by recalling the reduced green s function corresponding to eq . in the dirichlet limit , which can be derived by taking the @xmath119 limit in @xmath120 in eq . defined in region 2 ( @xmath121 ) in figure [ regions ] , @xmath122 where @xmath123 and @xmath124 stand for @xmath125 and @xmath126 respectively . it is trivially verified that the above function satisfies the dirichlet boundary conditions , @xmath127 and @xmath128 . using eq . we can evaluate @xmath129_{z_<=a_1,z_>=a_2 } = - \frac{\kappa}{\sinh \kappa a}.\ ] ] using the above result eq . can be expressed as the product of derivatives of two dirichlet green s functions . it is also worth mentioning that the result in eq . can also be derived by exclusively using @xmath120 in eq . after taking the limit @xmath119 . this is expected because the result in the dirichlet limit should depend only on the quantities between the plates . thus , all the results related to the dirichlet limit can be derived without relying on the averaging prescription for taking derivatives described in appendix [ ddz - green ] . however , for the more general case being considered here , we require the averaging prescription to derive eq . . for the weak coupling limit ( @xmath130 ) the expression for @xmath111 in eq . takes the very simple form @xmath131 where the subscript @xmath132 stands for the weak limit . we point out that in the case of weak coupling the averaging prescription for the evaluation of the @xmath133-kernels was not necessary because only the free green s functions come in . using the above expression in eq . we can evaluate the corresponding @xmath134-kernel as @xmath135 , \label{l2w}\end{aligned}\ ] ] where in the evaluation of the integral we used the change of variables @xmath136 , and the corresponding relation @xmath137 . for the particular case of sinusoidal corrugations , as described in figure [ corru ] , we will have @xmath138 , \\ h_2(y ) & = & h_2 \sin [ k_{0 } y],\end{aligned}\ ] ] [ sin - p ] where @xmath139 is the wavenumber corresponding to the corrugation wavelength @xmath5 . we get nonzero contributions , to the leading order , only for the case when both plates have the same corrugation wavelength . the fourier transforms @xmath103 for sinusoidal corrugations get contributions in the form of delta functions @xmath140 . \label{h - til}\ ] ] using the above expression in eq . , and after interpreting @xmath141 to be the infinite length in the @xmath52 direction , we write @xmath142 where we have used the symmetry property in the @xmath143-kernel , noted in eq . , and performed suitable rescaling in the integration variables . we have used the notations @xmath144 , and @xmath145 . the @xmath143-kernel in the above expression is provided by eq . . versus @xmath146 . , width=302 ] in the dirichlet limit , where the @xmath143-kernel is given by eq . , the interaction energy in eq . can be expressed in the form @xmath147 where the function @xmath148 is normalized such that @xmath149 , which becomes a convenience when we compare the results with those obtained by the proximity force approximation . the expression for @xmath148 is @xmath150 where the @xmath151-kernel , and the corresponding @xmath152-kernel , in the dirichlet limit were given in eq . and eq . respectively using the notations , @xmath153 , @xmath154 , and @xmath155 , we can write @xmath156 this expression can be numerically evaluated and has been plotted in figure [ a11-d - t0-fig ] . the value of the function at @xmath157 can be evaluated exactly by rewriting the integral in terms of spherical polar coordinates to yield @xmath158 ^ 2 = 1.\ ] ] it should be mentioned that the result corresponding to eq . for the electromagnetic case and the scalar dirichlet case have been evaluated exactly in @xcite . the lateral casimir force can be evaluated using eq . in the definition of lateral force in eq . which yields @xmath159 where @xmath160 is the magnitude of the normal scalar casimir force between two uncorrugated parallel dirichlet plates given as @xmath161 versus @xmath146 . , width=302 ] let us start by recalling the expression for the casimir energy and casimir force between uncorrugated parallel plates in the weak limit : @xmath162 in the weak limit we have seen that it is possible to evaluate the @xmath134-kernel in eq . without any effort . using this in eq . we have @xmath163,\ ] ] which after performing the integral can be written in the form @xmath164 where we have introduced the function @xmath165 = e^{-t_0 } \sum_{m=0}^2 \ , \frac{t_0^m}{m ! } = \frac{e_2(t_0)}{e^{t_0}},\ ] ] where @xmath166 is the the truncated exponential function which approximates to unity for @xmath167 . @xmath168 has been plotted in figure [ a2-w - t0-fig ] . using eq . we evaluate the lateral casimir force in this perturbation order as @xmath169 we start by collecting the terms in eq . in the form @xmath170 to economize typographical space and for bookkeeping purposes we introduce the notation @xmath171^{(m_1,m_2,\ldots , m_n)}_{(i_1,i_2,\ldots , i_n ) } = \text{tr } \big [ g^{(0 ) } \delta v_{i_1}^{(m_1 ) } g^{(0 ) } \delta v_{i_2}^{(m_2 ) } \cdots g^{(0 ) } \delta v_{i_n}^{(m_n ) } \big ] , \label{gv - not}\ ] ] where the superscripts @xmath172 denote the @xmath172-th order contribution in the power series expansion of the potential as defined in eq . . the subscripts @xmath173 identifies the potential that is contributing . in the case under consideration we have only two potentials coupling and thus @xmath174 for any @xmath90 . as an illustrative example we have @xmath171^{(2,1,1)}_{(1,1,2 ) } = \text{tr } \big [ g^{(0 ) } \delta v_1^{(2 ) } g^{(0 ) } \delta v_1^{(1 ) } g^{(0 ) } \delta v_2^{(1 ) } \big].\ ] ] using the above notation we can write @xmath175^{(2,1)}_{(1,2 ) } + [ g^{(0 ) } \delta v]^{(1,2)}_{(1,2 ) } - [ g^{(0 ) } \delta v]^{(1,1,1)}_{(1,1,2 ) } - [ g^{(0 ) } \delta v]^{(1,1,1)}_{(1,2,2 ) } \big\}.\end{aligned}\ ] ] we argue that for the case of sinusoidal corrugations of the same wavelength on both plates the third order perturbation does not contribute . ( contributions from third order are nonzero if the wavelength of the corrugations of one plate is double that of the other plate . ) since the plates under consideration are infinite in extent , the interaction energy is independent of translations in the @xmath52 direction . let us now make a translation of the amount @xmath176 in the potentials in eq . . this amounts to replacing @xmath177 . thus invariance under translation requires that the total power of @xmath85 s in the result should be even . this rules out the third order perturbation from contributing . starting from eq . and keeping terms contributing to the fourth order we have @xmath178 where the superscripts represent the powers of @xmath85 s . for example , the superscript @xmath179 represents terms involving @xmath180 . all terms except one gets contribution from @xmath181 in the logarithm expansion in eq . . since the @xmath182 term is distinct in structure we further separate it out by defining @xmath183 where the term involving the superscript @xmath184 is the contribution from @xmath182 in . using the notation in eq . we can thus collect the terms contributing to eq . and eq . as @xmath185^{(1,1,1,1)}_{(1,2,1,2 ) } \big\ } , \\ e_{12}^{(2,2)b } & = & \frac{i}{2\tau } \big\ { [ g^{(0)}\delta v]^{(1,1,1,1)}_{(1,1,2,2 ) } - [ g^{(0)}\delta v]^{(1,1,2)}_{(1,1,2 ) } - [ g^{(0)}\delta v]^{(2,1,1)}_{(1,2,2 ) } + [ g^{(0)}\delta v]^{(2,2)}_{(1,2 ) } \big\ } , \\ e_{12}^{(3,1 ) } & = & \frac{i}{2\tau } \big\ { [ g^{(0)}\delta v]^{(1,1,1,1)}_{(1,1,1,2 ) } - [ g^{(0)}\delta v]^{(1,2,1)}_{(1,1,2 ) } - [ g^{(0)}\delta v]^{(2,1,1)}_{(1,1,2 ) } + [ g^{(0)}\delta v]^{(3,1)}_{(1,2 ) } \big\ } , \\ e_{12}^{(1,3 ) } & = & \frac{i}{2\tau } \big\ { [ g^{(0)}\delta v]^{(1,1,1,1)}_{(1,2,2,2 ) } - [ g^{(0)}\delta v]^{(1,2,1)}_{(1,2,2 ) } - [ g^{(0)}\delta v]^{(1,1,2)}_{(1,2,2 ) } + [ g^{(0)}\delta v]^{(1,3)}_{(1,2 ) } \big\}.\end{aligned}\ ] ] [ de = gv - not ] in terms of the reduced green s function defined in eq . and eq . , and the fourier transform of the corrugations in eq . , we can write each of the terms in eq . in terms of a corresponding @xmath134-kernel . in terms of the notation introduced in eq . these will read as @xmath186^{(m_1,\ldots , m_n)}_{(i_1,\ldots , i_n ) } = l_x \int \frac{dk_1}{2\pi } \cdots \frac{dk_n}{2\pi } \,\tilde{h}_{i_1}^{m_1}(k_1-k_2 ) \cdots \tilde{h}_{i_n}^{m_n}(k_n - k_1 ) \,l^{(m_1,\ldots , m_n)}_{(i_1,\ldots , i_n)}(k_1,\ldots , k_n),\ ] ] where implicitly we have interpreted the powers of the fourier transformed corrugations as @xmath187 observe that each term contributing to the @xmath188-th order ( @xmath189 ) has exactly @xmath188 @xmath4 s . also , the total number of variables inside the @xmath134-kernel is @xmath190 which is in general less than or equal to @xmath188 . it might be appropriate to call them @xmath190-point kernels . proceeding further , in the spirit of the second order calculation in eq . , we introduce the corresponding @xmath133-kernels for the @xmath134-kernels as @xmath191 where the @xmath133-kernels are expressed in terms of derivatives operating on the reduced green s function defined in eq . as @xmath192 \bigg|_{z_n = a_{i_n}}. \label{imi}\ ] ] for clarification we illustrate the evaluation of a particular term which should also serve as an illustration of the notation . the term we consider is @xmath193^{(1,2,1)}_{(1,1,2 ) } & = & l_x \int \frac{dk_1}{2\pi } \frac{dk_2}{2\pi } \frac{dk_3}{2\pi } \,\tilde{h}_1(k_1-k_2 ) \tilde{h}_1 ^ 2(k_2-k_3 ) \tilde{h}_2(k_3-k_1 ) \,l^{(1,2,1)}_{(1,1,2)}(k_1,k_2,k_3 ) \nonumber \\ & = & l_x \int \frac{dk_1}{2\pi } \frac{dk_2}{2\pi } \frac{dk_3}{2\pi } \frac{dk_4}{2\pi } \,\tilde{h}_1(k_1-k_2 ) \,\tilde{h}_1(k_2-k_3 ) \tilde{h}_1(k_3-k_4 ) \,\tilde{h}_2(k_4-k_1 ) \,l^{(1,2,1)}_{(1,1,2)}(k_1,k_2,k_4 ) . \hspace{5mm}%\nonumber\end{aligned}\ ] ] notice how the particular @xmath194 dependence in the @xmath134-kernel is unambiguously specified . the corresponding @xmath133-kernel using eq . and eq . is given as @xmath195 \bigg|_{{\scriptstyle z_1=a_1 , z_2=a_1 , z_3=a_2}}.\ ] ] using the reciprocal symmetry in the green s functions we can learn the following symmetries in the @xmath133-kernels associated with the terms in eq . : @xmath196 [ i4sym ] we have not listed the 2-point @xmath133-kernels in the above list because they have the symmetry of the kind in eq . . in fact any 2-point kernel will have the following symmetry : @xmath197 in the above discussion involving very general notations we have in principle expressed each term in eq . . specific evaluation of each term involves the evaluation of the corresponding @xmath133-kernels which are given in terms of the derivatives of the green s functions . the derivatives are evaluated using the prescription described in appendix [ ddz - green ] . we can thus collect the terms in eq . into four distinct @xmath134-kernels in the form @xmath198 [ de12=l4 ] where it should be noted that different @xmath134 s combine with specific combination of @xmath4 s . the factor of one - half in eq . can be traced back to the coefficient of the second term in the expansion of logarithm in eq . . the respective kernels @xmath199 above are related to their corresponding @xmath200 by eq . , which are given by @xmath201 [ i4-cons ] proceeding in the spirit of section [ leading - order ] has brought us to the point of evaluation of the thirteen @xmath133-kernels on the right hand side of eq . . using the prescription described in appendix [ ddz - green ] we have evaluated all the thirteen kernels and explicit expressions analogous to eq . have been derived . this would have involved a lot more labor and bookkeeping if not for the facilitation achieved by the use of mathematica . we shall not display the explicit expressions here because they are too long . but , as in section [ leading - order ] , these expressions simplify considerably in the dirichlet limit and the weak coupling limit . observe that the reduced green s function defined by eq . has a well defined dirichlet limit . in the light of this observation in conjunction with the expression for the @xmath133-kernels in eq . taken at face value suggests that these kernels might not have a well - defined finite dirichlet limit . however , we evaluated the second order contribution in eq . . in fact it can be verified that all 2-point kernels have a well - defined finite dirichlet limit . further , @xmath202 also has a well - defined finite dirichlet limit . the rest of the nine @xmath133-kernels on the right hand side of eq . do not have a finite dirichlet limit . but , the sums of the @xmath133-kernels listed in eq . have finite dirichlet limits . the higher powers in @xmath203 in the numerator of each of these sums cancel identically to give a well - defined limit as @xmath119 . this seems to be a generic phenomena . here we list the @xmath133-kernels in eq . evaluated in the dirichlet limit : @xmath204 , \\ i^{(1,3)}_d(\kappa_1,\kappa_2,\kappa_3,\kappa_4 ) & = & -\frac{\kappa_1}{\sinh \kappa_1 a } \frac{\kappa_2}{\sinh \kappa_2 a } \frac{1}{4 } \left [ 4 \frac{\kappa_3}{\tanh \kappa_3 a } \frac{\kappa_4}{\tanh \kappa_4 a } + \frac{\kappa_1 ^ 2 + \kappa_2 ^ 2}{3 } - \kappa_3 ^ 2 - \kappa_4 ^ 2 \right].\end{aligned}\ ] ] [ i4d ] we can similarly evaluate the @xmath133-kernels in the weak limit by keeping terms to order @xmath205 in the general expressions for the @xmath133-kernels . but , we find it instructive , and simpler , to start over from eq . . we observe that only the 2-point kernels contribute in the weak limit . we further notice that only the contributions from the free green s function ( which is obtained by switching off the couplings @xmath203 in eq . ) contribute in the evaluation of the weak limit . these observations allows us to write the 2-point kernels in the weak limit as @xmath206_{z_1 = a_1 , z_2=a_2 } \nonumber \\ & = & \frac{(-1)^{m_1}}{m_1 ! \ , m_2 ! } \ , \frac{\lambda_1 \lambda_2}{2\kappa_1 2\kappa_2 } \ , ( \kappa_1 + \kappa_2)^{m } \ , e^{-a ( \kappa_1 + \kappa_2)},\end{aligned}\ ] ] where @xmath207 . note that the derivatives in the above expressions are well defined because they are evaluated at a point where @xmath208 . in particular using these in eq . with the observation that only the 2-point kernels contribute in the weak limit we get @xmath209 and @xmath210 [ i4w ] where notice how they differ in the particular dependence on the variables . we again ( see comments after eq . ) point out that the averaging prescription for the evaluation of the @xmath133-kernels was not necessary to arrive at the above expressions . using the change of variables introduced in evaluating eq . we can similarly evaluate the @xmath134-kernels using eq . in the weak limit as @xmath211 , \label{l2w - gen}\ ] ] which reproduces the result in eq . for @xmath212 we shall now specialize to the particular case of sinusoidal corrugations described by eq . . using the fourier transforms in eq . all but one of the @xmath194 integrals in eq . can be immediately performed . the symmetries listed in eq . lead to simplifications in the expressions . as in section [ leading - order ] the expressions boil down to two integrals in variables @xmath194 and @xmath213 . in particular eq . takes the form @xmath214 , \\ \frac{e^{(3,1)}_{12}}{l_xl_y } & = & - \cos ( k_0y_0 ) \ , \frac{h_1 ^ 3 h_2}{64\pi^2 } \int_{-\infty}^{\infty } dk\int_0^\infty \bar{\kappa } \,d\bar{\kappa } \big [ i^{(3,1)}(\kappa,\kappa_+,\kappa,\kappa_+ ) + i^{(3,1)}(\kappa,\kappa_+,\kappa,\kappa_- ) + i^{(3,1)}(\kappa,\kappa_-,\kappa,\kappa_+ ) \big ] , \hspace{8 mm } \\ \frac{e^{(1,3)}_{12}}{l_xl_y } & = & - \cos ( k_0y_0 ) \ , \frac{h_1 h_2 ^ 3}{64\pi^2 } \int_{-\infty}^{\infty } dk\int_0^\infty \bar{\kappa } \,d\bar{\kappa } \big [ i^{(1,3)}(\kappa,\kappa_+,\kappa,\kappa_+ ) + i^{(1,3)}(\kappa,\kappa_+,\kappa,\kappa_- ) + i^{(1,3)}(\kappa,\kappa_-,\kappa,\kappa_+ ) \big],\end{aligned}\ ] ] [ de12=i4 ] where the @xmath133-kernels are given from eq . . we use the notations @xmath215 and @xmath216 introduced after eq . . we note the factor of @xmath58 in the argument of cosine function of the first term above is a mark of the fourth order in perturbation theory . it should be mentioned here that in the above expressions we have omitted finite terms not having any dependence in the lateral shift variable @xmath55 . these terms do not contribute to the lateral force . we presented the expressions for the @xmath133-kernels in the dirichlet limit in eq . . using these in eq . and then adding the contributions from the three terms in eq . we can write the total contribution to interaction energy due to the presence of corrugations as @xmath217,\ ] ] where we have introduced the functions @xmath218 , \hspace{7 mm } \\ a^{(2,2)}_d(t_0 ) & = & \frac{1}{\pi^4 } \int_0^\infty \bar{s}\,d\bar{s } \int_{-\infty}^{\infty } dt \left [ \frac{s^2}{\sinh^2 s } \frac{s_-^2}{\sinh^2 s_- } + 2 \frac{s^2}{\tanh^2 s } \frac{s_+}{\sinh s_+ } \frac{s_-}{\sinh s_- } \right].\end{aligned}\ ] ] [ cols="^,^ " , ] a recent debate , see @xcite , involved the comparison of the pfa result and the perturbative result . we earlier noted that the pfa limit is obtained by taking the limit @xmath219 while keeping the ratio @xmath220 fixed . in figure [ pfa - and - exact - w - versus - hoa ] we plot the lateral force in the pfa limit and compare it with the non - perturbative result for various values of @xmath146 . this justifies our presumption that the pfa is a good approximation for @xmath221 . we note that the error in the pfa is less than 1% for @xmath221 for arbitrary @xmath220 . for @xmath222 , we observe that the pfa is a very good approximation for @xmath223 which satisfies @xmath224 . after viewing the plots for various values of @xmath225 we note that in general the pfa is a good approximation for @xmath223 and further beyond for offsets @xmath226 . it is , in fact , plausible that the pfa holds for large amplitude corrugations for small offsets because the corrugations fit together like fingers in a glove . we thank jef wagner for extensive collaborative assistance throughout this project . kvs would like to thank osama alkhouli , subrata bal , pravin chaubey , james dizikes , david hartnett , k. v. jupesh , sai krishna rao , s. shankar , and stphane valladier , for discussions , comments , and help with programming in mathematica . we thank steve fulling for constructive comments on the manuscript . we thank the us national science foundation ( grant no . phy-0554926 ) and the us department of energy ( grant no . de - fg02 - 04er41305 ) for partially funding this research . icp would like to thank the french national research agency ( anr ) for support through carnot funding . evaluation of eq . involves taking derivatives of the green s function which as we shall see involves evaluating a function at a point where it has jump discontinuities in the derivatives . we start by noting that the differential equation in eq . can be solved in terms of exponential functions in the different regions described in figure [ regions ] . explicitly the solutions are given by the following piecewise - defined functions , with subscripts denoting the regions they are defined in , @xmath227 , \\ g_2^{(0)}(z , z^\prime;\kappa ) & = & \frac{1}{2\kappa } \,e^{-\kappa |z - z^\prime| } - \frac{1}{\delta } \frac{1}{2\kappa } \left [ \frac{\lambda_2}{2\kappa } \big ( 1 + \frac{\lambda_1}{2\kappa } \big ) \,e^{\kappa ( z + z^\prime - 2a_2 ) } + \frac{\lambda_1}{2\kappa } \big ( 1 + \frac{\lambda_2}{2\kappa } \big ) \,e^{-\kappa ( z + z^\prime - 2a_1 ) } \right . \nonumber \\ & & \hspace{35 mm } \left . - \frac{\lambda_1}{2\kappa } \frac{\lambda_2}{2\kappa } \,e^{\kappa ( z - z^\prime - 2a ) } - \frac{\lambda_1}{2\kappa } \frac{\lambda_2}{2\kappa } \,e^{-\kappa ( z - z^\prime + 2a ) } \right ] , \label{g20zzp - sol } \\ g_3^{(0)}(z , z^\prime;\kappa ) & = & \frac{1}{2\kappa } \,e^{-\kappa |z - z^\prime| } - \frac{1}{\delta } \frac{1}{2\kappa } \,e^{-\kappa ( z+z^\prime ) } \left [ \frac{\lambda_1}{2\kappa } \,e^{2\kappa a_1 } + \frac{\lambda_2}{2\kappa } \,e^{2\kappa a_2 } - \frac{\lambda_1}{2\kappa } \frac{\lambda_2}{2\kappa } \big ( e^{2\kappa a_1 } - e^{2\kappa a_2 } \big ) \right ] , \\ g_4^{(0)}(z , z^\prime;\kappa ) & = & \frac{1}{\delta } \frac{1}{2\kappa } \left [ \big ( 1 + \frac{\lambda_2}{2\kappa } \big ) \,e^{\kappa ( z - z^\prime ) } - \frac{\lambda_2}{2\kappa } \,e^{\kappa ( z + z^\prime - 2a_2 ) } \right ] , \\ g_5^{(0)}(z , z^\prime;\kappa ) & = & \frac{1}{\delta } \frac{1}{2\kappa } \,e^{\kappa ( z - z^\prime ) } , \\ g_7^{(0)}(z , z^\prime;\kappa ) & = & \frac{1}{\delta } \frac{1}{2\kappa } \left [ \big ( 1 + \frac{\lambda_1}{2\kappa } \big ) \,e^{\kappa ( z - z^\prime ) } - \frac{\lambda_1}{2\kappa } \,e^{-\kappa ( z + z^\prime - 2a_1 ) } \right],\end{aligned}\ ] ] [ g0zzp - sol ] where @xmath228 using the reciprocal symmetry of the green s function we further have @xmath229 , @xmath230 , and @xmath231 . the above piecewise solution gives the complete green s function . domain on which the piecewise functions contributing to the green s function are defined.,width=188 ] it is of relevance to observe that the green s function represented above is continuous everywhere in the @xmath232 domain while its first derivatives have simple ( jump ) discontinuities along the lines , @xmath233 , @xmath234 , and @xmath235 . since the evaluation of @xmath133-kernels in eq . , and in eq . , involves taking derivatives of the green s function at the lines of discontinuity , a definite prescription for the evaluation of the derivatives is called for . for this purpose , we interpret the value of a function at the point where it has a jump discontinuity to be the average of all the possible limiting values . for example , the derivative of the green s function at the point @xmath236 in figure [ regions ] takes on four different values when approached from the four regions 2 , 6 , 8 , and 9 . the value of the derivative is defined as the average of these four values . we provide the following supporting arguments for the averaging prescription . firstly , we note that eq . was derived from eq after integrating over the complete @xmath237 domain . the potentials in eq involved derivatives of delta functions , and after integration by parts the integrals got contributions from a single point in the @xmath237 domain . since an integral is a sum we expect it to support the averaging prescription . the second justification we provide is in the way of verification for known examples . we made checks by using this prescription to analyze the green s function for the case of a single plate . we do not provide the details of the exercise here . evaluation of the derivatives using the averaging prescription is made convenient when we define the green s function around a point as the average of the suitable piecewise functions . thus , we evaluate @xmath238 = \frac{1}{\delta } \frac{1}{2\kappa } \,e^{-\kappa a } , \nonumber \\ g^{(0)}(a_2,a_1;\kappa ) & = & \frac{1}{4 } \big [ g_2^{(0)}(a_2,a_1;\kappa ) + g_6^{(0)}(a_2,a_1;\kappa ) + g_8^{(0)}(a_2,a_1;\kappa ) + g_9^{(0)}(a_2,a_1;\kappa ) \big ] = \frac{1}{\delta } \frac{1}{2\kappa } \,e^{-\kappa a } , \nonumber \\ g^{(0)}(a_1,a_1;\kappa ) & = & \frac{1}{4 } \big [ g_1^{(0)}(a_1,a_1;\kappa ) + g_2^{(0)}(a_1,a_1;\kappa ) + g_4^{(0)}(a_1,a_1;\kappa ) + g_6^{(0)}(a_1,a_1;\kappa ) \big ] = \frac{1}{\delta } \frac{1}{2\kappa } \left [ 1 + \frac{\lambda_2}{2\kappa } \,\big ( 1 - e^{-2\kappa a } \big ) \right ] , \nonumber \\ g^{(0)}(a_2,a_2;\kappa ) & = & \frac{1}{4 } \big [ g_2^{(0)}(a_2,a_2;\kappa ) + g_3^{(0)}(a_2,a_2;\kappa ) + g_7^{(0)}(a_2,a_2;\kappa ) + g_9^{(0)}(a_2,a_2;\kappa ) \big ] = \frac{1}{\delta } \frac{1}{2\kappa } \left [ 1 + \frac{\lambda_1}{2\kappa } \,\big ( 1 - e^{-2\kappa a } \big ) \right ] . \nonumber\end{aligned}\ ] ] the first derivatives evaluate to be @xmath239 where @xmath240 is derivative with respect to @xmath232 . the second derivatives involving two distinct variables evaluate to be @xmath241 \partial_z \partial_{z^\prime } \,g^{(0)}(z , z^\prime;\kappa ) \big|_{z = a_1,z^\prime = a_1 } & = & - \frac{1}{\delta } \frac{\kappa^2}{2\kappa } \left [ \big ( 1 + \frac{\lambda_1}{2\kappa } \big ) \big ( 1 + \frac{\lambda_2}{2\kappa } \big ) + \frac{\lambda_2}{2\kappa } \,e^{-2 \kappa a } \right ] , \\[2 mm ] \partial_z \partial_{z^\prime } \,g^{(0)}(z , z^\prime;\kappa ) \big|_{z = a_2,z^\prime = a_2 } & = & - \frac{1}{\delta } \frac{\kappa^2}{2\kappa } \left [ \big ( 1 + \frac{\lambda_1}{2\kappa } \big ) \big ( 1 + \frac{\lambda_2}{2\kappa } \big ) + \frac{\lambda_1}{2\kappa } \,e^{-2 \kappa a } \right].\end{aligned}\ ] ] the second derivatives involving the same variables evaluate to @xmath242 which when compared with eq . tells us that the averaging prescription practically throws away the delta function contributions in eq . . a more detailed study of this issue is being sought . since the second derivative returns the green s function back , all higher order derivatives are obtained in terms of the above expressions . here we wish to explicitly compare our results in the leading orderwith those in @xcite . in particular we show how the dirichlet limit of our expression for the interaction energy in the leading order [ eq . with eq . inserted ] matches with the results in @xcite [ eq . ( 11 ) with eqs . ( 13 ) and ( 17 ) there ] . to this end we begin from eq . and write the interaction energy in the form @xmath243 where we have used the notation introduced in @xcite : @xmath244 , @xmath245 , and switched to euclidean time by replacing @xmath246 . the @xmath247-kernel introduced above , following @xcite , is given as @xmath248 \bigg|_{z^\prime = a_1,z = a_2 } , \label{q = gg}\ ] ] where @xmath249 and @xmath250 is given in terms of eq . after switching to euclidean time as @xmath251 where @xmath252 and @xmath253 , @xmath254 . in the dirichlet limit the above expression should correspond to the @xmath247-kernel introduced in @xcite [ eq . ( 13 ) there ] . note that in @xcite the interaction energy is not isolated from the outset unlike in our eq . . this is achieved in @xcite by identifying and subtracting the contributions to the energy from the single plates . we identify the presence of our @xmath133-kernel [ see eq . ] in eq . which lets us express the dirichlet limit of the @xmath247-kernel , using eq . , in the form @xmath255 ^ 2 = \frac{1}{2 } \bigg [ \frac{1}{a^4 } p\big ( \frac{|{\bf y}-{\bf y}^\prime|}{a } \big ) \bigg]^2\ ] ] in terms of the integral @xmath256 this exact form for the @xmath247-kernel in the dirichlet limit reproduces the result in @xcite . this exact form for the @xmath247-kernel in the dirichlet limit suggests that we can consider a class of corrugations for which exact results , in the perturbative approximation , might be achievable . like the exact results achieved in @xcite for the weak coupling limit , we should be able to explore geometries in the dirichlet limit in the perturbative approximation starting from eq . . further , it should also be possible to extend these explorations to the next - to - leading orders . and , for very special geometries it might just be possible to transcend the perturbative approximation and get exact results in the dirichlet limit alone . we hope to be able to address some of these explorations in our forthcoming publications . h. b. g. casimir , kon . wetensch . proc . * 51 * , 793 ( 1948 ) . s. barash , izv . . radiofiz . * 16 * , 1086 ( 1973 ) [ sov . * 16 * , 945 ( 1973 ) ] . r. golestanian and m. kardar , phys . lett . * 78 * , 3421 ( 1997 ) [ arxiv : quant - ph/9701005 ] . r. golestanian and m. kardar , phys . a * 58 * , 1713 ( 1998 ) . t. emig , a. hanke , r. golestanian and m. kardar , phys . lett . * 87 * , 260402 ( 2001 ) [ arxiv : cond - mat/0106028 ] . t. emig , a. hanke , r. golestanian and m. kardar , phys . . a * 67 * , 022114 ( 2003 ) . r. buscher and t. emig , phys . rev . a * 69 * , 062101 ( 2004 ) [ arxiv : cond - mat/0401451 ] . a. roy and u. mohideen , phys . lett . * 82 * , 4380 ( 1999 ) . f. chen , u. mohideen , g. l. klimchitskaya , and v. m. mostepanenko , phys . lett . * 88 * , 101801 ( 2002 ) . f. chen , u. mohideen , g. l. klimchitskaya , and v. m. mostepanenko , phys . rev . a * 66 * , 032113 ( 2002 ) . r. b. rodrigues , p. a. maia neto , a. lambrecht , and s. reynaud , phys . lett . * 96 * , 100402 ( 2006 ) . f. chen , u. mohideen , g. l. klimchitskaya , and v. m. mostepanenko , phys . * 98 * , 068901 ( 2007 ) . r. b. rodrigues , p. a. maia neto , a. lambrecht , and s. reynaud , phys . * 98 * , 068902 ( 2007 ) . i. cavero - pelez , k. a. milton , p. parashar and k. v. shajesh , arxiv:0805.2777 [ hep - th ] . t. emig , n. graham , r. l. jaffe and m. kardar , phys . 025005 ( 2008 ) [ arxiv:0710.3084 [ cond-mat.stat-mech ] ] . o. kenneth and i. klich , phys . lett . * 97 * , 160401 ( 2006 ) [ arxiv : quant - ph/0601011 ] . k. a. milton and j. wagner , j. phys . a * 41 * , 155402 ( 2008 ) [ arxiv:0712.3811 [ hep - th ] ] . k. a. milton , j. phys . a * 37 * , 6391 ( 2004 ) [ arxiv : hep - th/0401090 ] . k. a. milton , p. parashar and j. wagner , in preparation . k. a. milton , p. parashar and j. wagner , arxiv:0806.2880 [ hep - th ] .

we present a sample of resolved galactic h ii regions and photodissociation regions ( pdrs ) observed with the _ spitzer _ infrared spectrograph ( irs ) in spectral mapping mode between the wavelengths of 515 . for each object we have spectral maps at a spatial resolution of in which we have measured all of the mid - infrared emission and absorption features . these include the pah emission bands , primarily at 6.2 , 7.7 , 8.6 , 11.2 and 12.7 , as well as the spectral emission lines of neon and sulfur and the absorption band caused by silicate dust at around 9.8 . in this work we describe the data in detail , including the data reduction and measurement strategies , and subsequently present the pah emission band intensity correlations for each of the objects and the sample as a whole . we find that there are distinct differences between the sources in the sample , with two main groups , the first comprising the h ii regions and the second the reflection nebulae ( rne ) . three sources , the reflection nebula ngc 7023 , the horsehead nebula pdr ( an interface between the h ii region ic 434 and the orion b molecular cloud ) and m 17 , resist this categorization , with the horsehead pdr points mimicking the rne and the ngc 7023 fluxes displaying unique bifurcated appearance in our correlation plots . these discrepancies seem to be due to the very low radiation field experienced by the horsehead pdr and the very clean separation between the pdr environment and a diffuse environment in the ngc 7023 observations .

1601.07906

over the past twenty years or so the sample of astronomical objects known to emit the mid - infrared ( mir ) aromatic bands present from 320 has increased dramatically , and new discoveries regarding the inter - relations between the different bands have been revealed ( e.g. , , peeters et al . 2016 ) . the observed emission of the mir bands is now commonly associated with the vibrational relaxation of ultraviolet - pumped polycyclic aromatic hydrocarbon ( pah ) molecules , with several tools now available to interpret the mir bands in terms of pah emission ( e.g. , pahfit , @xcite ; the nasa ames pah database fitting routines , @xcite ; and pahtat , ) . the pah bands are observed in most astronomical contexts , particularly ones in which there is a significant uv radiation field ( e.g. , h ii regions , rne and planetary nebulae , ; galactic cirrus and dark cloud surfaces , ; external galaxies , @xcite ) leading to the conclusion that pah molecules are ubiquitous in the universe . some of the strongest pah emission spectra are observed around h ii regions , where the region immediately beyond the ionized region ( often called a photodissociation region or pdr ; e.g. , @xcite ) has the right conditions for the existence and excitation of the molecules . many studies have examined the appearance of the pah bands in these environments , however there have only been a few which have done this in a spatially resolved way ( e.g. , @xcite ) . the ability to examine pah emission in a spatially resolved observation has become very important in developing the pah model because the emission from the pah molecules is sensitive to the physical conditions in which they reside and the variations seen therefore track changes in the local conditions . in this paper , which will be the first of a series investigating aspects of the data presented here in detail , we describe the acquisition , reduction and measurement of _ spitzer_/irs observations for a sample of observed h ii regions , as well as several famous rne for comparison . our analysis will focus on their mir emission characteristics , particularly the aromatic bands at 6.2 , 7.7 " , 8.6 , 11.2 and 12.7 . the remainder of this paper will be organized as follows : in section [ sec : data ] we will discuss all aspects of our data , from the targets themselves , to the observational properties and data reduction . section [ sec : ma ] contains a detailed description of all our measurement and analysis techniques , and section [ sec : res ] describes our results . finally , sections [ sec : disc ] and [ sec : cs ] provide some brief discussion of our results and then a summary of our conclusions . we have selected the mapped h ii regions from the _ spitzer _ archive , along with a subset of well - known rne . our objects span a variety of classes and physical properties , from the ultra - compact iras 12063 - 6259 to the giant star formation complex w49a ( summarized in table [ tables : objects ] ) . for each object we list its distance , the spectral type of its exciting star and its size in both angular and physical units . in addition we show the physical size of a _ spitzer_/irs pixel , as the wide spread of distances evident in table [ tables : objects ] corresponds to large changes ( a factor of 30 ) in the physical area represented by the 1.8@xmath1 _ spitzer _ irs pixels . each of the nebulae were observed using the spectral mapping mode of the irs instrument and its sl module , which covers the 515 region at a spectral resolution of @xmath0 128 . all of the pointing and exposure time information for each map is presented in table [ tables : observations ] . we also show the extent of each cube superimposed on optical / mir imagery in figure [ fig : pointings ] . rather than discuss each source in detail here , we refer to final column of table [ tables : observations ] , where we list the papers which have previously examined these specific spectral cubes and present discussion of the various objects . as the g37.55 - 0.11 , g11.94 - 0.62 , m 17 and ngc 1333 observations have not hitherto been presented , we give a brief overview of these sources here . [ [ g11.94 - 0.62-and - g37.55 - 0.11 ] ] g11.94 - 0.62 and g37.55 - 0.11 + + + + + + + + + + + + + + + + + + + + + + + + + + + these sources , which are also known as iras 18110 - 1854 and iras 18577 + 0358 respectively , are both ultra - compact h ii regions ( uc - h ii ) which have been extensively studied in the radio . the radio observations provide us with measures of the distance , radius and implied spectral type of the dominant exciting star ( e.g. , table [ tables : objects ] ) . neither has been spectroscopically investigated in the mir , although both objects have been imaged in the mir . @xcite presented observations of both sources using msx observations at a scale of around @xmath020 , which revealed that the structure of the emission for g11.94 did not mirror the radio emission ( which traces the ionized region ) , while for g37.55 the core - halo radio morphology was not contradicted by the ` extended ' appearance in the mir . for g11.94 this conclusion was confirmed by @xcite who presented much higher resolution images taken at ground based facilities . their 20.8 image clearly shows that there is a very strong peak in the emission close to the south edge of the cometary halo observed in the radio ( e.g. , @xcite ) with some extended emission to the north going well beyond the radio emission . in addition there appears to be a secondary peak in the mir emission on the northern edge of the radio halo . the spitzer imagery presented in figure [ fig : pointings ] agrees with the msx findings in that it shows distinct clumps of emission , in tension with the @xcite categorization of the region as ` cometary ' . on the other hand the image we present of g37.55 - 0.11 matches the expected core - halo morphology very well . [ [ ngc1333 ] ] ngc 1333 + + + + + + + + ngc 1333 is a large , active star forming region in the perseus molecular cloud ( see @xcite for a recent review ) . while the designation ngc 1333 has come to be used for the entire region , it originally refers to the bright reflection nebula in its core . the rne is illuminated by the b8 type star bd + 30@xmath2549 which lies around 3.5 ( @xmath00.23 pc ) to the north east of the rne and an early b star known as svs3 " @xcite . svs3 has been observed in the radio and mir , and found to have a very compact h ii region consistent with the central star being of spectral type b4 - 6 @xcite . the region around this source has also been spectroscopically observed in the mir , with several studies in the 90 s using such observations to make the first claims of the presence of ionized pah molecules in the ism ( e.g. , @xcite ) . both of these studies used ground based observations of the 813 region and so were limited in aperture . despite that limitation , both studies found that the 8.6 band peaked in the center of the region ( at around the location of svs3 ) while the 11.2 band peaked around 10 to the south ( corresponding to a physical distance of about 0.64 pc ) . l c c c c c c c c object & distance & exciting star & angular size@xmath3 & physical size & pixel size & references + & [ kpc ] & spectral type & @xmath4 & [ pc ] @xmath4 [ pc ] & [ pc ] & + + _ h ii regions : _ + w49a & 11.4 & o5 & 7 @xmath4 5 & 23 @xmath4 17 & 0.20 & 1 , 2 + iras 12063 - 6259 & 10.9 & @xmath5o5.5@xmath6 & 1 @xmath4 1 & 3.2 @xmath4 3.2 & 0.19 & 3 + g37.55 - 0.11 & 10.1 & o6 & 2 @xmath4 1.5 & 5.9 @xmath4 4.4 & 0.18 & 4 , 5 + g11.94 - 0.62 & 5.2 & o7.5 & 1.1 @xmath4 0.7 & 1.7 @xmath4 1.1 & 0.09 & 6 + m 17 & 1.6 & o4 & 10 @xmath4 10 & 4.7 @xmath4 4.7 & 0.03 & 7 , 8 + horsehead & 0.42 & o9 & 4 @xmath4 1.5 & 0.5 @xmath4 0.2 & 0.007 & 9 , 10 + + _ reflection nebulae : _ + ngc 7023 & 0.43 & b2.5 & 3 @xmath4 2.2 & 0.4 @xmath4 0.3 & 0.008 & 11 + ngc 2023 & 0.35 & b1.5 & 6 @xmath4 4.5 & 0.6 @xmath4 0.4 & 0.006 & 12 , 13 + ngc 1333 & 0.23 & b8 & 2 @xmath4 2 & 0.1 @xmath4 0.1 & 0.004 & 14 , 15 + + _ a _ : measured using irac 8 band imagery so as to be comparable to that expected in irs cubes . b _ : iras 12063 - 6259 is thought to be illuminated by a cluster of young stars ( e.g. , ) , however as we do not possess individual radio fluxes here we quote the spectral type of a single star producing the same radio luminosity ( using the model grids of @xcite ) as an upper limit . + l p1 cm l p1.5 cm c c c c c c object & irs & program & aorkey & position angle & steps & steps & step size & step size & reference + & module & i d & & [ @xmath2 ] & @xmath8 & @xmath9 & @xmath8 [ ] & @xmath9 [ ] & + + _ h ii regions : _ + w49a & sl & 63 & 16206848 & 84 & 5 & 100 & 40 & 1.8 & 1 + g37.55 - 0.11 & sl & 20517 & 14799104 & -93 & 1 & 9 & & 3.6 & + iras 12063 - 6259 & sl & 20517 & 14798848 & 157 & 1 & 7 & & 3.6 & 2 + g11.94 - 0.62 & sl & 20517 & 14799360 & 86 & 1 & 9 & & 3.6 & + m 17 & sl & + horsehead & sl & 3512 & 12011520 , 12017920 , 12018688 & 101 & 1 & 23 & & 1.8 & 9 , 10 + + _ reflection nebulae : _ + ngc 2023 s & sl & 30295 & 17977856 & 84 & 3 & 18 & 28 & 3.6 & 4 + ngc 7023 & sl1 , sl2 & 28 & 3871744 , 3871488 & -1.2 & 2 & 15 & 7.3 & 3.6 & 6 , 7 + ngc 1333 & sl & 20378 & 14587648 & -107 & 5 & 111 & 54.6 & 1.8 & + + the spectral maps for each object were processed using the _ spitzer _ science center reduction pipeline for irs / sl ( generally version 18.8 , 18.7 in some instances ) . cubism @xcite was employed to clean the individual bcds , primarily this meant masking any rogue or otherwise ` bad ' pixels , initially using the cubism functions ` autogen global bad pixels ' with settings ` sigma - trim ' = 7 , ` minbad - frac ' = 0.5 and ` autogen record bad pixels ' with settings ` sigma - trim ' = 7 and ` minbad - frac ' = 0.75 and subsequently by manual inspection of each cube . this process was repeated for each order ( sl1 , sl2 , sl3 ) of each observation for each object . in order to combine the cubes of different orders , sl2 and sl3 were resampled to the same pixel grid as sl1 using cubism , then the scaling factors were calculated between the three orders for the now aligned pixels by comparing the average fluxes within the overlap regions . subsequently the sl1 and sl2 cubes were stitched together using the derived scaling factors . for each of our maps the scale factors between orders were around 10% , with outliers being investigated and in some cases flagged as poor data . as the irs / sl pixels are around half the size of the beam , the final cubes were created by averaging 2@xmath42 pixel apertures , resulting in overlapping extraction apertures in which one in every four pixels is independent . this was performed by taking the average of the next three adjacent pixels , i.e. , @xmath10 , where @xmath11 is the resulting cube and @xmath12 is the observed cube . this process produces a new cube @xmath11 which has the same pixel scale as the old cube , however each pixel represents a local beam average . subsequent analysis masked out the non independent pixels . while several of the maps were not nyquist sampled , we adopted the same procedure such that we produced a uniform sample of spectral maps . the analysis of the spectra took place in several stages . firstly , we measured and corrected for extinction ( as will be described in section [ sec : ex ] ) . secondly , we defined the continua and aromatic emission plateaus using spline fits . thirdly , we fit the continuum subtracted emission features in a variety of ways , from direct integration for some pah bands , to gaussian fits for the emission lines . in general this combination of techniques is referred to as the ` spline ' method ( e.g. , ) . other methods of fitting the pah spectra involve fitting an array of drude profiles to model the pah emission ( e.g. , pahfit ; @xcite ) , or using templates based on principle component analysis ( e.g. , pahtat ; ) . it has been shown though that the trends present in the pah band intensity measurements do not depend on the measuring technique employed ( e.g. , @xcite ) , and so we are confident that our use of the spline method does not introduce any significant biases . before processing the data for each cube though , regions of poor data or contamination from unwanted sources were masked out . primarily this meant removing the pixels associated with young stellar objects ( yso s ) in m 17 and ngc 2023 ( visible as bright sources towards the middle of the ngc 2023 south cube and in the middle of the m 17 cube ) . in addition we removed sections of cubes which had fallen ` off - source ' , i.e. , pixels which did not contain significant flux such as the southern half of the ngc 2023 south cube , and the northern and southern 20% of the ngc 1333 cube . in general , molecular clouds and h ii regions contain copious amounts of dust , which combine to create both continuous extinction from the uv to the mid - infrared ( although this is generally a small effect in the mir ) as well as strong absorption features associated with dust grains of specific compositions . the most important of these , from the point of view of measuring pah band fluxes , is the silicate feature at 9.8 . this feature has significant width which overlaps two important pah bands at 8.6 and 11.2 . typically the 11.2 band is used to normalize measurements of the other bands , so its obscuration can strongly effect the interpretation of the pah spectrum . in order to correct our band measurements we adopt the ` modified spoon method ' from @xcite which is an updated version of the method used by @xcite which includes the silicate absorption profile given by @xcite . this procedure relies on the 5 - 15 continuum displaying a strong rising power law towards the red , which is appropriate for the h ii regions . however , the rne do not generally display spectra with rising power law continua , instead they seem to have roughly linear continua in physical ( i.e. w m@xmath13 @xmath14 vs. ) units . therefore for the rne we have used a modified version of our extinction measurement routine in which we assume a linear continuum between 5 and 15 . this assumption is based on the observational results of samples of pdrs ( e.g. , @xcite ) . the extinction maps for the h ii regions presented , i.e. g11.94 - 0.62 , g37.55 - 0.11 and m17 , are shown in figure [ fig : ext_maps ] ( along with the rne , which will be discussed in the following section ) . the horsehead pdr region was found to have essentially zero extinction measurable via the silicate feature . [ [ w49a - and - iras-12063 - 6259 ] ] w49a and iras 12063 - 6259 + + + + + + + + + + + + + + + + + + + + + + + + the w49a and iras 12063 - 6259 extinction maps have been shown and discussed at length by @xcite and @xcite , respectively . in general each of the maps was found to be consistent with prior measurements of stars on the line of sight or the nebulosity itself . [ [ m17 ] ] m 17 + + + + the m 17 extinction map displays reasonably consistent extinction levels throughout , ranging from a@xmath15 of 0.5 to a maximum of 2 in a region close to the main pdr in the northern half of the map . the north eastern corner of the map , which emits the majority of the pah flux has an average a@xmath15 of around 1 . in the southern half of the map we have masked out a protostar that dominated our extinction measurements . [ [ g11.94 - 0.62 ] ] g11.94 - 0.62 + + + + + + + + + + + extinction in g11.94 - 0.62 is strongly peaked around the core of the h ii region ( at around a@xmath15 = 4 ) , which drops off very strongly to less than a@xmath15 = 2 within a few pixels . the center of the map is affected by ch@xmath16oh absorption at 9.8 , which is much narrower than the silicate absorption and clearly evident in the spectra ( e.g. , @xcite ; ; also figure [ fig : g11_meth ] ) . in order to measure the silicate absorption for g11.94 - 0.62 , we modified the general method described in our previous papers ( e.g. , @xcite ) such that the wavelength of comparison between the observations and the interpolated power law spectra was outside the methanol band , in this case at 10.2 rather than 9.8 . the spectra strongly affected by methanol absorption were not used in the remaining analysis , as the methanol absorption created difficulties in defining an effective spline continuum for isolating the pah bands . oh absorption in the g11.94 - 0.62 spectra . the red spectrum is a ` typical ' h ii region spectrum with heavy silicate absorption and the black spectrum is similar except for prominent absorption bands at 6.8 and 9.7 , which are much narrower than the silicate absorption . the vertical dot - dashed lines mark the locations of ch@xmath16oh absorption bands given by @xcite and the pixels from which the spectra were drawn given in the key.,width=264 ] [ [ g37.55 - 0.11 ] ] g37.55 - 0.11 + + + + + + + + + + + an almost complete ring of silicate absorption is visible around the center of g37.55 - 0.11 . this could represent the projection of a spherical dust shell , as the central region still retains a high extinction level of around a@xmath15 = 2 compared to the silicate absorption ring , which peaks at around 3.5 [ [ ngc-2023 ] ] ngc 2023 + + + + + + + + extinction values found for ngc 2023 are very low ( of the order of a@xmath17 @xmath0 1 ; a@xmath15 @xmath0 0.1 ) , which is slightly lower than the average from previous measurements ( e.g. , @xcite and references therein ) . as the value of a@xmath15 recovered was very low , we did not deredden the ngc 2023 cube , as doing so would have introduced an additional source of uncertainty while only affecting the fluxes very minimally . used the pahtat tool and found values of a@xmath15 , which were systematically slightly higher than both our measurements and those discussed by @xcite . [ [ ngc-7023 ] ] ngc 7023 + + + + + + + + we find a@xmath17 values of up to around 20 in the north - west corner of the map behind the pdr front in agreement with that found by @xcite , who used a method similar to that employed by pahfit , albeit using a different extinction law . in contrast , the extinction map presented by finds values of up to a@xmath17 @xmath0 40 , a factor of two more than our measurement , albeit with the same general distribution . we also ran pahfit on each pixel of the cube and recovered the same pattern of extinction , except with a higher still peak value of around a@xmath17 = 60 . we dereddened the ngc 7023 cube using our extinction measurements . [ [ ngc-1333 ] ] ngc 1333 + + + + + + + + the ngc 1333 extinction map shows that the core of ngc 1333 ( the faint ring in the center of the map ) suffers from negligible extinction ( a@xmath15 @xmath18 0.5 ) , as we might expect given that it is a nearby rne . however , the western side of the map is dominated by a filament of the molecular cloud material , which obscures part of the field with a@xmath15 @xmath5 3 . in particular this ridge is inhabited by the protostar iras 03260 + 3111e which has the strongest silicate absorption in the spectral cube . spectra associated with this object were excluded from further analysis . for ngc 1333 we dereddened the cube using our measurements . the specifics of our implementation of the spline method are identical to those presented by @xcite . in short , this process involves computing a spline fit to a set of continuum points at wavelengths @xmath19 = 5.37 , 5.46 , 5.86 , 6.58 , 6.92 , 8.28 , 9.15 , 9.40 , 9.64 , 10.14 , 10.33 , 10.76 , 11.82 , 13.18 , 13.49 , 14.43 , 14.74 , and subtracting this spline from the measured data in order to create a continuum subtracted spectrum . in figure [ fig : decomp ] we show an example of the spline - method applied to a w49a spectrum , as well as examples of the techniques used to decompose the 7.7 complex and the blend between the 12.7 pah band and the [ ne ii ] 12.8 atomic line . in general two splines were defined for each spectrum , which differ from each other by the inclusion of a spline point at 8.28 ( between the peaks of the 7.7 and 8.6 pah bands ) . the first plateau , which we refer to as the 8 bump , is simply the difference of the two splines . the remaining plateaus that underlie the 510 and 1014 regions are defined by drawing a straight line underneath the features and integrating all of the flux under the general spline . this process is displayed graphically in figure [ fig : decomp ] . measurement of the narrower emission features proceeds using continuum subtracted spectra . [ fig : decomp ] measurement of the pah bands then proceeds in three distinct ways . the fluxes of the strong bands at 6.2 , ` 7.7 ' , 8.6 , and 11.2 are found by direct integration . for the 6.2 and 11.2 bands these measurements are supplemented with gaussian profile fits to determine the strengths of the weak 6.0 and 11.0 bands with which each are blended . the final 6.2 and 11.2 band fluxes are then found by subtracting the weaker bands from the total flux . while we have measured the 6.0 pah band , for the sake of brevity we do not discuss it , or the other weak pah bands at 5.7 , 12.0 , 13.5 and 14.0 . for the ` 7.7 ' complex the flux was directly integrated to produce measurements of the 7.7 and 8.6 bands . however , as here is some disagreement in the literature as to how the plateau underlying the ` 7.7 ' complex is related to the pah bands , this measurement was repeated for spectra which had been continuum subtracted using both the local and general spline producing two measurements of both the 7.7 and 8.6 bands , i.e. with and without a spline point at 8.3 . for the h ii regionsthe blend between the 12.7 pah band and 12.8 [ ne ii ] line is decomposed using the method proposed by @xcite ( and subsequently used by @xcite ) in which the 12.7 emission is assumed to have the same spectral profile as the average pah emission spectrum given by . the hony profile is scaled to the continuum subtracted emission in a small window between 12.4 and 12.6 where the emission is assumed to be dominated by the pah band ( shown in figure [ fig : decomp ] ( c ) ) . the pah flux can then be recovered by integrating under the feature profile and the 12.8 atomic line can then be measured using a simple gaussian profile fit after the pah emission profile has been subtracted . while it would have been possible to directly integrate the remaining flux to obtain the 12.8 line flux , we chose to use a gaussian profile fit in order to ensure that the the method of subtracting a scaled 12.7 pah emission band shape does not affect the shape of the 12.8 line . the possible systematics affecting this measurement ( i.e. the presence of very strong [ ne ii ] emission ) are discussed in detail by @xcite . for the rne , the 12.7 band flux was calculated by directly integrating the complex and subtracting the flux of the 12.3 h@xmath20 line which was fitted using a gaussian profile fit superposed on a rising polynomial continuum . the other atomic lines , where present , are all measured using gaussian profile fits with either flat or polynomial continua depending on the line . for example , the 8.99 [ ar iii ] line sits on the rising wing of the 8.6 pah band and so requires a second order polynomial continuum . in all cases uncertainties were computed by comparing the integrated feature flux and the rms noise of featureless areas of continuum between 9.3 and 9.5 , 13.3 and 13.5 and 13.7 , 13.9 respectively . we adopted this approach such that each flux measurement has an associated signal to noise ratio measurement calculated in a uniform way across the sample , rather than using , for example , the fit - parameter uncertainties for bands fit with gaussian profiles . in all subsequent discussion the uncertainties quoted or shown for a band measurement were calculated in this way . the standard correlations between pah ratios , obtained by ` normalizing ' each pah flux by dividing by the flux of another pah band in the same spectrum , are presented in figure [ fig : rat_corr ] , where we have chosen to use the 11.2 band to normalize the data . this procedure is performed to cancel out the effects of , for example , pah abundance and distance . in figure [ fig : rat_corr ] ( and the following figures [ fig : rat_corr_plat ] and [ fig : lm_corr_all ] we have chosen to present the majority of the data as contours for the sake of clarity . this was achieved by binning the data for each object into 50 bins in both @xmath21 and @xmath22 directions across the range of values present . subsequently these binned data were slightly smoothed using a 2 pixel boxcar average routine and plotted as contours . the contour levels are given in table [ tables : contlevs ] and were adjusted such that they only exclude isolated outliers . in figure [ fig : rat_corr ] we compare each mir emission component in turn to 7.7gs / 11.2 . for this figure we have included only the points where the signal - to - noise ratio for the combined bands is greater than three . in figure [ fig : rat_corr ] we have chosen to represent the data by grouping the points of several of the sources which display the same overall trends . the rationale for these groupings is clearer in the flux versus flux plots presented in the subsequent section , however the overall behavior within each group is also consistent in the traditional correlation plots . the first group represents the h ii regions and contains w49a , iras 12063 - 6259 , g11.94 - 0.62 and g37.55 - 0.11 . the second major group represents the rne and contains ngc 1333 and ngc 2023 . finally , there are three sources which resist classification into these groups , the rn ngc 7023 and the h ii region pdrs of m 17 and the horsehead nebula . .[tables : contlevs]contour levels [ cols="^,^,^ " , ] [ [ vs.-6.2 ] ] 7.7 vs. 6.2 + + + + + + + + + + + for the most well - known correlation , between the 7.7 and 6.2 bands , the expected linear relationship is recovered in each case ( albeit with a slightly different gradient for the h ii regions sample ) . for the ngc 7023 points in this plot we have shown that the majority of the points can be well explained by assuming that the 7.7 vs. 6.2 band intensity ratio is essentially constant and varying the 11.2 band intensity ( indicated by the black line ) . each of the other groups shown are also consistent with such a straight line , albeit with slight variations in the average 7.7 vs. 6.2 band intensity ratio . [ [ vs.-8.6-and-11.0 ] ] 7.7 vs. 8.6 and 11.0 + + + + + + + + + + + + + + + + + + + + for these correlations there are some splits by object type . for example for the 7.7 vs. 8.6 correlation the rne points have significantly better correlations than the h ii regions . m 17 and the horsehead pdr in particular display almost no correlation , while the grouped h ii regions display smaller variation than the rne group ( ngc 1333 , 2023 ) by around a factor of two . the latter is repeated by the 7.7 vs. 11.0 band correlation plot , where now the grouped h ii regions also show no correlation and the rne display strong , linear correlations . this may be a beam dilution effect whereby the range of variation in the h ii region data is squashed because the measurement represents the average over a much larger physical area because the distance to the h ii regions is much greater than our other sources ( see table [ tables : objects ] ) . interestingly , both of these correlations display some curvature in the correlation produced by the ngc 7023 points indicating that either a ) our decomposition is not totally isolating the 8.6 or 11.0 bands from the surrounding emission ( i.e. both are systematically low creating a slight upward curve on the plots ) or , b ) that the 8.6 and 11.0 bands do not follow the 7.7 band as well as the 6.2 band . [ [ vs.-12.7 ] ] 7.7 vs. 12.7 + + + + + + + + + + + + for the final pah band considered , at 12.7 , we see variations in behavior between the different groups . there is evidence of at least two different gradients , one of which is defined by the rne ( plus some of the w49a points ) and another displayed best by m 17 . curiously , the 12.7 band appears to be much weaker for the horsehead pdr , with most of the points having a 12.7/11.2 value of around 0.2 , which is well below the average of all the points ( 0.40.5 ) . interestingly , for both the 8.6 and 12.7 vs. 7.7 correlation plots there is also a significant number of points for which there is variation in the 12.7/11.2 and 8.6/11.2 band ratios without any progression in the 7.7/11.2 ratio , in fact , the 7.7/11.2 band ratio seems to be about 1.5 in both cases , with the 12.7/11.2 ratio varying from 0.2 to 0.9 and the 11.0/11.2 ratio varying from 0.01 to 0.18 . for the 11.0 ratio , this is exclusive to the rne sample , while for 12.7 this trend is also seen for the h ii regions sample . [ [ plateaus ] ] plateaus + + + + + + + + in figure [ fig : rat_corr_plat ] we show the behavior of the 510 plateau features with respect to the 7.7 pah band . here we see some consistency between object - types with the h ii regions sample , m 17 , the rne sample and ngc 7023 all showing a similar trend for the 7 plateau vs. the 7.7 band . there appears to be a difference in gradients between the rne sample and the h ii regions ( including m 17 ) though . interestingly the horsehead pdr points appear to support no correlation between the plateaus and the 7.7 pah flux . a subset of the ngc 7023 points coincides with the rne in the 510 plateau vs. 7.7 plot . the remainder of the points originate in the region of low 11.2 band fluxes , and hence are being drawn out to the top - right of the diagram . [ [ normalization ] ] normalization + + + + + + + + + + + + + @xcite showed that there is useful information regarding local physical conditions along the line of sight encoded into the pah band flux to pah band flux correlation . hence , we examine our sample in such a way as well . however , in order to visualize each of the sources in the same parameter space , we scaled the fluxes from each object , as the pixel by pixel band fluxes between the sources vary more than three orders of magnitude . the weakest and strongest emission is associated with the horsehead pdr and w49a , respectively . using the 11.2 pah band flux as an example , the w49a peak flux is 2.4 @xmath4 10@xmath23 w m@xmath13 , while the brightest horsehead 11.2 pah band measurement is 5.4 @xmath4 10@xmath24 w m@xmath13 . in order to consider the interrelationships of the fluxes of the pah bands , we scaled each source s fluxes by the _ peak _ 11.2 flux , such as to yield 11.2 fluxes in the range of 01 for each source while the 7.7 band fluxes , for example , vary from 0 @xmath04 . this process maintains the intrinsic relationships between each band and is very different from the usual process of normalization ( discussed in the previous section ) . so to summarize , the peak 11.2 pah flux was used as a scale factor for all of the flux measurements for each pixel of each source , rather than dividing each pixel , for example , by its own 11.2 flux measurement as done in the traditional way . the peak 11.2 band fluxes used as scaling factors are presented in table [ tables : norms ] . the flux relationships between the mir pah emission components are presented in figure [ fig : lm_corr_all ] . in these figures we have included only points for which the band was detected with a signal - to - noise ratio greater than three for both axes . c c object & peak 11.2 flux + & [ @xmath4 10@xmath25 w m@xmath13 ] + + _ h ii regions : _ + w49a & 241.12 + g37.55 - 0.11 & 126.14 + iras 12063 - 6259 & 294.22 + g11.94 - 0.62 & 225.04 + m 17 & 157.89 + horsehead pdr & 0.5466 + + _ reflection nebulae : _ + ngc 2023 s & 17.43 + ngc 7023 & 27.38 + ngc 1333 & 68.86 + + [ [ ngc-7023 - 1 ] ] ngc 7023 + + + + + + + + the reflection nebula ngc 7023 has been examined in terms of its flux - flux correlations by @xcite . therefore , we will not repeat the detailed findings of that paper here , although , we will make some comparisons later on . in general though , the ngc 7023 points agree with those found for the other rne , with the exception of an additional component in many cases . this component is defined by weak 11.2 emission and thus forms the lower trend in the top panels of figure [ fig : lm_corr_all ] . in the third panel of the top row of figure [ fig : lm_corr_all ] , we show the two general trends identified for ngc 7023 by @xcite , which they label the ` pdr ' line ( upper ) and the ` diffuse ' line ( lower ) . where appropriate , we will adopt this convention throughout . @xcite attributed the observed bifurcation to changes in the 11.2 complex profile , specifically the appearance of extra emission appearing redward of 11.3 which they refer to as the ` red wing ' . in general , this split is evident for all of the relationships between ionized pah bands ( at 6.2 , 7.7 , 8.6 and 12.7 ) and the neutral band at 11.2 , albeit very weakly for 12.7 vs. 11.2 . in addition @xcite found that the different gradients on this plot could be ascribed to different degrees of ionization , with the pdr line representing an ionized fraction of 0.4 and the diffuse correlation having an ionized fraction of around 0.7 . in our figure it can be seen that the horsehead points seem to adopt a slightly steeper gradient than the pdr line , perhaps suggesting that the pah molecules in the horsehead nebula possess an ionization fraction even lower than the 0.4 found for the pdr correlation . [ [ groupings ] ] groupings + + + + + + + + + in general , all of the h ii region points follow the ` pdr ' component identified by @xcite , with the exception of some outliers seen for m17 and the horsehead . there are some small inconsistencies in gradient though , with the h ii regions and m17 points forming slightly different gradients in the 11.2 vs. 6.2 plot , and the horsehead points forming an almost vertical trend in the 11.2 vs. 11.0 plot . the ngc 1333 , 2023 points ( grouped together as ` rne ' ) also follow the ` pdr ' component , albeit with a distinct bulge towards the ` diffuse ' component seen in the ngc 7023 points . in some cases , this bulge overlaps with the diffuse trend ( particularly when considering the 11.0 band ) . in terms of the @xcite interpretation of this diagram , this would imply that a subset of the rne points possess ionization fractions similar to that seen in ngc 7023 . finally , in the 11.2 vs. 12.7 plot , we see that the horsehead points actually follow the rne points much better than the h ii regions or m17 samples , contrary to what might be expected . [ [ trends - with - the-7.7-band ] ] trends with the 7.7 band + + + + + + + + + + + + + + + + + + + + + + + + + the middle - right panel and bottom row of figure [ fig : lm_corr_all ] feature the relationships between the 7.7 pah band fluxes ( 7.7gs ) and the other pah bands . for 6.2 and 8.6gs , all of the points are consistent with linear relationships between the bands , and for 11.0 we see what amounts to a curve , where every increase in 7.7 leads to a slightly smaller increase in 11.0 . this shape is effectively the same as that seen for the 11.2 vs. 11.0 correlation ( middle left panel of figure [ fig : lm_corr_all ] ) , implying that either the 7.7 or 11.0 band measurement features a neutral component . the most interesting result is that the slope of the trend in the 7.7 vs. 12.7 plot is different for each source , possibly reflecting the physical conditions . again , this suggests that the 12.7 pah band is comprised of emission from both neutral and ionized pah molecules . this idea has been suggested by several groups , particularly @xcite who showed that both ionized and neutral pahs with an ` armchair ' structure could contribute to emission at around 12.7 . subsequently @xcite found observational evidence of this effect in ngc 7023 , which we show here is a general behavior . our results can be summarized as follows : * the h ii regions behave differently from the rne . this was clearest when we consider the range of variations between the pah ratios ( figure [ fig : rat_corr ] ) where for the 8.6/11.2 and 11.0/11.2 ratios the rne span a dynamic range twice that of the h ii regions . we note that our h ii region sample is systematically more distant than the rne sample and suspect that this compresses the dynamic range of the h ii region data by spatial dilution . * there are systematic differences between the slopes recovered for the different groups present in many of the found correlations . for example , even in the most well - known correlation 6.2/11.2 versus 7.7/11.2 , the h ii region and m 17 samples have steeper slopes than the rne . * the data suggests that the 11.0 and 12.7 bands may arise from a combination of neutral and ionized pah species . for the 12.7 band this effect is very clear in figure [ fig : lm_corr_all ] , where the different groups have different slopes . * ngc 7023 has unique pah emission characteristics which seem to be driven by the weaker 11.2 band emission in the diffuse region ( around a factor of four weaker for a constant intensity of the 6 9 bands ) , which causes the traditional correlation plots to have extended ranges ( by at least a factor of two ) and the flux versus flux correlation plots to split into two branches ( labeled diffuse and pdr ) . in figure [ fig : lm_corr_all ] we showed that the scaled flux correlations between most of the pah bands had a variety of shapes , ranging from linear to bifurcated . the bifurcated correlations belong exclusively to ngc 7023 , and this is likely due to its unique geometry with respect to our line of sight , producing a very sharp edge between two distinct pah emitting regimes ( e.g. , @xcite ) . the shapes seen in the scaled correlation plots for the other objects seemed to take two forms : linear , and linear plus an offset peak ( this pattern is clearest in the 11.2 vs. 11.0 panel of figure [ fig : lm_corr_all ] , where the offset peak is centered around @xmath21 = 0.06 , @xmath22 = 0.5 for the rne ) . in the following discussion we follow the naming convention of @xcite and refer to the upper linear trend as the pdr material and the lower trend as the diffuse material . in these terms then , the linear component evident for most of the objects is very close to the ` pdr ' line defined by @xcite , although different sources seem to have slightly different slopes . while the second pattern consists of a section of the pdr line at lower intensity levels and an offset peak for the brighter regions . we can understand this pattern as being a mixture of environments along the line of sight , with the points within the offset peak being dominated ( although to what degree we can not say ) by ionized pah spectra from regions similar to the diffuse region seen for ngc 7023 , and the linear component dominated by colder material behind the pdr front ( i.e. experiencing a lower radiation field similar to the pdr regions of ngc 7023 ) . this observation can help us understand the seemingly uniform nature of the h ii region scaled flux correlations , which seem to all fall into the category of ` linear ' correlations , i.e. dominated by emission from the pdr surrounding the h ii regions . however , if we assume that the much more distant h ii regions might possess environments similar to the ngc 7023 ` diffuse ' component , then one may expect slight curvature in the linear correlations because the pixels containing the diffuse component would be diluted with normal ` pdr ' material to a greater extent with the overall effect being a slight shift rather than forming a separate peak . in this case the pdr material is still the major component of the spectrum . we can test this idea immediately by examining figure [ fig : lm_corr_all ] for evidence of curvature in the ` linear ' trends , with the expectation that the curvature would increase ( i.e. move towards the combination of pdr and diffuse components seen for the rne ) for the closer objects with pixels covering smaller spatial areas ( see table [ tables : objects ] for the area of 1 irs / sl pixel in square parsecs for each object ) . this trend is evident if we compare the h ii region sample ( generally very distant , dominated by w49a points at distance @xmath510 kpc ) and m17 ( almost a factor of ten closer ) in figure [ fig : lm_corr_all ] , indeed the m17 trend is curved towards the rne points ( this is especially evident in the top row where 11.2 band is plotted vs. 6.2 , 7.7 and 8.6 band measurements ) . interestingly , the horsehead points match this pattern for the 6.2gs and 7.7gs bands vs. the 11.2 band scaled flux correlation , but not the 8.6gs vs. 11.2 scaled flux correlation . we show this effect in figure [ fig : curvehii ] , in which we replicate the rne contours of the 11.2 vs. 8.6 scaled band intensity correlation plot and show best fits for the h ii regions and m 17 data using both straight line and polynomial ( @xmath27 ) models taking into account both @xmath21 and @xmath22 uncertainties calculated from the signal - to - noise ratios . for the h ii regions sample the curved fit follows the straight line almost exactly showing that there is almost no curvature , conversely for m17 a significant deviation from the straight line towards the secondary rne peak is seen ( with an accompanying increase in reduced @xmath28 ) . there is a significant amount of scatter around these trends , which shows that even in pixels which have similar dispositions of environments along the line of sight , there can be significant changes in flux . simple models like our polynomial fit , can not capture these changes ( in fact , the scatter for the horsehead points forces the best fitting curve model to turn over entirely ) . however our model does seem to capture the small changes present in more distant objects like m 17 . for the h ii regions , that are around a factor of ten more distant than m 17 , the diffuse component can not be detected in this way , as if it is present , its emission is overwhelmed by the pdr - like material dominating the beam . earlier we commented that the horsehead pdr points do not seem to follow the expected pattern of curvature despite being nearby , where it would be expected that the increased spatial resolution would yield a clearer detection of any diffuse component . the likely reason for this is , again , the size of the beam used to make our map , but in the opposite sense from the other h ii regions the horsehead region is so large on the sky that the diffuse region is not present in our map ! show irs / sl spectra for a selection of positions between the ic 434 / horsehead pdr interface and the ionizing star ( their figure 4 ) , in which they show that the spectrum of their region 6 , several arcminutes outside of our map , seems to have a pah spectrum reminiscent of that seen for the ` diffuse ' emission from ngc 7023 . we have seen that the observed pah band intensity ratios can be attributed mainly to changes in the overall ionization state of the emitting pah molecules within the beam of each observation . these feature ratios should therefore be related to the physical conditions in which the pah molecules find themselves . empirical calibrations between the band ratios and physical conditions have been attempted several times ( e.g. , @xcite , @xcite and @xcite ) . these three studies took very different approaches though , with @xcite estimating the physical conditions in their sample based on other information ( e.g. , spectral types and geometry ) ; @xcite using only well studied sources for which the physical conditions were known beforehand ; and @xcite using observations to calibrate the ionization fractions found using nasa ames pah database @xcite with @xmath29 , using only the pah spectra . we adopted a mixture of the first two approaches , using literature values where present , and estimating other parameters using simple relationships between , for example , the stellar spectral types , g@xmath30 , and continuum radio emission measurements and the nebular n@xmath31 . given the very different strategy employed by @xcite to compute physical conditions , we present their results as a point of comparison as we would not necessarily expect their methodology to produce the same results as those presented here or those by @xcite . firstly , we restrict our sample to those sources where we believe that a simple estimate is physically meaningful , i.e. where we believe that the simple estimates of physical conditions correspond to the location of the dominant emitting pah population . in some cases this is relatively straightforward , as the geometry of the system is obvious . for example , the ic 434 / horsehead pdr front is at a known ( large ) distance from its exciting star and hence experiences no strong gradients in g@xmath30 . conversely , objects such as m 17 are very difficult to treat in a spatially resolved way , as although there exists a prescription for the varying of g@xmath30 throughout our field of view ( based on the spectral types and distances of the stars in the central cluster ; @xcite ) we do not have corresponding spatially resolved measurements of density or temperature . as such , we concentrate on including the geometrically simplest sources in our sample : the uc - h ii regions , including those measured as part of the w49a map : the ultra - compact h ii regions cc and dd ( see , for example , @xcite ) . the uc - h ii regions can be characterized using only radio observations of the free - free hydrogen emission continuum . such observations typically provide us with : a ) the stellar spectral type ; b ) the radius of the h ii region ; and c ) the hydrogen density of the h ii region . in order to convert these values into a measurement of the physical conditions in the pdr region , we employ the following relationships . firstly , the density of a pdr ( in terms of the number of hydrogen nuclei n@xmath32 , regardless of ionization or molecular state ) is about a factor of 30 higher than the electron density of the ionized region ( e.g. , @xcite ) . we can convert this n@xmath32 into the election density , n@xmath31 , of the pdr by assuming that the dominant source of electrons is the ionization of neutral carbon atoms , e.g. , that n@xmath31 @xmath33 ( c / h)n@xmath32 @xmath33 1.6 @xmath4 10@xmath34n@xmath32 if we assume galactic abundances . here we are explicitly following the method of @xcite , although this relationship is a fundamental part of the structure of pdrs . an average value for g@xmath30 can be calculated from the stellar spectral type if we know the radius of the pdr . this relationship is given by @xcite as : @xmath35 where @xmath36 is the stellar luminosity , @xmath37 is the fraction of far - uv photons and @xmath38 is the diameter of the h ii region . using the relationships provided by @xcite we can translate the stellar spectral type into the fuv luminosity @xmath39 , and thereby arrive at the results presented in table [ tables : physconds ] for our uc - h ii regions . this process is performed routinely in order to calculate @xmath40 for pdrs ( e.g. , ) . the final parameter , t , only appears in its square root , and so , variations within the small range of temperatures physically plausible for pdrs , such as those studied here , only change @xmath29 by around 10% , with a typical temperature being around 400 k. therefore we adopted a value of 10% as the uncertainty of the calculated @xmath29 factors . for each source we calculated the 6.2/11.2 band intensity ratio by finding the average and standard deviations of the 6.2/11.2 band intensity ratios of all of the associated spectral pixels . c c c c c c c c object & spectral type & pdr radius [ pc ] & h ii density [ @xmath41 & log(g@xmath30 ) & n@xmath42 & t [ k ] & @xmath29 + + w49a / cc & o6 & 0.09 & 2700 & 4.67 & 13 & 400 & 2270 + w49a / dd & o6 & 0.09 & 3100 & 4.67 & 15 & 400 & 1977 + g37.55 - 0.11 & o6 & 0.06 & 12000 & 4.38 & 58 & 400 & 1070 + iras 12063 - 6259 & o6 & 0.13 & 1000 & 4.31 & 5 & 400 & 2746 + g11.94 - 0.62 & o7.5 & 0.02 & 61000 & 5.14 & 293 & 400 & 1192 + horsehead pdr@xmath3 & o9 & 3.7 & & 2 & 0.15 & 400 & 105 + + _ @xcite sources : _ + ngc 2023 & b1.5 & & & 4.17 & 16 & 750 & 800 + ngc 7027 & wd & & & 5.77 & 1600 & 2000 & 520 + orion bar & o6 & & & 4.60 & 8 & 500 & 3519 + + @xmath3 : all values from . + , see main text ) . the black points and line refer to the work of @xcite , where the thick line is the derived linear relationship between the black points and the thinner lines represent 1@xmath43 deviations . the red points represent the objects studied in this work , which seem to split into two groups , with four sources being consistent with the @xcite line and two sources which are not . , width=302 ] the data from table [ tables : physconds ] is shown in figure [ fig : gal ] along with the points and linear regression from @xcite . four of our uc - h ii regions are consistent with the @xcite data : w49a / cc , w49a / dd , g37.55 - 0.11 and iras 12063 - 6259 . the remaining sources ( g11.94 - 0.62 , and the horsehead ) deviate from the trend , appearing above the line ( by factors of 2 and 4 respectively ) . we first investigate possible systematics affecting our calculation of @xmath29 . the key parameter which could influence the @xmath29 values strongly is the size of the h ii region as measured from the radio data . this parameter strongly affects the value of @xmath40 without changing @xmath44 . for the uc - h ii region radii we used observations by @xcite for g11 and g37 , @xcite for the w49a sources and for iras 12063 - 6259 . iras 12063 - 6259 is a complex source with three separate peaks in the radio map . in order to include this source , we measured the approximate size of the northern peak , which shows the strongest emission and is roughly circular . for the w49a sources we did not use the radii quoted for the regions directly , as the low resolution radio observations of those regions gave very different values ( a factor of three larger ) than the radii given for the very similar sources ( in terms of stellar spectral type and density ) observed in high resolution mode . the numbers adopted and quoted in table [ tables : physconds ] refer to this average of the high resolution observations of similar sources in w49a ( i.e. , same spectral types and densities ) . in order to force the g11.94 - 0.62 point to match the @xcite trend we need to reduce the radius of the h ii region by 40% , and while @xcite do not quote uncertainties on their measurements , it is clear from inspection of their imagery ( @xcite , figure 5a ) that the emission region is larger than the 0.5 radius implied by a 40% reduction in size . in addition , for the horsehead pdr , the @xmath29 value which was calculated by , is even more strongly discrepant , with no similar arguments to be made regarding size , temperature or @xmath40 so it is clear that some objects can strongly deviate from the @xcite trend . this idea is further demonstrated by the @xcite calibration of the pah ratio and @xmath29 ( shown in figure [ fig : gal ] as a blue line ) . the relationship found by @xcite clearly does not match any of the points used by @xcite or from our h ii regions sample . that said , the horsehead and g11 are closer to the @xcite relationship than the @xcite fit . @xcite attribute this to differences in the decomposition of the bands , i.e. that @xcite include the 11.0 band flux in their 11.2 band measurement . however , our decomposition does differentiate between the two bands , and we find that the 11.0 band is on average around 510% of the strength of the 11.2 band ( e.g. , figure [ fig : rat_corr ] ) , which is clearly not enough to explain the difference in figure [ fig : gal ] , which is about 50% . the elevated ngc 7023 pah band ratios found by @xcite are consistent with those found for ngc 7023 in this paper ( and largely at greater 6.2/11.2 values than we have shown in figure [ fig : gal ] ) . this observation leaves us with two options : either the physical conditions in ngc 7023 are far more extreme than predicted by @xcite ; or that the relationships between the pah ratios and physical conditions are more complex than that expressed by @xmath29 . given that other sources in our sample , noticeably the horsehead pdr with its extremely low radiation field strength , do not fall on the @xcite trend , it seems that the second scenario is the most likely . we have reduced and analyzed a large sample of _ spitzer_-irs sl spectral cubes of h ii regions and rne , primarily in terms of their pah emission . each spectral pixel of each cube was processed using the same spline method with the same parameters , creating a uniform sample of measurements of the pah bands and the other emission features between 5 - 15 . several of these cubes have been investigated before ( e.g. , w49a , ngc 2023 , ngc 7023 ) and these datasets were used as a framework to understand the results for the other objects . we then attempted to understand the flux / flux correlation plots in terms of the average environment along the line of sight and showed that the observed shapes in the correlation plots represent different mixtures of diffuse and pdr environments . finally we investigated the relationship between the observed pah band ratios and the physical environments of our sources . our main findings are : 1 . the pah emission from h ii regions is distinct from that of rne in terms of the relationships between the band strengths we find that following general groups seem to have very similar pah properties the h ii regions ( w49a , g11.94 - 0.62 , g37.55 - 0.11 and iras 12063 - 6259 ) , the rne ( ngc 1333 , ngc 2023 ) and that some sources do not fall into either of these categories ( ngc 7023 , the horsehead pdr and m17 ) . traditional correlation plots reveal that the rne have a greater dynamic range for the pah band ratios than h ii - regions . in particular ngc 7023 which , due to the noted sharp divide between environments , has a strong reduction in 11.2 band emission across approximately half of the map . for the other rne the variations are still only around half of those seen for ngc 7023 . this effect may be due to the fact that the h ii regions are systematically more distant than the rne and , as such , the contents of each h ii region spectrum consist of the average of a wider variety of environments , especially as the physical area encompassed in each pixel increases . 4 . both the traditional , and the flux versus flux correlation plots reveal that the 11.0 and 12.7 bands behave differently from the other pah bands , in a manner consistent with being generated by a mixture of both neutral and ionized pah species the split seen between a ` diffuse ' and a ` pdr ' component observed by @xcite in flux versus flux correlation plots is only seen strongly for ngc 7023 , although the nearby h ii regions give hints that there may be some diffuse component contribution to their spectra . the presence of a subtle diffuse component can be deduced in some sources by examining the curvature in the distribution of data points for the flux versus flux correlations between a primarily ionized band and a primarily neutral band , e.g. , the 11.2 band versus the 8.6 band . 7 . our uc - h ii region sample is consistent with the @xcite relationship between physical conditions and pah ratios , however they are inconsistent with the @xcite relationship found in a spatially resolved study of ngc 7023 . this seems to indicate that there may not be one simple relationship between all of the physical conditions ( density , temperature and radiation field intensity ) and the pah band ratios . djs thanks christiaan boersma , jesse bregman and the anonymous referee for their constructive comments on the manuscript . studies of interstellar chemistry at leiden observatory are supported through advanced- erc grant 246976 from the european research council , through a grant by the dutch science agency , nwo , as part of the dutch astrochemistry network , and through the spinoza premie from the dutch science agency , nwo . this research has made use of nasa s astrophysics data system bibliographic services ; the simbad database , operated at cds , strasbourg , france ; aplpy , an open - source plotting package for python hosted at http://aplpy.github.com ; and the idl astronomy library @xcite .

wireless device - to - device ( d2d ) communication underlaying cellular network is a promising concept to improve user experience and resource utilization . unlike traditional d2d communication where two mobile devices in the proximity establish a direct local link bypassing the base station , in this work we focus on relay - aided d2d communication . relay - aided transmission could enhance the performance of d2d communication when d2d user equipments ( ues ) are far apart from each other and/or the quality of d2d link is not good enough for direct communication . considering the uncertainties in wireless links , we model and analyze the performance of a relay - aided d2d communication network , where the relay nodes serve both the cellular and d2d users . in particular , we formulate the radio resource allocation problem in a two - hop network to guarantee the data rate of the ues while protecting other receiving nodes from interference . utilizing time sharing strategy , we provide a centralized solution under bounded channel uncertainty . with a view to reducing the computational burden at relay nodes , we propose a distributed solution approach using stable matching to allocate radio resources in an efficient and computationally inexpensive way . numerical results show that the performance of the proposed method is close to the centralized optimal solution and there is a distance margin beyond which relaying of d2d traffic improves network performance . device - to - device ( d2d ) communication , lte - a l3 relay , uncertain channel state information , distributed resource allocation , stable matching .

1508.01459

we consider relay - assisted device - to - device ( d2d ) communication underlaying lte - a cellular networks where d2d user equipments ( ues ) are served by the relay nodes @xcite . when the link condition between two d2d ues is too poor for direct communication , the d2d traffic can be transmitted via a relay node , which performs scheduling and resource allocation for the d2d ues . we refer to this as _ relay - aided d2d communication _ which can be an efficient approach to provide better quality of service ( qos ) for communication between distant d2d ues . for this , we utilize the lte - a layer-3 ( l3 ) relay @xcite . we consider scenarios in which the potential d2d ues are located in the same macrocell ; however , the proximity and link condition may not be favorable for direct communication . therefore , they communicate via relays . the radio resources ( e.g. , resource blocks [ rbs ] and transmission power ) at the relays are shared among the d2d communication links and the two - hop cellular links . we formulate an optimization problem to allocate radio resources at a relay node in a muti - relay multi - user orthogonal frequency - division multiple access ( ofdma ) cellular network ( e.g. , lte - a network ) . due to the np - hardness of the resource allocation problem , we utilize time sharing strategy and provide an asymptotically optimal centralized solution . considering the random nature of wireless channels , we reformulate the resource allocation problem using the worst - case robust optimization theory . the uncertainties in link gains are modeled using ellipsoidal uncertainty sets . each relay node can centrally solve the problem taking channel uncertainty into consideration . however , considering the high ( e.g. , cubic to the number of ues and rbs ) computational overhead at the relay nodes , we provide a _ distributed _ solution based on _ stable matching _ theory which is computationally inexpensive ( e.g. , linear with the number of ues and rbs ) . we also analyze the stability , uniqueness , and optimality of the proposed solution . considering the computational and signalling overheads and lack of scalability of the centralized solutions , game theoretical models have been widely used for wireless resource allocation problems . however , the analytical tractability of equilibrium in such game - theoretical models requires special properties for the objective functions , such as convexity , which may not be satisfied for many practical cases @xcite . in this context , resource allocation using matching theory has several beneficial properties @xcite . for example , the stable matching algorithm terminates for every given preference profile . the outcome of matching provides suitable solutions in terms of stability and optimality , which can accurately reflect different system objectives . besides , with suitable data structures , a pareto optimal stable matching ( e.g. , allocation of resources to the ues ) can be obtained quickly for online implementation . the goal of this work is to design a practical radio resource scheme for relay - aided d2d communication in a multi - user multi - relay ofdma network . as opposed to most of the work in the literature where channel gain information is assumed to be perfect , we capture the dynamics of random and time - varying nature of wireless channels . to this end , we develop a low - complexity distributed solution based on the theory of stable matching and demonstrate how this scheme can be implemented in a practical lte - a system . the major contributions of this work can be summarized as follows : * we model and analyze the radio resource allocation problem for relay - aided d2d communication underlaying an ofdma cellular network considering uncertainties in channel gains . we formulate an optimization problem to maximize system capacity in a two - hop network while satisfying the minimum data rate requirement for each ue and limiting the interference to other receiving nodes . we show that the convexity of the optimization problem is conserved under bounded channel uncertainty in both the useful and interference links . * using the theory of stable matching , we develop a distributed iterative solution , which considers bounded channel uncertainty . the stability , uniqueness , optimality , and complexity of the proposed solution are analyzed . we also present a possible implementation approach of our proposed scheme in an lte - a system . * numerical results show that the proposed distributed solution performs close to the upper bound of the optimal solution obtained in a centralized manner ; however , it incurs a lower ( e.g. , linear compared to cubic ) computational complexity . through simulations , we also compare the performance of the proposed approach with a traditional underlay d2d communication scheme and observe that after a distant margin , relaying of d2d traffic improves network performance . we organize the rest of the paper as follows . we briefly review the related work in section [ sec : related_works ] . section [ sec : system_model ] presents the system model and related assumptions . followed by the formulation of the nominal resource allocation problem in section [ sec : rap_optimization ] , we reformulate the resource allocation problem considering wireless channel gain uncertainties in section [ sec : rap_uncert ] . we develop the stable matching - based distributed resource allocation algorithm in section [ sec : ra_sm ] . theoretical analysis of the proposed solution is presented in section [ sec : analysis_sm ] . in section [ sec : numerical_results ] , we present the performance evaluation results before we conclude the paper in section [ sec : conclusion_sm ] . despite the fact that the resource allocation problems in cellular d2d communication have been intensively studied in the recent literature , only a very few work consider relays for d2d communication and incorporate wireless link uncertainties in the formulation of the resource allocation problem . in @xcite , a greedy heuristic - based resource allocation scheme is proposed for both uplink and downlink scenarios where a d2d pair shares the same resources with cellular ue ( cue ) only if the achieved signal - to - interference - plus - noise ratio ( @xmath0 ) is greater than a given @xmath0 requirement . a resource allocation scheme based on column generation method is proposed in @xcite to maximize the spectrum utilization by finding the minimum transmission length ( i.e. , time slots ) for d2d links while protecting the cellular users from interference and guaranteeing qos . a new spectrum sharing protocol for d2d communication overlaying a cellular network is proposed in @xcite , which allows the d2d users to communicate bi - directionally while assisting the two - way communications between the enb and the cue . a graph - based resource allocation method for cellular networks with underlay d2d communication is proposed in @xcite . a two - phase resource allocation scheme for cellular network with underlaying d2d communication is proposed in @xcite . in @xcite , the mode selection and resource allocation problem for d2d communication underlaying cellular networks is investigated and the solution is obtained by particle swarm optimization . the works above do not consider relays for d2d communication . a distributed relay selection method for relay - assisted d2d communication system is proposed in @xcite . in @xcite , authors investigate the maximum ergodic capacity and outage probability of cooperative relaying in relay - assisted d2d communication considering power constraints at the enb . taking the advantage of l3 relays supported by the 3gpp standard , in @xcite , a centralized resource allocation approach is proposed for relay - assisted d2d communication assuming that perfect channel information is available . a gradient - based distributed resource allocation scheme is proposed in @xcite for relay - aided d2d communication in a multi - relay network under uncertain channel information . in this multi - relay network , the interference link gain between a ue and other relays ( to which the ue is not associated with ) is modeled with ellipsoidal uncertainty sets . however , the uncertainty in direct channel gain between relay and the ue is not considered . in this paper , we remodel the previous formulation and extend the work in @xcite by incorporating uncertainties in _ both _ the useful and interference links . in particular , we present a distributed resource allocation algorithm using stable matching considering the uncertainties in wireless channel gains ( e.g. , channel quality indicator [ cqi ] parameters according to the lte - a terminology ) . although not in the context of d2d communication , matching theory has been used in the literature to address the radio resource allocation problems in wireless networks . a spectrum allocation algorithm using matching theory is proposed in @xcite for a cognitive radio network ( crn ) under perfect channel assumption . in @xcite , a two - sided stable matching algorithm is applied for adaptive multi - user scheduling in an lte - a network . a distributed matching algorithm is proposed in @xcite for cooperative spectrum sharing among multiple primary and secondary users with incomplete information in a crn . in @xcite , a distributed algorithm is proposed to solve the user association problem in the downlink of small cell networks ( scns ) . a matching - based subcarrier allocation approach is proposed in @xcite for services with coupled uplink and downlink qos requirements . the radio resource ( e.g. , subcarrier and power ) allocation problem for a full - duplex ofdma network is modeled as a transmitter - receiver - subcarrier matching problem in @xcite . the matching - based solutions proposed in @xcite do not consider d2d - enabled networks . in the context of d2d communication , most of the works ( e.g. , @xcite ) provide centralized solutions . also , note that in @xcite , the effect of relaying on d2d communication is not investigated . moreover , the wireless link uncertainty is not considered in @xcite . different from the above works , we propose a stable matching - based distributed radio resource allocation approach considering the channel gain uncertainties in a multi - relay and multi - user relay - aided d2d communication scenario . a summary of the related work and comparison with our proposed approach is presented in table [ tab : summary ] . let @xmath1 denote the set of fixed - location l3 relays in the network ( fig . 1 in @xcite ) . the system bandwidth is divided into @xmath2 orthogonal rbs denoted by @xmath3 which are used by all the relays in a spectrum underlay fashion . the set of cues and d2d pairs are denoted by @xmath4 and @xmath5 , respectively . we assume that association of the ues ( both cellular and d2d ) to the corresponding relays are performed before resource allocation . prior to resource allocation , d2d pairs are also discovered and the d2d session is setup by transmitting known synchronization or reference signals @xcite . we assume that the cues are outside the coverage region of the enb and/or having bad channel condition , and therefore , the cue - enb communications need to be supported by the relays . communication between two d2d ues requires the assistance of a relay node due to poor propagation condition . the ues assisted by relay @xmath6 are denoted by @xmath7 . the set of ues assisted by relay @xmath6 is @xmath8 such that @xmath9 , @xmath10 , and @xmath11 . in the second hop , there could be multiple relays transmitting to their associated d2d ues . we assume that multiple relays transmit to the enb ( in order to forward cues traffic ) using orthogonal channels and this scheduling of relays is done by the enb . in our system model , taking advantage of the capabilities of l3 relays , scheduling and resource allocation for the ues is performed in the relay nodes to reduce the computational load at the enb . let @xmath12 denote the unit power @xmath0 for the link between ue @xmath13 and relay @xmath6 using rb @xmath14 in the first hop and @xmath15 be the unit power @xmath0 for the second hop . note that , in the second hop , when the relays transmit cues traffic ( i.e. , @xmath16 ) , @xmath15 denotes the unit power @xmath0 for the link between relay @xmath6 and the enb . on the other hand , when a relay transmits to a d2d ue ( i.e. , @xmath17 ) , @xmath15 refers to the unit power @xmath0 for the link between relay @xmath6 and the receiving d2d ue for the d2d - pair . let @xmath18 denote the transmit power in the link between @xmath19 and @xmath20 over rb @xmath14 and @xmath21 is the bandwidth of an rb . the achievable data rate . ] for @xmath7 in the first hop can be expressed as @xmath22 . similarly , the achievable data rate in the second hop is @xmath23 . since we consider a two - hop communication , the end - to - end data rate over rb @xmath14 can be expressed as @xmath24 , where @xmath25 , @xmath26 is the channel gain in the link between the d2d ues and @xmath27 denotes the set of ues transmitting using the same rb(s ) as @xmath28 . [ ftn : direct_d2d_rate ] ] for @xmath7 on rb @xmath14 is half of the minimum achievable data rate over two hops @xcite , i.e. , @xmath29 in the following , we present the formulation of the resource allocation problem assuming that perfect channel gain information is available . this formulation is referred to as the nominal problem since the uncertainties in channel gains are not considered . for each relay , the objective of radio resource ( i.e. , rb and transmit power ) allocation is to obtain the assignment of rb and power level to the ues that maximizes the system capacity , which is defined as the minimum achievable data rate over two hops . let the maximum allowable transmit power for ue ( relay ) is @xmath30 ( @xmath31 ) and let the qos ( i.e. , data rate ) requirement for ue @xmath7 be denoted by @xmath32 . the rb allocation indicator is a binary decision variable @xmath33 , where @xmath34 let @xmath35 denote the achievable sum - rate over allocated rb(s ) . we consider that the same rb(s ) will be used by the relay in both the hops ( i.e. , for communication between relay and enb and between relay and d2d ues ) . the objective of resource allocation problem is to maximize the end - to - end rate for each relay @xmath36 as follows : @xmath37 where the rate of ue @xmath7 over rb @xmath14 @xmath38 in ( [ eq : obj_func ] ) , the unit power @xmath0 for the first hop , @xmath39 where @xmath40 denotes the direct link gain between node @xmath19 and @xmath20 over rb @xmath14 for hop @xmath41 , @xmath42 in which @xmath43 denotes thermal noise . the interference link gain between relay ( ue ) @xmath19 and ue ( relay ) @xmath20 over rb @xmath14 in hop @xmath44 is denoted by @xmath45 , where ue ( relay ) @xmath20 is not associated with relay ( ue ) @xmath19 . similarly , the unit power @xmath0 for the second hop ) and ( [ eq : sinr2_w_x ] ) . consequently , interference from other transmitter nodes ( e.g. , ues associated to other relays in the first hop or other relays in the second hop ) will appear as a constant term in ( [ eq : sinr1_w_x ] ) and ( [ eq : sinr2_w_x ] ) . [ ftnt : x2_constant ] ] , @xmath46 where @xmath47 denotes the channel gain between relay - enb link for cues ( e.g. , @xmath16 ) or the channel gain between relay and receiving d2d ues ( e.g. , @xmath17 ) . from ( [ eqn : e2e_rate ] ) , the maximum data rate for ue @xmath7 over rb @xmath14 is achieved when @xmath48 . therefore , in the second hop , the power @xmath49 allocated for ue @xmath7 , can be expressed as a function of power allocated for transmission in the first hop , @xmath50 as follows : @xmath51 . hence , the data rate for @xmath7 over rb @xmath14 can be expressed as @xmath52 . considering the above , the objective function in ( [ eq : obj_func ] ) can be rewritten as @xmath53 for each relay @xmath36 in the network , the objective of resource allocation problem is to obtain the rb and power allocation vectors , i.e. , @xmath54^\mathsf{t}$ ] and @xmath55^\mathsf{t}$ ] , respectively , which maximize the data rate . in order to ensure the required data rate for the ues while protecting all receiver nodes from harmful interference , we define the following set of constraints . * the constraint in ( [ eq : con - bin ] ) ensures that each rb is assigned to only one ue , i.e. , @xmath56 * the following constraints limit the transmit power in each of the hops to the maximum power budget : @xmath57 * similar to @xcite , we assume that there is a maximum tolerable interference threshold limit for each allocated rb . the constraints in ( [ eq : con - intf-1 ] ) and ( [ eq : con - intf-2 ] ) limit the amount of interference introduced to the other relays and the receiving d2d ues in the first and second hop , respectively , to be less than some threshold , i.e. , @xmath58 * the minimum data rate requirements for the cue and d2d ues is ensured by the following constraint : @xmath59 * the binary decision variable on rb allocation and non - negativity condition of transmission power is defined by @xmath60 note that in constraint ( [ eq : con - intf-1 ] ) and ( [ eq : con - intf-2 ] ) , we adopt the concept of reference user . for example , to allocate the power level considering the interference threshold in the first hop , each ue @xmath7 associated with relay node @xmath6 obtains the reference user @xmath61 associated with the other relays and the corresponding channel gain @xmath62 for @xmath63 according to the following equation : @xmath64 similarly , in the second hop , for each relay @xmath6 , the transmit power will be adjusted accordingly considering interference introduced to the receiving d2d ues ( associated with other relays ) considering the corresponding channel gain @xmath65 for @xmath63 , where the reference user is obtained by @xmath66 [ cor : minlp ] the objective function in ( [ eq : obj_func - relx ] ) and the set of constraints in ( [ eq : con - bin])-([eq : x - s ] ) turn the optimization problem to a mixed - integer non - linear program ( minlp ) with non - convex feasible set . therefore , the formulation described in section [ sec : rap_optimization ] is np - hard . a well - known approach to solve the above problem is to relax the constraint that an rb is used by only one ue by using the time - sharing factor @xcite . in particular , we relax the optimization problem by replacing the non - convex constraint @xmath67 with the convex constraint @xmath68 . thus @xmath69 represents the sharing factor where each @xmath69 denotes the portion of time that rb @xmath14 is assigned to ue @xmath7 and satisfies the constraint @xmath70 . besides , we introduce a new variable @xmath71 , which denotes the actual transmit power of ue @xmath7 on rb @xmath14 @xcite . then the relaxed problem can be stated as follows : @xmath72 where @xmath73 , @xmath74 and @xmath75 . [ cor : relx - convex ] the objective function in ( [ eq : relx_objfunc ] ) is concave , the constraint in ( [ eq : con - qos - cue ] ) is convex , and the remaining constraints in ( [ eq : con - bin ] ) , ( [ eq : con - pow - ue - relx])-([eq : x - s - relx ] ) are affine . therefore , the optimization problem @xmath76 is convex . since @xmath76 is a non - linear convex problem , each relay can solve the optimization problem using standard algorithms such as interior point method ( * ? ? ? * chapter 11 ) . the centralized optimization - based solution is summarized in * algorithm [ alg : cent_soln]*. each relay locally solves the optimization problem @xmath76 and informs the other relays the allocation vectors using x2 interface . the process is repeated until the data rate is maximized , e.g. , @xmath77 for @xmath78 , where @xmath79 is the sum data rate for relay @xmath6 obtained by solving the optimization problem at iteration @xmath80 and @xmath81 is a small value . the duality gap of any optimization problem satisfying the time - sharing condition becomes negligible as the number of rbs becomes significantly large . the optimization problem @xmath76 satisfies the time - sharing condition , and therefore , the solution of the relaxed problem is asymptotically optimal @xcite . given the parameters of other relays ( e.g. , @xmath82 , @xmath83 ) , at each iteration of * algorithm [ alg : cent_soln ] * the allocation vectors ( e.g. , @xmath84 , @xmath85 ) obtained at each relay @xmath6 provide locally optimal solution for @xmath6 . in addition , * algorithm [ alg : cent_soln ] * allows the relays to perform allocation repeatedly with a view to finding the best possible allocation . if the data rate at the @xmath86-th iteration is not improved compared to that in the previous iteration @xmath87 , the algorithm terminates , and the allocation at iteration @xmath87 will be the resultant solution . each of the iteration of * algorithm [ alg : cent_soln ] * outputs the solution of relaxed version of the original np - hard optimization problem . since the solution of relaxed problem gives us the upper bound , and at the termination of * algorithm [ alg : cent_soln ] * , we obtain an upper bound of the achievable sum rate . t ] each relay @xmath36 estimates the cqi values from previous time slot and determines reference gains @xmath88 and @xmath89 . initialize @xmath90 . update @xmath91 . each relay @xmath36 : * solves the optimization problem @xmath76 and multicasts the allocation variables @xmath84 , @xmath85 to all relay @xmath92 over x2 interface . * calculates the achievable data rate based on current allocation as @xmath93 . data rate not maximized @xmath94 . allocate resources ( i.e. , rb and transmit power ) to associated ues for each relay . for worst - case robust resource optimization problems , the channel gain is assumed to have a bounded uncertainty of unknown distribution . an ellipsoid is often used ( e.g.,@xcite ) to approximate such an uncertainty region . let the variable @xmath95 denote the normalized channel gain which is defined as follows : @xmath96 in addition , let @xmath97 denote the uncertainty set that describes the perturbation of link gains for @xmath7 over rb @xmath14 . the normalized gain is then denoted by @xmath98 where @xmath99 is the nominal value and @xmath100 is the perturbation part . the uncertainty in the cqi values is modeled under an ellipsoidal approximation as follows : @xmath101 where @xmath102 is the uncertainty bound in each rb . using ( [ eq : ch_gain_f ] ) , we rewrite the rate expression for @xmath7 over rb @xmath14 as @xmath103 where @xmath104 and @xmath95 is given by ( [ eq : f_uncert ] ) . utilizing uncertainty sets similar to ( [ eq : uncert_set_f ] ) in the constraints ( [ eq : con - pow - rel - relx])-([eq : con - intf-2-relx ] ) , the optimization problem @xmath105 can be equivalently represented under channel uncertainty as follows : @xmath106 where for any parameter @xmath107 , @xmath108 denotes the nominal value and @xmath109 represents the corresponding deviation part ; @xmath110 , and @xmath111 are the maximum deviations ( e.g. , uncertainty bounds ) of corresponding entries in cqi values . in @xmath112 , @xmath113 is given by @xmath114 the above optimization problem is subject to an infinite number of constraints with respect to the uncertainty sets and hence becomes a semi - infinite programming ( sip ) problem @xcite . in order to solve the sip problem it is required to transform @xmath112 into an equivalent problem with finite number of constraints . similar to @xcite , we apply the cauchy - schwarz inequality @xcite and transform the sip problem . more specifically , utilizing cauchy - schwarz inequality , we obtain the following : @xmath115 similarly , @xmath116 @xmath117 note that , as presented in section [ subsec : uncert_set ] , to tackle the uncertainty in channel gains , we have considered the worst - case approach , e.g. , the estimation error is assumed to be bounded by a closed set ( uncertainty set ) . hence , from ( [ eq : worst_bound_rate])-([eq : worst_bound_intf2 ] ) , under the worst - case channel uncertainties , the optimization problem @xmath112 can be rewritten as @xmath118 , where @xmath113 is given by ( [ eq : rate_uncet_final ] ) . the transformed problem is a second - order cone program ( socp ) ( * ? ? ? * chapter 4 ) and the convexity of @xmath118 is conserved as shown in the following proposition . @xmath119 [ prop : convex_uncert ] @xmath118 is a convex optimization problem . using an argument similar to that in footnote [ ftnt : x2_constant ] , the objective function of @xmath118 in ( [ eq : relx_objfunc_uncrt_cs ] ) is concave . the constraints in ( [ eq : con - bin ] ) , ( [ eq : con - pow - ue - relx ] ) , ( [ eq : x - s - relx ] ) are affine and the constraint in ( [ eq : con - qos - cue ] ) is convex . in addition , the additional square root term in the left hand side of the constraints in ( [ eq : con - pow - rel - relx_uncrt - cs ] ) , ( [ eq : con - intf-1-relx_uncrt - cs ] ) , and ( [ eq : con - intf-2-relx_uncrt - cs ] ) is the linear norm of the vector of power variables @xmath120 with order @xmath121 , which is convex ( * ? ? ? * section 3.2.4 ) . therefore , the optimization problem @xmath118 is convex . @xmath118 is solvable using standard centralized algorithms such as interior point method . the joint rb and power allocation can be performed similar to * algorithm [ alg : cent_soln ] * and an upper bound for the solution can be obtained under channel uncertainty . it is worth noting that solving the above socp using interior point method incurs a complexity of @xmath122 at each relay node where @xmath123 denotes the length of vector @xmath124 . besides , the size of the optimization problem increases with the number of network nodes . despite the fact that the solution from * algorithm [ alg : cent_soln ] * outputs the optimal data rate , considering very short scheduling period ( e.g. , @xmath125 millisecond in lte - a network ) , it may not be feasible to solve the resource allocation problem centrally in practical networks . therefore , in the following , we provide a low - complexity distributed solution based on matching theory . that is , without solving the resource allocation problem in a centralized manner using any relaxation technique ( e.g. , time - sharing strategy as described in the preceding section ) , we apply the method of two - sided stable many - to - one matching @xcite . the resource allocation approach using stable matching involves multiple decision - making agents , i.e. , the available rbs and the ues ; and the solutions ( i.e. , matching between ue and rb ) are produced by individual actions of the agents . the actions , i.e. , matching requests and confirmations or rejections are determined by the given _ preference profiles_. that is , for both the rbs and the ues , the lists of preferred matches over the opposite set are maintained . for each rb , the relay holds its preference list for the ues . the matching outcome yields mutually beneficial assignments between rbs and ues . _ stability _ in matching implies that , with regard to their initial preferences , neither rbs nor ues have an incentive to alter the allocation . a _ matching _ ( i.e. , allocation ) is given as an assignment of rbs to ues forming the set of pairs @xmath126 . note that a ue can be allocated more than one rb to satisfy its data rate requirement ; however , according to the constraint in ( [ eq : con - bin ] ) , one rb can be assigned to only one ue . this scheme corresponds to a _ many - to - one _ matching in the theory of stable matching . more formally , we define the matching as follows @xcite . a matching @xmath127 for @xmath128 is defined as a function , i.e. , @xmath129 such that a. @xmath130 and @xmath131 b. @xmath132 and @xmath133 where the integer @xmath134 , @xmath135 for @xmath136 and @xmath137 denotes the cardinality of matching outcome @xmath138 . the above definition implies that @xmath127 is a one - to - one matching if the input to the function is an rb . on the other hand , @xmath127 is a one - to - many function , i.e. , @xmath139 is not unique if the input to the function is a ue . in order to satisfy the data rate requirement for each ue , we introduce the parameter @xmath140 denoting the number rb(s ) which are sufficient to satisfy the minimum rate requirement @xmath32 . consequently , the constraint in ( [ eq : con - qos - cue ] ) is rewritten as @xmath141 . generally this parameter is referred to as _ quota _ in the theory of matching @xcite . each user @xmath7 will be subject to an acceptance quota @xmath140 over rb(s ) within the range @xmath142 and allowed for matching to at most @xmath140 rb(s ) . the outcome of the matching determines the rb allocation vector at each relay @xmath6 , e.g. , @xmath143 . let us consider the utility matrix @xmath144 under the worst - case uncertainty , which denotes the achievable data rate for the ues in different rbs , defined as follows : @xmath145\ ] ] where @xmath146 $ ] denotes the entry of @xmath19-th row and @xmath20-th column in @xmath144 , and @xmath147 is given by ( [ eq : rate_sm_uncrt ] ) . @xmath148 each of the ues and rbs holds a list of preferred matches where a preference relation can be defined as follows ( * ? ? ? * chapter 2 ) . let @xmath149 be a binary relation on any arbitrary set @xmath150 . the binary relation @xmath149 is * complete * if for @xmath151 , either @xmath152 or @xmath153 or both . a binary relation is * transitive * if @xmath152 and @xmath154 implies that @xmath155 for @xmath156 . the binary relation @xmath149 is a ( weak ) * preference * relation if it is complete and transitive . the preference profile of a ue @xmath13 over the set of available rbs @xmath157 is defined as a vector of linear order @xmath158_{i\in \mathcal{n}}$ ] . the ue @xmath7 prefers rb @xmath159 to @xmath160 if @xmath161 , and consequently , @xmath162 > \mathfrak{u}_l[u_l , n_2]$ ] . likewise , the preference profile of an rb @xmath163 is given by @xmath164_{j\in \mathcal{u}_l}$ ] . based on the discussions in the previous section , we utilize an improved version of matching algorithm ( adapted from ( * ? ? ? * chapter 1.2 ) ) to allocate the rbs . the allocation subroutine , as illustrated in * algorithm [ alg : rb_alloc ] * , executes as follows . while an rb @xmath14 is unmatched ( i.e. , unallocated ) and has a non - empty preference list , the rb is temporarily assigned to its first preference over ues , i.e. , @xmath7 . if the allocation does not exceed @xmath140 , the allocation will persist . otherwise , the worst preferred rb from @xmath7 s matching will be removed even though it was previously allocated . the iterations are repeated until there are unallocated pairs of rb and ue . the iterative process dynamically updates the preference lists and hence leads to a stable matching . [ ! t ] the preference profiles @xmath165 , @xmath166 ; @xmath167 . the rb allocation vector @xmath84 . initialize @xmath168 . @xmath169 most preferred ue from the profile @xmath166 . set @xmath170 . @xmath171 least preferred resource allocated to @xmath172 . set @xmath173 . @xmath171 least preferred resource allocated to @xmath172 . remove @xmath174 from @xmath175 . remove @xmath172 from @xmath176 . once the optimal rb allocation is obtained , the transmit power of the ues on assigned rb(s ) is obtained as follows . we couple the classical generalized distributed constrained power control scheme ( gdcpc ) @xcite with an autonomous power control method @xcite which considers the data rate requirements of ues while protecting other receiving nodes from interference . more specifically , at each iteration @xmath87 , the transmission power for each allocated rb is updated as follows : @xmath177 where @xmath178 and @xmath179 is obtained as @xmath180 in ( [ eq : power_p_tilde ] ) , the parameter @xmath181 is chosen arbitrarily within the range of @xmath182 and @xmath183 is given by @xmath184 based on the rb allocation , the relay informs the parameter @xmath185 and each ue updates its transmit power in a distributed manner using ( [ eq : pow_alloc ] ) . each relay independently performs resource allocation and allocates resources to corresponding associated ues . the joint rb and power allocation algorithm is given in * algorithm [ alg : joint_alloc]*. [ ! t ] each relay @xmath36 estimates the nominal cqi values from previous time slot and determines reference gains @xmath186 and @xmath187 . initialize @xmath90 , @xmath188 @xmath189 and @xmath144 based on cqi estimates . update @xmath91 . build the preference profile @xmath166 for each rb @xmath163 based on utility matrix and inform corresponding entries of @xmath144 to ues . each ue @xmath13 builds the preference profile @xmath165 . obtain rb allocation vector using * algorithm [ alg : rb_alloc]*. update the transmission power using ( [ eq : pow_alloc ] ) for @xmath189 and update the utility matrix @xmath144 . inform the allocation variables @xmath84 , @xmath85 to each relay @xmath92 and calculate the achievable data rate based on current allocation as @xmath93 . data rate not maximized @xmath94 . for each relay , allocate resources ( i.e. , rb and transmit power ) to the associated ues . in the following , we analyze the performance of our proposed distributed resource allocation approach under bounded channel uncertainty . more specifically , we analyze the stability , optimality , and uniqueness of the solution , and its computational complexity . [ def : rational ] a. the pair of ue and rb @xmath190 in @xmath191 is * acceptable * if @xmath7 and @xmath14 prefer each other ( to be matched ) to being remain unmatched . b. a matching @xmath127 is called * individually rational * if no agent ( i.e. , ue or rb ) @xmath192 prefers to remain unmatched to @xmath193 . [ def : blocking ] a matching @xmath127 is * blocked * by a pair of agents @xmath194 if they each prefer each other to the matching they obtain by @xmath127 , i.e. , @xmath195 and @xmath196 . from * definition [ def : rational ] * , * [ def : blocking ] * , the matching @xmath127 is blocked by rb @xmath14 and ue @xmath7 if @xmath14 prefers @xmath7 to @xmath197 and either @xmath7 prefers @xmath14 to some @xmath198 , or @xmath199 and @xmath14 is acceptable to @xmath7 . using the above definitions , the stability of matching can be defined as follows ( * * chapter 5 ) . [ def : blocking2 ] a matching @xmath127 is * stable * if it is individually rational and there is no pair @xmath190 in the set of acceptable pairs such that @xmath7 prefers @xmath14 to @xmath139 and @xmath14 prefers @xmath7 to @xmath197 , i.e. , not blocked by any pair of agents . [ prop : stability ] the assignment performed in * algorithm [ alg : rb_alloc ] * abides by the preferences of the ues and rbs and it leads to a stable allocation . see * appendix [ appsec : stability]*. note that the allocation of rbs is stable at each iteration of * algorithm [ alg : joint_alloc]*. since after evaluation of the utility , the preference profile of ues and rbs are updated and the routine for rb allocation is repeated , a stable allocation is obtained . [ prop : uniq ] if there are sufficient number of rbs ( i.e. , @xmath200 ) , and the preference lists of all ues and rbs are determined by the @xmath201 utility matrix @xmath144 whose entries are all different and obtained from given uncertainty bound , then there is a unique stable matching . see * appendix [ appsec : uniq]*. [ def : pareto ] a matching @xmath127 is weak pareto optimal if there is no other matching @xmath202 that can achieve a better sum - rate , i.e. , @xmath203 , where the inequality is component - wise and strict for one user . [ prop : optimality ] the proposed resource allocation algorithm is weak pareto optimal under bounded channel uncertainty . see * appendix [ appsec : optimality]*. [ cor : algo_bound ] since @xmath204 satisfies the binary constraint in ( [ eq : bin_con ] ) , and the optimal allocation @xmath205 satisfies all the constraints in the optimization problem @xmath118 , for a sufficient number of available rbs , the data rate obtained by * algorithm [ alg : joint_alloc ] * gives a lower bound of the solution under channel uncertainty . [ prop : time ] the subroutine for rb allocation terminates after some finite number of steps @xmath206 . let the finite set @xmath207 represent all possible combinations of ue - rb matching where each element @xmath208 denotes that rb @xmath20 is allocated to ue @xmath19 . since no ue rejects the same rb more than once ( see line @xmath209 in * algorithm [ alg : rb_alloc ] * ) , the finiteness of the set @xmath207 ensures the termination of rb allocation subroutine in finite number of steps . note that the distributed approach replaces the optimization routine in the centralized approach with the matching algorithm ( * algorithm 2 * ) . since by * proposition [ prop : time ] * we show matching algorithm terminates after finite number of iterations , ultimately * algorithm 3 * will end up with a ( local ) pareto optimal solution after some finite number of iterations . = " 2d in line @xmath210 of * algorithm [ alg : joint_alloc ] * , the complexity to output the ordered set of preference profiles for the rbs using any standard sorting algorithm is @xmath211 and for each ue , the complexity to build the preference profile is @xmath212 . let @xmath213 be the total length of input preferences in * algorithm [ alg : rb_alloc ] * , where @xmath214 denotes the length of the profile vector @xmath215 . from * proposition [ prop : time ] * and ( * ? ? ? * chapter 1 ) it can be shown that , if implemented with suitable data structures , the time complexity of rb allocation subroutine is linear in the size of input preference profiles , i.e. , @xmath216 . since * phase ii * of * algorithm [ alg : joint_alloc ] * runs at most fixed @xmath217 iterations , at each relay node @xmath6 , the complexity of the proposed solution is linear in @xmath218 and @xmath219 . assuming that the relays obtain the cqi prior to resource allocation , the centralized approach does not require any exchange of information between a relay node and the associated ues to perform resource allocation . however , in the distributed approach , the relay node and the ues need to exchange information to update the preference profiles and transmit power . in both the approaches , the relay nodes need to exchange the allocation variables among themselves ( e.g. , over x2 interface ) in order to calculate the interference levels at the receiving nodes . in the distributed approach , the exchange of information between a ue and the relay node during execution of the resource allocation algorithm can be mapped onto the standard lte - a scheduling control messages . for scheduling in lte - a networks , the exchanges of messages over control channels are as follows @xcite . the ues will periodically sense the physical uplink control channel ( pucch ) by transmitting known sequences as sounding reference signals ( srs ) . when data is available for uplink transmission , the ue sends the scheduling request ( sr ) over pucch . the relay , in turn , uses the scheduling grant ( sg ) over physical downlink control channel ( pdcch ) to allocate the appropriate rb(s ) to the ue . once the allocation of rb(s ) is received , the ue regularly sends buffer status report ( bsr ) using pucch in order to update the resource requirement , and in response , the relay sends the acknowledgment ( ack ) over the physical hybrid - arq indicator channel ( phich ) . given the above scenario , the ues may provide the preference profile @xmath165 with the sr and bsr messages . the relays may provide the corresponding values in the utility matrix , e.g. , @xmath220_{j = 1 , \cdots , n}$ ] and inform the parameter @xmath185 using sg and ack messages . once the rb and power allocation is performed , the relays multicast the allocation information over x2 interface . in what follows , we analyze signalling overhead for our proposed solution . for a sufficient number of available rbs , we consider two cases : number of available rbs at each relay @xmath6 is equal to the number of ues ( e.g. , @xmath221 and @xmath222 ) ; and the number of rbs is greater than the number of ues ( e.g. , @xmath223 ) . for the first case , once * algorithm 2 * terminates , all rbs are allocated to the ues . this is because , by the definition of individual rationality ( see * definition 3 * ) , none of the agents ( i.e. , ue or rb ) wants to remain unallocated . hence , at the end of any iteration @xmath224 of * algorithm 2 * , there are @xmath225 unallocated rbs at each relay @xmath6 . therefore , the maximum number of iterations , say @xmath226 are required when all the rbs are allocated , e.g. , @xmath227 , and therefore , @xmath228 . since at each iteration @xmath224 , @xmath229 messages are exchanged , the total number of messages exchanged in * algorithm 2 * can be quantified as @xmath230 in * algorithm 3 * , each relay @xmath6 exchanges the allocation parameters ( e.g. , @xmath84 , @xmath85 ) over x2 interface . if * algorithm 3 * executes @xmath231 iterations , the overall signalling overhead ( e.g. , number of messages exchanged ) is given by @xmath232 likewise , for the second case ( e.g. , when @xmath223 ) , * algorithm 2 * terminates when there are no unallocated ues with less than their quota requirement ( e.g. , @xmath233 ) . hence the maximum number of iterations @xmath234 , and the numnber of messages exchanged in * algorithm 2 * for the second case is given by @xmath235 therefore , the overall signalling overhead for the second case can be expressed as follows : @xmath236 as can be seen from ( [ eq : sig-1 ] ) and ( [ eq : sig-2 ] ) , the overhead of signalling increases with the number of ues and the available rbs . however , the proposed distributed approach has significantly lower computational complexity than the centralized approach ( linear compared to cubic ) and it offers performance improvement over the existing solutions ( see section [ sec : res ] ) . we develop a discrete - time simulator in matlab and evaluate the performance of our proposed solution . we simulate a single three - sectored cell in a rectangular area of @xmath237 , where the enb is located in the center of the cell and three relays are deployed , i.e. , one relay in each sector . the cues are uniformly distributed within the relay cell . the d2d ues are located according to the clustered distribution model @xcite . in particular , the d2d transmitters are uniformly distributed over a radius @xmath238 ; and the d2d receivers are distributed uniformly in the perimeter of the circle with radius @xmath239 centered at the corresponding d2d transmitter ( fig . 2 in @xcite ) . both @xmath238 and @xmath239 are varied as simulation parameters and the values are specified in the corresponding figures . the simulation results are averaged over @xmath240 network realizations of user locations and channel gains . we consider a snapshot model and all the network parameters are assumed to remain unchanged during a simulation run . for propagation modeling , we consider distance - dependent path - loss , shadow fading , and multi - path rayleigh fading ( see section vii - a in @xcite ) . we measure the uncertainty in channel gains as percentages and assume similar uncertainty bounds in the cqi parameters for all the ues . for example , uncertainty bound @xmath241 refers that uncertainty ( e.g. , estimation error ) in the cqi parameters for @xmath242 is not more than @xmath243 of their nominal values . the simulation parameters are similar to those in table ii in @xcite . a summary of the observations from the performance evaluation results is provided in table [ tab : obsrv ] . .summery of performance results [ cols= " < , < , < " , ] in fig . [ fig : convergence_sm ] , we show the convergence behavior of our proposed distributed algorithm . in particular , we plot the average achievable data rate for the ues in different network realizations versus the number of iterations . the algorithm starts with uniform power allocation over rbs , which provides a higher data rate at the first iteration ; however , it may cause severe interference to other receiving nodes . as the algorithm executes , the allocations of rb and power are updated considering the interference threshold and data rate constraints . from this figure it can be observed that the solution converges to a stable data rate very quickly ( e.g. , in less than @xmath244 iterations ) . we compare the performance of our proposed scheme with a dual - decomposition based suboptimal resource allocation scheme proposed in @xcite . we refer to this scheme as _ existing algorithm_. in this scheme , the relay node allocates rbs considering the data rate requirement and the transmit power is updated in an iterative manner by updating the lagrange dual variables . for details refer to ( * ? ? ? * algorithm 2 ) . the complexity of this algorithm is of @xmath245 , where @xmath246 denotes the number of iterations it takes for the power allocation vector to converge @xcite . in fig . [ fig : compare_ub ] , we show the performances of the proposed distributed scheme and the existing algorithm , and the upper bound of the optimal solution which can be obtained in a centralized manner using * algorithm [ alg : cent_soln]*. we use the matlab optimization toolbox to obtain this upper bound . we plot the average achievable data rate for the ues versus the total number of ues . the average data rate is given by @xmath247 , where @xmath248 is the achievable data rate for ue @xmath28 . note that , for a given number of rbs , increasing the number of ues decreases the data rate . recall that , the complexity of both the proposed and reference schemes is linear with the number of rbs and ues ; and for the optimal solution , the complexity is cubic to the number of rbs and ues . as can be seen from this figure , the proposed approach outperforms the existing algorithm and performs close to the optimal solution . in order to obtain more insights into the performance , in fig [ fig : efficiency ] , we plot the efficiency of the proposed scheme and existing algorithm for different number of ues . similar to ( * chapter 3 ) , we measure the efficiency as @xmath249 , where @xmath250 is the network sum - rate for optimal solution . the parameters @xmath251 and @xmath252 denote the data rate for the proposed and existing schemes , respectively , which are used to calculate the corresponding efficiency metric @xmath253 and @xmath254 . the closer the value of @xmath255 to @xmath125 , the nearer the solution is to the optimal solution . clearly , the efficiency of the existing algorithm is lower compared to the proposed scheme . from the figure we observe that even in a dense network scenario ( i.e. , @xmath256 ) the proposed scheme performs @xmath257 close to the optimal solution ( compared to @xmath258 for the existing algorithm ) ; however , with much less computational complexity . we compare the performance of the proposed method for relay - aided d2d communication with a conventional underlay d2d communication scheme . in this scheme@xcite , an rb allocated to cue can be shared with at most one d2d link . the d2d ues share the same rb(s ) ( allocated to a cue using * algorithm [ alg : joint_alloc ] * ) and communicate directly with their peers _ without _ a relay if the data rate requirements for both the cues and d2d ues are satisfied ; otherwise , the d2d ues refrain from transmitting . we refer to this underlay d2d communication scheme @xcite as the _ reference scheme_. notice that this conventional ( e.g. , direct ) d2d communication approach can save half of the rbs . therefore , as mentioned in footnote [ ftn : direct_d2d_rate ] , data rate of @xmath7 in the reference scheme is given by @xmath259 . on the contrary , data rate of @xmath7 in the proposed relay - aided approach is given by @xmath260 . a. _ average achievable data rate vs. distance between d2d ues : _ the average achievable data rates of d2d ues for both the proposed and reference schemes are illustrated in fig . [ fig : rate_vs_d2d_dist ] . although the reference scheme outperforms when the distance between the d2d ues is small ( i.e. , @xmath261 ) , our proposed approach , which uses relays for d2d traffic , can greatly improve the data rate especially when the distance increases . this is due to the fact that when the distance increases , the performance of direct communication deteriorates due to increased signal attenuation . besides , when the d2d ues share resources with only one cue , the spectrum may not be utilized efficiently , and therefore , the achievable rate decreases . as a result , the gap between the achievable rate with our proposed algorithm and that with the reference scheme widens when the distance increases . b. _ gain in aggregate achievable data rate vs. varying distance between d2d ues : _ the gain in terms of aggregate achievable data rate under both uncertain and perfect cqi is shown in fig . [ fig : gain_vs_d2d_dist ] . we calculate the rate gain as follows : @xmath262 , where @xmath251 and @xmath263 denote the aggregate data rate for the d2d ues in the proposed scheme and the reference scheme , respectively . the figure shows that , compared to direct communication , with the increasing distance between d2d ues , relaying provides considerable gain in terms of achievable data rate and hence spectrum utilization . as expected , the gain reduces under channel uncertainty since the algorithm becomes cautious against channel fluctuations and allocates rbs and power accordingly to protect the receiving nodes in the network . note that there is a trade - off between performance gain and robustness against channel uncertainty . for example , when the distance @xmath264 , the performance gain of relaying under perfect cqi is @xmath265 . in the case of uncertain cqi , the gain reduces to @xmath266 and @xmath267 for the uncertainty bound parameter @xmath268 and @xmath269 , respectively . as the uncertainty bounds increase , the system becomes more roust against uncertainty ; however , the achievable data rate degrades . effect of relay - ue distance and distance between d2d ues on rate gain : _ the performance gain in terms of the achievable aggregate data rate under different relay - d2d ue distances is shown in fig . [ fig : rel_var_sm ] . it is clear from the figure that , even for relatively large relay - d2d ue distances , e.g. , @xmath270 , relaying d2d traffic provides considerable rate gain for distant d2d ues . , @xmath271 . uncertainty in cqi parameters is assumed to be not more than @xmath243 . for every @xmath238 , there is a distance threshold ( i.e. , upper position of the light - shaded surface ) beyond which relaying provides significant gain in terms of aggregate achievable rate . ] we have investigated the radio resource allocation problem in a relay - aided d2d network considering uncertainties in wireless channels and provided an iterative distributed solution to this problem using stable matching . we have analyzed the stability , uniqueness , and optimality of the proposed solution . we have also analyzed the complexity of the proposed approach . numerical results have shown that the distributed solution is close to the centralized optimal solution with significantly lower computational complexity . we have compared the proposed relay - aided d2d communication scheme with an underlay d2d communication scheme . as an extension of this work , design and analysis of a unified mdium access control ( mac ) protocol incorporating mode selection , device discovery , and such relay - aided d2d communication in the context of lte - a network will be worth investigating . note that any arbitrary matching is not necessarily stable . in the following , we show that for any given preference profiles , each iteration of * algorithm [ alg : joint_alloc ] * ends up with a stable matching ( i.e. , there is no blocking pair ) . we prove the proposition by contradiction . let @xmath127 be a matching obtained by * algorithm [ alg : rb_alloc ] * at any step @xmath87 of * algorithm [ alg : joint_alloc]*. let us assume that rb @xmath14 is not allocated to ue @xmath7 , but it has a higher order in the preference list . according to this assumption , the @xmath190 pair will block @xmath127 . since the position of @xmath7 in the preference profile of @xmath14 is higher compared to the user @xmath272 that is matched by @xmath127 , i.e. , @xmath273 , rb @xmath14 must select @xmath7 before the algorithm terminates . however , the pair @xmath190 does not match each other in the matching outcome @xmath127 . this implies that @xmath7 rejects @xmath14 ( e.g. , line @xmath209 in * algorithm [ alg : rb_alloc ] * ) and @xmath274 is a better assignment . as a result , the pair @xmath190 will not block @xmath127 , which contradicts our assumption . consequently , the matching outcome @xmath127 leads to a stable matching since no blocking pair exists and the proof concludes . the proof is followed by the induction of number of users @xmath275 , that are supported by relay @xmath6 . for instance , let @xmath276 ( the proof for @xmath277 can be done analogously introducing dummy rows [ i.e. , ues ] in the utility matrix ) . the basis ( i.e. , @xmath278 ) is trivial , since the only user definitely gets the best rbs according to her preference . when @xmath279 , let us consider @xmath280 to be the maximal entity of the utility matrix @xmath144 . for instance , let the matrix @xmath281 be obtained by removing the @xmath19-th row and @xmath20-th column from the utility matrix @xmath144 . if @xmath127 is a stable matching for @xmath144 , then by definition @xmath282 and hence @xmath283 must be a stable matching for @xmath281 . by induction , there exists a unique stable matching @xmath284 for the smaller matrix @xmath281 . therefore , the proof is concluded due to the fact that @xmath285 is the unique stable matching for the utility matrix @xmath144 . without loss of generality , let @xmath286 denote the data rate achieved by ue @xmath7 for any matching @xmath127 for given uncertainty bounds and @xmath287 is the sum - rate of all ues . on the contrary , let @xmath284 denote an arbitrary unstable outcome better than @xmath127 , i.e. , @xmath284 can achieve a better sum - rate . there are two cases that make @xmath284 unstable : _ case 1 ( lack of individual rationality ) : _ if rb @xmath14 is not individually rational , then the utility of @xmath14 can be improved by removing user @xmath288 with any arbitrarily user @xmath289 . hence , the utility of @xmath7 increases and @xmath290 . if condition _ ( i ) _ is true , we can obtain a stable matching @xmath127 by interchanging @xmath14 and @xmath293 for @xmath7 as follows : @xmath294 hence , the new data rate of ue @xmath7 is @xmath295 where @xmath147 is given by ( [ eq : rate_sm_uncrt ] ) . since @xmath7 strictly prefers rb @xmath14 to @xmath293 and the data rates for other ues remain unchanged , for condition _ ( i ) _ , it can be shown that @xmath296 . when condition _ ( ii ) _ is true , @xmath297 let @xmath298 with data rate @xmath299 . then @xmath300 from ( [ eq : sm_case21 ] ) and ( [ eq : sm_case22 ] ) , neither @xmath301 nor @xmath302 . since for both cases _ 1 ) _ and _ 2 ) _ there is no outcome @xmath284 better than @xmath127 , by * definition [ def : pareto ] * , @xmath127 is an optimal allocation and the proof follows . m. hasan and e. hossain , `` resource allocation for network - integrated device - to - device communications using smart relays , '' in _ ieee global communications conference workshops ( gc wkshps ) _ , 2013 , pp . 597602 . m. hasan , e. hossain , and d. i. kim , `` resource allocation under channel uncertainties for relay - aided device - to - device communication underlaying lte - a cellular networks , '' _ ieee transactions on wireless communications _ , vol . 13 , no . 4 , pp . 23222338 , april 2014 . d. i. kim , w. choi , h. seo , and b .- h . kim , `` partial information relaying and relaying in 3gpp lte , '' in _ cooperative cellular wireless networks_.1em plus 0.5em minus 0.4emcambridge university press , 2011 . m. zulhasnine , c. huang , and a. srinivasan , `` efficient resource allocation for device - to - device communication underlaying lte network , '' in _ ieee 6th international conference on wireless and mobile computing , networking and communications ( wimob ) _ , 2010 , pp . 368375 . p. phunchongharn , e. hossain , and d. kim , `` resource allocation for device - to - device communications underlaying lte - advanced networks , '' _ ieee wireless communications _ , vol . 20 , no . 4 , pp . 91100 , 2013 . y. pei and y .- c . liang , `` resource allocation for device - to - device communications overlaying two - way cellular networks , '' _ ieee transactions on wireless communications _ , vol . 12 , no . 7 , pp . 36113621 , 2013 . r. zhang , x. cheng , l. yang , and b. jiao , `` interference - aware graph based resource sharing for device - to - device communications underlaying cellular networks , '' in _ ieee wireless communications and networking conference ( wcnc ) _ , 2013 , pp . 140145 . l. su , y. ji , p. wang , and f. liu , `` resource allocation using particle swarm optimization for d2d communication underlay of cellular networks , '' in _ ieee wireless communications and networking conference ( wcnc ) _ , 2013 , pp . 129133 . x. ma , r. yin , g. yu , and z. zhang , `` a distributed relay selection method for relay assisted device - to - device communication system , '' in _ ieee 23rd international symposium on personal indoor and mobile radio communications ( pimrc ) _ , sept 2012 , pp . 10201024 . d. lee , s .- kim , j. lee , and j. heo , `` performance of multihop decode - and - forward relaying assisted device - to - device communication underlaying cellular networks , '' in _ international symposium on information theory and its applications ( isita ) _ , 2012 , pp . 455459 . c. an , l. zhang , and w. liu , `` a spectrum allocation algorithm based on matching game , '' in _ 5th international conference on wireless communications , networking and mobile computing ( wicom ) _ , sept 2009 , pp . 13 . x. feng , g. sun , x. gan , f. yang , x. tian , x. wang , and m. guizani , `` cooperative spectrum sharing in cognitive radio networks : a distributed matching approach , '' _ ieee transactions on communications _ , 2014 . o. semiari , w. saad , s. valentin , m. bennis , and b. maham , `` matching theory for priority - based cell association in the downlink of wireless small cell networks , '' in _ ieee 39th international conference on acoustics , speech and signal processing ( icassp ) _ , 2014 . a. el - hajj , z. dawy , and w. saad , `` a stable matching game for joint uplink / downlink resource allocation in ofdma wireless networks , '' in _ ieee international conference on communications ( icc ) _ , june 2012 , pp . 53545359 . b. di , s. bayat , l. song , and y. li , `` radio resource allocation for full - duplex ofdma networks using matching theory , '' in _ ieee conference on computer communications workshops ( infocom wkshps ) _ , april 2014 , pp . 197198 . g. fodor , e. dahlman , g. mildh , s. parkvall , n. reider , g. mikls , and z. turnyi , `` design aspects of network assisted device - to - device communications , '' _ ieee communications magazine _ , vol . 50 , no . 3 , pp . 170177 , march 2012 . m. sikora , j. laneman , m. haenggi , d. costello , and t. fuja , `` bandwidth- and power - efficient routing in linear wireless networks , '' _ ieee transactions on information theory _ , 52 , no . 6 , pp . 26242633 , 2006 . a. abdelnasser and e. hossain , `` joint subchannel and power allocation in two - tier ofdma hetnets with clustered femtocells , '' in _ ieee international conference on communications ( icc ) _ , june 2013 , pp . 60026007 . z. shen , j. andrews , and b. evans , `` adaptive resource allocation in multiuser ofdm systems with proportional rate constraints , '' _ ieee transactions on wireless communications _ , vol . 4 , no . 6 , pp . 27262737 , 2005 . m. tao , y .- c . liang , and f. zhang , `` resource allocation for delay differentiated traffic in multiuser ofdm systems , '' _ ieee transactions on wireless communications _ , vol . 7 , no . 6 , pp . 21902201 , 2008 . a. j. g. anandkumar , a. anandkumar , s. lambotharan , and j. chambers , `` robust rate - maximization game under bounded channel uncertainty , '' in _ ieee international conference on acoustics speech and signal processing ( icassp ) _ , march 2010 , pp . 31583161 . b. rastegari , a. condon , n. immorlica , and k. leyton - brown , `` two - sided matching with partial information , '' in _ 14th acm conference on electronic commerce_.1em plus 0.5em minus 0.4emnew york , ny , usa : acm , 2013 , pp . 733750 . f. berggren , r. jantti , and s .- l . kim , `` a generalized algorithm for constrained power control with capability of temporary removal , '' _ ieee transactions on vehicular technology _ 50 , no . 6 , pp . 16041612 , 2001 . h. zhang , c. jiang , n. beaulieu , x. chu , x. wen , and m. tao , `` resource allocation in spectrum - sharing ofdma femtocells with heterogeneous services , '' _ ieee transactions on communications _ , vol . 62 , no . 7 , pp . 23662377 , july 2014 .

non - supersymmetric so(10 ) grand unified theories provide a framework in which the stability of dark matter is explained while gauge coupling unification is realized . in this work , we systematically study this possibility by classifying weakly interacting dark matter candidates in terms of their quantum numbers of , , and . we consider both scalar and fermion candidates . we show that the requirement of a sufficiently high unification scale to ensure a proton lifetime compatible with experimental constraints plays a strong role in selecting viable candidates . among the scalar candidates originating from either a * 16 * or * 144 * of so(10 ) , only su(2) singlets with zero hypercharge or doublets with satisfy all constraints for and intermediate scale gauge groups . among fermion triplets with zero hypercharge , only a triplet in the * 45 * with intermediate group leads to solutions with and a long proton lifetime . we find three models with weak doublets and as dark matter candidates for the and intermediate scale gauge groups assuming a minimal higgs content . we also discuss how these models may be tested at accelerators and in dark matter detection experiments . minn15/40 + umn th3501/15 + ipmu150142 + natsumi nagata , keith a. olive , and jiaming zheng 0.5 cm _william i. fine theoretical physics institute , school of physics and astronomy , university of minnesota , minneapolis , mn 55455 , usa _ + _ kavli ipmu ( wpi ) , utias , university of tokyo , kashiwa , chiba 277 - 8583 , japan _ 1.5 cm

1509.00809

various cosmological observations have now established that more than 80% of the energy density of matter in the universe is composed of non - baryonic dark matter ( dm ) @xcite . the standard model ( sm ) of particle physics , however , can not explain this observation , and therefore there should be new physics beyond the sm which contains a dm candidate . one of the most promising class of candidates for dm is the so - called weakly - interacting massive particle ( wimp ) . these are electrically neutral and colorless particles which have masses of @xmath15 gev and couple to sm particles via weak - scale interactions . their thermal relic abundance can explain the current energy density of dm . such particles are predicted in many new - physics models ; for example , the lightest neutralino in the supersymmetric ( susy ) sm is a well - known candidate for wimp dm @xcite . for a wimp to be dm , it should be stable or have a sufficiently long lifetime compared to the age of the universe . to assure that , it is usually assumed that there is a symmetry which stabilizes the dm particle . for instance , in the minimal susy sm ( mssm ) , @xmath16-parity makes the lightest susy particle stable and thus a candidate for dm in the universe @xcite . similarly , kaluza - klein parity in universal extra dimensional models @xcite and @xmath17-parity in the littlest higgs model @xcite yield stable particles , which can also be promising dm candidates . the ultraviolet ( uv ) origin of such a symmetry is , however , often obscure ; thus it would be quite interesting if a theory which offers a dm candidate and simultaneously explains its stability can be realized as a uv completion rather than introducing the additional symmetry by hand . in fact , grand unified theories ( guts ) can provide such a framework . suppose that the rank of a gut gauge group is larger than four . in this case , the gut symmetry contains extra symmetries beyond the sm gauge symmetry . these extra symmetries should be spontaneously broken at a high - energy scale by a vacuum expectation value ( vev ) of a higgs field . then , if we choose the proper representation for the higgs field , there remains discrete symmetries , which can be used for dm stabilization @xcite . the discrete charge of each representation is uniquely determined , and thus we can systematically identify possible dm candidates for each symmetry . in this work , we discuss the concrete realization of this scenario in non - susy so(10 ) gut models . it is widely known that so(10 ) guts @xcite have a lot of attractive features . firstly , all of the sm quarks and leptons , as well as right - handed neutrinos , can be embedded into @xmath18 representations of so(10 ) . secondly , the anomaly cancellation in the sm is naturally explained since so(10 ) is free from anomalies . thirdly , one obtains improved gauge coupling unification @xcite and improved fermion mass ratios @xcite if partial unification is achieved at an intermediate mass scale . in addition , since right - handed neutrinos have masses of the order of the intermediate scale , small neutrino masses can be explained via the seesaw mechanism @xcite if the intermediate scale is sufficiently high . so(10 ) includes an additional u(1 ) symmetry , which is assumed to be broken at the intermediate scale . if the higgs field that breaks this additional u(1 ) symmetry belongs to a @xmath19 dimensional representation , then a discrete @xmath20 symmetry is preserved at low energies . one also finds that as long as we focus on relatively small representations ( @xmath21 ) , the @xmath22 higgs field leaving a @xmath20 symmetry is the only possibility for a discrete symmetry @xcite . we focus on this case in the following discussion . dm candidates appearing in such models can be classified into two types ; one class of dm particles have effectively weak - scale interactions with the sm particles so that they are thermalized in the early universe , while the other class contains sm singlets which are never brought into thermal equilibrium . in the latter case , dm particles are produced out of equilibrium via the thermal scattering involving heavy ( intermediate scale ) particle exchange processes . this type of dm is called non - equilibrium thermal dm ( netdm ) @xcite , whose realization in so(10 ) guts was thoroughly discussed in ref . netdm is necessarily fermionic as scalar dm would naturally couple to the sm higgs bosons . depending on the choice of the intermediate - scale gauge group , candidates for netdm may originate from several different so(10 ) representations such as @xmath23 or @xmath24 . although the netdm candidate itself does not affect the running of the gauge couplings from the weak scale to the intermediate scale , part of the original so(10 ) multiplet has a mass at the intermediate scale and does affect the running up to the gut scale . demanding gauge coupling unification with a gut scale above @xmath25 gev leaves us with a limited set of potential netdm candidates . when we further demand the splitting of the gut scale or intermediate scale multiplets so that only a singlet survives at low energies only two candidates were left : in @xmath26 notations there were the dirac ( @xmath27 ) originating in a @xmath28 or a weyl ( @xmath29 ) also originating from a @xmath28 or a @xmath24 in so(10 ) . in this paper , we study the remaining possibility , namely , that of wimp dm candidates in so(10 ) gut models . we systematically classify wimp dm candidates in terms of their quantum numbers and embed them in so(10 ) representations . as noted above and discussed in ref . @xcite , the presence of dm multiplets significantly affects the running of the gauge coupling constants . in this case , since the dm candidates are no longer sm singlets , the running of the gauge couplings is also affected between the weak and intermediate scales and may spoil gauge coupling unification realized in non - susy so(10 ) guts . we list wimp dm models in which gauge coupling unification is achieved with appropriate gut and intermediate scales . then , we study the phenomenology of these models , such as the relic abundance of dm , the dm direct detection rate , the proton decay lifetime , and neutrino masses . it is found that the condition of gauge coupling unification , as well as the proton decay bounds , severely restricts the wimp dm models in so(10 ) guts . still , we obtain some promising candidates , which can be probed in future dm searches and proton decay experiments . this paper is organized as follows . in the next section , we show the model setup for the so(10 ) wimp dm scenario . the realization of a @xmath20 symmetry and the classification of wimp dm candidates are discussed there . then , we analyze the scalar and fermionic dm models in sec . [ sec : scalardm ] and sec . [ sec : fermiondm ] , respectively . section [ sec : conclusion ] is devoted to conclusion and discussion . we begin with an overview of the basic so(10 ) model needed to accommodate a dm candidate . as mentioned above , in this work , we consider so(10 ) gut models and restrict ourselves to a two step simultaneous symmetry breaking chain , in which the so(10 ) gauge group is broken to an intermediate gauge group @xmath30 at the gut scale @xmath31 , and subsequently broken to the sm gauge group @xmath32 and a @xmath20 symmetry at the intermediate scale @xmath33 : @xmath34 the higgs multiplets that break so(10 ) and @xmath30 are labeled by @xmath35 and @xmath36 , respectively . as discussed in the introduction , this @xmath20 symmetry is a remnant of an extra u(1 ) symmetry in so(10 ) @xcite and is used to stabilize dm candidates @xcite . a brief introduction to the intermediate subgroups and @xmath20 symmetry will be given in sec . [ sec : intdisc ] . possible so(10 ) multiplets that contain an electric and color neutral component for a wimp dm candidate are summarized in sec . [ sec : wimp ] . for a group theoretical argument on the classification of these dm candidates , see appendix [ app : dmcandidates ] . among them , those who have a non - zero hypercharge are severely restricted by the dm direct search experiments . we consider this class of dm candidates in sec . [ sec : ydmbound ] and discuss conditions for the dm models to evade the direct search bound . to keep our model concise , in the following discussion , we only consider so(10 ) irreducible representations with dimensions up to 210 . we start by giving a brief description of the ingredients in our model . in an so(10 ) unification theory , a generation of sm fermions and a right - handed neutrino are embedded in a @xmath18 chiral representation , while the sm higgs boson usually lies in a @xmath37 representation . to obtain a realistic yukawa sector , it is necessary to take the @xmath38 to be complex @xcite . we will keep this sector unchanged in most of what follows . in addition to the sm particles , the @xmath35 and @xmath36 higgs representations are added to break so(10 ) and @xmath39 , respectively . the last ingredient of our model is the dm multiplet , whose lightest component is targeted to be the dm in the universe . the stability of the dm is guaranteed by a remnant @xmath20 symmetry of the extra u(1 ) gauge symmetry of so(10 ) as we will discuss soon . possible representations for the dm multiplet are determined below . here , we assume that only a minimal set of the higgs and dm multiplets which are necessary for the symmetry breaking and mass generation of dm lie in the low - energy regime and other components have masses of the order of the symmetry breaking scale at which their masses are generated . for example , among the @xmath38 representation , only the electroweak doublet components remains light to break the electroweak symmetry , while the other components have gut - scale masses . also , to obtain the right relic abundance , the mass of the dm particle is taken to be of order the tev scale , while the masses of the other components are either of @xmath40 or @xmath41 . such a hierarchical mass spectrum is obtained with fine - tunings similar to the doublet - triplet splitting needed for the @xmath38 . in principle , it may be possible to achieve the splitting of the dm multiplets with a more elaborate scheme of particle representations as in the higgs doublet - triplet separation @xcite . so(10 ) is a rank - five group so it contains an additional u(1 ) symmetry besides the sm gauge symmetry . this additional u(1 ) can be broken into a @xmath20 symmetry by a vev of an appropriate higgs field . if we restrict our attention to representation of dimension @xmath24 or less , the only choice of an irreducible @xmath36 that ensures the @xmath20 symmetry is a @xmath22 .. in addition , we note that a @xmath42 symmetry can be obtained by a * 672 * higgs field @xcite . ] this @xmath20 symmetry is equivalent to matter parity @xmath43@xcite , under which the sm fermions are odd , while the sm higgs field is even . thus a fermion ( boson ) is stable if it has an even ( odd ) matter parity . we list in table [ tab : intgauge ] all possible rank - five subgroups and corresponding @xmath35 whose vev breaks so(10 ) to the subgroups . here @xmath44 denotes the so - called @xmath44-parity @xcite , that is , a @xmath20 symmetry with respect to the exchange of su(2)@xmath45 . @xmath44-parity can be related to an element of so(10 ) @xcite under which a fermion field transforms into its charge conjugate . in cases where @xmath44-parity is not broken by @xmath35 , it is subsequently broken by @xmath36 at the intermediate scale @xmath33 . in this work we only consider the subgroups without an explicit su(5 ) factor . since the dm is necessarily a color singlet , the running of the strong gauge coupling is unaltered by the presence of a new dm particle below the intermediate scale . thus even though the addition of a dm multiplet yields unification of the gauge couplings , the unification scale @xmath33 is always less than @xmath46 gev as the contribution to the u(1)@xmath47 beta function is always positive . if we now associate @xmath33 with su(5 ) , this low partial unification is heavily disfavored on the basis of proton lifetime constraints . flipped su(5 ) usually has a high intermediate scale and a high gut scale close to the planck scale . in this case higher dimension operators suppressed by planck scale become important , and one may also need to rely on a double seesaw for the explanation of neutrino masses @xcite . these bring extra complication into our model and we do not consider these possibilities here . other intermediate gauge groups in the table are subgroups of @xmath48 , and @xmath49 is a subgroup of @xmath50 . the relationship among hypercharge @xmath51 , the u(1)@xmath52 charge @xmath1 , and the third component of the su(2)@xmath53 generators @xmath54 is very useful for determining the quantum numbers of dm candidates : @xmath55 the convention we are using for hypercharge is such that electric charge is given by @xmath56 , with @xmath57 denoting the third component of the su(2)@xmath3 generators . .__candidates for the intermediate gauge group @xmath30 . _ _ [ cols="<,<",options="header " , ] after an exhaustive calculation for different choices of @xmath58 and @xmath39 , we find several possible models listed in table [ tab : nethyperchargedmodels ] that survive the direct detection and proton decay constraints . viable minimal models only exist for doublet dm when @xmath59 , @xmath60 and @xmath61 . the intermediate scale of each model is larger than @xmath62 gev , large enough to evade the direct detection bound indicated by eq . . as can be seen , most of the models in table [ tab : nethyperchargedmodels ] can be probed in future proton decay experiments , though it is hard to detect these dm candidates in direct detection experiments . the success of the standard model is now well established . nevertheless , we know that the standard model is incomplete . neutrinos have masses , there is a non - zero baryon asymmetry in the universe , and dark matter makes up a sizable component of the total matter density . many extensions to the standard model have been studied to explain these phenomena . but rarely can a single extension explain all three . so(10 ) grand unification is one such example . in most models , so(10 ) symmetry breaking involves an intermediate gauge group , whose unknown scale presumably lies between the weak scale and the grand unified scale ( defined by the renormalization scale where the gauge couplings are equal ) . standard model fermions are neatly contained in a * 16 * of so(10 ) which includes all of the known sm fermions plus a right - handed neutrino per generation . as the right - handed neutrino is a sm singlet , it easily picks up a mass of order the intermediate scale during the second phase of symmetry breaking , and leads to a natural realization of the see - saw mechanism @xcite for the generation of neutrino masses . if produced ( thermally or non - thermally ) after inflation , these same right - handed neutrinos can decay out of equilibrium and produce a lepton asymmetry which can be converted through electroweak effects to a baryon asymmetry , a process known as leptogenesis @xcite . furthermore the existence of an intermediate scale makes gauge coupling unification feasible in the absence of supersymmetry . here we studied in detail one of the often unheralded features of so(10 ) grand unification . namely its ability to provide for a wimp dark matter candidate in addition to the benefits described above . since so(10 ) includes an additional u(1 ) symmetry beyond hypercharge , if the higgs field that breaks this symmetry at the intermediate scale belongs to a @xmath19 dimensional representation , then a discrete @xmath20 symmetry is preserved at low energies . this discrete symmetry ( equivalent to matter parity @xmath43 ) naturally provides the stability for dark matter candidates . we considered all possible intermediate gauge group with broken su(5 ) . stable so(10 ) scalar ( fermion ) dm candidates must be odd ( even ) under the @xmath20 symmetry . therefore fermions must originate in either a * 10 * , * 45 * , * 54 * , * 120 * , * 126 * , * 210 * or * 210@xmath63 * representation , while scalars are restricted to either a * 16 * or * 144 * of so(10 ) . these multiplets must be split and we gave explicit examples of fine - tuning mechanisms in order to retain a 1 tev wimp candidate which may be a su(2)@xmath3 , singlet , doublet , triplet , or quartet with or without hypercharge . fermions which are su(2)@xmath3 singlets with no hypercharge are not good wimp candidates but are netdm candidates and these were considered elsewhere @xcite . our criteria for a viable dark matter model required : gauge coupling unification at a sufficiently high scale to ensure proton stability compatible with experiment ; a unification scale greater than the intermediate scale ; and elastic cross sections compatible with direct detection experiments . the latter criterion often requires additional higgs representations to split the degeneracy of the fermionic intermediate scale representations if dm is hypercharged . despite the potential very long list of candidates ( when one combines the possible different so(10 ) representations and intermediate gauge groups ) , we found only a handful of models which satisfied all constraints . among the scalar candidates , the @xmath64 singlet and @xmath4 doublet ( often referred to as an inert higgs doublet @xcite ) are possible candidates for @xmath65 and @xmath66 ( with or without a left - right symmetry ) intermediate gauge groups . these originate from either the * 16 * or * 144 * of so(10 ) . the latter group ( without the left - right symmetry ) is also consistent with a state originating from the * 144 * being a triplet under su(2)@xmath53 . to avoid immediate exclusion from direct detection experiments , a mass splitting of order 100 kev implies that the intermediate scale must be larger than about 3 @xmath67 @xmath68 for a nominal 1 tev hypercharged scalar dm particle . some of these models imply proton lifetimes short enough to be testable in on - going and future proton decay experiments . the fermion candidates were even more restrictive . models with @xmath64 must come from a su(2)@xmath3 triplet ( singlets are not wimps ) . in this case only one model was found using the @xmath65 intermediate gauge group and requiring additional higgses ( already present in @xmath35 ) at the intermediate scale . models with @xmath4 doublets were found for @xmath69 with a singlet fermion required for mixing , and @xmath65 with a triplet fermion for mixing . in both cases , additional higgses from @xmath35 are required at the intermediate scale . more possibilities can be found if the additional higgs are taken outside @xmath35 . so(10 ) almost always involves rather large representations ( at least when compared with minimal su(5 ) for example ) . so(10 ) itself can be broken by either a * 45 * , * 54 * or * 210 * representation ( @xmath35 ) which determines the intermediate scale gauge group and a * 126 * is needed to break down to the sm ( and preserve the needed @xmath20 symmetry ) . the sm higgs originates from a * 10 * and matter fields reside in three copies of * 16 * s . one additional representation is needed to account for dm . we have delineated the possible representations and necessary intermediate gauge groups needed to account for wimp - like dark matter , proton stability , and gauge coupling unification . some of these models are accessible for experimental tests . we thank yann mambrini and mikhail b. voloshin for valuable discussions . the work of n.n . is supported by research fellowships of the japan society for the promotion of science for young scientists . the work of k.a.o . was supported in part by doe grant de - sc0011842 at the university of minnesota . here we give a group theoretical argument to classify possible dm candidates in so(10 ) models . we basically follow the notation of ref . @xcite in the following discussion . see also refs . @xcite . since so(10 ) is a rank - five group , we have five linearly independent cartan generators . we denote them by @xmath70 @xmath71 . in the dual basis , they are expressed in terms of five - dimensional vectors as follows : @xmath72 ~ , \nonumber \\ h_2 & = \frac{1}{2\sqrt{3}}[1~0~0~-1~1 ] ~ , \nonumber \\ h_3 & = \frac{1}{2}[0~0~1~1~1 ] ~ , \nonumber \\ h_4 & = \frac{1}{6}[-2~0~3~-1~1 ] ~ , \nonumber \\ h_5 & = [ 2~0~2~1~-1 ] ~.\end{aligned}\ ] ] here , @xmath73 and @xmath74 are the su(3)@xmath75 cartan generators . @xmath76 and @xmath77 are the weak isospin and hypercharge , @xmath78 and @xmath51 , respectively . @xmath79 is given by @xmath80 . every component of an so(10 ) multiplet is specified by a weight vector @xmath81 , which is expressed by a set of five integers called dynkin labels as @xmath82 . its eigenvalues of @xmath70 are given by @xmath83 with @xmath84 $ ] . the dm particle should have zero eigenvalues of @xmath73 , @xmath74 , and @xmath85 . this condition is satisfied by the following set of weight vectors characterized by two integers @xmath86 and @xmath87 : @xmath88 the hypercharge and @xmath1 charge of the weight vector are @xmath89 we find that the @xmath90 cases correspond to the hypercharged dm candidates . for @xmath91 , the weight vector agrees to @xmath92 discussed in ref . @xcite . a convenient way to determine the matter parity of the dm candidates @xmath93 is to consider congruence classes of so(10 ) @xcite . two representations @xmath16 and @xmath94 of so(10 ) belong to the same congruence class if the difference between any weights of @xmath16 and @xmath94 is always written as a linear combination of simple roots of so(10 ) with integer coefficients @xcite . so(10 ) has four congruence classes . each congruence class is characterized by a congruence number . the congruence number of a weight vector @xmath95 is defined by @xmath96 for @xmath93 , we thus have @xmath97 then , we find that the congruence number of @xmath93 is @xmath98 if and only if @xmath86 is an odd integer . considering eq . , we conclude that a dm particle has the odd matter parity if it has a congruence number of @xmath98 , while it has the even matter parity if its congruence number is @xmath99 or @xmath100 . in this section , we describe how we calculate proton decay lifetimes in the intermediate - scale scenarios . in these scenarios , proton decay is induced by the exchange of the gut - scale gauge bosons @xcite . the relevant part of the so(10 ) gauge interaction is given by @xmath101 ~,\end{aligned}\ ] ] where @xmath102 @xmath103 represents the gut gauge bosons which induce proton decay , @xmath104 is the unified gauge coupling constant , @xmath105 are su(3)@xmath75 indices , @xmath106 are su(2)@xmath3 indices , @xmath107 are su(2)@xmath53 indices , and @xmath108 are the chirality projection operators . the exchange of the @xmath103 fields generates dimension - six proton decay operators . these operators are expressed in a form that respects the intermediate gauge symmetries . between the gut and intermediate scales , the renormalization factors for the effective operators are in general different among the choices of @xmath30 . below the intermediate scale , the low - energy effective theory is described by the @xmath109 gauge theory , and thus after matching the theories above and below the intermediate scale , the prescription for the calculation is common to all of the cases . for this reason , we first describe the calculation below the intermediate scale . after that , we discuss each intermediate gauge theory showing the matching conditions at the gut and intermediate scales as well as the rges between them . in the @xmath109 gauge theory , the effective lagrangian for proton decay is generically written as @xmath110 with the effective operators given by @xcite @xmath111 up to dimension six . we then run down the coefficients to the electroweak scale . we will see below that the coefficients @xmath112 and @xmath113 vanish in all of the cases we consider in this paper , and thus we focus on @xmath114 and @xmath115 . their renormalization factors are @xcite @xmath116^{-\frac{2}{b_3 } } \biggl[\frac{\alpha_2(\mu)}{\alpha_2 ( m_{\text{int}})}\biggr]^{-\frac{9}{4b_2 } } \biggl[\frac{\alpha_1(\mu)}{\alpha_1 ( m_{\text{int}})}\biggr]^{-\frac{11}{20b_1 } } c_1(m_{\text{int } } ) ~ , \\ c_2(\mu ) & = \biggl[\frac{\alpha_3(\mu)}{\alpha_3 ( m_{\text{int}})}\biggr]^{-\frac{2}{b_3 } } \biggl[\frac{\alpha_2(\mu)}{\alpha_2 ( m_{\text{int}})}\biggr]^{-\frac{9}{4b_2 } } \biggl[\frac{\alpha_1(\mu)}{\alpha_1 ( m_{\text{int}})}\biggr]^{-\frac{23}{20b_1 } } c_2(m_{\text{int } } ) ~,\end{aligned}\ ] ] where @xmath117 denote the one - loop beta - function coefficients for the gauge couplings @xmath118 and @xmath119 is an arbitrary scale . we need to change the beta - function coefficients appropriately when we across the dm mass threshold . below the electroweak scale , the qcd corrections are the dominant contribution . by using the two - loop rge given in ref . @xcite , we compute the wilson coefficients at the hadronic scale @xmath120 as @xmath121^{\frac{6}{25}}\biggl [ \frac{\alpha_s(m_b)}{\alpha_s(m_z ) } \biggr]^{\frac{6}{23 } } \biggl [ \frac{\alpha_s(\mu_{\text{had}})+\frac{50\pi}{77 } } { \alpha_s(m_b)+\frac{50\pi}{77 } } \biggr]^{-\frac{173}{825 } } \biggl [ \frac{\alpha_s(m_b)+\frac{23\pi}{29 } } { \alpha_s(m_z)+\frac{23\pi}{29 } } \biggr]^{-\frac{430}{2001}}c_i(m_z)~,\ ] ] with @xmath122 . in non - susy guts , the dominant decay mode of proton is @xmath123 . the partial decay width of the mode is computed as @xmath124 ~,\ ] ] where @xmath125 and @xmath126 are the masses of the proton and the neutral pion , respectively , and @xmath127 the hadron matrix elements are evaluated with the lattice qcd simulations in ref . we have @xmath128 with @xmath129 gev . here , the first and second parentheses indicate statistical and systematic errors , respectively . for @xmath130 , the dimension - six effective operator is given by ] and thus the operator in eq . is the unique choice . ] @xmath131 where @xmath132 denote the su(4 ) indices , and the dirac field @xmath133 is defined by @xmath134 here , the indices represent the su(3)@xmath75 color and @xmath135 indicates charge conjugation . at tree level , the coefficient of the effective operator is evaluated as @xmath136 with @xmath137 the mass of the heavy gauge field @xmath103 . in this paper , we neglect fermion flavor mixings @xcite for simplicity . the wilson coefficient is evolved down to the intermediate scale using the rge . the renormalization factor is computed to be @xcite @xmath138 ^{-\frac{15}{4b_4}}\biggl[\frac{\alpha_{2l } ( m_{\text{int}})}{\alpha_{\text{gut}}}\biggr]^{-\frac{9}{4b_{2l } } } \biggl[\frac{\alpha_{2r}(m_{\text{int}})}{\alpha_\text{gut}}\biggr ] ^{-\frac{9}{4b_{2r } } } c_{422}(m_{\text{gut}})~.\ ] ] then , the effective operator is matched onto the operators in eq . . the wilson coefficients @xmath139 are given by @xmath140 in the case of @xmath142 , the effective lagrangian is written as @xmath143 with @xmath144 the gut - scale matching condition for the operator is @xmath145 and the renormalization factor is given by @xcite @xmath146 ^{-\frac{15}{4b_4}}\biggl[\frac{\alpha_{2l } ( m_{\text{int}})}{\alpha_{\text{gut}}}\biggr]^{-\frac{9}{4b_{2l } } } \biggl[\frac{\alpha_{r}(m_{\text{int}})}{\alpha_\text{gut}}\biggr ] ^{-\frac{3}{4b_{r } } } c_{421}(m_{\text{gut}})~.\ ] ] for the intermediate - scale matching conditions , we have @xmath147 when @xmath148 , there are four independent effective operators @xcite , @xmath149 and thus the effective lagrangian is expressed as @xmath150 for the gut - scale matching condition , we have @xmath151 the renormalization factors for the coefficients @xmath152 and @xmath153 are given in refs . @xcite : @xmath154 ^{-\frac{2}{b_3}}\biggl[\frac{\alpha_{2l } ( m_{\text{int}})}{\alpha_{\text{gut}}}\biggr]^{-\frac{9}{4b_{2l } } } \biggl[\frac{\alpha_{2r}(m_{\text{int}})}{\alpha_\text{gut}}\biggr ] ^{-\frac{9}{4b_{2r } } } \biggl[\frac{\alpha_{b - l}(m_{\text{int}})}{\alpha_\text{gut}}\biggr ] ^{-\frac{1}{4b_{b - l } } } ~,\ ] ] for @xmath155 and @xmath153 . then the wilson coefficients at the electroweak scale are matched onto those of the operators as @xmath156 to show the process of mass fine - tuning explicitly , in this section , we consider the case of @xmath157 with @xmath158 as an example . we take @xmath159 , which contains two independent sm singlet components that might develop vevs ; one is in a @xmath160 while the other is in a @xmath161 under @xmath162 . we refer to these vevs as @xmath163 and @xmath164 , respectively , and other notation is taken from eq . . since the components of a scalar @xmath18 have the same quantum numbers as those of a generation of the sm fermions , we denote them by the same symbol as for the corresponding sm fermions with a tilde , just like the notation for sfermions in supersymmetric models let us first study the @xmath165 coupling . since @xmath35 is the adjoint representation of so(10 ) , the decomposition of this coupling in terms of the component fields has a similar form to the gauge interaction for a @xmath18 spinor representation . we have @xmath166 ~ , \label{eq:45mass}\end{aligned}\ ] ] where the contraction of the su(3)@xmath75 and su(2)@xmath3 indices is implicit . when @xmath167 and @xmath168 , the mass spectrum preserves the @xmath169 symmetry , while when @xmath170 and @xmath171 , then it is @xmath172 symmetric . if both of the vevs have non - zero values , then the low - energy theory is invariant under the @xmath173 symmetry . the coefficients of @xmath164 for left and right doublets have different signs , which indicates the breaking of left - right symmetry . here , we choose @xmath174 and @xmath175 to obtain @xmath158 . next we consider the mass terms generated by @xmath176 . the sm singlet in @xmath177 transforms as @xmath178 under @xmath179 , which acquires a vev @xmath180 to break @xmath39 into the sm gauge group . according to the result in ref . @xcite , the resultant mass terms are contains a @xmath28 , there is a contribution to the mass corresponding to eq . at the intermediate scale proportional to @xmath181 with independent coefficients @xmath182 and @xmath183 . the result shown is obtained from eq . by taking @xmath184 and @xmath185 , up to an overall factor . ] @xmath186 ~.\end{aligned}\ ] ] notice that the right - hand side of the expression can be grouped in terms of su(5 ) multiplets . this is expected since @xmath180 is invariant under the su(5 ) transformation . from the above equations , it is found that we can ensure that only the dm component has a mass around tev scale by fine - tuning the parameters @xmath187 , @xmath188 and @xmath181 . for example , to obtain the model @xmath189 , we can take @xmath190 then , @xmath191 acquires a tev - scale mass , while the mass of @xmath192 is @xmath193 . the rest of the components lie around the gut scale . on the other hand , if we take @xmath194 then we can make only the @xmath195 component have a tev - scale mass and the other components have gut - scale masses . thus we obtain the @xmath196 model . to simplify our argument , in the above discussion , we have taken into account only the contribution of the @xmath187 , @xmath188 , and @xmath181 terms , and neglected that of the other terms in eq . . even in the presence of the other contributions , we can always perform a similar fine - tuning among the parameters to realize desired mass spectrum for our dm models . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1502.01589 [ astro-ph.co ] . h. goldberg , phys . * 50 * ( 1983 ) 1419 ; j. ellis , j.s . hagelin , d.v . nanopoulos , k.a . olive and m. srednicki , nucl . * b238 * ( 1984 ) 453 . t. appelquist , h. c. cheng and b. a. dobrescu , phys . d * 64 * , 035002 ( 2001 ) [ hep - ph/0012100 ] ; h. c. cheng , k. t. matchev and m. schmaltz , phys . d * 66 * , 036005 ( 2002 ) [ hep - ph/0204342 ] ; g. servant and t. m. p. tait , nucl . phys . b * 650 * , 391 ( 2003 ) [ hep - ph/0206071 ] ; h. c. cheng , j. l. feng and k. t. matchev , phys . lett . * 89 * , 211301 ( 2002 ) [ hep - ph/0207125 ] ; m. kakizaki , s. matsumoto and m. senami , phys . d * 74 * , 023504 ( 2006 ) [ hep - ph/0605280 ] ; g. belanger , m. kakizaki and a. pukhov , jcap * 1102 * , 009 ( 2011 ) [ arxiv:1012.2577 [ hep - ph ] ] . n. arkani - hamed , a. g. cohen and h. georgi , phys . b * 513 * , 232 ( 2001 ) [ hep - ph/0105239 ] ; n. arkani - hamed , a. g. cohen , e. katz , a. e. nelson , t. gregoire and j. g. wacker , jhep * 0208 * , 021 ( 2002 ) [ hep - ph/0206020 ] ; n. arkani - hamed , a. g. cohen , e. katz and a. e. nelson , jhep * 0207 * , 034 ( 2002 ) [ hep - ph/0206021 ] ; h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ hep - ph/0405243 ] ; i. low , jhep * 0410 * , 067 ( 2004 ) [ hep - ph/0409025 ] ; j. hubisz and p. meade , phys . rev . d * 71 * , 035016 ( 2005 ) [ hep - ph/0411264 ] ; a. birkedal , a. noble , m. perelstein and a. spray , phys . d * 74 * , 035002 ( 2006 ) [ hep - ph/0603077 ] . t. w. b. kibble , g. lazarides and q. shafi , phys . b * 113 * , 237 ( 1982 ) . l. m. krauss and f. wilczek , phys . lett . * 62 * , 1221 ( 1989 ) . l. e. ibanez and g. g. ross , phys . b * 260 * , 291 ( 1991 ) ; l. e. ibanez and g. g. ross , nucl . b * 368 * , 3 ( 1992 ) . s. p. martin , phys . d * 46 * , 2769 ( 1992 ) [ hep - ph/9207218 ] . m. kadastik , k. kannike and m. raidal , phys . d * 81 * , 015002 ( 2010 ) [ arxiv:0903.2475 [ hep - ph ] ] ; m. kadastik , k. kannike and m. raidal , phys . d * 80 * , 085020 ( 2009 ) [ erratum - ibid . d * 81 * , 029903 ( 2010 ) ] [ arxiv:0907.1894 [ hep - ph ] ] . m. frigerio and t. hambye , phys . d * 81 * , 075002 ( 2010 ) [ arxiv:0912.1545 [ hep - ph ] ] ; t. hambye , pos idm * 2010 * , 098 ( 2011 ) [ arxiv:1012.4587 [ hep - ph ] ] . h. georgi , aip conf . * 23 * , 575 ( 1975 ) ; h. fritzsch and p. minkowski , annals phys . * 93 * , 193 ( 1975 ) . m. s. chanowitz , j. r. ellis and m. k. gaillard , nucl . b * 128 * , 506 ( 1977 ) ; h. georgi and d. v. nanopoulos , nucl . b * 155 * , 52 ( 1979 ) . h. georgi and d. v. nanopoulos , nucl . b * 159 * , 16 ( 1979 ) ; c. e. vayonakis , phys . b * 82 * , 224 ( 1979 ) [ phys . lett . * 83b * , 421 ( 1979 ) ] . a. masiero , phys . b * 93 * , 295 ( 1980 ) . q. shafi , m. sondermann and c. wetterich , phys . b * 92 * , 304 ( 1980 ) . f. del aguila and l. e. ibanez , nucl . b * 177 * , 60 ( 1981 ) . r. n. mohapatra and g. senjanovic , phys . d * 27 * , 1601 ( 1983 ) . s. rajpoot , phys . d * 22 * , 2244 ( 1980 ) ; m. yasue , prog . * 65 * , 708 ( 1981 ) [ erratum - ibid . * 65 * , 1480 ( 1981 ) ] ; j. m. gipson and r. e. marshak , phys . d * 31 * , 1705 ( 1985 ) ; d. chang , r. n. mohapatra , j. gipson , r. e. marshak and m. k. parida , phys . d * 31 * , 1718 ( 1985 ) ; n. g. deshpande , e. keith and p. b. pal , phys . d * 46 * , 2261 ( 1993 ) ; n. g. deshpande , e. keith and p. b. pal , phys . d * 47 * , 2892 ( 1993 ) [ hep - ph/9211232 ] ; s. bertolini , l. di luzio and m. malinsky , phys . rev . d * 81 * , 035015 ( 2010 ) [ arxiv:0912.1796 [ hep - ph ] ] . m. fukugita and t. yanagida , in * fukugita , m. ( ed . ) , suzuki , a. ( ed . ) : physics and astrophysics of neutrinos * 1 - 248 . and kyoto univ . - yitp - k-1050 ( 93/12,rec.feb.94 ) 248 p. c. l. di luzio , arxiv:1110.3210 [ hep - ph ] . g. lazarides , q. shafi and c. wetterich , nucl . b * 181 * , 287 ( 1981 ) . p. minkowski , phys . b * 67 * , 421 ( 1977 ) ; t. yanagida , conf . c * 7902131 * , 95 ( 1979 ) ; m. gell - mann , p. ramond and r. slansky , conf . c * 790927 * , 315 ( 1979 ) [ arxiv:1306.4669 [ hep - th ] ] ; s. l. glashow , nato sci . b * 59 * , 687 ( 1980 ) ; r. n. mohapatra and g. senjanovic , phys . lett . * 44 * , 912 ( 1980 ) ; r. n. mohapatra and g. senjanovic , phys . d * 23 * , 165 ( 1981 ) . m. de montigny and m. masip , phys . d * 49 * , 3734 ( 1994 ) [ hep - ph/9309312 ] . y. mambrini , n. nagata , k. a. olive , j. quevillon and j. zheng , phys . d * 91 * , 095010 ( 2015 ) [ arxiv:1502.06929 [ hep - ph ] ] . b. bajc , a. melfo , g. senjanovic and f. vissani , phys . d * 73 * , 055001 ( 2006 ) [ hep - ph/0510139 ] . t. fukuyama , a. ilakovac , t. kikuchi , s. meljanac and n. okada , eur . j. c * 42 * , 191 ( 2005 ) [ hep - ph/0401213 ] . a. masiero , d. v. nanopoulos , k. tamvakis and t. yanagida , phys . b * 115 * , 380 ( 1982 ) ; b. grinstein , nucl . b * 206 * , 387 ( 1982 ) . s. dimopoulos and f. wilczek , print-81 - 0600 ( santa barbara ) , nsf - itp-82 - 07 . k. inoue , a. kakuto and h. takano , prog . * 75 * , 664 ( 1986 ) ; a. a. anselm and a. a. johansen , phys . b * 200 * , 331 ( 1988 ) . g. r. farrar and p. fayet , phys . b * 76 * , 575 ( 1978 ) ; s. dimopoulos and h. georgi , nucl . b * 193 * , 150 ( 1981 ) ; s. weinberg , phys . d * 26 * , 287 ( 1982 ) ; n. sakai and t. yanagida , nucl . b * 197 * , 533 ( 1982 ) ; s. dimopoulos , s. raby and f. wilczek , phys . b * 112 * , 133 ( 1982 ) . v. a. kuzmin and m. e. shaposhnikov , phys . b * 92 * , 115 ( 1980 ) ; t. w. b. kibble , g. lazarides and q. shafi , phys . d * 26 * , 435 ( 1982 ) ; d. chang , r. n. mohapatra and m. k. parida , phys . lett . * 52 * , 1072 ( 1984 ) ; d. chang , r. n. mohapatra and m. k. parida , phys . d * 30 * , 1052 ( 1984 ) ; d. chang , r. n. mohapatra , j. gipson , r. e. marshak and m. k. parida , phys . d * 31 * , 1718 ( 1985 ) . i. antoniadis , j. r. ellis , j. s. hagelin and d. v. nanopoulos , phys . b * 194 * , 231 ( 1987 ) . i. antoniadis , j. r. ellis , j. s. hagelin and d. v. nanopoulos , phys . b * 208 * , 209 ( 1988 ) [ addendum - ibid . b * 213 * , 562 ( 1988 ) ] ; j. r. ellis , j. l. lopez and d. v. nanopoulos , phys . lett . b * 292 * , 189 ( 1992 ) [ hep - ph/9207237 ] ; j. r. ellis , d. v. nanopoulos and k. a. olive , phys . lett . b * 300 * ( 1993 ) 121 [ hep - ph/9211325 ] ; j. r. ellis , j. l. lopez , d. v. nanopoulos and k. a. olive , phys . b * 308 * , 70 ( 1993 ) [ hep - ph/9303307 ] . r. slansky , phys . rept . * 79 * , 1 ( 1981 ) . m. cirelli , n. fornengo and a. strumia , nucl . b * 753 * , 178 ( 2006 ) [ hep - ph/0512090 ] ; m. cirelli , a. strumia and m. tamburini , nucl . b * 787 * , 152 ( 2007 ) [ arxiv:0706.4071 [ hep - ph ] ] ; m. cirelli and a. strumia , new j. phys . * 11 * , 105005 ( 2009 ) [ arxiv:0903.3381 [ hep - ph ] ] . r. essig , phys . d * 78 * , 015004 ( 2008 ) [ arxiv:0710.1668 [ hep - ph ] ] . t. hambye , f .- s . ling , l. lopez honorez and j. rocher , jhep * 0907 * , 090 ( 2009 ) [ erratum - ibid . * 1005 * , 066 ( 2010 ) ] . j. hisano , d. kobayashi , n. mori and e. senaha , phys . b * 742 * , 80 ( 2015 ) [ arxiv:1410.3569 [ hep - ph ] ] . n. nagata and s. shirai , jhep * 1501 * , 029 ( 2015 ) [ arxiv:1410.4549 [ hep - ph ] ] . n. nagata and s. shirai , phys . d * 91 * , 055035 ( 2015 ) [ arxiv:1411.0752 [ hep - ph ] ] . s. m. boucenna , m. b. krauss and e. nardi , phys . b * 748 * , 191 ( 2015 ) [ arxiv:1503.01119 [ hep - ph ] ] . k. harigaya , k. ichikawa , a. kundu , s. matsumoto and s. shirai , arxiv:1504.03402 [ hep - ph ] . j. heeck and s. patra , arxiv:1507.01584 [ hep - ph ] . m. cirelli , t. hambye , p. panci , f. sala and m. taoso , arxiv:1507.05519 [ hep - ph ] ; c. garcia - cely , a. ibarra , a. s. lamperstorfer and m. h. g. tytgat , arxiv:1507.05536 [ hep - ph ] . c. w. chiang and e. senaha , arxiv:1508.02891 [ hep - ph ] . b. feldstein , m. ibe and t. t. yanagida , phys . lett . * 112 * , 101301 ( 2014 ) [ arxiv:1310.7495 [ hep - ph ] ] . v. silveira and a. zee , phys . b * 161 * , 136 ( 1985 ) ; j. mcdonald , phys . rev . d * 50 * , 3637 ( 1994 ) [ hep - ph/0702143 ] ; c. p. burgess , m. pospelov and t. ter veldhuis , nucl . b * 619 * , 709 ( 2001 ) [ hep - ph/0011335 ] ; h. davoudiasl , r. kitano , t. li and h. murayama , phys . lett . b * 609 * , 117 ( 2005 ) [ hep - ph/0405097 ] . n. g. deshpande and e. ma , phys . d * 18 * , 2574 ( 1978 ) ; e. ma , phys . d * 73 * , 077301 ( 2006 ) [ hep - ph/0601225 ] ; r. barbieri , l. j. hall and v. s. rychkov , phys . d * 74 * , 015007 ( 2006 ) [ hep - ph/0603188 ] ; l. lopez honorez , e. nezri , j. f. oliver and m. h. g. tytgat , jcap * 0702 * , 028 ( 2007 ) [ hep - ph/0612275 ] . a. arhrib , y. l. s. tsai , q. yuan and t. c. yuan , jcap * 1406 * , 030 ( 2014 ) [ arxiv:1310.0358 [ hep - ph ] ] ; a. ilnicka , m. krawczyk and t. robens , arxiv:1508.01671 [ hep - ph ] . t. w. kephart and t. c. yuan , arxiv:1508.00673 [ hep - ph ] . m. farina , d. pappadopulo and a. strumia , jhep * 1308 * , 022 ( 2013 ) [ arxiv:1303.7244 [ hep - ph ] ] . j. hisano , s. matsumoto and m. m. nojiri , phys . lett . * 92 * , 031303 ( 2004 ) [ hep - ph/0307216 ] ; j. hisano , s. matsumoto , m. m. nojiri and o. saito , phys . d * 71 * , 063528 ( 2005 ) [ hep - ph/0412403 ] . m. shiozawa , talk presented at taup 2013 , september 813 , asilomar , ca , usa . k. s. babu , e. kearns , u. al - binni , s. banerjee , d. v. baxter , z. berezhiani , m. bergevin and s. bhattacharya _ et al . _ , arxiv:1311.5285 [ hep - ph ] . j. m. cline , k. kainulainen , p. scott and c. weniger , phys . d * 88 * , 055025 ( 2013 ) [ arxiv:1306.4710 [ hep - ph ] ] ; m. duerr , p. fileviez perez and j. smirnov , arxiv:1508.04418 [ hep - ph ] . t. abe , r. kitano and r. sato , phys . d * 91 * , no . 9 , 095004 ( 2015 ) [ arxiv:1411.1335 [ hep - ph ] ] . d. s. akerib _ et al . _ [ lux collaboration ] , phys . lett . * 112 * , 091303 ( 2014 ) [ arxiv:1310.8214 [ astro-ph.co ] ] . g. belanger , b. dumont , u. ellwanger , j. f. gunion and s. kraml , phys . d * 88 * , 075008 ( 2013 ) [ arxiv:1306.2941 [ hep - ph ] ] . j. hisano , s. matsumoto , m. nagai , o. saito and m. senami , phys . b * 646 * , 34 ( 2007 ) [ hep - ph/0610249 ] . g. lazarides , q. shafi and c. wetterich , nucl . b * 181 * , 287 ( 1981 ) ; k. s. babu and r. n. mohapatra , phys . lett . * 70 * , 2845 ( 1993 ) [ hep - ph/9209215 ] ; k. matsuda , y. koide and t. fukuyama , phys . d * 64 * , 053015 ( 2001 ) [ hep - ph/0010026 ] ; t. fukuyama , k. ichikawa and y. mimura , arxiv:1508.07078 [ hep - ph ] . v. khachatryan _ et al . _ [ cms collaboration ] , jhep * 1504 * , 025 ( 2015 ) [ arxiv:1412.6302 [ hep - ex ] ] ; g. aad _ et al . _ [ atlas collaboration ] , phys . d * 90 * , no . 5 , 052005 ( 2014 ) [ arxiv:1405.4123 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , arxiv:1508.04735 [ hep - ex ] ; cms collaboration [ cms collaboration ] , cms - pas - exo-12 - 041 ; [ cms collaboration ] , cms - pas - exo-12 - 042 . m. ibe , s. matsumoto and r. sato , phys . b * 721 * , 252 ( 2013 ) [ arxiv:1212.5989 [ hep - ph ] ] . g. aad _ et al . _ [ atlas collaboration ] , phys . d * 88 * , 112006 ( 2013 ) [ arxiv:1310.3675 [ hep - ex ] ] . k. shingo , cern - thesis-2014 - 163 ; m. low and l. t. wang , jhep * 1408 * , 161 ( 2014 ) [ arxiv:1404.0682 [ hep - ph ] ] ; m. cirelli , f. sala and m. taoso , jhep * 1410 * , 033 ( 2014 ) [ erratum - ibid . * 1501 * , 041 ( 2015 ) ] [ arxiv:1407.7058 [ hep - ph ] ] . m. ibe , s. matsumoto , s. shirai and t. t. yanagida , phys . d * 91 * , no . 11 , 111701 ( 2015 ) [ arxiv:1504.05554 [ hep - ph ] ] ; k. hamaguchi , t. moroi and k. nakayama , phys . b * 747 * , 523 ( 2015 ) [ arxiv:1504.05937 [ hep - ph ] ] . t. cohen , m. lisanti , a. pierce and t. r. slatyer , jcap * 1310 * , 061 ( 2013 ) [ arxiv:1307.4082 ] ; j. fan and m. reece , jhep * 1310 * , 124 ( 2013 ) [ arxiv:1307.4400 ] ; a. hryczuk , i. cholis , r. iengo , m. tavakoli and p. ullio , jcap * 1407 * , 031 ( 2014 ) [ arxiv:1401.6212 [ astro-ph.he ] ] . a. abramowski _ et al . _ [ h.e.s.s . collaboration ] , phys . lett . * 110 * , 041301 ( 2013 ) [ arxiv:1301.1173 ] . m. ackermann _ et al . _ [ fermi - lat collaboration ] , phys . rev . d * 89 * , 042001 ( 2014 ) [ arxiv:1310.0828 [ astro-ph.he ] ] . b. bhattacherjee , m. ibe , k. ichikawa , s. matsumoto and k. nishiyama , jhep * 1407 * , 080 ( 2014 ) [ arxiv:1405.4914 [ hep - ph ] ] . j. hisano , s. matsumoto , m. m. nojiri and o. saito , phys . d * 71 * , 015007 ( 2005 ) [ hep - ph/0407168 ] ; j. hisano , k. ishiwata and n. nagata , phys . b * 690 * , 311 ( 2010 ) [ arxiv:1004.4090 [ hep - ph ] ] ; j. hisano , k. ishiwata and n. nagata , phys . d * 82 * , 115007 ( 2010 ) [ arxiv:1007.2601 [ hep - ph ] ] ; j. hisano , k. ishiwata , n. nagata and t. takesako , jhep * 1107 * , 005 ( 2011 ) [ arxiv:1104.0228 [ hep - ph ] ] . j. hisano , k. ishiwata and n. nagata , jhep * 1506 * , 097 ( 2015 ) [ arxiv:1504.00915 [ hep - ph ] ] . j. billard , l. strigari and e. figueroa - feliciano , phys . d * 89 * , no . 2 , 023524 ( 2014 ) [ arxiv:1307.5458 [ hep - ph ] ] . r. j. hill and m. p. solon , phys . b * 707 * , 539 ( 2012 ) [ arxiv:1111.0016 [ hep - ph ] ] ; r. j. hill and m. p. solon , phys . lett . * 112 * , 211602 ( 2014 ) [ arxiv:1309.4092 [ hep - ph ] ] ; r. j. hill and m. p. solon , phys . d * 91 * , 043504 ( 2015 ) [ arxiv:1401.3339 [ hep - ph ] ] ; r. j. hill and m. p. solon , phys . d * 91 * , 043505 ( 2015 ) [ arxiv:1409.8290 [ hep - ph ] ] . s. weinberg , phys . lett . * 43 * , 1566 ( 1979 ) . f. wilczek and a. zee , phys . lett . * 43 * , 1571 ( 1979 ) . l. f. abbott and m. b. wise , phys . d * 22 * , 2208 ( 1980 ) . t. nihei and j. arafune , prog . phys . * 93 * , 665 ( 1995 ) [ hep - ph/9412325 ] . y. aoki , e. shintani and a. soni , phys . d * 89 * , 014505 ( 2014 ) [ arxiv:1304.7424 [ hep - lat ] ] . p. fileviez perez , phys . b * 595 * , 476 ( 2004 ) [ hep - ph/0403286 ] . c. munoz , phys . b * 177 * , 55 ( 1986 ) . w. e. caswell , j. milutinovic and g. senjanovic , phys . d * 26 * , 161 ( 1982 ) . nath and r. m. syed , nucl . b * 618 * , 138 ( 2001 ) [ hep - th/0109116 ] ; t. fukuyama , a. ilakovac , t. kikuchi , s. meljanac and n. okada , j. math . phys . * 46 * , 033505 ( 2005 ) [ hep - ph/0405300 ] .

this paper is the first part in our series on the influence of tidal interactions and minor mergers on the radial and vertical disk structure of spiral galaxies . we report on the sample selection , our observations , and data reduction . surface photometry of the optical and near infrared data of a sample of 110 highly - inclined / edge - on disk galaxies are presented . this sample consists of two subsamples of 61 non - interacting galaxies ( control sample ) and of 49 interacting galaxies / minor merging candidates . additionally , 41 of these galaxies were observed in the near infrared . we show that the distribution of morphological types of both subsamples is almost indistinguishable , covering the range between . an improved , 3-dimensional disk modelling- and fitting procedure is described in order to analyze and to compare the disk structure of our sample galaxies by using characteristic parameters . we find that the vertical brightness profiles of galactic disks respond very sensitive even to small deviations from the perfect edge - on orientation . hence , projection effects of slightly inclined disks may cause substantial changes in the value of the disk scale height and must therefore be considered in the subsequent study .

astro-ph0007103

considering the fact that the majority of ( spiral ) galaxies is not completely isolated but located in an environment which enables repeated close encounters or even merging processes with small companions it seems to be meaningful to systematically investigate the properties of galaxies affected by such processes . the investigation of their structural and dynamical changes caused by tidal interactions or low - mass satellite infall hence `` minor merger '' can help to clarify how far the evolution of disk galaxies was modified or even dominated by environmental effects . several n - body simulations were performed during the last decade in order to study the influence of minor mergers on galactic disks in greater detail ( e.g. quinn et al . @xcite ; mihos et al . @xcite ; walker et al . it was possible to use more realistic , multiple - component models for the galaxy - satellite system usually consisting of disk , bulge , and halo as well as a large number of particles ( @xmath1 ) . one of the main conclusions was that even merging processes in the range between @xmath2 can cause a vertical thickening of the stellar disk component by a factor between 2 and 4 , depending on the galactocentric distance . it was found that this vertical heating is due to a gain of kinetic energy of the disk stars by enhanced two - body relaxation . according to a series of papers on the frequency of these so called `` soft merging '' events ( e.g. toth & ostriker @xcite ; zaritsky @xcite , @xcite ) a large number of present - day ( disk- ) galaxies were affected by merging- or accretion processes of this magnitude since they have formed . as a consequence , interactions and minor mergers within this mass range might modify our picture of galaxy formation and evolution . however , the enormous parameter space of such a complex scenario makes it difficult to derive general conclusions from a set of few specific simulations . the quantitative results still crucially depend on the chosen parameters such as the content and behaviour of gas in the disk , the mass ratio between bulge and disk , induced star formation , or the satellite orbit ( quinn et al . @xcite ; mihos et al . @xcite ; velazquez & white @xcite ) . statistical studies of galaxy interactions based on optical photometry of disk galaxies ( reshetnikov & combes @xcite , @xcite ) focused on the effects of tidally - triggered disk thickening between systems of comparable mass . they found that the ratio @xmath3 of the radial exponential scale length @xmath4 to the constant scale height @xmath5 is only about twice smaller for interacting galaxies a lower value than derived from the minor mergers simulations . however , the small number of objects in their sample ( 7 non - interacting and 24 interacting galaxies ) did not permit to study these questions in detail . therefore , we started a project based on a larger sample of edge - on disk galaxies in both optical and near infrared passbands . this combination offers a number of advantages : first , observations in the near infrared particularly benefit from the much lower dust extinction near the galactic plane , i.e. at small @xmath6 . second , the presence of a dust lane along the major axis of most edge - on disk galaxies still presents one of the best methods to determine precise inclinations of the disks two facts that will become very important in order to derive reliable scale parameters from a disk fitting procedure . third , this combination enables us to make conclusions on disk populations of different ages . the main questions of this study can be summarized as follows : * are interactions / minor mergers able to change the radial and vertical structure of affected galactic disks ? * is there a substantial vertical disk thickening ? * of which order are the differences and similarities in the disk parameter distribution for a sample of interacting / non - interacting galaxies , respectively ? * to what extent are the disk properties of galaxies in the local universe influenced by interactions / minor mergers ? due to the complexity of these questions the paper is split into three parts : in this first part ( paper i ) we present deep optical and near infrared photometric data of a total sample of 110 highly - inclined / edge - on disk galaxies . this sample consists of two subsamples of 61 non - interacting galaxies ( control sample ) and of 49 minor merging candidates . additionally , 41 of these galaxies were observed in the near infrared . in sect . 2 the criteria of the sample selection will be described briefly . 3 gives an overview on the observations and data reduction . the disk modelling- and fitting procedure applied to derive the disk parameters will be reviewed in sect . 4 . in sect . 5 we summarize and conclude the paper . in the second part ( schwarzkopf & dettmar @xcite , paper ii ) the results of a detailed analysis of the structure of galactic disks will be presented . the third part ( schwarzkopf & dettmar in preparation , paper iii ) will be focused on the influence of accompanying minor merger features like disk `` warping '' and `` flaring '' on the vertical disk structure . 1.55 mm llccccccccccccccccc & & + & & + & & -5 & -4 & -3 & -2 & -1 & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 + & & e & & e - s0 & s0 & & s0/a & sa & sa - b & sb & sb - c & & sc & & sc - irr & & irri & + & & e & & e - s0 & s0 & & s0-a & sa & sa - b & sb & sb - c & s ... & sc , sc - d & s ... /irr & sd & & irr & unknown + ( 8.8,7.7 ) ( 8.8,7.7 ) * fig . 1a and b. * the distribution of morphological types * ( a ) * for the total galaxy sample and * ( b ) * for both subsamples of non - interacting ( normal ) and interacting / merging galaxies ( merger ) . the selection of two separate subsamples of highly - inclined disk galaxies consisting of a large number of interacting / merging candidates in the mass range @xmath7 , and of relatively isolated disk galaxies is a crucial point for the comparison of disk parameters . since presumably not all of the merging processes of this order of magnitude compared to large merging events are able to change the structure of affected galaxy disks completely , two facts must be considered in order to classify both subsamples : first , only merging events in a more progressed phase will have any appreciable effect on the disk structure . we therefore did not consider candidates in an early stage of interaction . instead , the sample contains some highly - inclined disk galaxies that show no separate satellite merging with the disk component , but strong evidence for accretion in the recent past ( indicated by both a disturbed disk structure and characteristic effects such as warping and tidal tails ) . most of these objects are located in galaxy groups with a few members , e.g. ngc 3628 , ngc 4634 , or ngc 4762 . due to their perturbed structure a morphological classification is difficult . second , a control sample of non - interacting edge - on disk galaxies must include all morphological types in the range @xmath0 ( table [ types ] ) . the galaxies should be isolated without indication of interaction and accretion . in particular , it is important to include the latest type spirals like sd so called `` superthin '' galaxies since they belong to a class of disk galaxies having the smallest known axis ratios ( between 1:9 and 1:20 ) . these galaxies are characterized by velocity curves of modest gradient ( goad & roberts @xcite ; griv & peter @xcite,@xcite,@xcite ) , indicating that the `` superthins '' represent kinematically almost unheated galaxy disks . beyond these `` limitations '' and the specific criteria described in the next section . it seems difficult to give a generalized definition of the selection criteria . as a consequence , it is unavoidable that the sample of non - interacting galaxies is slightly `` polluted '' by some galaxies that were affected by interactions / minor mergers during their past but show no indication of these effects today . this effect would , however , only influence the non - interacting galaxies and lead to an underestimation of the actual differences between both samples . in this study , the notation `` interacting / merging '' refers closely to the classification scheme introduced by arp & madore ( @xcite ) and will therefore be used as a synonym for all galaxies which fulfil the following criteria : + galaxies with interacting companions , interacting doubles , galaxies with peculiar disks , galaxies with tails , loops of material or debris , irregular or disturbed galaxies , chains and groups of galaxies . a more detailed description of these classifications is given in arp & madore ( @xcite ) . with these limitations in mind , we selected interacting / merging candidates with an inclination @xmath8 ( derived from a first visual inspection ) . during this pre - selection we did not impose any morphological restrictions except that their type should be not much earlier than @xmath9 . the candidates were chosen from optical prints in `` a catalogue of southern peculiar galaxies and associations '' ( arp & madore @xcite ) , `` atlas of peculiar galaxies '' ( arp @xcite ) , and the nasa `` atlas of galaxies '' ( sandage & bedke @xcite ) . we also selected some systems from the catalogue `` satellites of spiral galaxies '' ( zaritsky et al . @xcite , @xcite ) . another selection criterion for the minor mergers and those galaxies in the subsample with very close companions was the mass ratio between the main bodies . for all the candidates which could be separated into two individuals ( i.e. for @xmath10 galaxies of the interacting / merging sample ) this ratio was checked by an estimation of their total fluxes within a certain aperture or ellipse that contains all intensities down to the sky brightness . the resulting mean ratio is @xmath11 . typical examples ( with errors of @xmath12 ) are ngc 1531/32 ( @xmath13 ) , ngc 128 ( @xmath14 ) or ngc 1888 ( @xmath15 ) . for interacting galaxies located a less dense group or for those with a remote companion this mass ratio can be larger . the selection criteria for disk inclination and morphological types were the same as described for the interacting / merging galaxies . our principal sources for the subsample of highly - inclined , non - interacting galaxies were the eso - uppsala catalogue ( lauberts & valentijn @xcite ) , the rc3- ( de vaucouleurs et al . @xcite ) and ugc ( nilson @xcite ) catalogues , the `` carnegie atlas of galaxies '' ( sandage & bedke @xcite ) , and `` the hubble atlas of galaxies '' ( sandage @xcite ) . to check their isolation , larger fields @xmath16 , depending on the distance of individual objects ) were inspected visually using the digitized sky survey . in order to benefit from a better spatial resolution of some closer objects there was only a lower limit in apparent angular disk diameter of @xmath17 for both galaxy samples . finally , the sample of non - interacting galaxies was filled up by 11 edge - on galaxies of the barteldrees & dettmar ( @xcite ) data set . several studies of the properties of edge - on galaxies argued that some of the disk parameters of spiral galaxies , e.g. the ratio of disk scale length to scale height , might be correlated with the morphological type ( de grijs & van der kruit @xcite ; de grijs @xcite ; schwarzkopf & dettmar @xcite ) . therefore , it was also important to ensure that both subsamples investigated in this study are not affected by selection effects . this applies , in particular , to the distribution of morphologigal types and to the redshifts / distances of selected objects . due to their relation to the absolute properties of galactic disks the latter will be discussed in detail in paper ii . the distribution of morphological types ( according to ned ) is shown in fig . 1a for the complete galaxy sample , and in fig . 1b for both subsamples of non - interacting and interacting / merging galaxies , respectively . the statistical test of kolmogorov & smirnov ( darling @xcite ; sachs @xcite ) hereafter ks quantifies the similarity of both subsamples with a result of 0.04 . that is significantly lower than the value necessary for the 20%-limit ( 0.2 ) , which is the strongest of the ks - criteria . considering also the unavoidable errors ( @xmath18 ) introduced by an automated type classification ( corwin et al . @xcite ; lauberts & valentijn @xcite ) both distributions are statistically indistinguishable . the gross of galaxies covers the range between @xmath19 , with a strong peak at @xmath20 . this peak can also be observed in the frequency distribution of galaxy types in the available catalogues . the effect is caused by the selection criteria of the classification programs used in the catalogues ( sect . 6 in lauberts & valentijn ( 1989 ) ) . due to the large number of galaxies needed for reliable statistics the optical observations were obtained with different telescopes and during several observing runs between february 1996 and june 1998 . the following telescopes were used : 1.54 m danish and 61 cm bochum telescope on la silla , chile ; 42-inch telescope at lowell observatory , flagstaff ; 1.23 m telescope at calar alto observatory , spain ; 1.06 m telescope at hoher list observatory , germany . the used passbands were johnson @xmath21 and thuan & gunn @xmath22 , with a central wavelength at @xmath23 nm and 670 nm , bandpass @xmath24 nm and 103 nm , respectively . all observing runs and the used telescope / detector characteristics are listed in table [ telescopes ] . the seeing conditions are given as averaged fwhm values . details of the optical observations are listed in table [ samples ] . additionally for most of the sample galaxies the total blue surface brightness , @xmath25 , and the morphological type @xmath26 ( revised hubble type , according to lauberts & valentijn ( @xcite ) , table [ types ] ) are given . the near infrared observations were obtained with the magic camera of the mpia attached to the 2.2 m and 1.23 m telescopes of the calar alto observatory , spain , and with the irac2b camera on the eso / mpia 2.2 m telescope of the european southern observatory ( eso ) , chile , respectively . both the magic and irac2b cameras are equipped with a rockwell @xmath27 @xmath28 nicmos3 hgcdte array . the data were acquired during several observing runs between february 1996 and june 1998 . for all runs we used the @xmath29 filter ( central wavelength @xmath30 , bandpass @xmath31 ) , or the @xmath32 filter ( @xmath33 ) . for some objects images in the @xmath34 filter were obtained ( @xmath35 ) . we recorded object and sky frames alternately , with typical single integration times of @xmath36s per sequence , and spatially separated by @xmath37 . to eliminate bad pixels , the telescope `` on source''-positions were dithered by a few arcseconds between subsequent exposures . since many of the sample galaxies are faint objects it was required to reach at least @xmath38 min integration time `` on source '' , which corresponds to @xmath39 cycle repetitions and a resulting observing time of @xmath40 hours . therefore , only one third of our total optical galaxy sample was observed in the near infrared . since our study is aiming at an investigation of the structure and geometrical properties of galactic disks , we could make use of observations obtained under non - photometric conditions . for that reason the optical and near infrared contour maps of the sample galaxies shown in figs . 3 and 4 are not flux calibrated . as for the optical observations , the telescope / detector characteristics as well as a list of all galaxies observed in the near infrared are given in tables [ telescopes ] and [ samples ] , respectively . 0.8 mm llclrcllc & & & & & & & + & & & & & & & + & & & & & & & + + mar & 1996 & & 1.1 lo & 4.9 & & 0.729@xmath41 & ti 800 & 2.0 + may & 1996 & & 1.0 hl & 27.0 & & 0.825 & loral 2048 & 3.0 + jun & 1996 & & 1.2 ca & 8.6 & & 0.503 & tek 1024 & 2.0 + aug & 1996 & & 1.0 hl & 27.0 & & 0.825 & loral 2048 & 3.0 + sep & 1996 & & 1.2 ca & 8.6 & & 0.503 & tek 1024 & 2.0 + dec & 1996 & & 0.6 bo & 4.8 & & 0.496 & th 7882 & 1.8 + jan & 1997 & & 1.2 ca & 8.6 & & 0.503 & tek 1024 & 2.0 + feb & 1997 & & 1.0 hl & 6.8 & & 0.800@xmath41 & loral 2048 & 3.2 + apr & 1997 & & 1.5 da & 13.7 & & 0.390 & loral 2048 & 1.5 + may & 1997 & & 1.1 lo & 9.9 & & 0.730@xmath41 & site 2k & 2.0 + sep & 1997 & & 1.0 hl & 27.0 & & 0.825 & loral 2048 & 3.0 + jun & 1998 & & 1.5 da & 13.7 & & 0.390 & loral 2048 & 1.6 + + mar & 1996 & & 1.2 ca & 5.1 & & 1.200 & nicmos3 & 1.8 + sep & 1996 & & 2.2 ca & 2.7 & & 0.642 & nicmos3 & 1.8 + jan & 1997 & & 2.2 ca & 2.7 & & 0.642 & nicmos3 & 1.5 + apr & 1997 & & 2.2 eso & 3.0 & & 0.708 & nicmos3 & 1.6 + feb & 1998 & & 2.2 ca & 2.7 & & 0.642 & nicmos3 & 1.8 + may & 1998 & & 2.2 eso & 3.0 & & 0.708 & nicmos3 & 1.5 + pixel binning @xmath42 . + the optical data were reduced using the midas software package , developed by eso . following the standard reduction procedures ( bias subtraction , flat fielding with sky flats ) the remaining gradients in the background of galaxies that covered a major fraction of the field of view were removed using a two - dimensional polynomial . for some of the frames that were affected by bad columns these columns were also removed by the standard midas fitting routine . in order to increase the signal - to - noise ratio s / n ( important for an investigation of faint disk features ) the images of fainter objects were binned ( @xmath42 ) . finally , the frames were rotated in such a way that the galaxy planes are in a horizontal position ( assuming symmetrical light distribution of the vertical disk profiles ) . it should be stressed that the remaining small rotation error which is typically @xmath43 for most of the relatively uniform disks of non - interacting galaxies can be considered by the disk fitting routine ( sect . 4 ) . for galactic disks affected by strong disk warping ( mostly interacting objects ) precise rotation was more difficult . for these cases , only the inner parts of the galaxy disk were considered to determine the rotation . for standard reduction of the near infrared data , the iraf software package was used . in particular , the sky frame subtraction , the flat fielding , and the combining of the flat - fielded images was done with the arnica ( arcetri near - infrared camera ) add - on package . the sky frames used for the flat field subtraction were obtained from a set of the nearest frames in time , filtered by a median . the median filtering also removed the stars in the sky frames . to produce a final source frame , the reduced , flat - fielded images were combined using the arnica standard `` mosaic '' task ( this task includes both median filtering , and centering of frames by stars in the field ) . due to the small detector size the resulting field of view was @xmath44 on average ( table [ telescopes ] ) images of larger fields were produced by mosaicing . since this is a time - expensive procedure , it was only applied in order to obtain images of some larger objects . for precise adjustment of each of the frames we used either stars in the field or the sharp central bulge regions of the galaxies themselves . image rotation was applied as explained for the optical images . isophote maps of the complete samples of interacting / merging and non - interacting galaxies , respectively , are shown in figs . 3 and 4 of this paper . in order to analyze and to compare the radial and vertical disk structure of a large sample of highly - inclined / edge - on spiral galaxies , it is necessary to apply a disk model that enables both a good quantitative and very flexible description of the 3-dimensional luminosity distribution . although mathematical simplicity is also a desired property of disk models , there should be a firm physical basis . therefore , a disk modelling- and fitting procedure was developed based on the results of a fundamental study of edge - on spiral galaxies by van der kruit & searle ( @xcite,@xcite ; @xcite,@xcite ) as well as on other observational studies thereafter ( e.g. van der kruit @xcite ; barteldrees & dettmar @xcite ; de grijs & van der kruit @xcite ; just et al . the disk model presented here also considers the effects of an inclined disk ( @xmath45 ) as well as 3 different vertical luminosity distributions . using the following notation for disk model parameters @xmath46 = @xmath47 = central luminosity density + @xmath48 @xmath47 inclination angle of the disk + @xmath49 @xmath47 disk cut - off radius + @xmath4 @xmath47 disk scale length + @xmath5 @xmath47 disk scale height + @xmath50 @xmath47 vertical distribution , see ( 2 ) to ( 4 ) + @xmath51 @xmath47 truncation ( heaviside- ) function + ( 1 for @xmath52 ; 0 for @xmath53 ) the 3-dimensional luminosity distribution @xmath54 ( @xmath55 ) of the disk can be described in cylindrical coordinates by @xmath56 a set of 3 functions @xmath50 with the same asymptotic behaviour for @xmath57 proposed by van der kruit ( @xcite ) ; wainscoat et al . ( @xcite , @xcite ) ; burkert & yoshii ( @xcite ) is used to describe the vertical luminosity distribution @xmath58 : @xmath59 @xmath60 @xmath61 thus , for large @xmath6 , a comparison between different scale heights is possible via @xmath62 the use of this combination of 3 different vertical luminosity distributions meets the mentioned specifications and allows also very flexible description of a large variety of vertical profiles of galactic disks . in order to obtain a final disk model , i.e. a two - dimensional intensity distribution , integration of ( 1 ) along the line - of - sight through the disk is required . but there exist only two possible disk models the projection face - on ( @xmath63 ) and edge - on ( @xmath64 ) , combined with the isothermal disk model ( 4 ) which would allow an analytical solution of ( 1 ) using a modified bessel function ( van der kruit & searle @xcite ) . therefore , integration of ( 1 ) must be calculated numerically . hence , the intensity @xmath65 of one point ( pixel ) of the disk as it can be seen by an observer at the projected coordinates @xmath66 and @xmath67 on the ccd can be written in cartesian coordinates in the following form ( for example , only the double exponential disk model ( 2 ) is given here in its explicit form , for @xmath68 ; integration of models ( 3 ) and ( 4 ) is analogous ) : @xmath69 the integration limits are , in a direction of the line - of - sight , given by the geometry of the `` galaxy - cylinder '' : @xmath70 and can be approximated for @xmath71 by @xmath72 i.e. by a cylinder with a height of about 10 scale heights . at larger distances above the disk plane the contribution of luminosity , acc . to ( 2)(4 ) , becomes negligibly . the integration of ( 6 ) along the line - of - sight through the disk-``cylinder '' is realized by a modified newton - cotes rule . according to @xmath73 , and ( 7 ) the intensity @xmath74 of any disk model can be described by an integral of a family of 3 different density laws @xmath54 , depending on 5 free parameters each : @xmath75 in practice , this is still a large parameter space . it therefore requires a further , step - by - step restriction , realized in the following disk fitting procedure . ( 14,10 ) + * fig . * example of the disk fitting procedure : set of typical radial ( left panels ) and vertical ( right panels ) profiles used for disk fitting of the sc - galaxy eso 461-g06 ( @xmath22-band , see also fig . 5 ) . the galaxy profiles ( averaged ) are displayed together with the final disk model , the used parameters are : @xmath76 ( further explanation see text ) . ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) ( 8,7.3 ) * fig . 3 . * the influence of inclination on the shape of vertical disk profiles , shown for 3 different edge - on galaxies and their corresponding models ( columns from left to right ) : @xmath77 exp ( ic 2531 ) ; @xmath77 sech ( ngc 1886 ) ; @xmath78 ( eso 187-g08 ) . although the disk inclination is varied between @xmath79 , for each of the vertical profiles a quantitative good fit can be found by changing only the scale height @xmath5 and the central luminosity density @xmath46 ( as indicated ; further explanation see text ) . in order to derive relible disk parameters we developed two independent disk fitting procedures that were designed to comply the special tasks proposed for the first ( paper i+ii ) and second ( paper iii ) part of this study ( see sect . 1 ) . the first fitting procedure was realized semi - automatically . it uses different graphs enabling a direct comparison by eye of a set of radial and vertical profiles within a preselected disk area with those of an underlying disk model . to take advantage of symmetrical light distribution of galactic disks the profiles used for modelling are averaged over two quadrants . they are usually displayed equidistantly and cover the whole fitting area . the size of the fitting area vastly depends on the properties of the individual image , i.e. on both the image quality and some special features of the galaxy disk itself . therefore the program allows after a first inspection of the profiles an interactive selection of a qualified fitting area . depending on the s / n ratio and the spatial resolution of the images it is also possible to use a desired pixel binning for both image and model ( as an example fig . 2 shows one of the sets of radial and vertical disk profiles ) . this kind of modelling allows fast and flexible , but reliable disk fitting of a large number of galaxies in particular of interacting / merging galaxies whose disk profiles do often show considerable deviations from the simple model . such galaxies are therefore difficult to handle with conventional least squares fitting methods . the second fitting procedure uses the same set of disk models as described before but a least square fitting in order to fit the scale height as a function of the galactocentric distance . the properties of the program will be described in detail in paper iii . to ensure homogenity of the fitting results the fit quality reached with this method was compared with that of the above described semi - automated disk modelling ( for this purpose the data of non - disturbed disk profiles of galaxies in the non - interacting sample were used ) . it was found that both methods give consistent results with errors on ( or below ) a 10%-level . the errors found for the fits of the scale height only are even smaller than 5% . the remaining discrepancies are not due to the individual method itself but rather to the following two reasons : first , the fitting areas are slightly different ( i.e. they are usually more restricted for the least square method as a result of its sensibility against the error sources described in the following ) . second , the scale height of a large fraction of galaxy disks investigated shows variations of different absolute size ( both irregularly and systematically , i.e. gradients ) along the galactocentric distance . therefore this point will be discussed in detail in paper iii . a detailed comparison of the semi - automated disk modelling with another , independent developed least square fitting routine will be given in pohlen et al . ( @xcite ) . after comparison of both fitting methods we choose to use the semi - automated disk modelling for this part of the study because it combines the advantages of fast and flexible fitting with a high accuracy . in view of the existing data the point of flexible fitting is very important since despite all structural similarities the radial and vertical profiles of induvidual ( disk ) galaxies are unique . the profiles investigated here are often heavily contaminated by light from , e.g. , a bulge and/or a bar , a nearby companion , other disk components , foreground stars or reflections from bright stars . in addition , the modelling may be complicated by strong dust extinction along the galactic plane , by a low s / n ratio , or a considerably warped disk ( mainly interacting / merging galaxies ) . since most of the foregoing deviations can not be easily quantified their complete consideration by an automated fitting procedure thus seems almost impossible and would cause unpredictably errors . in the following step - by - step procedure the disk parameters of the semi - automated disk fitting procedure are reduced systematically : as a first step , the inclination is determined by using the axis ratio of the dust lane in the disk plane of optical images . given a relatively sharp dust lane , an accuracy of less than @xmath80 can be reached . the central luminosity density , @xmath81 , is calculated automatically by the fitting program for each new parameter set by using a number of preselected reference points along the disk ( outside the bulge- or bar - light contaminated regions ) . given a sufficient s / n ratio , the cut - off radius @xmath49 can be fitted to the major axes profiles with an accuracy between @xmath82 . the cut - off is determined by a significant decrease of the intensity extrapolated to @xmath83 ( left panels in fig . 2 ) . for highly - inclined disks such as the ones studied here , the effect of a variation of @xmath49 on the slope of radial disk profiles is negligible . thus , the remaining `` real '' fitting parameters are the disk scale length and height , @xmath4 and @xmath5 , as well as the set of 3 functions @xmath84 . within the following procedure both the scale length and the scale height are fitted in an iterative process until a first good convergence of the `` global '' fit is achieved . during this process , the quality of the corresponding fits can be checked simultaneously using a small set of radial and vertical disk profiles ( usually 3 profiles each , see fig . 2 as an example ) . for further small corrections , if necessary , both parameters can be considered as independent and thus separated without any loss of accuracy . for this `` fine tuning '' a set with more disk profiles ( usually 6 - 8 ) is used for both parameters . during the iterative process of modelling the scale length is fitted to a set of averaged major axes profiles in a radial region typically between @xmath85 and vertically outside strong dust extinction ( left panels in fig . 2 ) . the fit quality of the vertical profiles along the disk can be used as a cross - check for the scale length ( right panels in fig . 2 ) . as a result of the error sources mentioned at the beginning the disk scale length of a galaxy can be reproduced ( with the same method ) with an accuracy of @xmath82 . in contrast to this , the scale lengths derived with different methods can in some cases differ by @xmath86 . the disk scale height @xmath5 is estimated by fitting the @xmath6-profiles inside the previously selected radial regions , which are outside the bulge- or bar contamination ( right panels in fig . 2 ) . the vertical region used for the fits is typically between @xmath87 , but depends strongly on the individual characteristic of the dust lane . if the disk inclination is known precisely , this fitting method works reliably . otherwise , additional errors may be introduced ( see next sect . ) . the vertical disk profiles of most of the galaxies investigated in optical passbands enable a reliable choice of the quantitatively best fitting function @xmath84 . this is because deviations between different models become visible at vertical distances larger than that of the most sharply - peaked dust regions . for those disks that are affected by strong dust extinction , the choice was made easier by using a combination of both optical and near infrared profiles . for these cases profiles of both passbands were , if available , used for fitting . if a large sample of disk galaxies is investigated statistically as is the case in this study moderate deviations from edge - on orientation of the order of @xmath88 are common ( sect . 2 ) . it was found in this study and in schwarzkopf & dettmar ( @xcite ) that , if reliable values for the scale height are desired , deviations from @xmath89 larger than @xmath90 can not be neglegted . therefore in this section the influence of small changes of inclination on vertical disk profiles and on the resulting parameters will be investigated in detail . for that purpose we selected 3 @xmath22-band images of galaxies in the non - interacting sample with well known disk parameters , but with different vertical profiles : @xmath77 exp ( ic 2531 ) ; @xmath77 sech ( ngc 1886 ) ; and @xmath78 ( eso 187-g08 ) . at 3 different inclinations ( @xmath91 ; @xmath92 ; and @xmath93 ) the vertical profiles of each model were best - fitted to the observed disk profiles ( see left , middle , and right panels of fig . 3 ) . to compensate for the effect of inclination , both the disk scale height , @xmath5 , and the central luminosity density , @xmath46 , must be changed . as can be seen in the disk parameters ( middle raw in fig . 3 and table [ inclination ] ) the effect for @xmath5 is , at @xmath94 , around the @xmath95-level for all vertical models , whereas for @xmath96 ( fig . 3 bottom ) the error amounts to @xmath97 for the exp - model and , after all , to @xmath98 for the @xmath99 and sech - models . at smaller inclinations the quality of all vertical fits decreases rapidly and allows no more qualitative good fits . as expected , the effect of slight changes of inclination on both the scale height and the shape of vertical profiles near the disk plane is the strongest for the exp - model ( left panels in fig . 3 ) . the effect shown above is due to the fact that the slope of the vertical disk profiles in a region that is relevant for fitting ( previous section ) remains nearly unchanged in the range between @xmath100 , whereas this is not true for the width of the vertical profiles . hence , to obtain reliable disk parameters for @xmath5 and @xmath46 it is required that both * the inclination of the disk is known precisely ( within @xmath101 ) and * a disk model with a flexible inclination is used . otherwise , substantial errors for @xmath5 and @xmath46 are introduced ( table [ inclination ] lists averaged errors @xmath102 and @xmath103 , obtained by fitting 5 galaxies with 3 different vertical disk models each ) . 1.2 mm ccrrccrrccrr & & & & & & & & + & & & & & & & & & & & + & & & & & & & & & & & + @xmath92 & & @xmath104 & @xmath105 & & & @xmath105 & @xmath105 & & & @xmath105 & @xmath106 + @xmath93 & & @xmath107 & @xmath108 & & & @xmath109 & @xmath109 & & & @xmath110 & @xmath111 + optical and near infrared photometric data of a sample of 110 highly - inclined / edge - on disk galaxies are presented . this sample consists of two subsamples of 61 non - interacting galaxies and 49 minor merging candidates . additionally , 41 of these galaxies were observed in the near infrared . the sample selection , observations , and data reduction are described . we show that although the subsamples are naturally slightly polluted due to unavoidable selection effects the distribution of their morphological types is almost indistinguishable , covering the range between @xmath0 . this is important for the forthcoming detailed statistical study focused on the influence of interaction and minor merger on the radial and vertical disk structure of spiral galaxies . moreover , a 3-dimensional disk modelling- and fitting procedure is described in order to analyze and to compare the disk structure of our sample galaxies by using characteristical parameters . we find that the vertical brightness profiles of modelled galaxy disks respond very sensitive even to small changes of inclination around perfect edge - on orientation . therefore , projection effects of highly - inclined disks must be considered . rllllcccrrc & & & & & & & & & & + & & & & & & & & & & + & & & & & & & & & & + + 1 & ngc 7 & 00 08 20.6 & @xmath11229 54 54 & 0.6 bo & dec 27 96 & @xmath21 & 75 & 14.40 & 4.5 & 1 + 2 & ugc 260 & 00 27 03.0 & + 11 35 03 & 1.0 hl & sep 04 97 & @xmath21 & 65 & 13.71 & 6.0 & 1 + 3 & ngc 128 & 00 29 15.0 & + 02 51 55 & 1.2 ca & sep 08 96 & @xmath21 & 30 & 12.77 & -2.0 & 1 + & & & & 2.2 ca & sep 03 96 & @xmath29 & 4 & & -2.0 & 1 + 4 & am 0107 - 375 & 01 09 42.0 & @xmath11237 42 27 & 0.6 bo & dec 28 96 & @xmath21 & 65 & & 3.5 & 1 + 5 & eso 296-g17 & 01 23 55.0 & @xmath11238 00 44 & 0.6 bo & dec 29 96 & @xmath21 & 60 & 16.38 & 3.0 & 1 + 6 & eso 354-g05 & 01 52 07.0 & @xmath11233 31 46 & 0.6 bo & jan 05 97 & @xmath21 & 75 & 15.95 & 4.0 & 1 + 7 & eso 245-g10 & 01 56 44.0 & @xmath11243 58 23 & 0.6 bo & dec 31 96 & @xmath21 & 75 & 14.28 & 3.0 & 1 + 8 & eso 417-g08 & 02 58 47.0 & @xmath11232 05 52 & 0.6 bo & jan 04 97 & @xmath21 & 75 & 13.64 & 0.7 & 1 + 9 & eso 199-g12 & 03 03 25.0 & @xmath11250 29 43 & 0.6 bo & dec 28 96 & @xmath21 & 88 & 15.52 & 8.0 & 1 + 10 & eso 357-g16 & 03 19 34.0 & @xmath11232 27 53 & 0.6 bo & jan 02 97 & @xmath21 & 75 & 14.34 & 3.0 & 1 + 11 & eso 357-g26 & 03 23 56.0 & @xmath11236 27 50 & 1.2 ca & jan 01 97 & @xmath21 & 75 & 11.37 & -1.0 & 1 + 12 & eso 418-g15 & 03 39 23.0 & @xmath11231 19 19 & 0.6 bo & dec 28 96 & @xmath21 & 55 & 12.40 & 4.0 & 1 + 13 & ngc 1531/32 & 04 11 59.0 & @xmath11232 50 57 & 1.5 da & apr 08 97 & @xmath22 & 30 & 10.65 & 2.7 & 1 + 14 & eso 202-g04 & 04 17 46.0 & @xmath11250 09 51 & 0.6 bo & dec 26 96 & @xmath21 & 80 & 13.47 & 2.0 & 1 + 15 & eso 362-g11 & 05 16 39.0 & @xmath11237 06 00 & 0.6 bo & jan 05 97 & @xmath21 & 75 & 13.04 & 4.0 & 1 + 16 & ngc 1888 & 05 22 35.0 & @xmath11211 29 58 & 1.5 da & apr 08 97 & @xmath22 & 40 & 12.83 & 5.0 & 1 + & & & & 2.2 eso & apr 10 97 & @xmath32 & 40 & & 5.0 & 1 + 17 & eso 363-g07 & 05 33 13.0 & @xmath11236 23 59 & 0.6 bo & dec 30 96 & @xmath21 & 75 & 13.25 & 5.5 & 1 + 18 & eso 487-g35 & 05 42 01.0 & @xmath11222 56 43 & 0.6 bo & jan 01 97 & @xmath21 & 90 & 13.39 & 7.8 & 1 + 19 & ngc 2188 & 06 10 10.0 & @xmath11234 06 22 & 1.5 da & apr 09 97 & @xmath22 & 75 & 12.14 & 9.0 & 1 + & & & & 2.2 eso & apr 12 97 & @xmath32 & 33 & & 9.0 & 1 + 20 & ugc 3697 & 07 11 21.3 & + 71 50 06 & 1.1 lo & mar 16 96 & @xmath21 & 33 & 13.50 & 7.0 & 1 + & & & & 1.2 ca & mar 01 96 & @xmath34 & 53 & & 7.0 & 1 + 21 & eso 060-g24 & 09 02 40.3 & @xmath11268 13 38 & 1.5 da & jun 03 98 & @xmath22 & 30 & 13.95 & 2.5 & 1 + 22 & eso 497-g14 & 09 07 42.0 & @xmath11223 37 15 & 0.6 bo & dec 26 96 & @xmath21 & 75 & 14.16 & 3.0 & 1 + 23 & ngc 2820 & 09 21 47.0 & + 64 15 29 & 1.2 ca & mar 01 96 & @xmath34 & 53 & 13.28 & 5.0 & 1 + 24 & ngc 3044 & 09 53 40.0 & + 01 34 46 & 1.5 da & apr 08 97 & @xmath22 & 45 & 12.46 & 5.0 & 1 + & & & & 2.2 eso & apr 10 97 & @xmath32 & 40 & & 5.0 & 1 + 25 & ngc 3187 & 10 17 48.0 & + 21 52 25 & 1.5 da & apr 09 97 & @xmath22 & 30 & 13.91 & 5.0 & 1 + & & & & 2.2 eso & apr 11 97 & @xmath32 & 28 & & 5.0 & 1 + 26 & eso 317-g29 & 10 27 44.0 & @xmath11240 26 08 & 0.6 bo & dec 28 96 & @xmath21 & 50 & 13.74 & 1.0 & 1 + 27 & eso 264-g29 & 10 40 12.0 & @xmath11247 06 11 & 0.6 bo & jan 06 96 & @xmath21 & 60 & 15.68 & 5.6 & 1 + & & & & 2.2 eso & apr 12 97 & @xmath32 & 11 & & 5.6 & 1 + 28 & ngc 3432 & 10 52 31.0 & + 36 37 08 & 1.1 lo & may 31 97 & @xmath21 & 40 & 11.67 & 9.0 & 1 + & & & & 2.2 ca & feb 13 98 & @xmath32 & 60 & & 9.0 & 1 + 29 & ngc 3628 & 11 20 16.0 & + 13 35 22 & 1.5 da & apr 09 97 & @xmath22 & 95 & 10.28 & 3.0 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 13 & & 3.0 & 1 + 30 & eso 378-g13 & 11 37 08.0 & @xmath11232 49 13 & 0.6 bo & jan 07 97 & @xmath21 & 60 & 15.45 & 1.0 & 1 + & & & & 2.2 eso & apr 12 97 & @xmath32 & 38 & & 1.0 & 1 + 31 & eso 379-g20 & 12 00 59.0 & @xmath11235 11 36 & 0.6 bo & jan 08 97 & @xmath21 & 75 & 15.44 & 1.0 & 1 + & & & & 2.2 eso & apr 11 97 & @xmath32 & 36 & & 1.0 & 1 + 32 & ngc 4183 & 12 13 18.0 & + 43 41 55 & 1.1 lo & jun 02 97 & @xmath21 & 25 & 12.86 & 6.0 & 1 + & & & & 2.2 ca & feb 14 98 & @xmath32 & 40 & & 6.0 & 1 + 33 & ngc 4631 & 12 42 08.0 & + 32 32 28 & 1.1 lo & may 31 97 & @xmath21 & 30 & 9.75 & 7.0 & 1 + & & & & 2.2 ca & feb 16 98 & @xmath32 & 13 & & 7.0 & 1 + rllllcccrrc & & & & & & & & & & + & & & & & & & & & & + & & & & & & & & & & + 34 & ngc 4634 & 12 42 40.4 & + 14 17 47 & 1.5 da & jun 03 98 & @xmath22 & 45 & 13.16 & 6.0 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 25 & & 6.0 & 1 + 35 & ngc 4747 & 12 51 45.0 & + 25 46 27 & 1.1 lo & jun 02 97 & @xmath21 & 72 & 12.96 & 6.0 & 1 + 36 & ngc 4762 & 12 52 56.3 & + 11 13 48 & 1.5 da & apr 08 97 & @xmath22 & 30 & 11.12 & -2.0 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 27 & & -2.0 & 1 + 37 & eso 443-g21 & 12 59 46.0 & @xmath11229 35 58 & 1.5 da & apr 08 97 & @xmath22 & 45 & 14.41 & 6.0 & 1 + & & & & 2.2 eso & apr 11 97 & @xmath32 & 15 & & 6.0 & 1 + 38 & ngc 5126 & 13 24 54.0 & @xmath11230 20 00 & 0.6 bo & jan 10 97 & @xmath21 & 40 & 14.06 & 0.0 & 1 + 39 & eso 324 - 23 & 13 27 29.0 & @xmath11238 10 26 & 1.5 da & apr 08 97 & @xmath22 & 45 & 13.07 & 6.5 & 1 + & & & & 2.2 eso & apr 10 97 & @xmath32 & 32 & & 6.5 & 1 + 40 & eso 383-g05 & 13 29 23.8 & @xmath11234 16 23 & 1.5 da & jun 03 98 & @xmath22 & 45 & 14.21 & 3.7 & 1 + 41 & ngc 5297 & 13 46 24.0 & + 43 52 25 & 1.1 lo & jun 03 97 & @xmath21 & 3 & 12.47 & 4.5 & 1 + 42 & eso 445-g63 & 13 52 07 & @xmath11230 49 41 & 1.5 da & jun 03 98 & @xmath22 & 40 & 15.78 & 5.3 & 1 + 43 & ngc 5529 & 14 15 34.1 & + 36 13 36 & 1.1 lo & jun 01 97 & @xmath21 & 50 & 12.75 & 5.0 & 1 + & & & & 1.2 ca & mar 05 96 & @xmath34 & 12 & & 5.0 & 1 + 44 & ngc 5965 & 15 34 02.0 & + 56 41 10 & 1.1 lo & jun 03 97 & @xmath21 & 60 & 12.60 & 3.0 & 1 + 45 & ngc 6045 & 16 05 08.0 & + 17 45 22 & 1.2 ca & sep 06 96 & @xmath21 & 30 & 14.87 & 5.0 & 1 + & & & & 2.2 ca & sep 03 96 & @xmath29 & 30 & & 5.0 & 1 + 46 & ngc 6361 & 17 18 40.0 & + 60 36 32 & 1.2 ca & jun 02 96 & @xmath21 & 37 & 13.87 & 3.0 & 1 + & & & & 1.0 hl & may 20 96 & @xmath113 & 5 & & 3.0 & 1 + & & & & 2.2 ca & sep 03 96 & @xmath29 & 20 & & 3.0 & 1 + 47 & arp 121 & 00 59 24.0 & @xmath11204 48 13 & 1.2 ca & sep 04 96 & @xmath21 & 60 & & 2.0 & 1 + & & & & 2.2 ca & sep 04 96 & @xmath29 & 16 & & 2.0 & 1 + 48 & eso 462-g07 & 20 18 23 & @xmath11227 27 18 & 2.2 eso & apr 12 97 & @xmath32 & 41 & 15.48 & 4.0 & 1 + 49 & ic 4991 & 20 18 23.0 & @xmath11241 03 01 & 1.5 da & apr 08 98 & @xmath22 & 30 & 11.56 & -2.0 & 1 + + 1 & ugc 231 & 00 24 02.6 & + 16 29 09 & 1.0 hl & aug 19 96 & @xmath21 & 30 & 13.91 & 6.0 & 1 + & & & & 2.2 ca & sep 05 96 & @xmath29 & 40 & & 6.0 & 1 + 2 & eso 150-g07 & 00 25 37.0 & @xmath11257 11 28 & 1.5 da & jun 04 98 & @xmath22 & 30 & 15.28 & 1.0 & 1 + 3 & eso 112-g04 & 00 28 04.0 & @xmath11258 06 13 & 2.2 eso & june 87 & @xmath22 & & 15.86 & 5.6 & 2 + 4 & eso 150-g14 & 00 36 38.0 & @xmath11256 54 24 & 2.2 eso & june 87 & @xmath22 & & 14.90 & 0.4 & 2 + 5 & ugc 711 & 01 08 37.0 & + 01 38 29 & 0.6 bo & dec 25 96 & @xmath21 & 118 & 14.39 & 6.7 & 1 + 6 & eso 244-g48 & 01 39 09.0 & @xmath11247 07 42 & 1.5 da & jun 03 98 & @xmath22 & 30 & 15.55 & -2.0 & 1 + 7 & ugc 1839 & 02 22 30.2 & @xmath11200 37 07 & 1.2 ca & sep 08 96 & @xmath21 & 75 & 15.26 & 7.3 & 1 + 8 & ngc 891 & 02 22 33.1 & + 42 20 48 & 1.0 hl & sep 05 97 & @xmath21 & 30 & 10.81 & 3.0 & 1 + & & & & 2.2 ca & feb 14 98 & @xmath32 & 20 & & 3.0 & 1 + 9 & eso 416-g25 & 02 48 41.0 & @xmath11231 32 10 & 2.2 eso & june 87 & @xmath22 & & 14.64 & 3.0 & 2 + 10 & ugc 2411 & 02 58 00.9 & + 75 45 00 & 1.1 lo & mar 17 96 & @xmath21 & 20 & 16.50 & 8.5 & 1 + 11 & ic 1877 & 03 03 10.0 & @xmath11250 30 43 & 0.6 bo & dec 28 96 & @xmath21 & 88 & 16.30 & 3.0 & 1 + 12 & eso 201-g22 & 04 09 00.4 & @xmath11248 43 35 & 0.6 bo & dec 25 96 & @xmath21 & 148 & 14.69 & 5.0 & 1 + 13 & ngc 1886 & 05 12 48.7 & @xmath11223 48 45 & 0.6 bo & jan 06 97 & @xmath21 & 30 & 13.60 & 3.5 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 20 & & 3.5 & 1 + 14 & ugc 3474 & 06 32 00.6 & + 71 33 00 & 1.1 lo & mar 17 96 & @xmath21 & 10 & 15.40 & 6.0 & 1 + & & & & 2.2 ca & feb 14 98 & @xmath32 & 27 & & 6.0 & 1 + 15 & ngc 2310 & 06 53 53.6 & @xmath11240 51 44 & 0.6 bo & jan 05 07 & @xmath21 & 30 & 12.74 & -2.0 & 1 + 16 & ugc 4278 & 08 13 59.0 & + 45 54 43 & 1.1 lo & mar 16 96 & @xmath21 & 10 & 13.07 & 7.0 & 1 + & & & & 1.2 ca & mar 02 96 & @xmath34 & 67 & & 7.0 & 1 + 17 & eso 564-g27 & 09 11 54.4 & @xmath11220 07 03 & 2.2 eso & june 87 & @xmath22 & & 14.35 & 6.3 & 2 + 18 & ugc 4943 & 09 19 58.1 & + 37 11 27 & 1.1 lo & mar 16 96 & @xmath21 & 6 & 14.80 & 3.0 & 1 + & & & & 1.2 ca & mar 06 96 & @xmath34 & 8 & & 3.0 & 1 + 19 & ic 2469 & 09 23 00.9 & @xmath11232 26 59 & 1.5 da & jun 02 08 & @xmath22 & 5 & 13.03 & 2.0 & 1 + 20 & ugc 5341 & 09 56 36.6 & + 20 38 53 & 0.6 bo & jan 02 97 & @xmath21 & 134 & 15.03 & 6.0 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 27 & & 6.0 & 1 + rllllcccrrc & & & & & & & & & & + & & & & & & & & & & + & & & & & & & & & & + 21 & ic 2531 & 09 59 55.7 & @xmath11229 36 55 & 1.5 da & apr 08 97 & @xmath22 & 30 & 12.90 & 5.3 & 1 + 22 & ngc 3390 & 10 48 04.0 & @xmath11231 31 57 & 1.5 da & jun 02 98 & @xmath22 & 30 & 12.85 & 3.0 & 1 + 23 & eso 319-g26 & 11 30 20.0 & @xmath11241 03 57 & 2.2 eso & june 87 & @xmath22 & & 14.58 & 5.3 & 2 + 24 & ngc 3957 & 11 54 01.1 & @xmath11219 34 06 & 1.5 da & jun 03 98 & @xmath22 & 15 & 12.81 & -1.0 & 1 + 25 & ngc 4013 & 11 58 31.7 & + 43 56 48 & 1.1 lo & jun 01 97 & @xmath21 & 18 & 12.19 & 3.0 & 1 + & & & & 2.2 ca & feb 15 98 & @xmath32 & 33 & & 3.0 & 1 + 26 & eso 572-g44 & 12 01 09.0 & @xmath11220 29 18 & 1.5 da & jun 03 98 & @xmath22 & 20 & 15.04 & 3.0 & 1 + 27 & ugc 7170 & 12 10 37.0 & + 18 49 24 & 1.1 lo & mar 17 96 & @xmath21 & 15 & 14.96 & 6.0 & 1 + 28 & eso 321-g10 & 12 11 42.0 & @xmath11238 32 53 & 2.2 eso & june 87 & @xmath22 & & 14.22 & 1.4 & 2 + 29 & ngc 4217 & 12 15 50.9 & + 47 05 32 & 1.1 lo & jun 01 97 & @xmath21 & 30 & 12.04 & 3.0 & 1 + 30 & ngc 4244 & 12 17 30.0 & + 37 48 27 & 1.1 lo & jun 01 97 & @xmath21 & 20 & 10.88 & 6.0 & 1 + 31 & ugc 7321 & 12 17 34.1 & + 22 32 21 & 1.1 lo & mar 16 96 & @xmath21 & 20 & 14.15 & 7.0 & 1 + & & & & 1.2 ca & mar 02 96 & @xmath34 & 59 & & 7.0 & 1 + 32 & ngc 4302 & 12 21 42.5 & + 14 36 05 & 1.0 hl & may 19 96 & @xmath21 & 10 & 12.50 & 5.0 & 1 + & & & & 1.2 ca & mar 02 96 & @xmath34 & 53 & & 5.0 & 1 + 33 & ngc 4330 & 12 23 16.5 & + 11 22 06 & 1.5 da & apr 09 97 & @xmath22 & 40 & 13.09 & 6.0 & 1 + & & & & 2.2 eso & apr 10 97 & @xmath32 & 32 & & 6.0 & 1 + 34 & ngc 4565 & 12 36 20.6 & + 25 59 05 & 1.0 hl & may 20 96 & @xmath21 & 10 & 10.42 & 3.0 & 1 + & & & & 2.2 ca & feb 16 98 & @xmath32 & 13 & & 3.0 & 1 + 35 & ngc 4710 & 12 49 39.0 & + 15 09 55 & 1.1 lo & jun 03 97 & @xmath21 & 15 & 11.91 & -1.0 & 1 + & & & & 2.2 ca & feb 16 98 & @xmath32 & 20 & & -1.0 & 1 + 36 & ngc 5170 & 13 29 49.0 & @xmath11217 57 59 & 1.5 da & apr 08 97 & @xmath22 & 45 & 12.06 & 5.0 & 1 + & & & & 1.2 ca & mar 03 96 & @xmath34 & 11 & & 5.0 & 1 + & & & & 1.2 ca & mar 05 96 & @xmath29 & 8 & & 5.0 & 1 + 37 & eso 510-g18 & 13 55 32.0 & @xmath11227 24 47 & 1.5 da & jun 03 98 & @xmath22 & 30 & 16.21 & 1.0 & 1 + 38 & ugc 9242 & 14 25 20.9 & + 39 32 22 & 1.1 lo & mar 15 96 & @xmath21 & 21 & 14.09 & 7.0 & 1 + & & & & 1.2 ca & mar 04 96 & @xmath34 & 60 & & 7.0 & 1 + 39 & ngc 5775 & 14 53 57.7 & + 03 32 40 & 1.5 da & jun 03 98 & @xmath22 & 15 & 12.24 & 5.0 & 1 + 40 & ngc 5907 & 15 15 53.8 & + 56 19 46 & 1.2 ca & jun 02 96 & @xmath21 & 30 & 11.12 & 5.0 & 1 + & & & & 1.2 ca & jun 04 96 & @xmath34 & 6 & & 5.0 & 1 + 41 & ngc 5908 & 15 16 43.5 & + 55 24 40 & 1.2 ca & sep 08 96 & @xmath21 & 45 & 12.79 & 3.0 & 1 + & & & & 2.2 ca & feb 16 98 & @xmath32 & 33 & & 3.0 & 1 + 42 & eso 583-g08 & 15 57 50.5 & @xmath11222 29 47 & 1.5 da & jun 04 98 & @xmath22 & 35 & & 4.0 & 1 + 43 & ngc 6181 & 16 32 20.9 & + 19 49 30 & 1.1 lo & jun 03 97 & @xmath21 & 15 & 12.49 & 5.0 & 1 + 44 & eso 230-g11 & 18 46 24.0 & @xmath11252 09 23 & 1.5 da & jun 04 98 & @xmath22 & 30 & 13.74 & 4.0 & 1 + 45 & ngc 6722 & 19 03 40.0 & @xmath11264 53 41 & 2.2 eso & june 87 & @xmath22 & & 13.54 & 3.0 & 2 + 46 & eso 461-g06 & 19 51 55.9 & @xmath11231 58 52 & 1.5 da & jun 04 98 & @xmath22 & 40 & 16.21 & 5.0 & 1 + 47 & eso 339-g16 & 20 00 07.0 & @xmath11240 43 03 & 2.2 eso & june 87 & @xmath22 & & 16.50 & 1.0 & 2 + 48 & ic 4937 & 20 05 18.0 & @xmath11256 15 20 & 2.2 eso & june 87 & @xmath22 & & 14.86 & 3.0 & 2 + 49 & eso 187-g08 & 20 43 25.2 & @xmath11256 12 17 & 1.5 da & jun 04 98 & @xmath22 & 30 & 15.69 & 6.0 & 1 + 50 & ic 5052 & 20 52 06.3 & @xmath11269 12 14 & 1.5 da & jun 03 98 & @xmath22 & 35 & 11.16 & 7.0 & 1 + & & & & 2.2 eso & apr 10 97 & @xmath32 & 31 & & 7.0 & 1 + 51 & ic 5096 & 21 18 22.0 & @xmath11263 45 41 & 1.5 da & jun 03 98 & @xmath22 & 30 & 13.30 & 4.0 & 1 + 52 & eso 466-g01 & 21 42 32.0 & @xmath11229 22 10 & 2.2 eso & june 87 & @xmath22 & & 14.63 & 2.0 & 2 + 53 & eso 189-g12 & 21 55 38.7 & @xmath11254 52 33 & 1.5 da & jun 04 98 & @xmath22 & 30 & 15.59 & 5.0 & 1 + 54 & ugc 11859 & 21 58 07.3 & + 01 00 34 & 2.2 eso & june 87 & @xmath22 & & 15.16 & 4.0 & 2 + 55 & eso 533-g04 & 22 14 03.2 & @xmath11226 56 18 & 1.5 da & jun 03 98 & @xmath22 & 70 & 14.18 & 4.8 & 1 + 56 & ic 5199 & 22 19 33.0 & @xmath11237 32 01 & 1.5 da & jun 04 98 & @xmath22 & 30 & 15.00 & 3.0 & 1 + 57 & ugc 11994 & 22 20 53.4 & + 33 17 34 & 1.0 hl & sep 07 97 & @xmath21 & 54 & 14.85 & 4.0 & 1 + 58 & ugc 12281 & 22 59 12.4 & + 13 36 21 & 1.2 ca & sep 07 96 & @xmath21 & 60 & 14.79 & 8.0 & 1 + & & & & 2.2 ca & sep 04 96 & @xmath29 & 20 & & 8.0 & 1 + 59 & ugc 12423 & 23 13 06.0 & + 06 24 00 & 1.2 ca & sep 07 96 & @xmath21 & 60 & 14.53 & 5.0 & 1 + & & & & 2.2 ca & sep 03 96 & @xmath29 & 27 & & 5.0 & 1 + 60 & ngc 7518 & 23 13 12.9 & + 06 19 16 & 1.1 lo & jun 02 97 & @xmath21 & 40 & 14.24 & 1.0 & 1 + 61 & eso 604-g06 & 23 14 54.0 & @xmath11220 59 44 & 1.5 da & jun 03 98 & @xmath22 & 40 & 15.00 & 4.0 & 1 +

reservoir computing is a recently introduced , highly efficient bio - inspired approach for processing time dependent data . the basic scheme of reservoir computing consists of a non linear recurrent dynamical system coupled to a single input layer and a single output layer . within these constraints many implementations are possible . here we report an opto - electronic implementation of reservoir computing based on a recently proposed architecture consisting of a single non linear node and a delay line . our implementation is sufficiently fast for real time information processing . we illustrate its performance on tasks of practical importance such as nonlinear channel equalization and speech recognition , and obtain results comparable to state of the art digital implementations .

1111.7219

the remarkable speed and multiplexing capability of optics makes it very attractive for information processing . these features have enabled the telecommunications revolution of the past decades . however , so far they have not been exploited insomuch as computation is concerned . the reason is that optical nonlinearities are very difficult to harness : it remains challenging to just demonstrate optical logic gates , let alone compete with digital electronics@xcite . this suggests that a much more flexible approach is called for , which would exploit as much as possible the strenghts of optics without trying to mimic digital electronics . reservoir computing @xcite , a recently introduced , bio - inspired approach to artificial intelligence , may provide such an opportunity . here we report the first experimental reservoir computer based on an opto - electronic architecture . as nonlinear element we exploit the sine nonlinearity of an integrated mach - zehnder intensity modulator ( a well known , off - the - shelf component in the telecommunications industry ) , and to store the internal states of the reservoir computer we use a fiber optics spool . we report results comparable to state of the art digital implementations for two tasks of practical importance : nonlinear channel equalization and speech recognition . reservoir computing , which is at the heart of the present work , is a highly successful method for processing time dependent information . it provides state of the art performance for tasks such as time series prediction @xcite ( and notably won a financial time series prediction competition@xcite ) , nonlinear channel equalization @xcite , or speech recognition @xcite . for some of these tasks reservoir computing is in fact the most powerful approach known at present . the central part of a reservoir computer is a nonlinear recurrent dynamical system that is driven by one or multiple input signals . the key insight behind reservoir computing is that the reservoir s response to the input signal , i.e. , the way the internal variables depend on present and past inputs , is a form of computation . experience shows that in many cases the computation carried out by reservoirs , even randomly chosen ones , can be extremely powerful . the reservoir should have a large number of internal ( state ) variables . the exact structure of the reservoir is not essential : for instance , in some works the reservoir closely mimics the interconnections and dynamics of biological neurons in a brain @xcite , but many other architectures are possible . to achieve useful computation on time dependent input signals , a good reservoir should be able to compute a large number of different functions of its inputs . that is , the reservoir should be sufficiently high - dimensional , and its responses should not only depend on present inputs but also on inputs up to some finite time in the past . to achieve this , the reservoir must operate at the threshold of dynamical instability which ensures that the system exhibits a `` fading memory '' , meaning that it will gradually forget previous inputs . reservoir computing is a versatile and flexible concept . this follows from two key points : 1 ) many of the details of the nonlinear reservoir itself are unimportant except for the dynamic regime ( threshold of instability ) which can be tuned by some global parameters ; and 2 ) the only part of the system that is trained is a linear output layer . because of this flexibility , reservoir computing is amenable to a large number of experimental implementations . thus proof of principle demonstrations have been realized in a bucket of water @xcite and using an analog vlsi chip @xcite , and arrays of semiconductor amplifiers have been considered in simulation @xcite . however , it is only very recently that an analog implementation with performance comparable to digital implementations has been reported : namely , the electronic implementation presented in @xcite . our experiment is based on a similar architecture as that of @xcite , namely a single non linear node and a delay line . the main differences are the type of non linearity and the desynchronisation of the input with respect to the period of the delay line . these differences highlight the flexibility of the concept . the performance of our experiment on two benchmark tasks , isolated digit recognition and non linear channel equalization , is comparable to state of the art digital implementations of reservoir computing . compared to@xcite , our experiment is almost 6 orders of magnitude faster , and a further 2 - 3 orders of magnitude speed increase should be possible with only small changes to the system . the flexibility of reservoir computing and its success on hard classification tasks makes it a promising route for realizing computation in physical systems other than digital electronics . in particular it may provide innovative solutions for ultra fast or ultra low power computation . in supplementary material 1 we describe reservoir computing in more detail and provide a road map for building high performance analog reservoir computers . in the discussion we discuss how to measure the speed of experimental reservoir computers . understanding how physical systems can process information is a major scientific challenge . the tremendous progress that has been accomplished during the past century has given rise to whole new fields of science , such as computer science and digital ( silicon ) electronics , or quantum information . information processing by classical analog systems is a case apart because remarkable examples are found in the biological sciences ( e.g. the brain ) , but our understanding is still very incomplete . thus , we understand at the basic level many aspects of how information is processed using biochemical reactions in cells , or how information is processed by neurons in a brain , to take two examples . but how these elementary processes are organized at a higher level , and what is the large scale architecture of these systems , still escapes us . understanding these issues would be of great conceptual interest . it could also have huge technological repercussions , as it could completely change the way we build information processing machines . that this tremendous scope for progress exists is illustrated by the approximately 6 orders of magnitude gap in energy consumption between a brain and a present day silicon computerwatt . for a presentation of the simulation , see @xcite ] . so far most work on information processing in analog systems has been based on imitating biological systems . this has given rise to the field of artificial neural networks . more abstract approaches have also been developed , such as hidden markov models , or support vector machines . reservoir computing @xcite provides an alternative line of attack . the interest of reservoir computing is that 1 ) it proposes architectures that are quite different from those studied up to now ; and 2 ) despite a relatively recent history -the first papers on the topic date from 2002- , its performances are comparable to , and sometimes even exceed , those of other approaches to machine learning . a deeper understanding of reservoir computing could provide new insights into how analog classical systems , and in particular biological systems , process information . additionally it could give rise to novel approaches for implementing computation and enable new applications . the aim of the present text is not to duplicate existing reviews of reservoir computing @xcite . rather , we wish to present the subject from the point of view of the physicist or engineer who wishes to build an experimental reservoir computer . we will argue that reservoir computing provides a quite detailed , but also quite flexible , road map towards building physical systems that process information with architectures very different from those used up to now . before introducing our implementation , we recall a few key features of reservoir computing ; for a more detailed treatment of the underlying theory , we refer the reader to supplementary material 1 . as is traditional in the literature , we will consider tasks that are defined in discrete time , e.g. , using sampled signals . we denote by @xmath0 the input signal , where @xmath1@xmath2 is the discretized time ; by @xmath3 the internal states of the system used as reservoir ; and by @xmath4 the output of the reservoir . a typical evoultion law for @xmath3 is @xmath5 , where @xmath6 is a nonlinear function , @xmath7 is the time independent connection matrix and @xmath8 is the time independent input mask . note that in our work we will use a slightly different form for the evolution law , as explained below . in order to perform the computation one needs a readout mechanism . to this end we define a subset @xmath9 ( also in discrete time ) of the internal states of the reservoir . it is these states which are observed and used to build the output . the time dependent output is obtained in an output layer by taking a linear combination of the internal states of the reservoir @xmath10 . the readout weights @xmath11 are chosen to minimize the mean square error ( mse ) between the estimator @xmath4 and a target function @xmath12 : @xmath13 over a set of examples ( the training set ) . because the mse is a quadratic function of the @xmath11 the optimal weights can be easily computed from the knowledge of @xmath14 and @xmath12 . in a typical run , the quality of the reservoir is then evaluated on a second set of examples ( the test set ) . after training , the @xmath11 are kept fixed . in the present work we use an architecture related to that used in @xcite and to the minimum complexity networks studied in @xcite . as in @xcite , the reservoir is based on a non - linear system with delayed feedback ( a class of systems widely studied in the nonlinear dynamics community , see e.g. @xcite ) and consists of a single nonlinear node and a delay loop . the information about the previous internal state of the reservoir up to some time @xmath15 in the past is stored in the delay loop . after a period @xmath15 of the loop , the entire internal state has been updated ( processed ) by the nonlinear node . in contrast to the work described in @xcite , the nonlinear node in our implementation is essentially instantaneous . hence , in the absence of input , the dynamics of our system can be approximated by the simple recursion @xmath16 where @xmath17 ( the _ feedback gain _ ) and @xmath18 ( the _ bias _ ) are adjustable parameters and we have explicitly written the sine nonlinearity used in our implementation . we will use this system to perform useful computation on input signals @xmath0 evolving in discrete time @xmath19 as the system itself operates in continuous time , we need to define ways to convert input signal(s ) to continuous time and to convert the system s state back to discrete time . the first is achieved by using a sample and hold procedure . we obtain a piecewise constant function @xmath20 of the continuous variable @xmath21 : @xmath22 . the time @xmath23 is taken to be less than or equal to the period @xmath15 of the delay loop ; when @xmath24 we are in the unsynchronised regime ( see below ) . to discretize the system s state , we note that the delay line acts as a memory , storing the delayed states of the nonlinearity . from this large - dimensional state space , we take @xmath25 samples by dividing the input period @xmath26 into @xmath25 segments , each of duration @xmath27 and sampling the state of the delay line at a single point with periodicity @xmath27 . this provides us with @xmath25 _ snapshots _ of the nonlinearity s reponse to each input sample @xmath0 . from these snapshots , we construct @xmath25 discrete - time sequences @xmath28 ( @xmath29 ) to be used as reservoir states from which the required ( discrete - time ) output is to be constructed . without further measures , all such recorded reservoir states would be identical , so for computational purposes our system is one - dimensional . in order to use this system as a reservoir computer , we need to drive it in such a way that the @xmath14 represent a rich variety of functions of the input history . it is often helpful @xcite to use an `` input mask '' that breaks the symmetry of the system . in @xcite good performance was improved by using a nonlinear node with an intrinsic time scale longer than the time scale of the input mask . in the present work we also use the `` input mask '' , but as our nonlinearity is instantaneous , we can not exploit its intrinsic time scale . we instead chose to desynchronize the input and the reservoir , that is , we hold the input for a time @xmath26 which differs slightly from the period @xmath15 of the delay loop . this allows us to use each reservoir state at time @xmath30 for the generation of a new different state at time @xmath31 ( unlike the solution used in @xcite where the intrinsic time scale of the nonlinear node makes the successive states highly correlated ) . we now explain these important notions in more detail . the input mask @xmath32 is a periodic function of period @xmath26 . it is piecewise constant over intervals of length @xmath27 , i.e. , @xmath33 when @xmath34 , for @xmath35 . the values @xmath36 of the mask are randomly chosen from some probability distribution . the reservoir is driven by the product @xmath37 of the input and the mask , with @xmath38 an adjustable parameter ( the _ input gain _ ) . the dynamics of the driven system can thus be approximated by @xmath39 it follows that the reservoir states can be approximated by @xmath40 when @xmath41 ( the synchronized regime ) or more generally as @xmath42 when @xmath43 , ( @xmath44 ) ( the unsynchronized regime ) . in the synchronized regime , the reservoir states correspond to the responses of @xmath25 uncoupled discrete - time dynamical systems which are similar , but slightly different through the randomly chosen @xmath36 . in the unsynchronized regime , with a desynchronization @xmath45 , the state equations become coupled , yielding a much richer dynamics . with an instantaneous nonlinearity , desynchronisation is necessary to obtain a set of state transformations that is useful for reservoir computing . we beleive that it will also be useful when the non linearity has an intrinsic time scale , as it provides a very simple way to enrich the dynamics . in summary , by using an input mask , combined with desynchronization of the input and the feedback delay , we have turned a system with a one - dimensional information representation into an n - dimensional system . the above architecture is implemented in the experiment depicted in fig . [ fig : experimental - set - up . ] . the sine nonlinearity is implemented by a voltage driven intensity modulator ( lithium niobate mach zehnder interferometer ) , placed at the output of a continuous light source , and the delay loop is a fiber spool . a photodiode converts the light intensity @xmath46 at the output of the fiber spool into a voltage ; this is mixed with an input voltage generated by a function generator and proportional to @xmath47 , amplified , and then used to drive the intensity modulator . the feedback gain @xmath17 is set by adjusting the average intensity @xmath48 of the light inside the fiber loop with an optical attenuator . by changing @xmath17 we can bring the system to the edge of dynamical instability . the nonlinear dynamics of this system have already been extensively studied , see @xcite . the dynamical variable @xmath49 is obtained by rescaling the light intensity to lie in the interval @xmath50 $ ] through @xmath51 . then , neglecting the effect of the bandpass filter induced by the electronic amplifiers , the dynamics of the system is given by eq . ( [ eq : dynamics ] ) where @xmath17 is proportional to @xmath48 . equation ( [ eq : dynamics ] ) , as well as the discretized versions thereof , eqs . ( [ eq : discretedynamics ] ) and ( [ eq : discreteunsynchronizeddynamics ] ) , are derived in the supplementary material ; the various stages of processing of the reservoir nodes and inputs are shown in fig . [ fig : diagram ] . in our experiment the round trip time is @xmath52s and we typically use @xmath53 internal nodes . the parameters @xmath17 and @xmath38 in eq . ( [ eq : dynamics ] ) are adjusted for optimal performance ( their optimal value may depend on the task ) , while @xmath18 is usually set to 0 . the intensity @xmath46 is recorded by a digitizer , and the estimator @xmath4 is reconstructed offline on a computer . we illustrate the operations of our reservoir computer in fig.[fig : signalclass ] where we consider a very simple signal recognition task . here , the input to the system is taken to be a random concatenation of sine and square waves ; the target function @xmath12 is @xmath54 for a sine wave and @xmath55 for a square wave . the top panel of fig . [ fig : signalclass ] shows the input to the reservoir : the blue line is the representation of the input in continuous time @xmath20 . in the bottom panel , the output of the network after training is shown with red crosses , against the desired output represented by a blue line . the performance on this task is essentially perfect : the normalized mean square error @xmath56 reaches @xmath57 , which is significantly better than the results reported using simulations in @xcite . we have checked the performance of this system extensively in simulations . first of all , if we neglect the effects of the bandpass filters , and neglect all noise introduced in our experiment , we obtain a discretized system described by eq . ( [ eq : discreteunsynchronizeddynamics ] ) which is similar to ( but nevertheless distinct from ) the minimum complexity reservoirs introduced in @xcite . we have checked that this discretized version of our system has performance similar to usual reservoirs on several tasks . this shows that the chosen architecture is capable of state of the art reservoir computing , and sets for our experimental system a performance goal . secondly we have also developed a simulation code that takes into account all the noises of the experimental components , as well as the effects of the bandpass filters . these simulations are in very good agreement with the experimentally measured dynamics of the system . they allow us to efficiently explore the experimental parameter space , and to validate the experimental results . further details on these two simulation models are given in the supplementary information . we apply our optoelectronic reservoir to three tasks . these tasks are benchmarks which have been widely used in the reservoir computing community to evaluate the performance of reservoirs . they therefore allow comparison between our experiment and state of the art digital implementations of reservoir computing . for the first task , we train our reservoir computer to behave like a nonlinear auto regressive moving average equation of order 10 , driven by white noise ( narma10 ) . more precisely , given the white noise @xmath0 , the reservoir should produce an output @xmath4 which should be as close as possible to the response @xmath12 of the narma10 model to the same white noise . the task is described in detail in the methods section . the performance is measured by the normalised mean square error ( nmse ) between output @xmath4 and target @xmath12 . for a network of 50 nodes , both in simulations and experiment , we obtain a @xmath58 . this is similar to the value obtained using digital reservoirs of the same size . for instance a nmse value of @xmath59 is reported in @xcite also for a reservoir of size 50 . for our second task we consider a problem of practical relevance : the equalization of a nonlinear channel . we consider a model of a wireless communication channel in which the input signal @xmath60 travels through multiple paths to a nonlinear and noisy receiver . the task is to reconstruct the input @xmath60 from the output @xmath0 of the receiver . the model we use was introduced in @xcite and studied in the context of reservoir computing in @xcite . our results , given in fig . [ fig : channelequalization ] , are one order of magnitude better than those obtained in @xcite with a nonlinear adaptive filter , and comparable to those obtained in @xcite with a digital reservoir . at 28 db of signal to noise ratio , for example , we obtain an error rate of @xmath61 , while the best error rate obtained in @xcite is @xmath62 and in @xcite error rates between @xmath63 and @xmath64 are reported . finally we apply our reservoir to isolated spoken digits recognition using a benchmark task introduced in the reservoir computing community in @xcite . the performance on this task is measured using the word error rate ( wer ) which gives the percentage of words that are wrongly classified . performances reported in the literature are a wer of @xmath65 using a hidden markov model @xcite ; wers of @xmath66 @xcite , of @xmath67 @xcite , of @xmath68 @xcite for reservoir computers of different sizes and with different post processing of the output . the experimental reservoir presented in @xcite reported a wer of @xmath67 . our experiment yields a wer of @xmath69 , using a reservoir of @xmath70 nodes . further details on these tasks are given in the methods section and in supplementary material 2 . we have reported the first demonstration of an opto - electronic reservoir computer . our experiment has performance comparable to state of the art digital implementations on benchmark tasks of practical relevance such as speech recognition and channel equalization . our work demonstrates the flexibility of reservoir computers that can be readily reprogrammed for different tasks . indeed by re - optimising the output layer ( that is , choosing new readout weights @xmath71 ) , and by readjusting the operating point of the reservoir ( changing the feedback gain @xmath17 , the input gain @xmath38 , and possibly the bias @xmath18 ) one can use the same reservoir for many different tasks . using this procedure , our experimental reservoir computer has been used successively for tasks such as signal classification , modeling a dynamical system ( narma10 task ) , speech recognition , and nonlinear channel equalization . we have introduced a new feature in the architecture , as compared to the related experiment reported in @xcite . namely by desynchronizing the input with respect to the period of the reservoir we conserve the necessary coupling between the internal states , but make a more efficient use of the internal states as the correlations introduced by the low pass filter in @xcite are not necessary . our experiment is also the first implementation of reservoir computing fast enough for real time information processing . it can be converted into a high speed reservoir computer simply by increasing the bandwidth of all the components ( an increase of at least 2 orders of magnitude is possible with off - the - shelf optoelectronic components ) . we note that in future realisations it will be necessary to have an analog implementation of the pre - processing of the input ( digitisation and multiplication by the input mask ) and of the post - processing of the output ( multiplication by output weights ) , rather than the digital pre- and post - processing used in the present work . from the point of view of applications , the present work thus constitutes an important step towards building ultra high speed optical reservoir computers . to help achieve this goal , in the supplementary material we present guidelines for building experimental reservoir computers . whether optical implementations can eventually compete with electronic implementations is an open question . from the fundamental point of view , the present work helps understanding what are the minimal requirements for high level analog information processing . we would like to thank j. van campenhout whose suggestions initiated this research project . s.m . acknowledges a helpful discussion at the beginning of this project in which ingo fisher . all authors would like to thank the researchers of the photonics@be network working on reservoir computing for numerous discussions over the duration of this project . the authors acknowledge financial support by interuniversity attraction poles program ( belgian science policy ) project photonics@be iap6/10 and by the fonds de la recherche scientifique frs - fnrs . auto regressive models and moving average models , and their generalisation nonlinear auto regressive moving average models ( narma ) , are widely used to simulate time series . the narma10 model is given by the recurrence @xmath72 where @xmath0 is a sequence of random inputs drawn from an uniform distribution over the interval @xmath73 $ ] . the aim is to predict the @xmath12 knowing the @xmath74this task was introduced in @xcite . it has been widely used as a benchmark in the reservoir computing community , see for instance @xcite this task was introduced in @xcite , and used in the reservoir computing community in @xcite and @xcite . the input to the channel is an i.i.d . random sequence @xmath60 with values from @xmath75 . the signal first goes through a linear channel , yielding @xmath76 it then goes through a noisy nonlinear channel , yielding @xmath77 where @xmath78 is an i.i.d . gaussian noise with zero mean adjusted in power to yield signal - to - noise ratios ranging from 12 to 32 db . the task is , given the output @xmath0 of the channel , to reconstruct the input @xmath60 . the performance on this task is measured using the symbol error rate , that is the fraction of inputs @xmath60 that are misclassified ( ref . @xcite used another error metric on this task ) . the data for this task is taken from the nist ti-46 corpus @xcite . it consists of ten spoken digits ( 0 ... 9 ) , each one recorded ten times by five different female speakers . these 500 spoken words are sampled at 12.5 khz . this spoken digit recording is preprocessed using the lyon cochlear ear model @xcite . the input to the reservoir @xmath79 consists of an 86-dimensional state vector ( @xmath80 ) with up to 130 time steps . the number of variables is taken to be @xmath81 . the input mask is taken to be a @xmath82 dimensional matrix @xmath83 with elements taken from the the set @xmath84 with equal probabilities . the product @xmath85 of the mask with the input is used to drive the reservoir . ten linear classifiers @xmath86 ( @xmath87 ) are trained , each one associated to one digit . the target function for @xmath88 is + 1 if the spoken digit is @xmath89 , and -1 otherwise . the classifiers are averaged in time , and a winner - takes - all approach is applied to select the actual digit . using a standard cross - validation procedure , the 500 spoken words are divided in five subsets . we trained the reservoir on four of the subsets , and then tested it on the fifth one . this is repeated five times , each time using a different subset as test , and the average performance is computed . the performance is given in terms of the word error rate , that is the fraction of digits that are misclassified . we obtain a wer of @xmath69 ( which correspond to 2 errors in 500 recognized digits ) . scheme for the experimental set - up . the red and green parts represent respectively the optical and electronic components . the optical part of the setup is fiber based , and operates around 1550 nm ( standard telecommunication wavelength ) . `` m - z '' : lithium niobate mach - zehnder modulator . `` @xmath18 '' : dc voltage determining the operating point of the m - z modulator . `` combiner '' : electronic coupler adding the feedback and input signals . `` awg '' : arbitrary waveform generator . a computer generates the input signal for a task and feeds it into the system using the arbitrary waveform generator . the response of the system is recorded by a digitiser and retrieved by the computer which optimizes the read - out function in a post processing stage . the feedback gain @xmath17 is adjusted by changing the average intensity inside the loop with the optical attenuator . the input gain @xmath38 is adjusted by changing the output voltage of the function generator by a multiplicative factor . the bias @xmath18 is adjusted by using a dc voltage to change the operating point of the m - z modulator . the operation of the system is fully automated and controlled by a computer using matlab scripts . ] schematic diagram of the information flow in the experiment depicted in fig . [ fig : experimental - set - up . ] . on the plot we have represented four reservoir nodes at different stages of processing , labelled according to equation [ eq : discreteunsynchronizeddynamics ] with @xmath90 starting from the bottom , and going clockwise , a input value @xmath0 gets multiplied by an input gain @xmath38 and a mask value @xmath91 , then mixed with the previous node state @xmath92 . the result goes through the sine function to give the new state of the reservoir @xmath14 , which then gets amplified by a factor @xmath17 and , after the delay , will get mixed with a new input @xmath93 . all the network states @xmath14 are also collected by the readout unit , multiplied by their respective weights @xmath11 and added together to give the desired output @xmath4 . ] , a stepwise constant function resulting from the discretization of successive step and sine functions . the bottom panel shows in red crosses the output of the reservoir @xmath94 . the target function ( dashed line in the lower panel ) is equal to 1 when the input signal is a step function and to 0 when the input signal is a sine function . the normalized mean square error , evaluated over @xmath95 inputs , is @xmath96 . [ fig : signalclass ] ] ) ( blue rhombs ) , and the continuous simulations that take into account noise and bandpass filters in the experiment ( red squares ) . all three sets of results agree within the statistical error bars . error bars on the experimental points relative to 24 , 28 and 32 db might be overestimated ( see supplementary material 2 ) . the results are practically identical to those obtained using a digital reservoir in @xcite . [ fig : channelequalization ] ] to understand the potentialities of reservoir computing , it is best to start with an example , and then examine to what extent the example can be generalized . finally , on the basis of this discussion , we outline a road map for building experimental reservoir computers . the archetypal reservoir computer is constructed as follows . it consists of a finite number @xmath25 of internal variables @xmath14 @xmath97 evolving in discrete time @xmath98 . the evolution of the variables @xmath14 is perturbed by an external input @xmath0 . the evolution equations for the reservoir is : @xmath99,\label{eq : restanh}\ ] ] where the ( time independent ) coefficients @xmath100 and @xmath101 are chosen independently at random from a simple distribution , for instance gaussian distributions with mean zero and variances @xmath102 and @xmath103 . the dynamics of this system ( called the `` reservoir '' ) is fixed ( i.e. the coefficients @xmath100 and @xmath101 are fixed ) . the dimensionality @xmath25 of the reservoir is typically much larger than the input dimensionality . given proper parameters @xmath102 and @xmath103 , the reservoir will also have memory of past inputs . the use of the @xmath104 nonlinearity in eq . ( [ eq : restanh ] ) is traditional , but by no means essential , see below . the aim of reservoir computing is to perform a computation on the input @xmath0 . to this end one maps the internal state of the reservoir to an output signal denoted @xmath4 . the output should be as close as possible to a given target function @xmath12 . an example of this would be the detection of a stereotypical pattern in the input sequence such as spoken digits in a stream of audio , or a certain header in digital communication . another example would be trying to predict future inputs such as electrical load prediction or financial time - series prediction . the output @xmath4 of the reservoir computer at time @xmath30 is a linear combination of the high dimensional internal states of the reservoir at time @xmath30 : @xmath105 the @xmath11 are usuallly chosen as follows . for some time interval , @xmath106 $ ] we simulate the reservoir such that one knows both the inputs @xmath0 and the target function @xmath12 . we minimize , over this training interval , the mean square error between @xmath4 and @xmath12 : @xmath107 this can be efficiently performed using standard linear regression techniques . [ in cases where a classification of the inputs is desired , one may train several outputs ( one for each class ) and then use a winner takes all approach , or one may try to optimize the miss - classification rate using techniques such as logistic regression , gaussian discriminant analysis , regression to the indicator function or a linear support vector machine . ] note that in general one should apply some form of regularization ( to reduce the model complexity ) to the readout function eq . ( [ eq : output ] ) . to this end one uses optimisation techniques that prefer models with small parameters @xmath11 . this is accomplished by using e.g. ridge regression . the coefficients @xmath11 are now fixed and the reservoir computer is ready for use : one can input into the reservoir new inputs @xmath0 and obtain the corresponding output @xmath4 . because of the training procedure where we ensured proper generalization , the output will continue to be close to the target function , even for new data . it is considered good practice to check this . to this end one estimates the performance of the reservoir in a test phase during which new inputs are fed into the reservoir , and one compares , using the previously chosen @xmath11 and an appropriate error metric , how close the output is to the target function . several remarks are now in order : 1 . for good performance , it is necessary to scale the coefficients by a multiplicative factor @xmath108 ( in other words , if the coefficients are drawn from normal distributions with mean zero , one adjusts the variances of the normals ) . these parameters are often called the feedback gain and input gain respectively . the necessity for this scaling can be understood intuitively as follows : if the coefficients @xmath100 are too small , then the state equation ( [ eq : restanh ] ) is strongly damped ; if the coefficients @xmath100 are too large , then the system has highly complex dynamics or is even chaotic , and its behavior is very sensitive to small changes in the input . the coefficients @xmath101 are adjusted in order to have an appropriate ratio between the contribution of the terms @xmath109 and the source term @xmath110 . they also determine how non - linearly the inputs are expanded in the reservoir states . 2 . finding the optimal values of @xmath11 is immediate in the case of linear regression . indeed inserting eq . ( [ eq : output ] ) into eq . ( [ eq : mse ] ) and extremizing with respect to @xmath11 , one finds that the optimal choice is @xmath111 where @xmath112 is the inverse of the correlation matrix @xmath113 and @xmath114 . this should be contrasted to what would happen if one tried to optimize the coefficients @xmath100 and @xmath101 . in this case there are many more coefficients to optimize , and furthermore it is difficult to find the optimal value , as one may very easily end up in local optima . the use of a linear readout eq . ( [ eq : output ] ) is thus highly advantageous . many variants on the above architecture have been investigated with success . we can list : 1 . using other nonlinearities than @xmath104 , such as the biological inspired models which approximate the way real neurons interact with each other ( spiking neuron models ) . 2 . using sparse connection matrices @xmath100 ( i.e. matrices with most of the elements set to zero ) , so as to decrease the computational resources required to simulate the evolution . 3 . using multiple inputs @xmath115 , @xmath116 . in this case one simply adds an index to the coefficients @xmath101 . the source term in eq . ( [ eq : restanh ] ) thus becomes @xmath117 . 4 . using the output @xmath4 as input to the reservoir . this modification allows training the reservoir to behave as a given dynamical system @xcite.[enu : using - the - output ] 4 . the above approach is extremely powerful . to take two examples : 1 given a time series @xmath0 for times @xmath118 , the task is to guess the subsequent values of the time series @xmath119 the time series could for instance come from the evolution of a chaotic dynamical system , such as the mackey - glass system . using the trick [ enu : using - the - output ] , results which outperform all other known approaches for predicting chaotic time series were obtained @xcite . reservoir computing also won an international competition on prediction of future evolution of financial time series @xcite . reservoir computing has been extensively applied to speech recognition . using well known data sets , reservoir computing outperforms other approaches for isolated digit recognition @xcite . recently reservoir computing has been applied to a more complex task : phoneme recognition . performances comparable to the best alternative methods were obtained using 3 coupled reservoirs , each with @xmath120 variables @xcite . the basic setup of reservoir computing , although typically implemented in software , suggests many promising new avenues to implement computation in analog dynamic systems . the theoretical requirements for reservoir computing to be computationally universal ( in the analog sense ) @xcite are very loose : the reservoir is required to have fading memory , to be excitable by the input and a high dimensional readout must be possible . many physical systems could be conceived that adhere to these rules and could thus potentially be turned in universal computing machines . however turning these general ideas into a working machine in more difficult . if one wishes to build an experimental reservoir computer , then it is essential to understand what are the constraints , but also the design freedom . in this respect , a number of important lessons can be learned from the example presented in the previous section . 1 . the fact that the interconnection matrix @xmath100 and the coefficients @xmath101 in eq . ( [ eq : restanh ] ) are chosen at random is extremely important . it means that fine tuning of a large number of coefficients is not necessary for a good reservoir computer . rather almost all interconnection matrices will give good performance . 2 . the fact that the tanh in eq . ( [ eq : restanh ] ) can be changed into other non - linearities is also extremely important . it means that one does not have to imitate specific non - linear behavior ( such as the specific dynamics of neurons in a biological brain ) , but one can use the non - linearities that are easily accessible experimentally . 3 . the fact that the only coefficients that are task specific are the @xmath11 ( the weights of the readout function ) implies that a given dynamical system can be used for many different information processing tasks . it also implies that one can separate the design and analysis of the reservoir itself from the design and analysis of the readout function . on the basis of these remarks , a reservoir computer can be built out of a dynamical system that satisfies the following constraints : 1 . it should consist of a large number ( say 50 , or more , but this depends on the task and on the specifics of the dynamic system ) dynamical variables which are coupled together by a non - linear evolution equation . the evolution of the dynamical system can be perturbed by the external input . 3 . as much as possible , one should try to break all symmetries of the system ( this is the message coming from the fact that the @xmath100 and @xmath101 are random : there is no residual structure / symmetry in the dynamics ) . to this end one should privilege dynamical systems that depend on a large number of random parameters . a few global parameters must be experimentally tunable . these global parameters are used to adjust the operating point of the system ( typically to just below the threshold of instability ) , and to adjust the overall weight given to the external inputs ( this corresponds to the scaling of the coefficients @xmath100 and @xmath101 ) . it should be possible to read - out the state of a large number ( better all ) of the dynamical variables , or at least to construct the readout function @xmath121 with adjustable weights @xmath11 . note that a reservoir computer can function satisfactorily even if only a subset of the dynamical variables are read out . once these constraints are satisfied , one can proceed to test the system ( either using numerical simulations , or using the experimental realization ) . in the reservoir computing literature , there exists a series of ( somewhat ) standardized tests on which to evaluate the performance of the system . some of these tests have a theoretical justification ( e.g. linear memory capacity ) ; others are interesting tasks which have been often used by the community . these tasks thus provide benchmarks with which to compare the performances of different reservoir computers . possible tasks to study include : 1 . linear memory capacity and memory function @xcite . in this task the target function @xmath122 is simply the input @xmath89 time steps in the past . performance on this task indicates how well the reservoir s state can be used to recover past inputs . if the linear memory capacity is small , then the reservoir will be unable to carry out tasks that require long memory of the input . non - linear memory capacity , see @xcite . in this task the target function is a non - linear function of the past inputs , for instance the product @xmath123 . performance on this task measures how much non - linear processing of the input is carried out by the reservoir . 3 . simulating specific nonlinear auto regressive moving average ( narma ) systems . for the narma tasks , the aim is to use the reservoir to simulat the response of a narma system driven by a random input . two variants are widely used : a 10th order system ( narma10 ) introduced in @xcite , and the more difficult 30th order system ( narma30 ) ( see for instance @xcite for a definition ) . 4 . predicting the evolution of the trajectory of chaotic attractors such as the attractor of the mackey - glass system @xcite . speech recognition : several benchmarks have been published , going from japanese vowel recognition @xcite , to isolated spoken digits @xcite , to phoneme recognition @xcite . this ( non - exhaustive ) list of benchmarks provides a natural road - map for future experimental reservoirs . first master easy tasks such as signal classification , or isolated digit recognition . then go on to moderately harder tasks such as narma10 or narma30 . finally , one can imagine tackling hard tasks such as phoneme recognition . in all cases , measure the linear and nonlinear memory functions , as they will give task independent information on how the reservoir is processing information . this road - map is appealing . steps along it have been followed using reservoirs based on water waves in a bucket @xcite , on the cerebral cortex of a cat @xcite , on analog vlsi chips @xcite , using a ( numerically simulated ) array of coupled semi - conductor optical amplifiers @xcite . however it is only very recently that it has been possible , using an electronics implementation and a delayed feedback loop , to demonstrate an analog reservoir computer with performance comparable to that of digital reservoirs on a non trivial task ( in this case isolated digit recognition ) @xcite . the above road map leaves open a number of important questions . we outline the most obvious : 1 . effect of noise . in experimental realizations of reservoir computing , there will inevitably be noise and imperfections . how will these affect the performance of the scheme ? there has not been an exhaustive analysis of this issue , but a few remarks are in order . one should distinguish between noise within the reservoir itself and noise in the readout . 1 . noise in the reservoir itself is deleterious . however , _ _ reservoir computers can continue to work even with moderate levels of noise . indeed the noise can be viewed as an unwanted input , and the aim is to perform the desired task while ignoring this second unwanted input . a possible approach to counteract the effect of noise would simply be to increase the size of the reservoir . but this remains to be studied in detail . 2 . noise in the readout can be beneficial . indeed adding noise to the readout is a trick , equivalent to ridge regularization , often used in numerical simulations to increase the robustness of reservoirs . best non - linearity . experience with digital simulations of reservoirs shows that sigmoidal non - linearities , such as @xmath104 , give very good performance . however in experimental realizations , other non - linearities may be much easier to implement . experience suggests that the best non - linearity depends to some extent on the task at hand . also sub - optimal non - linearities can presumably be compensated by increasing the size of the reservoir . 3 . continuous time and continuous space . in numerical simulations , it is by far easier to work with a discrete set of variables @xmath14 , @xmath124 evolving in discrete time @xmath98 . but in experimental systems , it is often much more natural to work with continuous variables and continuous time . a theory of reservoir computing operating with continuous variables evolving in continuous time remains to be written , although some theoretical progress has been made along these lines @xcite . finally , what is the future of experimental reservoir computing ? on the one hand the quest is fundamental : can we build analog systems that perform high - level information processing ? will we in this way gain new insights into how biological information processing systems operate ? on the other hand experimental reservoir computers could ultimately have practical applications . indeed they are based on architectures completely different from those used in present day digital electronics . presumably the place to look for applications is at the frontier where digital electronics has difficulty coping , such as information processing at ultra high speeds , or with ultra low energy consumption , or for tasks which are hard to code using standard programming methods . here we describe in more detail our experimental system ( schematized in fig . 1 of the main text ) , the algorithm used to simulate it numerically , and the tasks we study to validate its performance as a reservoir . we first consider the free running dynamics of our system , in the absence of input . the light source consists of a laser at the wavelength 1564 nm ( in the standard c band of optical telecommunication ) . it produces a time independent intensity @xmath48 . the light passes through an intensity modulator ( mach - zehnder modulator ) driven by a time dependent voltage @xmath125 . the light intensity just after the modulator is @xmath126 where @xmath127 is a constant voltage ( the voltage which is needed to go from a maximum to the next minimum of light at the output of the modulator ) and @xmath128 can be adjusted by applying a dc bias voltage to the modulator . taking @xmath129 allows us to rewrite @xmath130 the light passes trough a tunable optical attenuator , enabling the adjustment of the loop s gain through the variation of the value @xmath48 . it propagates through a long spool of fiber ( 1.7 km of single mode optical fiber in our experiment ) , and is then detected by a photodiode integrated with a transimpedance amplifier . the resulting voltage is amplified again ( by a rf amplifier ) to produce the voltage @xmath125 that drives the intensity modulator . in our experiment , the photodiode ( with its transimpedance amplifier ) and rf amplifier operate in a linear regime , hence @xmath125 is proportional to @xmath131 where @xmath15 is the round trip time of the oscillator . however , one must take into account that the photodiode acts as a lowpass filter and that the rf amplifiers act as highpass filters . both the photodiode and the amplifiers also add noise . thus , approximating the filters by first order filters , we have @xmath132 where @xmath133 is the imaginary unit , tilde denotes the fourier transform , @xmath134 denotes the amplification factor , @xmath135 are the cutoff frequencies of the resulting bandpass filter , and @xmath30(t ) is the white noise . in the time domain we have @xmath136 where @xmath137 denotes the linear bandpass filter . because of the high pass filter , we can use as variable the fluctuations around the mean intensity . we define @xmath138 ( i.e. , we rescale the intensity i(t ) to lie in the interval @xmath50 $ ] ) whereupon eq . ( [ eq:1 ] ) becomes @xmath139 where @xmath140 is experimentally adjustable by changing the input light intensity @xmath48 . experimentally it can be tuned over the range @xmath141 $ ] . typical values of the parameters are @xmath142s , @xmath143mhz , @xmath144khz , and the amplitude of @xmath145 is approximately @xmath146% . at the output of the intensity modulator , an optical splitter enables us to take approximatively 3% of the optical signal in order to measure the optical intensity in the fiber by the mean of a second amplified photodiode . the resulting voltage is digitized with a national instrument pxi card at the sample rate of 200 megasamples per second , or can be measured with a digital oscilloscope . if we increase @xmath48 gradually , the intensity @xmath147 undergoes a bifurcation diagram typical of nonlinear dynamical systems . figure [ fig : bifurcationdiagram ] shows the excellent agreement between the experimentally observed bifurcation diagram and the one obtained by numerical simulations of the evolution equations ( with the bias @xmath148 ) . in this bifurcation diagram , the number of bifurcations before reaching chaotic behavior is strongly affected by the amount of noise in the system . comparing this bifurcation diagram to simulations is the most precise way we have to estimate the amount of noise in our experimental setup . the estimated value is in agreement with measurements carried out on each component of the setup separately . we also verified that the thickness of the branches inside the bifurcation diagram is mainly due to the noise level of oscilloscope . we now consider the addition of an input term . the input is characterized by a new time scale @xmath23 . the scalar input @xmath149 ( evolving in discrete time @xmath150 and the input mask @xmath101 are transformed into a continuous input @xmath151 as follows : @xmath152,\quad i=0,\ldots , n-1\quad,\quad n\in\mathbb{z}\label{eq : source}\ ] ] where @xmath25 is the number of nodes in the reservoir . the input mask values @xmath101 , @xmath124 are randomly chosen from a given distribution ( which may depend on the task ) . the time scale @xmath27 over which the input @xmath151 changes is : @xmath153 a voltage proportional to @xmath151 is generated by a function generator ( at a sample rate of 200 msamples / s ) and added to the output voltage of the amplified photodiode using a rf combiner placed at the entrance of the rf amplifier . so the voltage @xmath125 that drives the intensity modulator is a combination of the light intensity and the input signal . the dynamical equations thus become @xmath154 where @xmath38 is experimentally adjustable by varying the output voltage amplitude of the function generator . in this equation , we take into account that the rf combiner is placed before the rf amplifier , and therefore that the source term is affected by the highpass filter @xmath155 of the rf amplifier . because of the difference between the time scales @xmath15 and @xmath156 , the effect of the filter @xmath155 is almost negligible . its main effect is to ensure that the effective source signal @xmath157 has mean value zero . in our experiments we take @xmath158 . the number @xmath25 of variables is taken in the range @xmath159 . we take @xmath160 ( see figure [ fig : inputoutputand discretization ] ) . thus for @xmath161 we have @xmath162 . the performance of the reservoir does not depend on the exact value of @xmath26 chosen , as long as @xmath163 is not a simple fraction ( such as 1 , 3/4 , or 1/2 ) , which would divide our reservoir into different independent subsystems . to obtain the discretized dynamics , we discretise the intensity along the fiber according to @xmath164 where we suppose that @xmath165 with @xmath89 integer ( see figure [ fig : inputoutputand discretization ] ) . we thus have that the physical time @xmath21 is related to @xmath166 through @xmath167 upon neglecting the effects of the filters @xmath168 and @xmath155 we obtain the discretized version of eq . ( [ eq : xsinxsource ] ) . note that the absence of synchronisation ( @xmath24 , or equivalently @xmath169 ) completely modifies the dynamics by coupling the discretised variables @xmath170 to each other . note also that the wrap arround effect ( the second line in eq . ( [ eq : discreteunsynchronizeddynamics ] ) ) does not appear in traditional reservoir computing . two different numerical models were developed to study the capabilities of the network : a discretized model , closer to the standard formulation of echo state networks , and a continuous model , which is as close as possible to our experimental apparatus . in the discretized version of the model we implement the discretization described by eq . ( [ eq : discreteunsynchronizeddynamics ] ) . no noise is considered , and the bandpass effects of the various components are neglected . the sine nonlinearity and the topology of the network are preserved . the optimal operating point of the system is found by tuning the parameters @xmath17 and @xmath38 in eq . ( [ eq : discreteunsynchronizeddynamics ] ) . this model is used to set a performance goal for our experimental system : if the performance of the experiment is close to the one of the model , then our system is robust enough with respect to the noise and the effects introduced by each of its components . moreover , the performance of the discretized model is the same , within the experimental error , to the one of traditional networks as reported in @xcite , allowing us to validate the chosen nonlinearity and topology as good choices for a reservoir computer . the continuous matlab model we developed is instead as close as possible to the experiment . all the signals in the simulation are discretized at 200 msamples / s . this corresponds to the sample rate of the arbitrary waveform generator and the digitizer . all the components are represented by their transfer function at their respective operating point ( sine function for the mach - zehnder modulator , responsivity and transimpedance gains of the photodiodes , gain of the amplifier ) . the collective frequency response of all the components is represented by a bandpass filter with first order slopes . this is a reasonable approximation to the exact slope of the bandpass filters which was measured using a vector network analyzer in open loop configuration . noise is added to the signal at each noisy element of the system ( dark current of the photodiode , noise added by each amplifier ... ) . the dynamics of the model correspond very closely to the experiment . this is illustrated for instance by the clear agreement of the simulated and measured bifurcation diagram ( see fig . [ fig : bifurcationdiagram ] ) and by the clear agreement of the simulated and measured performances on the channel equalization task , see fig . [ fig : channelequalization ] in the main text . the continuous model allows us to easily explore the sensitivity to parameters , such as noise level or the shape of the nonlinearity , which ca nt always be reached with the experiment . the light intensity @xmath46 in the fiber loop is converted to a voltage by a photodiode and recorded by the digitizer operating at 200ms / s . from the intensity @xmath46 recorded during a time @xmath26 we extract @xmath25 discrete variable values @xmath14 , @xmath124 . this is carried out as follows . the intensity @xmath46 is divided into @xmath25 pieces of duration @xmath27 . we neglect the first quarter and the last quarter of the data points recorded over the duration @xmath27 and associate to @xmath14 the average of the remaining data points . this procedure in which the beginning and end of each interval @xmath27 is omitted allows us to not be affected by the transients as the intensity goes from one value to the other , and also allows an efficient synchronization of our system . the estimator @xmath4 is then obtained by taking a linear combination @xmath171 where the weights @xmath11 are optimized . this post processing is carried out offline , on a computer . [ fig : inputoutputand discretization ] shows an example of the input sent to the reservoir , the reservoir response and the discretization operated on the reservoir output . in our work we considered several tasks . we review them in detail . this is a simple task that we use for a first evaluation of the performance of the reservoir . the input @xmath0 to the system consists of random sequences of sine and square waves discretized into @xmath172 points per period . the mask values @xmath101 are drawn from the uniform distribution over the interval @xmath173 $ ] . the reservoir size is taken to be @xmath53 . the output @xmath4 should be @xmath55 when the signal is the square wave , and @xmath54 when it is the sine . the weights @xmath11 are obtained by minimizing the mse between @xmath4 and the ideal output . experimentally we obtain @xmath96 , which corresponds to essentially perfect operation for this task . these experimental results are in close agreement with those obtained using numerical simulations of our reservoir . where @xmath78 represents i.i.d . gaussian noise with zero mean , adjusted in power to yield signal - to - noise ratios ranging from 12 to 32 db . this task was introduced in @xcite and it was shown in @xcite that reservoir computing could significantly improve performance on this task . in our study , the input mask is taken to be uniformly distributed over the interval @xmath50 $ ] . the reservoir size is taken to be @xmath53 . to perform the task , we first obtain an estimator @xmath4 of the input by minimizing the mse between @xmath4 and @xmath60 . we then obtain an estimator @xmath177 by replacing @xmath4 by the discretized value @xmath178 to which it is closest . finally we estimate the symbol error rate ( ser ) , i.e. , the fraction of time that @xmath177 differs from @xmath60 . the performance of the network is calculated as the average performance over 10 different input sequences , in which the first 3000 samples form the training set and the following 6000 samples form the test set . the ser on the test set is then studied as a function of signal to noise ratio at the input . the values reported in figure [ fig : channelequalization ] in the main text are the average ser for 10 different trials , while the error bars represent the standard deviation of the ser for the same trials . it should be noted that 6000 test steps are the maximum number of steps that the arbitrary waveform generator in our setup allows . hence , when sers approach @xmath63 , our average sers include trials where two errors , one error , or no error at all have been made by the reservoir . this means that the error bars on the experimental data where sers are close to @xmath63 might be overestimated . in contrast , data from simulations do not suffer from this effect , as we can arbitrarily increase the number of test samples for a more precise measurement . for comparison , at a snr ratio of 28db , the three models studied in @xcite gave ser of @xmath179,@xmath62,@xmath180 , while the reservoir studied in @xcite gave sers of @xmath181 to @xmath182 . at the same snr of 28db , our experimental system gives a ser of @xmath61 . in this task the aim is to reproduce the behavior of a nonlinear , tenth - order system with random input drawn from a uniform distribution over the interval @xmath73 $ ] . the equation defining the target system is given by for this task , the mask is uniformly distributed over @xmath173 $ ] and the reservoir size is @xmath53 . we first train our system using @xmath95 time steps , then we test the system on a new sequence of @xmath95 inputs . the performance of the reservoir is measured by the normalized mean square error of the estimator @xmath4 , averaged over 10 different pairs of train and test sequences . for this narma10 task , the best performances ( @xmath184 for the discretized simulation , @xmath185 for the continuous simulation and @xmath186 for the experiment ) are obtained in a very linear regime . for the isolated spoken digit recognition task , the data is taken from the nist ti-46 corpus @xcite . it consists of ten spoken digits ( 0 ... 9 ) , each one recorded ten times by five different female speakers . these 500 spoken words are sampled at 12.5 khz . this spoken digit recording is preprocessed using the lyon cochlear ear model @xcite . the input to the reservoir @xmath79 consists of an 86-dimensional state vector ( @xmath80 ) with up to 130 time steps . the number of variables is taken to be @xmath81 . the input mask is taken to be a @xmath82 dimensional matrix @xmath83 with elements taken from the the set @xmath84 with equal probabilities . the product @xmath85 of the mask with the input is used to drive the reservoir . ten linear classifiers @xmath86 ( @xmath87 ) are trained , each one associated to one digit . the target function for @xmath88 is + 1 if the spoken digit is @xmath89 , and -1 otherwise . the classifiers are averaged in time , and a winner - takes - all approach is applied to select the actual digit . in our study , the 500 spoken words are divided in five subsets . we trained the reservoir on four of the subsets , and then tested it on the fifth one . this is repeated five times in order to use each subset as test part . in this way we can test our system over all the speakers and digits , and compute an average performance . the performance is given in terms of the word error rate , that is the fraction of digits that are misclassified . we obtain a wer of @xmath69 which correspond to 2 errors in 500 recognized digits . for comparison , in @xcite , where reservoir computing was first used on this spoken digit benchmark , a wer of 4.3% was reported for a reservoir of size 1232 . in @xcite a wer of 0.2% was obtained for a reservoir of size 308 and using the winner - takes - all approach . in @xcite a wer of 1.3% for reservoirs of size 200 is reported . the experimental reservoir reported in @xcite gives a wer of 0.2% using a reservoir of size 400 . the sphinx-4 system @xcite uses a completely different method based on hidden markov models and achieves a wer of 0.55% on the same data set . simulated ( left panel ) and measured ( right panel ) bifurcation diagram . the vertical axis ( output states ) is proportional to the optical intensity in the optical fiber for the simulations , and to the voltage at the output of the readout photodiode for the measurements ( this voltage is equal to de optical intensity multiplied by the gain of the amplified photodiode ) . the gray scale represents the histogram of optical intensities inside the system as a function of the feedback gain . when the feedback gain is lower than unity , only one value of the light intensity is possible . for feedback gain slightly larger than unity , the light intensity oscillates between two values . for even larger feedback gain ( around 2 ) , the nonbijective nature of the mach - zehnder modulator s transfer function leads to oscillation between multiple light intensity levels or even to a chaotic behavior . the number of bifurcations before reaching the chaotic behavior is determined by the amount of noise inside the system . the thickness of the branches in the measured bifurcation diagram is due to the noise added by the oscilloscope . ] operation of the experimental reservoir . the blue line represents the masked input driving the reservoir ; the red line represents the reservoir response to the depicted input as measured by the oscilloscope . both have been normalized so that their values lie between -1 and 1 for the whole length of the experiment . the white squares along the red line represent the discretized values of the reservoir states , obtained by averaging the reservoir output on a time interval of amplitude @xmath187 . the upper part of the figure shows the reservoir operation for the first 8 inputs of the reservoir ; note that the input is zero up to @xmath188 . the bottom left panel shows a detail from the measurement on the response to the very first input value , when no feedback is present yet , and we only see the instantaneous response of the mach - zehnder . the panel on the bottom right depicts the response of the system to the 5th input , where the effect of the feedback is clearly visible : the amplitudes of the reservoir states are no longer just related to the instantaneous input , but are influenced by previous inputs as well . ] 10 h. john caulfield and shlomi dolev . . , 4(5):261263 , 2010 . h. jaeger . . technical report , technical report gmd report 148 , german national research center for information technology , 2001 . h. jaeger . . technical report , technical report gmd report 152 , german national research center for information technology , 2001 . herbert jaeger and harald haas . , 304(5667):7880 , 2004 . robert legenstein and wolfgang maass . , chapter what makes a dynamical system computationally powerful ? , pages 127154 . mit press , 2005 . wolfgang maass , t. natschlager , and henry markram . . , 14(11):25312560 , 2002 . steil . . , pages 843848 . david verstraeten , benjamin schrauwen , m dhaene , and dirk stroobandt . , 20(3):391403 , 2007 . mantas lukoeviius and herbert jaeger . . , 3(3):127149 , 2009 . barbara hammer and benjamin schrauwen . . in _ proceedings of the european symposium on artificial neural networks _ , pages 213216 , 2009 . . david verstraeten , benjamin schrauwen , and dirk stroobandt . . in _ the 2006 ieee international joint conference on neural network proceedings _ , pages 10501053 . ieee , 2006 . fabian triefenbach , a. jalalvand , benjamin schrauwen , and j. martens . . , 23:19 , 2010 . herbert jaeger , mantas lukosevicius , dan popovici , and udo siewert . , 20(3):33552 , 2007 . chrisantha fernando and sampsa sojakka . pattern recognition in a bucket . in wolfgang banzhaf , jens ziegler , thomas christaller , peter dittrich , and jan kim , editors , _ advances in artificial life _ , volume 2801 of _ lecture notes in computer science _ , pages 588597 . springer berlin / heidelberg , 2003 . felix schrmann , karlheinz meier , and johannes schemmel . edge of chaos computation in mixed - mode vlsi - a hard liquid . in _ advances in neural information processing systems_. mit press , 2005 . kristof vandoorne , wouter dierckx , benjamin schrauwen , david verstraeten , roel baets , peter bienstman , and jan van campenhout . , 16(15):1118292 , 2008 . l. appeltant , m.c . soriano , g. van der sande , j. danckaert , serge massar , j. dambre , benjamin schrauwen , c. r. mirasso , and i. fischer . . , ali rodan and peter tino . , 22(1):13144 , 2011 . thomas erneux . . springer , 2009 . laurent larger , pierre - ambroise lacourt , stphane poinsot , and marc hanna . . , 95(4):14 , 2005 . yanne k. chembo , pere colet , laurent larger , and nicolas gastaud . . , 95(20):25 , 2005 . michael peil , maxime jacquot , yanne k. chembo , laurent larger , and thomas erneux . . , 79(2):115 , 2009 . ali rodan and peter tino . . , pages 267274 , 2010 . v. john mathews and junghsi lee . . , 2296(1):317327 , 1994 . david verstraeten , benjamin schrauwen , and dirk stroobandt . . in _ proceedings of the 13th european symposium on artificial neural networks ( esann ) _ , pages 435440 , 2005 . willie walker , paul lamere , philip kwok , bhiksha raj , rita singh , evandro gouvea , peter wolf , and joe woelfel . sphinx-4 : a flexible open source framework for speech recognition . technical report , mountain view , ca , usa , 2004 . amir f. atiya and alexander g. parlos . , 11(3):697709 , 2000 . h. jaeger . . in _ advances in neural information processing systems _ , volume 8 , pages 593600 . mit press , 2002 . , 1991 . r. lyon . . in _ icassp 82 . ieee international conference on acoustics , speech , and signal processing _ , pages 12821285 . institute of electrical and electronics engineers , 1982 . rajagopal ananthanarayanan , steven k. esser , horst d. simon , and dharmendra s. modha . . acm , new york , ny , usa , 2009 . david verstraeten , joni dambre , benjamin schrauwen , and serge massar . linear and nonlinear memory capacity of dynamical systems . . david verstraeten , joni dambre , xavier dutoit , and benjamin schrauwen . . in _ the 2010 international joint conference on neural networks ( ijcnn ) _ , pages 18 . ieee , july 2010 . benjamin schrauwen , m wardermann , david verstraeten , j.j . steil , and dirk stroobandt . . , 71(7 - 9):11591171 , 2008 . danko nikolic , stefan husler , wolf singer , and wolfgang maass . temporal dynamics of information content carried by neurons in the primary visual cortex . , 19:10411048 , 2007 . michiel hermans and benjamin schrauwen . , 23(3):34155 , 2010 .

astrometric surveys such as gaia and lsst will measure parallaxes for hundreds of millions of stars . yet they will not measure a single distance . rather , a distance must be estimated from a parallax . in this didactic article , i show that doing this is not trivial once the fractional parallax error is larger than about 20% , which will be the case for about 80% of stars in the gaia catalogue . estimating distances is an inference problem in which the use of prior assumptions is unavoidable . i investigate the properties and performance of various priors and examine their implications . a supposed uninformative uniform prior in distance is shown to give very poor distance estimates ( large bias and variance ) . any prior with a sharp cut - off at some distance has similar problems . the choice of prior depends on the information one has available and is willing to use concerning , for example , the survey and the galaxy . i demonstrate that a simple prior which decreases asymptotically to zero at infinite distance has good performance , accommodates non - positive parallaxes , and does not require a bias correction .

1507.02105

the parallax of an object is its observed angular displacement with respect to a reference frame due to the movement of the observer over a baseline . stellar parallaxes are measured using twice the earth sun separation as a baseline . from geometry and the definition of the parallax , @xmath0 , the distance of a star from the sun , @xmath1pc , is @xmath2 to a very good approximation . parallaxes are one of the few distance measures in astronomy which do not make assumptions about the intrinsic properties of the object . hipparcos blazed the trail when it measured the parallaxes of over @xmath3 stars to an accuracy of around 1mas down to @xmath4 in the 1990s ( perryman et al . gaia is currently extending this to @xmath5 objects with expected accuracies as good as 0.01mas ( and still much less than a mas at its magnitude limit of @xmath6 ) ( lindegren et al . 2008 ) , and lsst hopes to obtain parallaxes with standard deviations of order 1mas for billions of stars down to much fainter magnitudes ( ivezi et al . 2011 ) . it is therefore of considerable importance to know how to estimate distances ( and their uncertainties ) from parallaxes . despite the simple relation between the two , inverting a parallax to give a distance is only appropriate when we have no measurement errors . as we always have measurement errors , determining the distance given a parallax becomes an inference problem . here i show that this inference is not trivial when the errors are even still quite small , necessarily involves assumptions , and can lead to surprisingly large errors . i then set out to solve the following problem : given a measured parallax , @xmath0 , and its uncertainty , @xmath7 , how can we obtain a good estimate of the distance and its uncertainty ? numerous studies in the astronomical literature have examined estimating physical quantities from parallaxes and possibly additional data ( such as colours ) . these quantities may be individual distances , absolute magnitudes , distances to clusters , etc . example studies include lutz & kelker 1973 , smith & eichhorn 1996 , brown et al . 1997 , verbiest et al . 2012 , palmer et al . 2014 . my objective is not to repeat such analyses , but rather to illustrate the issues involved using what is arguably the simplest of these problems : estimating distance from a single parallax . more complex problems will need to consider these same issues , even though in many cases we will not want to estimate distances explicitly . we should instead infer the quantity of interest directly . if we want to compare model predictions with data ( which is not the topic of this paper ) , then it is usually better to project the predictions ( and their distributions ) into the data space , where the noise ( measurement ) model is better defined , rather than inferring quantities and then trying to determine an appropriate noise model to use in the comparison . i use @xmath8 to denote parallax ( always in ) and @xmath1 to denote distance ( always in pc ) . where it is necessary to make a distinction , i use @xmath9 to indicate the true distance , and @xmath10 to indicate an estimate of this , such as the mode @xmath11 , or median , @xmath12 . i will frequently refer to the _ fractional parallax error _ , the ratio of the ( 1@xmath13 ) parallax uncertainty to the parallax . this can either be the empirical quantity based only on the measured data , @xmath14 , or a quantity based on the true distance , @xmath15 . in a real application @xmath9 and @xmath16 are of course unknown . for a star at true distance @xmath1 , its true but unknown parallax is @xmath17 . the measured parallax , @xmath0 , is a noisy measurement of @xmath17 . i will assume that @xmath0 is normally distributed with unknown mean @xmath17 and known standard deviation @xmath7 . that is , i assume @xmath0 has been drawn from the distribution @xmath18 } \hspace*{1em}{\rm where}\hspace*{1em } \parsd \geq 0 \label{eqn : likelihood}\ ] ] which is gaussian in @xmath0 , but of course not in @xmath1 . this function is the measurement model , or likelihood . it gives the probability density function ( pdf ) probability per unit parallax for any @xmath0 , given values of @xmath1 and @xmath7 . this is the measurement model used in the hipparcos and gaia data processing . @xmath7 depends in particular on the centroiding accuracy of each of the individual position measurements which are used in the astrometric solution . this in turn depends primarily on the number of photons ( @xmath19 ) received from the star ( @xmath20 ) , and thus on its brightness , the observing time per observation , and the number of observations ( assuming the observations have an appropriate cadence to determine a parallax at all ) . depending on the source brightness , other terms and systematic errors may also be significant . the two most important points here are ( 1 ) @xmath7 is independent of @xmath0 , and ( 2 ) @xmath7 can be estimated from another measurement model using other measured data ( de bruijne 2012 ) .. but this will generally be a small uncertainty compared to the noise in the parallax , so i do nt consider it here . ] equation [ eqn : likelihood ] has a finite probability of giving non - positive parallaxes , and this probability gets larger with increasing @xmath16 . an astrometric reduction can indeed produce a negative parallax , because the `` measured '' parallax is the result of reducing a set of noisy angular measurements . it corresponds to the parallactic displacement being in the opposite direction on the sky from that expected due to the movement of the observer along the baseline . it does _ not _ correspond to a negative distance , because @xmath21 by definition . non - positive parallaxes are perfectly reasonable measurements , and we will see that they can deliver useful distance information . strictly speaking the likelihood must be changed at extremely small distances , because then the definition that the ( noise - free ) parallax is the reciprocal of distance breaks down . but this only occurs for parallaxes on degree scales , whereas all measured stellar parallaxes are less than an arcsecond . suppose we have a measurement @xmath22 . one may be tempted to report @xmath2 as the distance estimate , and use a first order taylor expansion to give @xmath23 as the uncertainty . but we will now see that this quickly becomes a poor approximation . from the definition of the gaussian , the intervals @xmath24 \hspace{1em}{\rm and}\hspace{1em } 1/r \,=\ , [ \parallax , \parallax+2\parsd]\ ] ] each includes a fraction @xmath25 of the total probability of the distribution @xmath26 . the transformation from @xmath17 to @xmath1 is monotonic and so preserves the probability . thus the intervals @xmath27 \hspace{1em}{\rm and}\hspace{1em } r \,=\ , [ 1/(\parallax+2\parsd ) , 1/\parallax]\ ] ] must each also contain a fraction 0.477 of the total probability over the distance . but whereas these intervals are equally sized in @xmath17 ( the gaussian is symmetric ) , they are not in @xmath1 . for example , with @xmath28 and @xmath29 , these intervals are @xmath30 $ ] and @xmath31 $ ] respectively . the uncertainties do not transform symmetrically because of the nonlinear transformation from @xmath17 to @xmath1 . we see that the first order taylor expansion suggested above is a poor approximation even for relatively small errors ( here @xmath32=@xmath33 ) . but what if the errors are larger , with @xmath34 ? the upper distance interval in the example becomes @xmath35 $ ] . if @xmath32 is even larger , then this interval becomes undefined and we seem to `` lose '' some of the probability . as the gaussian distribution has infinite support for all values of @xmath0 and @xmath7 , some finite amount of probability in the likelihood function will always correspond to an undefined distance . the problem here is that we are trying to make a probabilistic statment about @xmath1 using just equation [ eqn : likelihood ] , yet this is a probability distribution over @xmath0 , not @xmath1 . the solution is to pose the problem correctly . given @xmath0 and @xmath7 , we want to infer @xmath1 . as there is noise involved , we know we can not infer @xmath1 exactly , so we want a probability distribution over the possible values of @xmath1 . that is , we want to find @xmath36 . from bayes theorem ( which follows from the axioms of probability ) this is related to the likelihood by @xmath37 where @xmath38 is the normalization constant @xmath39 which is not a function of @xmath1 . @xmath40 is the prior , and expresses our knowledge of or assumptions about the distance , independent of the parallax we have measured . @xmath36 is the posterior distribution . by multiplying the likelihood by a prior , we essentially transform an expression ( the likelihood ) for the probability of the known data ( parallax ) given the unknown parameter ( distance ) to an expression ( the posterior ) for the probability of the parameter given the data . we see that we can only infer the distance in a probabilistic sense if we adopt a prior . some people object to this on philosophical grounds ( _ how can science depend on assumptions ? _ ) , others on practical grounds ( _ how can i know the prior if i havent yet measured any distances ? _ ) . the latter is a valid objection and will be discussed later . yet without a prior , we run into the problems we just saw in the previous section . a common strategy for dealing with prior discomfort is to adopt a uniform prior over the full range of the parameter , on the grounds that this does not prefer one value over another . in the present context we should go one step further and use @xmath41 so as to introduce the definition that distances must be positive . the @xmath42 symbol is used to indicate that the pdf is unnormalized . the subscript is an abbreviation of the distribution . because it extends to infinity , this prior can not be normalized . such priors are referred to as _ improper_. from equation [ eqn : bayes ] , the ( unnormalized ) posterior , @xmath43 , in this case is the likelihood ( equation [ eqn : likelihood ] ) but now considered as a function of @xmath1 rather than @xmath0 , and subject to the additional constraint that @xmath1 must be positive , i.e.@xmath44 examples of this posterior are shown in figure [ fig : ud.post_runifprior ] for @xmath45 and various values of @xmath32 . this demonstrates the skewness discussed in section [ sec : intuition ] . ( improper uniform prior ) for @xmath45 and four values of @xmath46 . the posteriors have been scaled to all have their mode at @xmath47 . as the prior is improper , the posterior remains finite out to infinite distance for all values of @xmath48.[fig : ud.post_runifprior],scaledwidth=50.0% ] inspection of equation [ eqn : likelihood ] shows that @xmath49 the posterior does not converge , has an infinite area and so can not be normalized . consequently it has no mean , no standard deviation , no median , and no quantiles . the only plausible estimator of the distance is the mode . ] , which we see from figure [ fig : ud.post_runifprior ] is well defined for all values of @xmath32 , and is equal to @xmath2 for @xmath50 . for non - positive parallaxes the posterior increases monotonically from 0 at @xmath51 to a finite asymptote ( equation [ eqn : likelihood ] ) putting the mode at @xmath52 , which is physically implausible in a finite universe . this is bad , as negative parallaxes are valid measurements . we should also be concerned that this estimator ignores the parallax uncertainty , even though we have seen that its magnitude determines the skewness of the distance distribution ( section [ sec : intuition ] ) . an uncertainty could be derived from the full width at half maximum ( fwhm ) ( or similar ) , but without a normalizable posterior there is no useful probabilistic interpretation of this . the above mentioned problems notwithstanding , we can still ask how good the mode of this posterior is as an estimator . recall that the posterior is a function of the _ measured _ parallax , which due to noise is not equal to the true parallax , so @xmath2 is not equal to the true distance . thus to assess the quality of any estimator we need to test it using noisy simulated data . in particular , we would like to see how its accuracy varies as a function of the expected fractional parallax error . i do this with the following empirical test procedure : assign a fractional parallax error , @xmath16 assign a true distance , @xmath9 ( which together with @xmath16 determines @xmath7 ) [ listitem : assign_distance ] simulate a parallax measurement , @xmath0 , by drawing a value at random from the likelihood with @xmath53[listitem : draw_parallax ] use @xmath0 in the posterior to evaluate the distance estimator , @xmath10 calculate the _ scaled residual _ , @xmath54 repeat steps 25 numerous times to compile a set of values @xmath55 at a single value of @xmath16 calculate the bias , @xmath56 , and sample standard deviation , @xmath57 repeat steps 17 for different values of @xmath16 . the bias and standard deviation are standard tests of the quality of an estimator . a good estimator will have small values of both , where `` small '' is , of course , a relative term . by using the scaled residuals ( rather than just the residuals ) i can collate information on many different true distances into a single value of @xmath58 and @xmath59 for a given @xmath16 . in the first test i assign true distances in step [ listitem : assign_distance ] by drawing at random from the distribution @xmath60 which corresponds to stars being uniformly distributed in three - dimensional space , @xmath61=const . , of finding a star in a shell with inner and outer radii @xmath62 is proportional to the volume of that shell , @xmath63 . @xmath64 , so @xmath65 . ] i use @xmath66 . for each value of @xmath16 i sample @xmath67 true distances . i do this for 100 values of @xmath16 ranging from 0.01 to 1.0 in steps of 0.01 . the estimator in this case is the mode , which is just @xmath2 . if the parallax is non - positive ( in step [ listitem : draw_parallax ] ) , then for the improper uniform prior i throw away the sample , because an infinite distance would give an infinite bias and standard deviation ( telling us straight away that it is a bad estimator ! ) . the procedure for other priors i will discuss as we encounter them . for the mode estimator of the posterior @xmath43 ( equation [ eqn : urunifpost ] , which uses the improper uniform prior ) for data drawn from the constant volume density prior ( equation [ eqn : r2prior ] ) . the right panel is a zoom of the left panel . the red line in the right panel shows the fraction of samples which had to be rejected because they had non - positive parallaxes . [ fig : scaledresiduals_mode_runifprior_r2trueprior_rmax1e3],title="fig:",scaledwidth=49.0% ] for the mode estimator of the posterior @xmath43 ( equation [ eqn : urunifpost ] , which uses the improper uniform prior ) for data drawn from the constant volume density prior ( equation [ eqn : r2prior ] ) . the right panel is a zoom of the left panel . the red line in the right panel shows the fraction of samples which had to be rejected because they had non - positive parallaxes . [ fig : scaledresiduals_mode_runifprior_r2trueprior_rmax1e3],title="fig:",scaledwidth=49.0% ] the results are shown in figure [ fig : scaledresiduals_mode_runifprior_r2trueprior_rmax1e3 ] . we see that the standard deviation increases steadily until @xmath68 , then it explodes . the bias also increases sharply , in this case reaching a value of about 4 at @xmath16=1 . the reason for this appearance is that as @xmath16 increases , the probability increases that the measured parallax which is drawn from the likelihood becomes arbitrarily close to zero , in which case @xmath69 becomes arbitrarily large . this also explains why both the bias and standard deviation become very noisy for @xmath70 : the exact appearance of the curves depends on the sample drawn in the simulation .. below this value we get a very smooth dependence of bias and standard deviation on @xmath16 . at values of @xmath16 well below this `` explosion '' we could even use the taylor approximation mentioned in section [ sec : intuition ] to estimate the distance and a ( symmetric ) uncertainty . ] although noisy , the plot is independent of @xmath71 , because we are calculating functions of the scaled residuals , @xmath72 , for fixed @xmath16 . doubling @xmath71 would double both @xmath9 and @xmath10 , leaving @xmath72 unchanged . one might argue that by drawing true distances from a distribution which is very different from the uniform prior adopted in the posterior , we are bound to get poor results . to some extent this is true . to test this , i repeat the above procedure , but drawing from the prior in equation [ eqn : urunifprior ] . of course , to be able to draw from this i have to set an upper limit ( otherwise all draws will be infinite ) , and i again use @xmath66 ( although the exact value is irrelevant , because the scaled residuals are independent of it ) . the results are shown in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3 ] . we see that they are hardly any better than before . the problem is therefore not prior mismatch , but the fact that @xmath2 ( the mode ) is a very noisy estimator once @xmath70 . because this posterior is not normalizable , no other estimator is available . , but now drawing data from the truncated uniform prior.[fig : scaledresiduals_mode_runifprior_rmax1e3],title="fig:",scaledwidth=49.0% ] , but now drawing data from the truncated uniform prior.[fig : scaledresiduals_mode_runifprior_rmax1e3],title="fig:",scaledwidth=49.0% ] this example shows that an apparently innocuous approach to inference is in fact highly problematic . it is the kind of thing which may be done by those who reject the bayesian approach to inference on the grounds that it depends on prior assumptions . they want to rely only on the likelihood , yet have to adopt some kind of prior to get logically consistent answers , and tacitly choose an unbounded uniform prior in @xmath1 in the belief that it is `` uninformative '' . yet one can equally argue that uniform in @xmath73 ( i.e. in distance modulus ) is uninformative . in fact , by adopting a uniform prior in @xmath1 , one assumes that the volume density of stars varies as @xmath74 ( see the discussion around equation [ eqn : r2prior ] ) , which is highly informative . corresponds to uniform in @xmath17 and so @xmath75 . ] clearly , this approach is not very robust : it will only give distance estimates for @xmath76 , it can not give self - consistent uncertainty estimates ( only a poor approximation using a taylor expansion ) , and it breaks down entirely with non - positive parallaxes . can we do better ? ( truncated uniform prior with @xmath66 ) for @xmath45 and four values of @xmath77 ( black lines ) . the red line shows the posterior for @xmath78 and @xmath79 . the posteriors have been scaled to all have their mode at at @xmath80 . right : the same five posterior pfds but now normalized . [ fig : ud.post_runiftruncprior],title="fig:",scaledwidth=49.0% ] ( truncated uniform prior with @xmath66 ) for @xmath45 and four values of @xmath77 ( black lines ) . the red line shows the posterior for @xmath78 and @xmath79 . the posteriors have been scaled to all have their mode at at @xmath80 . right : the same five posterior pfds but now normalized . [ fig : ud.post_runiftruncprior],title="fig:",scaledwidth=49.0% ] an obvious improvement is to truncate the prior at some value . this prevents the inferred distance becoming very large , so should reduce the explosion of the bias and standard deviation we saw with the improper prior for larger @xmath16 . the normalized prior is @xmath81 where @xmath71 is the largest distance we expect for any star in our survey . the posterior is the same shape as in figure [ fig : ud.post_runifprior ] but set to zero for @xmath82 @xmath83 this is shown in figure [ fig : ud.post_runiftruncprior ] . note the effect of the normalization . the right panel illustrates how the posterior is a combination of the likelihood and prior . when the data are good ( @xmath16 is small ) , the likelihood is narrow compared to the prior , so the former dominates the posterior . as the data become less informative ( larger @xmath16 ) , the posterior gets broader and the prior plays more of a role . for non - positive parallaxes the posterior increases monotonically from 0 at @xmath51 to a maximum at @xmath84 ( the red line shows an example ) . the mode of the resulting posterior serves as our distance estimator , and is @xmath85 i call a mode solution `` extreme '' if the estimate from the likelihood alone would place it beyond @xmath71 , and so is truncated by the prior to @xmath71 . that non - postitive parallaxes are set to @xmath71 is consistent with the nature of the parallax measurement ( see section [ sec : likelihood ] ) . i redo the empirical test with data drawn from the uniform prior ( with a common value of @xmath71 ) . to allow a fairer comparison with the results obtained with the improper uniform prior , i again reject simulated parallax measurements in step [ listitem : draw_parallax ] which are non - positive . the results are shown in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_truncatemode ] . they are again independent of @xmath71 . the bias and standard deviation are significantly reduced . this is not surprising , because we are now clipping all those large distance estimates to @xmath71 . the dashed line shows that relatively few objects have their distances clipped in this way only 10% at @xmath16=0.25 and 20% at @xmath16=1 which in turn shows that the explosion in the bias and standard deviation seen previously was due to a minority of objects . an alternative to clipping these distances would be to reject them , which means we decide we can not estimate distances . the results of the empirical test in that case ( for the same data ) are shown in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode ] . as we would expect , rejecting these cases reduces the bias and standard deviation , but not by much . for the mode estimator ( equation [ eqn : postut_mode ] ) of the posterior @xmath86 ( equation [ eqn : runifpost ] , which uses the truncated uniform prior ) for data drawn from a truncated uniform prior in @xmath1 . the right panel is a zoom of the left panel . the solid red line shows the fraction of samples which were rejected because they had non - positive parallaxes . the dashed red line shows the fraction of samples with an extreme mode ( @xmath87 ) which was truncated to @xmath71 . [ fig : scaledresiduals_mode_runifprior_rmax1e3_truncatemode],title="fig:",scaledwidth=49.0% ] for the mode estimator ( equation [ eqn : postut_mode ] ) of the posterior @xmath86 ( equation [ eqn : runifpost ] , which uses the truncated uniform prior ) for data drawn from a truncated uniform prior in @xmath1 . the right panel is a zoom of the left panel . the solid red line shows the fraction of samples which were rejected because they had non - positive parallaxes . the dashed red line shows the fraction of samples with an extreme mode ( @xmath87 ) which was truncated to @xmath71 . [ fig : scaledresiduals_mode_runifprior_rmax1e3_truncatemode],title="fig:",scaledwidth=49.0% ] , but now rejecting all samples which have @xmath87 . [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode],title="fig:",scaledwidth=49.0% ] , but now rejecting all samples which have @xmath87 . [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode],title="fig:",scaledwidth=49.0% ] compared to the improper uniform prior at least , the results look quite good . in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode ] , the bias and standard deviation increase approximately linearly for @xmath88 with gradients of 0.10 and 1.1 respectively ; their values are 0.025 and 0.29 respectively at @xmath89 . one might think about correcting the bias ( see section [ sec : biascorrection ] ) , but the need for this will be later obviated by the use of a better prior . now that the posterior is normalized we could investigate using the mean or median . but these are strongly influenced by the choice of @xmath71 for large values of @xmath16 , which is just the domain where we want a more robust estimator than the mode . the results in figures [ fig : scaledresiduals_mode_runifprior_rmax1e3_truncatemode ] and [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode ] are for the situation when the real data come from the same distribution as the prior . the shape of this prior aside , individual distance estimates for small parallaxes and/or large values of @xmath16 will be sensitive to the choice of @xmath71 . astrophysically this corresponds to knowing the maximum distance of any star . this could be set for each star individually according to its measured apparent magnitude , @xmath90 , and the fact that stellar evolution limits the absolute magnitude , @xmath91 , to about @xmath92mag . the flux continuity relation then tells us that @xmath93 , where @xmath94 is the interstellar absorption in the observed band . setting @xmath95 gives us the maximum distance ( although this will be an overestimate for observations at low galactic latitudes ) . a more stringent upper limit on @xmath91 could be set if the star s colour were known . truncating the prior has helped , but it still corresponds to assuming that the volume density of stars varies as @xmath74 . this is not only a strong prior , it is also scale independent , which the galactic stellar distribution most certainly is not . one could adopt a more complex prior based on a galaxy model . we may want to do this in practice ( see section [ sec : choice ] ) , but this gives us little insight into the role of the prior . let us instead take at least an astrophysically reasonable yet less informative prior , namely a constant volume density of stars ( equation [ eqn : r2prior ] ) . this too is truncated at @xmath84 in order to be normalizable . the unnormalized posterior is @xmath96 } & \:{\rm if}~~ 0 < r \leq \rlim \\ 0 & \:{\rm otherwise } \ . \end{dcases } \label{eqn : upostr2}\ ] ] examples of this posterior are shown in figure [ fig : post_r2prior ] for @xmath45 , @xmath66 , and various values of @xmath32 . for low @xmath32 the posterior looks unimodal , but then appears to develop a minimum ( and thus a second mode at @xmath84 ) , until for @xmath97 the lower mode ( and hence the minimum ) disappears , leaving a posterior which increases monotonically from zero up to @xmath71 . ( truncated constant volume density prior with @xmath66 ) for @xmath45 and eight values of @xmath98 . the posteriors have been scaled to all have their mode at at @xmath99 . right : the same posterior pdfs but now normalized . the curves with the clear maxima around @xmath100 are @xmath101 in decreasing order of the height of the maximum.[fig : post_r2prior],title="fig:",scaledwidth=49.0% ] ( truncated constant volume density prior with @xmath66 ) for @xmath45 and eight values of @xmath98 . the posteriors have been scaled to all have their mode at at @xmath99 . right : the same posterior pdfs but now normalized . the curves with the clear maxima around @xmath100 are @xmath101 in decreasing order of the height of the maximum.[fig : post_r2prior],title="fig:",scaledwidth=49.0% ] we can find the turning points of the posterior by solving @xmath102 for @xmath1 . this gives a quadratic equation with solutions @xmath103 the solution with the positive sign is the minmum ( @xmath104 ) and that with the negative sign is the mode ( @xmath11 ) . they are plotted as a function of @xmath32 in figure [ fig : roots_deriv_d.post_r2prior ] . there is an additional mode at @xmath84 due to the discontinuous cut - off in the prior . we see that for @xmath32@xmath105@xmath106 ( = 0.354 ) neither turning point is defined : the posterior transitions from having two modes and a minimum to the single mode at @xmath84 . thus for larger values of @xmath32 we would need to resort to another estimator ; @xmath71 itself would be consistent . ) of the posterior @xmath107 ( equation [ eqn : upostr2 ] , which uses the constant volume density prior ) as a function of @xmath32 . the two roots are the minimum ( @xmath104 , left ) and the mode ( @xmath11 , right ) . @xmath0 is the measured parallax . [ fig : roots_deriv_d.post_r2prior],scaledwidth=98.0% ] equation [ eqn : r2post_roots ] shows that the mode scales linearly with @xmath2 : the function in the right panel of figure [ fig : roots_deriv_d.post_r2prior ] is the factor which we multiply @xmath2 by to estimate the distance . it depends on the parallax error ( through @xmath32 ) , but not on @xmath71 . contrary to expectations ( perhaps ) , the posterior would have a minimum for all ( positive ) values of @xmath108 , but if it occurs at @xmath82 it will not be seen due to the cut off from the prior . in the case shown @xmath109 , so the minimum will only drop below @xmath71 when @xmath110 . for @xmath32 ( @xmath111 ) with @xmath112 . the solution is @xmath113 ] the minimum goes to infinity in the limit @xmath114 . ( truncated constant volume density prior with @xmath66 ) for @xmath115 and four values of @xmath116 ( black lines ) . the red line shows the posterior for @xmath117 for @xmath118 ( @xmath32 is then undefined).[fig : post_r2prior_nonposw],scaledwidth=49.0% ] the posterior is also defined when @xmath119 , as can be seen by inspection of equation [ eqn : upostr2 ] and shown in figure [ fig : post_r2prior_nonposw ] . this demonstrates that when we use a prior , negative or zero parallaxes provide information in agreement with our intuition : they are just noisy measurements which happened to end up non - positive . they imply the distance is likely to be large ( up to the constraint imposed by prior knowledge ) , and the range of probable solutions depends on the size of the parallax error ( a smaller error means more constraint ) . this intuition is expressed quantitatively by the probabilistic approach . let us now test this estimator with the same empirical test as used before ( section [ sec : empirical_test ] ) . i draw the data from the constant volume density prior ( equation [ eqn : r2prior ] ) with the same value of @xmath71 as used in the posterior ( @xmath120 ) . my distance estimator is the lower mode of the posterior when it is defined , i.e. @xmath11 in equation [ eqn : r2post_roots ] if @xmath108 and @xmath121 , and @xmath71 otherwise . note that this decision can be made using only measured data ( it involves @xmath32 , not @xmath16 ) . if @xmath122 an extreme mode then i set @xmath123 , because this is what the prior tells us . thus @xmath124 for the mode estimator ( equation [ eqn : postr2t_mode ] ) of the posterior @xmath107 ( constant volume density prior ; equations [ eqn : upostr2 ] and [ eqn : r2post_roots ] ) for data drawn from the same prior . the right panel is a zoom of the left panel . the dashed red line shows the fraction of samples with an extreme mode ( @xmath125 ) . the dotted red line shows the fraction of samples with an undefined mode ( @xmath126 ) . in both these cases the distance estimates for the samples are set to @xmath71 . [ fig : scaledresiduals_mode_r2prior_rmax1e3_truncatemode],title="fig:",scaledwidth=49.0% ] for the mode estimator ( equation [ eqn : postr2t_mode ] ) of the posterior @xmath107 ( constant volume density prior ; equations [ eqn : upostr2 ] and [ eqn : r2post_roots ] ) for data drawn from the same prior . the right panel is a zoom of the left panel . the dashed red line shows the fraction of samples with an extreme mode ( @xmath125 ) . the dotted red line shows the fraction of samples with an undefined mode ( @xmath126 ) . in both these cases the distance estimates for the samples are set to @xmath71 . [ fig : scaledresiduals_mode_r2prior_rmax1e3_truncatemode],title="fig:",scaledwidth=49.0% ] the results are shown in figure [ fig : scaledresiduals_mode_r2prior_rmax1e3_truncatemode ] . note that the curve is defined for all positive value of @xmath16 ( the horizontal axis ) , in particular beyond @xmath106 . at any given value of @xmath16 the samples will have a range of values of @xmath32 due to the parallax sampling in step [ listitem : draw_parallax ] of the empirical test . but these still contribute to the sample for the specific value of @xmath16 . we see in the figure how the fraction of samples with an undefined mode ( dotted red line ) increases steadily with @xmath16 . the majority of the samples fall in this category once @xmath127 . this largely explains why the bias and standard deviation saturate : setting most distances to @xmath71 limits the distance error incurred . the fraction of samples with an extreme mode ( dashed red line ) initially increases , but then decreases , because for large parallax errors the measured parallax is increasingly likely either to be negative ( which is not considered an extreme mode ) or positive but so small that @xmath128 , in which case the mode is undefined and the sample contributes to the dotted curve instead . , but now the samples with extreme and undefined modes are rejected rather than set to @xmath71 . [ fig : scaledresiduals_mode_r2prior_rmax1e3_rejectmode],title="fig:",scaledwidth=49.0% ] , but now the samples with extreme and undefined modes are rejected rather than set to @xmath71 . [ fig : scaledresiduals_mode_r2prior_rmax1e3_rejectmode],title="fig:",scaledwidth=49.0% ] using this prior and strategy , we get much better results than with the improper uniform prior ( figure [ fig : scaledresiduals_mode_runifprior_rmax1e3 ] ) . we also achieve a lower standard deviation than with the truncated uniform prior ( figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_truncatemode ] ) for larger values of @xmath16 , but at the price of a much larger bias . note that these results are for data drawn from the same prior as that used in the posterior . with a mismatched prior we will get different results . instead of truncating the extreme and undefined mode values to @xmath71 , we could just reject them , as we did with the truncated uniform prior in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode ] . the results of this are shown in figure [ fig : scaledresiduals_mode_r2prior_rmax1e3_rejectmode ] . the dotted and dashed lines are the same ( they are statistically the same data ) , but now the standard deviation and especially the bias look quite different . indeed , the bias now turns negative for some values of @xmath16 . this is because we are rejecting exclusively samples which would otherwise be assigned the largest distance possible . this also helps to lower the standard deviation . but once @xmath16 increases nearly to 1 ( 100% expected parallax errors ) , the standard deviation and bias are large again . even worse , we have had to reject over 95% of the data . while a constant volume density prior is more desirable from a physical point of view than the prior uniform in @xmath1 , it suffers from the same problem : a discontinuous cut - off in @xmath10 . for lower accuracy parallaxes , this demands that we either throw away those data or assign a fixed distance , @xmath71 . both are undesirable . we would like to have a posterior which yields a distance estimate for any value of @xmath32 and @xmath0 , including non - positive parallaxes . this can be achieved by replacing the sharp cut - off of the previous priors with something which drops asymptotically to zero as @xmath129 . here i investigate a prior which produces an exponential decrease in the volume density of stars , @xmath130 , namely @xmath131 where @xmath132 is a length scale . the unnormalized posterior is @xmath133 } & \:{\rm if}~~ r > 0 \\ 0 & \:{\rm otherwise } \ . \end{dcases } \label{eqn : upostr2e-2}\ ] ] examples of this posterior are shown in figure [ fig : ud.post_r2e - rprior ] . we see a dependence on @xmath32 which is similar to that of the @xmath134 prior for small @xmath1 , but now with a smooth peak at large @xmath1 and a continuous decline beyond . depending on the value of @xmath32 , we may have one or two modes . in this example there is a single mode for @xmath135 , two modes for @xmath136 , and one mode for @xmath137 . note that although @xmath138 is a characteristic length scale of the posterior , the mode corresponding to this is at a somewhat larger value of @xmath1 . ( exponentially decreasing volume density prior ; equation [ eqn : upostr2e-2 ] ) for @xmath139 , @xmath45 and seven values of @xmath140 . the red line is the posterior for @xmath78 and @xmath79 . the green curve is the prior . the right panel is a zoom of the left one and also shows an additional posterior for @xmath141 . all curves have been scaled to have their highest mode at @xmath142 ( outside the range for some curves in the right panel ) . [ fig : ud.post_r2e - rprior],title="fig:",scaledwidth=49.0% ] ( exponentially decreasing volume density prior ; equation [ eqn : upostr2e-2 ] ) for @xmath139 , @xmath45 and seven values of @xmath140 . the red line is the posterior for @xmath78 and @xmath79 . the green curve is the prior . the right panel is a zoom of the left one and also shows an additional posterior for @xmath141 . all curves have been scaled to have their highest mode at @xmath142 ( outside the range for some curves in the right panel ) . [ fig : ud.post_r2e - rprior],title="fig:",scaledwidth=49.0% ] the larger @xmath32 , the less informative the data , and in the limit @xmath143 the posterior just becomes the prior for any value of the parallax ( positive , zero , or negative ) . indeed , already at @xmath144 the posterior is almost indistinguishable from the prior ( the green line ) . the red line in figure [ fig : ud.post_r2e - rprior ] shows the posterior for a negative parallax of @xmath145 with @xmath79 . if @xmath146 were smaller this posterior would shift to the right . this makes sense , because a smaller @xmath146 means we are more confident that the true parallax is close to zero . as @xmath146 gets larger the posterior shifts to the left , eventually converging to the prior . to find the mode we set @xmath147 which gives @xmath148 this is a cubic equation which generally has three complex roots , but fewer solutions here due to the physical limitations on the signs of the variables . the roots are a function not only of @xmath14 , but also of @xmath138 and @xmath7 . depending on these values , the posterior may have three real roots , corresponding to two modes and one minimum , or just one real root , corresponding to a single mode . inspection of the roots leads to the following strategy for assigning the distance estimator , @xmath11 , from the modes : if there is one real root , it is a maximum : select this as the mode . if there are three real roots , there are two maxima : if @xmath149 , select the smallest root ( value of @xmath1 ) as the mode . if @xmath150 , select the mode with @xmath21 ( there is only one ) . the other two possibilities ( zero or two real roots ) do not occur for @xmath151 . , of the posterior @xmath152 ( which uses the exponentially decreasing volume density prior ; equation [ eqn : upostr2e-2 ] ) shown as @xmath153 as a function of @xmath16 . each line is for a different value of the parallax , @xmath0 , and labelled with @xmath154 . the left panel is for @xmath139 and the right panel for @xmath155 . note the ( different ) log scales on the vertical axes . [ fig : roots_mainmode_d.post_r2e - rprior],scaledwidth=98.0% ] the variation of @xmath11 as a function of @xmath32 for different @xmath0 and @xmath138 is shown in figure [ fig : roots_mainmode_d.post_r2e - rprior ] . and @xmath0 , the variation of @xmath156 with respect to @xmath16 is independent of @xmath0 . ] let us follow the curve labelled `` @xmath157 '' ( @xmath158 ) in the left panel , which corresponds to the posteriors plotted in figure [ fig : ud.post_r2e - rprior ] . at small values of @xmath16 ( below 0.30 ) , the posterior has a single mode , and the value of @xmath11 ( for a given @xmath0 ) increases slowly and smoothly with increasing fractional parallax error . as @xmath16 rises above about 0.30 , a second mode appears , but i continue to use the mode at the lower value of @xmath1 as the distance estimator , because we can think of this one as being dominated by the data : it continues to evolve smoothly from the data - dominated regime ( small @xmath16 ) . once @xmath16 rises above 0.373 , there is a sudden increase in the value of @xmath11 . this corresponds to the `` data - dominated '' mode disappearing , leaving only the other,``prior - dominated '' , mode for all larger values of @xmath16 . if the measured parallax were smaller , say @xmath159 , but @xmath138 the same , then we see from figure [ fig : roots_mainmode_d.post_r2e - rprior ] ( left panel ) that the posterior only ever has one mode for all @xmath16 . this is because the data and the prior are now indicating distances on a similar order of magnitude . if we instead used a larger @xmath138 for @xmath159 ( right panel , line labelled `` @xmath160 '' ) , the measured parallax is again quite different from the prior , so we again see two modes for smaller @xmath16 , and the transition to a single mode for larger @xmath16 . we only and always get such transitions when @xmath161 . whenever we have two modes , it is always the smaller one which is dictated by the data , so we can always make the correct choice of distance estimator . for the mode distance estimator of the posterior @xmath152 ( exponentially decreasing volume density prior ; equation [ eqn : upostr2e-2 ] ) with @xmath139 for data drawn from the same prior . the black circles and blue triangles are the bias and standard deviation respectively of the median of the posterior . the right panel is a zoom of the left panel . [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3],title="fig:",scaledwidth=49.0% ] for the mode distance estimator of the posterior @xmath152 ( exponentially decreasing volume density prior ; equation [ eqn : upostr2e-2 ] ) with @xmath139 for data drawn from the same prior . the black circles and blue triangles are the bias and standard deviation respectively of the median of the posterior . the right panel is a zoom of the left panel . [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3],title="fig:",scaledwidth=49.0% ] to assess the performance of this posterior and estimator , i perform the empirical test drawing data from the same prior ( equation [ eqn : r2e-2prior ] ) as used in this posterior , both with @xmath139 . the results are shown in figure [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3 ] . the variation of the standard deviation is similar to what we have seen before with proper priors , and similar in scale to that obtained with the @xmath134 prior . the standard deviation increases linearly when @xmath88 with a gradient of 1.0 , slightly better than the @xmath134 prior in figure [ fig : scaledresiduals_mode_runifprior_rmax1e3_rejectmode ] , and now without having to reject any data . but in the present case we get essentially no bias even for large @xmath32 . this lack of bias was not seen with the previous priors , even when we had a match between the prior and the distribution of true distances . the explanation for this is that we are now using a prior which decreases continuously and asymptotically after some distance . the particular choice of @xmath162 is not important to achieve this . the problem with using a sharp cut - off in a prior is that a star with a sufficiently large fractional parallax error will , before applying the cut - off , have a significant probability for distances greater than the cut - off distance . if the mode lies beyond this value , applying the cut - off will cause a `` build up '' of inferred distances at the cut - off , resulting in a strong bias . this can occur even for stars with very large parallaxes ( small distances ) : not only stars with a true distance near to the cut - off distance are affected . if the fractional parallax error is large enough , the sharp cut - off will always cause a problem , and increasing the cut - off distance will not help . this shows that truncating the prior at the largest distance expected based on the observed magnitude will not avoid the bias . one might argue that truncating the prior at very large distances will have little impact in practice , but with an asymptotically decreasing prior this is even unnecessary , because at @xmath163 there is vanishingly small probability anyway . returning to figure [ fig : roots_mainmode_d.post_r2e - rprior ] , we may be tempted to conclude that using @xmath164 would be a good idea , because it would give @xmath165 . but the parallax is a noisy measurement , so @xmath166 . we are not aiming to get @xmath2 as our estimator ( we saw in earlier sections how bad it is ) . the empirical test confirms that using @xmath164 in the prior gives poor results . so far i have only investigated the mode of this posterior . finding the mode is generally easy , because it involves differentiation and solving a polynomial equation . but it is not necessarily the best estimator in terms of bias and variance . finding the mean and median ( or any other quantile ) involves integrals ( semi - infinite and finite ) of the form @xmath167 over @xmath1 for @xmath168 ( and @xmath169 for the standard deviation , if we wanted this as an error estimate ) . these integrals have no simple solution , so we must resort to a numerical approach . where possible i use adaptive quadrature techniques . for some combinations of parameters this approach can find no solution , in which case i use the metropolis algorithm . samples and a short burn - in . a longer chain would reduce the noise , but takes longer to compute . ] this is slow compared to adaptive quadrature , so i have only computed the bias and standard deviation of the median at a few values of @xmath16 . these are plotted in figure [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3 ] as black circles and blue triangles ( respectively ) . the median has a significant bias for larger @xmath16 . the mean has a similar profile , but with slightly larger values of both bias and standard deviation . so of all the obvious estimators , the mode appears to be the best . , but now drawing data from the @xmath134 prior with @xmath66 ( and the left panel now has a different vertical scale ) . [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3_r2trueprior_rmax1e3],title="fig:",scaledwidth=49.0% ] , but now drawing data from the @xmath134 prior with @xmath66 ( and the left panel now has a different vertical scale ) . [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3_r2trueprior_rmax1e3],title="fig:",scaledwidth=49.0% ] what happens if we have a mismatch between the distribution the data are drawn from , and the prior assumed in the model ? figure [ fig : scaledresiduals_mode_r2e - rprior_rlen1e3_r2trueprior_rmax1e3 ] shows an example of this , where data are drawn from the truncated constant volume density prior ( @xmath65 ) with @xmath66 . as the posterior extends to considerably larger distances than the maximum _ true _ distance of any star in the sample , we inevitably overestimate distances when @xmath16 is large : the estimation of course only sees the noisy measured parallax , which for large @xmath16 has a high probability of being much smaller than @xmath170 . if we repeat this experiment but now drawing data from a prior which extends to larger distances , e.g. @xmath171 , then the posterior estimator will tend to underestimate distances . there is not much more you can do about this other than to try to construct a well - matched prior _ and _ to report confidence intervals on your distance estimates by calculating quantiles on the posterior : i suggest 5% and 95% . this interval will be large for large @xmath32 ( see the left panel of figure [ fig : ud.post_r2e - rprior ] ) , but large fractional parallax errors necessarily mean that our distance estimates are poor and that we are dependent on the prior . yet this is arguably preferable to throwing away data once @xmath32 rises above some arbitrary value . while this prior overcomes many of the problems we saw with other priors , i am not suggesting that we _ always _ use it . the point of the examples in the previous few sections was to show that the choice of prior ( and estimator ) can have a significant impact on the inferred distances , and that some of the obvious choices may be poor . one should always think about what is an appropriate prior , given the available information . this is discussed further in section [ sec : choice ] . the main application of the exponentially decreasing volume density prior will be when we have very little information other than the parallax , and/or want to make minimal assumptions . some of the literature on astrometry has been concerned with trying to apply bias corrections to using @xmath2 as a distance estimator ( e.g. lutz & kelker 1973 , smith 2003 , francis 2013 ) . unfortunately , many of these corrections are only valid under very restricted circumstances , only apply to small ( @xmath172 ) fractional parallax errors , or are even wrong ( brown et al . 1997 , brown 2012 ) . furthermore , we have just seen that there are priors which give unbiased estimates , so do we need to ever make bias corrections ? if there is a mismatch between the prior and the true distribution the stars are drawn from , then we might . but this requires that we know what the mismatch is , in which case we could improve the prior . it seems far better to always improve the prior than to make ad hoc corrections . suppose we did have a desirable estimator which is possibly biased ( e.g. one which gives a lower standard deviation than an unbiased estimator ) . is there a useful correction which could be applied which does not simultaneously increase the standard deviation by too much ? this is an issue , because any bias correction is itself a function of measured and therefore noisy data . the bias and standard deviation plots i have shown , plotted against @xmath16 ( @xmath173 ) , show the bias and standard deviation we expect to obtain on average as a function of the expected fractional parallax accuracy . they are useful to predict the quality of an estimator . but they can not be used to calculate a bias correction , because they depend on an unknown quantity . we would instead have to plot the bias against @xmath32 ( @xmath174 ) . this could be done , but it is questionable whether the use of a suitable prior will ever bring up the need to make a bias correction . there are many more priors and posterior - based estimators one could think of . a good prior is one which is as closely matched to the expected data as possible . in practice it should contain all the relevant information we have which is independent of individual measurements . this could be a combination of both the expected intrinsic distribution of stars in the galaxy , how they enter the sample ( as influenced by the choice of observational wavelength and interstellar extinction ) , and how they are selected by the survey ( e.g. the survey detection efficiency and magnitude limit ) . this could be quite complex , and will generally vary with the direction of observation through the galaxy . the exponentially decreasing volume density prior ( section [ sec : expdec ] ) may be suitable when looking well out of the disk , where for a sufficiently deep survey the decrease in stellar density is caused mostly by the galaxy itself rather than the survey . there are of course many variations on the theme of this prior . for example , we could produce a sharper decrease with increasing @xmath1 by using @xmath175 for @xmath176 . but with @xmath177 replaced by @xmath178 . ] although stellar physics implies a maximum distance for any star , we have seen that discontinuities in the prior have the unpleasant property of leading to rejection of data or the cumulation of all poor data at the same maximum distance . for smooth priors , arbitrarily large distances receive asymptotically small probabilities , so are preferred . if the survey includes extragalactic objects , one will want to allow for very large ( but not infinite ) distances . this could be achieved using a two component prior ( one for galactic , one for extragalactic objects ) with two modes . when the posterior has two modes , one should generally report both , but may choose only to report one if it is significantly higher than the other . setting an informative prior which is wrong could be disastrous . when faced with a choice , i recommend using the simplest prior possible consistent with the constraints , as its influence on the results will be easier to interpret . as with all cases of data analysis and interpretation whether bayesian or not results will depend on personal choices . a good approach is to test the sensitivity of the results to the use of different but equally acceptable priors . the poor quality data ( high @xmath32 ) are more affected by the choice of prior , and these will anyway have broader posteriors . thus it is imperative that confidence intervals are reported on every distance estimate . i suggest to use 5% and 95% quantiles to define a 90% confidence interval . the standard deviation , @xmath13 , should not be used , because @xmath1 is strictly positive . quoting @xmath179 implies a gaussian and negative distances . given a posterior , one is still left with the choice of estimator . for the priors i have tested , the mode ( taking care to deal with multimodality as described ) is more accurate ( lower bias and variance ) and faster to calculate ( no numerical integration ) than the mean or median . ) for gaia , calculated using the gums catalogue with the post - launch , sky - averaged , 5-year mission astrometric accuracy model . the red line is the measured fractional parallax error ( i.e. @xmath32 ) for the hipparcos catalogue . [ fig : gums_hipparcos_cumf],title="fig:",scaledwidth=49.0% ] ) for gaia , calculated using the gums catalogue with the post - launch , sky - averaged , 5-year mission astrometric accuracy model . the red line is the measured fractional parallax error ( i.e. @xmath32 ) for the hipparcos catalogue . [ fig : gums_hipparcos_cumf],title="fig:",scaledwidth=49.0% ] we have seen that when the fractional parallax errors are larger than about 20% , use of a prior is imperative in order to make reasonable distance estimates at all and to assign confidence limits . simply rejecting data with larger errors is in the best case a waste of hard - earned survey data , and in the worst case can severely bias analyses , because we will tend to reject fainter and/or more distant stars . while the accuracy of the ongoing gaia survey will surpass all previous surveys in parallax accuracy on a large number of stars ( accuracy of order 0.02mas at @xmath180 magnitude ) , it remains that many of the stars in the gaia catalogue will have large fractional parallax errors . using a published simulation of what gaia is expected to observe ( robin et al . 2012 ) together with the first post - launch estimates of gaia s end - of - mission , sky - averaged parallax accuracy ( de bruijne et al . 2015 ) , i have calculated the fraction of stars with expected fractional parallax errors ( @xmath16 ) less than some amount : see the black line in figure [ fig : gums_hipparcos_cumf ] . only about 20% of the gaia catalogue will have @xmath181 . this leaves at least 800 million stars requiring use of sensible prior information if we are to obtain a meaningful and not totally deviant distance estimate from the astrometry . the situation is better with the hipparcos catalogue ( red line figure [ fig : gums_hipparcos_cumf ] ) , yet even here only about 40% of the stars have @xmath182 . for this reason , we should make use of astrophysical parameters which will be estimated from the gaia spectrophotometry and spectroscopy ( bailer - jones et al . 2013 ) to improve our distance estimates . specifically , the interstellar extinction and absolute magnitude , combined with our knowledge of stellar astrophysics embodied in the hertzsprung russell diagram , give an independent , probabilistic determination of the distance . this posterior pdf can be combined with the astrometric - based posterior pdf to yield a distance estimate which is more accurate than either estimate alone . i have shown that estimating the distance given a parallax is not trivial . the naive approach of reporting @xmath183 fails for non - positive parallaxes , is extremely noisy for fractional parallax errors larger than about 20% , and gives an incorrect ( symmetric ) error estimate . probability - based inference for the general case is unavoidable . adopting an `` uninformative '' improper uniform prior over all positive @xmath1 does not solve any of these problems , and is in fact both informative and implausible ( viz . a volume density of stars varying as @xmath184 ) . the problems can be avoided by using a properly normalized prior which necessarily decreases after some distance . the use of a prior with a sharp cut - off is not recommended , because it introduces significant biases for low accuracy measurements at all distances ( not just those near the cut - off ) . instead , a prior which converges asymptotically to zero as distance goes to infinity should be used . perhaps the simplest case is @xmath185 , which corresponds to an exponentially decreasing volume density with a length scale @xmath138 . it is relatively neutral in that it is broad , decreases slowly , and has only one parameter . the mode of the corresponding posterior is an unbiased estimator ( for the case that the population observed conforms to the prior ) and it provides meaningful estimates for arbitrarily large parallax errors as well as non - positive parallaxes . bias corrections are not necessary . the variance of this estimator behaves as well as one could expect given the irreducible measurement errors . the median and mean perform less well than the mode . distance estimates must be accompanied by uncertainty estimates . for fractional parallax errors larger than about 20% , the posterior over distance is significantly asymmetric , so we should always report confidence intervals using quantiles ( e.g. 5% and 95% ) rather than standard deviations . the performance of any distance estimator depends , of course , on how well the prior is matched to the true distribution of distances . one must consider how much one is willing to tune the prior to previous knowledge of the galaxy . in any case , priors with discontinuities should be avoided , and the sensitivity of results to the adopted priors should always be tested and reported . limiting inference to objects with fractional parallax errors much better than 20% significantly diminishes the dependence of results on the choice of prior . but this would limit one to using only 20% of the gaia catalogue , which may introduce truncation biases into subsequent astrophysical analyses . its appealing properties notwithstanding , i am not advocating exclusive use of the exponentially decreasing volume density prior . my main message is rather that one should think carefully about what is an appropriate prior , given the available information , and what impact this has on the results . i would like to thank morgan fouesneau , hans - walter rix , anthony brown , and jan rybizki , as well as two anonymous referees , for useful discussions and comments . this work was supported in part by the sfb 881 project `` the milky way system '' of the german research foundation ( dfg ) . bailer - jones c.a.l . , andrae r. , arcay b. et al . , 2013 , _ the gaia astrophysical parameters inference system ( apsis ) . pre - launch description _ , a&a 559 , a74 brown a.g.a , 1997 , arenou f. , van leeuwen f. , lindegren l. , luri x. , 1997 , _ considerations in making full use of the hipparcos catalogue _ , proceedings of the esa symposium ` hipparcos - venice ' 97 , esa sp-402 ( july 1997 ) , pp . 6368 lindegren l. , babusiaux c. , bailer - jones c. a. l. , et al . 2008 , _ the gaia mission : science , organization and present status _ , in iau symposium vol . 248 , ed . jin , i. platais , & m. a. c. perryman , 217-223

grb050904 at redshift , discovered and observed by _ swift_and with spectroscopic redshift from the subaru telescope , is the first gamma - ray burst to be identified from beyond the epoch of reionization . since the progenitors of long gamma - ray bursts have been identified as massive stars , this event offers a unique opportunity to investigate star formation environments at this epoch . apart from its record redshift , the burst is remarkable in two respects : first , it exhibits fast - evolving and optical flares that peak simultaneously ats in the observer frame , and may thus originate in the same emission region ; and second , its afterglow exhibits an accelerated decay in the near - infrared ( nir ) froms tos after the burst , coincident with repeated and energetic flaring activity . we make a complete analysis of available , nir , and radio observations , utilizing afterglow models that incorporate a range of physical effects not previously considered for this or any other grb afterglow , and quantifying our model uncertainties in detail via markov chain monte carlo analysis . in the process , we explore the possibility that the early optical and flare is due to synchrotron and inverse compton emission from the reverse shock regions of the outflow . we suggest that the period of accelerated decay in the nir may be due to suppression of synchrotron radiation by inverse compton interaction of flare photons with electrons in the forward shock ; a subsequent interval of slow decay would then be due to a progressive decline in this suppression . the range of acceptable models demonstrates that the kinetic energy and circumburst density of grb050904 are well above the typical values found for low - redshift grbs .

astro-ph0612256

one of the most exciting results from the first year of operations of nasa s _ swift _ satellite mission @xcite has been the discovery and observation of grb050904 @xcite at redshift @xmath0 @xcite . this burst was initially detected by the _ bat at 01:51:44 ut on september 4 , 2005 , and was quickly followed up by pointed observations with the telescope ( xrt ) and uv / optical telescope on _ swift _ , and by numerous ground - based facilities . early afterglow photometry provided the first indications for a very high redshift for this event , @xmath4 @xcite , prompting a global observing campaign that culminated in the spectroscopic observations by subaru that provided the redshift @xcite and enabled the first investigation of the reionization epoch via grb afterglow light @xcite . within a day of the burst , the discovery of an associated prompt @xmath5mag optical / nir flash was reported @xcite . the timing of this flare , which peaks at @xmath1s after the burst , is coincident with ( xrt ) and @xmath6-ray ( bat ) flares observed by _ swift _ @xcite . this is the second such bright optical flash observed so far , after grb990123 @xcite , and may even exceed that event in optical luminosity @xcite . subsequent optical and nir observations of the fading afterglow were reported by @xcite and @xcite , and provide evidence for a slow decay phase during the first day after the burst , and a jet break at @xmath7days . a campaign of radio observations at the vla yielded multiple detections of the afterglow at 8ghz , which exhibited a slow evolution consistent with a high circumburst density and extremely high kinetic energy @xcite . the lightcurve observed by _ swift _ exhibits numerous interesting features , including the early flare at @xmath1s , vigorous flaring activity , and a possible jet break @xcite . finally , observations with the _ hubble space telescope _ and the _ spitzer space telescope _ by @xcite have yielded late - time detections of the afterglow and host galaxy . studies of grbs at low - redshift , @xmath8 @xcite , have established that the circumburst densities for these events range between @xmath9 and @xmath10 , with their kinetic energies having a relatively narrow distribution with a peak around @xmath11 @xcite . it is therefore interesting to investigate whether the quantities for the high - redshift grbs follow these results or not . several questions seem particularly pertinent . what density range is found in the environments of grbs at high redshift ? how do their kinetic energies compare to those of low - redshift events ? are other properties of high - redshift grbs their beaming angles and shock microphysical parameters the same as for low - redshift grbs ? the answers to these questions can potentially cast light on outstanding mysteries of the grbs themselves , and reveal important aspects of the early universe . these interesting questions provide the motivation for our detailed investigation of the properties of grb050904 . in this paper , we attempt a complete model of the full set of , optical / nir , and radio afterglow observations of grb050904 . our derivation of physical parameters from afterglow observations is carried out in the context of the fireball model ( @xcite , and references therein ) . we make explicit consideration of two scenarios for the origin of the and optical flares at @xmath1s . our first scenario ( a ) attributes the flares to internal shocks or engine activity , and excludes the flares from afterglow fits , with only later observations considered . this is consistent with the approach of @xcite , who argued on the basis of the fast decay of the flares that they could not be due to reverse shock emission . in our alternate scenario ( b ) , however , we attribute the flares to emission from the reverse shock . in order to accommodate their fast decay , we use a starting time for the asymptotic blandford - mckee solution which is near the start of the flare , later than the burst trigger time , which serves to flatten the post - flare decay . in this scenario , the optical flare comes from synchrotron radiation in the reverse shock , and the and @xmath6-ray flares are produced by synchrotron self - compton scattering ( ssc ) of photons in the reverse shock region . a cosmology where @xmath12 is assumed in calculating the luminosity distance @xmath13 . for grb050904 at @xmath0 , the luminosity distance is @xmath14 cm . our model and its supporting analytical formulae are presented in [ sec : obsandtheory ] , with several derivations reserved for the appendices . our numerical simulation procedure and the data set used in our fits are described in [ sec : numsim ] , while in [ sec : analysisres ] we analyze the results , including the @xmath15-band light curve ( [ sec : earlyag ] ) , the radio light curve ( [ sec : radio_light ] ) , and our ultimate energy and density constraints ( [ sec : energy_density ] ) . a discussion of the results and our conclusions are presented in [ sec : disc ] and [ sec : conclusions ] , respectively . the _ swift _ bat observations of the prompt emission give a @xmath6-ray duration of @xmath16 s , a spectrum with power - law photon index @xmath17 , and a fluence of @xmath18 @xcite . given the burst redshift of @xmath19 , the isotropic - equivalent gamma - ray energy is @xmath20 . xrt observations began 161s after the burst and continued for 10 days after the burst trigger , overlapping with bat observations for about 300s before the high energy emission faded below the bat threshold @xcite . thanks to the _ swift _ prompt alert , the early afterglow was also observed promptly by the tarot robotic telescope . the tarot observations started at 86s after trigger and lasted for more than 1500s ; by making a spectrophotometric calibration of the field , boer et al . ( 2006 ) are able to present their data as flux densities at 9500 , which we use and shall refer to as the tarot @xmath21-band observations . other larger ground - based telescopes started imaging the field 3 hours later @xcite . we present our compilation of the observational data in the ( xrt ) , hard ( bat ) , and optical / nir in fig . [ fig : all_the_data ] . here and throughout the paper we convert measurements to flux density measurements at the frequencies of 5 kev ( xrt ) or 50 kev ( bat ) ; these energies correspond to observing frequencies of @xmath22hz ( xrt ) and @xmath23hz ( bat ) , respectively . the conversion factors from photon counts per second ( cps ) to flux are 3.31 @xmath24 for xrt pc data , 1.82 @xmath24 for xrt wt data , 154.6 @xmath24 and 86.2 @xmath24 for bat masktag - lc data , respectively ( see @xcite @xcite for the details of xrt pc , wt and bat data ) . as seen in fig . [ fig : all_the_data ] , the flux and spectral evolution of the burst and afterglow divide the lightcurve into six distinct segments , determined by inspection and motivated by the physical model put forth herein . these segments are : ( a ) @xmath25 seconds : in the band , the flux decays as @xmath26 with an index of @xmath27 @xcite . we follow the conventional definition for the flux @xmath28 . in the @xmath21-band observation , there are two observational data points , the earlier of which is only an upper limit , apparently indicating an increasing tendency of the flux with time . ( b ) @xmath29 seconds : flares are observed in both the @xmath21 and bands , and also in the bat energy range . they peak around 470 seconds after the burst trigger time . the spectral index evolves from @xmath30 to @xmath31 over this time interval . ( c ) @xmath32 seconds : a power - law decay is shown in the band . during the same time interval , there is no optical / nir detection , except for two upper limit flux values at 9500 @xmath33 . ( d ) @xmath34 seconds : many irregular fluctuations are observed in the band . the j band shows a decay which can be described with @xmath35 @xcite . ( e ) 0.5 days @xmath36 days : there is no effective xrt observation within this period , and no further fluctuations are detected . the flux in the j band is flattened a little bit with a temporal index of @xmath37 @xcite or @xmath38 @xcite . ( f ) @xmath39 days : the flux decay index in j band is around @xmath40 . only one data point in the band is available . the j - band data shows a sharp break , which is thought to be the jet break . thanks to the rapid and precise alerts generated by _ swift _ , and its extensive multiwavelength follow - up campaigns , the afterglow data collected during the _ swift _ era has resulted in some necessary modifications to the standard afterglow model . data from _ swift _ have provided the greatest advance over earlier datasets in the band , as _ swift _ responds to the burst orders of magnitudes faster than previous generations of satellites and can often track the afterglow for up to 10 days . many new features of the afterglow have thus emerged , leading to the identification of a canonical x - ray afterglow behavior . in addition to the prompt emission phase , this involves at least five components of the afterglow ( nousek et al . 2006 ; zhang et al . 2006a ) , which are : ( 1 ) a steep decay phase , often interpreted as the tail of the prompt emission , and thus , part of the grb internal shock ; ( 2 ) a shallow decay phase of uncertain origin , with several theoretical models proposed , including energy injection , jet inhomogeneities , or varying shock microphysical parameters ; ( 3 ) a normal decay phase , familiar from pre-_swift _ observations ; ( 4 ) a post jet break phase ; and ( 5 ) the flares , which are superposed on the various power - law segments of the afterglow s decay , and are fast - evolving in the sense that the flare s rise and decay timescales @xmath41 are much smaller than the time since the burst @xmath42 , that is , @xmath43 . the current interpretation ascribes the flares to the same cause as the prompt gamma - ray emission energy dissipation in internal shocks @xcite . since the flares are thus a manifestation of central engine activity , it is necessary to exclude them from afterglow model fits ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the accumulation of _ swift _ afterglow observations has also raised questions about the collimated or `` jet '' interpretation of afterglow lightcurves . traditionally , jets have been invoked to explain a late time ( @xmath44day ) , broadband and achromatic steepening in the afterglow decay by an increment of + 1 in the power - law index , from @xmath45 to @xmath46 . specifically , the late - time decay index is predicted on robust grounds to be @xmath47 , where @xmath48 is the power - law index of the energy distribution of the synchrotron - emitting electrons that generate the afterglow light . current challenges to the jet break picture are two - fold . first , there are few jet breaks seen in the light curves of afterglows observed by _ swift _ @xcite . as of 7 october 2006 , the xrt had observed 145 long grbs and 16 short grbs , with the afterglows of almost all long bursts followed for days to weeks with _ swift _ , in campaigns that typically last until 10 days after the burst . unexpectedly , among these bursts , only 4 long bursts and 2 short bursts show the jet break signature ( see table 1 , * ? ? ? second , where breaks of the appropriate steepness are seen , they are often observed to be chromatic , exhibiting a different evolution in uvot ( or ground - based optical ) observations . thus the jet break picture is being reevaluated , with new possibilities , such as energy injection or the evolution of microphysical shock parameters , under active consideration @xcite . despite these results , we believe that the jet break picture remains the better interpretation for most bursts , even in the _ swift _ era . first , the energetics of grbs are extremely difficult to reproduce from stellar - mass progenitors without beaming corrections of some sort . second , the successes of the standard jet break picture in modeling bursts from before _ swift _ are too numerous and significant to be discounted ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? third , several candidate jet breaks have been identified in the _ swift _ era , and when these are seen , they exhibit the expected properties ( e.g. , * ? ? ? * ; * ? ? ? finally , several arguments support the presence of a jet break in the evolution of the afterglow of interest to us here , grb050904 . the break that is observed in the nir lightcurve of the afterglow at @xmath49 day @xcite is achromatic , being seen in multiple nir bands , and is followed by a steep post - break decay with power - law index @xmath50 , consistent with the closure relation expected from the standard jet model @xcite . separately , combining the ghirlanda relation @xcite , and expression for the jet break time @xcite , a jet break is expected in the range between 0.64 and 3.0 days . the peak energy of grb050904 is @xmath51 kev and its isotropic - equivalent gamma - ray energy lies between @xmath52 and @xmath53 erg @xcite , so we take @xmath54 kev and @xmath55 100 . in addition , the circumburst density lies between @xmath56 @xcite . setting @xmath57 and the radiation efficiency in the prompt emission phase @xmath58 , one can obtain the jet break ranges of 0.64 and 3.0 days from eqn . 3 of @xcite , which is consistent with the observed break time . it has been argued by @xcite that the flares at @xmath1s are due to internal shocks , rather than a reverse shock , based on an apparent steep decay of the light curve right after the peak , with a temporal index of @xmath59 when referenced to the burst trigger time ( see @xcite for a discussion on the onset of grb afterglow ) . this is because the fastest decay must not be steeper than that from the high latitude emission , whose temporal index is @xmath60 , in this case , @xmath61 . this leads them to favor internal shocks for both the optical and flares , or a combination of reverse shock for the optical and internal shocks for the . consistent with this argument , we investigate a model of the forward - shock emission only , that is , we consider only data points in regions ( c)-(f ) of fig . 1 . separately , we consider a scenario in which the flares peaking at @xmath1 seconds in the optical and are due to the reverse shock . this is motivated by the following argument . first , from the observational point of view , the flares of grb050904 have a great similarity with the behavior of grb990123 , which is thought to be due to the reverse shock ( e.g. , @xcite ) . second , the blandford - mckee ( bm ) solution @xcite describing the afterglow is parameterized with a time origin @xmath62 given by the trigger time . however , the bm solution is an asymptotic description of the afterglow evolution , which is valid for times substantially longer than any of the timescales associated with the prompt emission . clearly , however , there is a transition from the initial prompt phase , when the bulk lorentz factor is relatively constant , to the steady deceleration phase when the asymptotic bm solution applies . numerical simulations have so far not provided specific guidance on the most appropriate value of the reference time for bm ( afterglow ) evolution . naturally , however , the steepness of the light curve decay , as parameterized by @xmath63 , depends on the reference time @xmath62 that is adopted . this has been discussed most recently in the context of the flares seen by _ swift _ , which exhibit steep decays and so are widely attributed to internal shocks @xcite . here we are dealing with a somewhat different situation , in that the light curve exhibits an initial flare and a subsequent decay , which we propose to attribute to a reverse shock . in the absence of guidance from numerical simulations , we adopt a purely phenomenological approach , based on the constraint that , whether for internal shocks or for a reverse external shock , the temporal decay can not be steeper than @xmath64 , where @xmath65 is the reference time which fits the high latitude decay @xmath66 , where @xmath67 is the spectral index . for late central engine internal shocks , @xmath65 is found to be near the onset of the last spike @xcite . here , in the same spirit , since the initial flare is assumed to be due to the reverse shock , after which the bm solution is asymptotically reached , the new reference time @xmath65 is determined from the constraint that the decay slope be equal to that expected from the high latitude mechanism . having set a new reference time @xmath65 , all the time - related quantities for the afterglow evolution , such as @xmath68 and so on , are now referred to @xmath69 . we emphasize , however , that although the reference time is shifted from the burst trigger time to a new time point , the physical deceleration timescale remains the usual one . for grb050904 , the best - fitting reference time satisfying the high latitude condition is @xmath70 , where the ( usual ) deceleration timescale is 468s ( and the latter is measured relative to the burst trigger time ) . the deceleration time is a critical point in the afterglow evolution . after that , both the reverse shock and the forward shock evolve into the asymptotic regime of the bm solution . in model ( b ) , the peak time should represent the deceleration time of the shock . in the discussion below all the formulae after the deceleration time are taken relative to new reference time @xmath65 ( we have inserted the exact value of @xmath65 into the equations below ) . a difference between models ( a ) and ( b ) lies in that @xmath71 for model ( a ) and @xmath72 for model ( b ) . the determination of @xmath65 for model ( b ) will be described in [ sec : dataconst ] . the flares in the optical @xcite and @xcite are observed to peak simultaneously at @xmath1seconds after the burst . the data has better time resolution , with an estimated peak time of @xmath73s . we consider fits for two generic models , as described above . in model ( a ) the flares are ignored , as they may come from other regions ( possibly internal shocks ) rather than the reverse shock region . in model ( b ) , however , we consider the flares to be due to the reverse shock . in this case , since the peak of the flare is clearly separated from the grb itself , the reverse shock must be in the thin - shell regime @xcite . spectral analysis @xcite shows that the flares in the and optical bands must be due to different mechanisms . here , we assume that the optical flare is due to synchrotron radiation in reverse shock region , and that the flare is produced by synchrotron - self - compton ( ssc)in the reverse shock region ( e.g. , @xcite ) . other observed components at late times , except for the flares , are assumed to be produced by synchrotron radiation from the forward shock . we use the following reference values for the main parameters involved : the isotropic - equivalent kinetic energy @xmath74 , the external circumstellar density @xmath75 , and the magnetic field ratio of the reverse and forward shocks @xmath76 . in addition , we have assumed equality of the energy equipartition parameter for electrons in the forward and reverse shock regions in model ( b ) . a transition from the temporal index @xmath77 to @xmath50 , interpreted as the jet break , is observed at @xmath78 days @xcite , and the sharpness of this break makes the burst not likely to have occurred in a stellar wind - type density environment @xcite . therefore , we will focus on afterglow evolution in a uniform density environment throughout the paper . usually the synchrotron emission spectrum of the forward shock is described by a broken power - law with three critical quantities @xmath79 , @xmath80 , and @xmath81 which are the synchrotron frequencies for the electron energies at the injection , cooling and peak flux emission , respectively @xcite . in modeling the radio emission , it is necessary to also take into account the self - absorption frequency , @xmath82 . we present the formulae including self - absorption in appendix [ self_ab ] , because there are different forms depending on various afterglow regimes , six in all , for the spectra given by @xcite . the main quantities of interest are : @xmath83 where the convention @xmath84 is used , kinetic energy @xmath85 erg , and density @xmath86 . @xmath87 is the magnetic field equipartition parameter , and @xmath88 is the electron equipartition parameter . the subscript `` @xmath89 '' denotes the forward shock , and `` @xmath90 '' denotes the reverse shock . the parameter @xmath91 refers to the first - order inverse compton effect and is defined as @xmath92 @xcite . here @xmath93 , the fraction of the electron energy that is radiated away , expresses the magnitude of the radiative correction : @xmath94 for fast cooling and @xmath95 for slow cooling , where @xmath96 is the cooling lorentz factor and @xmath97 is the typical lorentz factor for the electrons . for @xmath98 , the luminosity distance in a concordance cosmology is @xmath99 cm . after the deceleration time , relative to the new reference time , the afterglow will evolve asymptotically as the bm solution . thus , the evolution relation for each quantity is @xmath100 , and @xmath101 . the quoted typical values for @xmath79 , @xmath80 and @xmath102 are consistent with the ones in @xcite , but the values for @xmath79 and @xmath80 are 2 times smaller , and the typical value for @xmath103 is 4 times smaller than that in @xcite , which are fitted for the numerical simulation results . in the reverse shock at the deceleration time , the typical quantities are @xcite : @xmath104 where @xmath105 , @xmath106 , @xmath107 is the compton parameter in the reverse shock region , and the factor @xmath108 , where @xmath109 is the lorentz factor at deceleration time . for the thin - shell case , @xmath110 , so we have @xmath111 @xcite . after the deceleration time , since the shock is in the thin - shell case , each quantity evolves as @xmath112 where @xmath113 , @xmath114 where @xmath115 , and @xmath116 where @xmath117}$ ] @xcite where @xmath118 is the evolution index for @xmath119 where @xmath120 is the lorentz factor of the afterglow , and @xmath121 is the radius of the afterglow . both models include inverse compton effects . the formulae for the normal case @xmath122 and @xmath122 in the forward shock case are listed in @xcite . we have derived the formulae for additional cases where the self - absorption frequency is above the typical frequency @xmath79 or cooling frequency @xmath80 in appendix [ app : ic ] . for inverse compton effects in the reverse shock , the self - absorption frequency has a similar form as that of the forward shock , the only difference being that the forward shock - related quantities are replaced by the corresponding reverse - shock quantities . before the bulk lorentz factor drops below the inverse of the opening half - angle , @xmath123 , each of the typical quantities follows the evolution given above ( see sections [ sec : forsy ] and [ sec : revsy ] ) . after @xmath124 , those typical quantities follow a different evolution given by @xcite . we also consider the non - relativistic ( nr ) evolution of the afterglow in its end - state . the time for the afterglow to enter into the nr stage is calculated by the condition that the lorentz factor of the shell @xmath125 . after the afterglow evolves into the nr stage , its dynamics are described by the self - similar sedov - von neumann - taylor solution , for which @xcite have given a detailed description . we have considered the host galaxy extinction in the optical band in the rest frame of the host galaxy , using a milky way extinction curve . we note that the particular type of extinction curve used is not expected to make a difference in this case ; @xcite have tested the application of smc , lmc , and milky way extinction curves to the composite j - band data , and all three models suggest minimal extinction . in our fitting , we treat the visual extinction @xmath127 in the host galaxy as a free parameter . in addition , because the extinction affects the spectral shape , we have considered the spectral index correction due to the host galaxy extinction in our fitting . besides the normal extinction by the host galaxy , the emission close to the wavelength of lyman-@xmath126 at @xmath0 , including @xmath128 band emission ( effective wavelength @xmath129 and band width is @xmath130 ) , has undergone neutral hydrogen absorption in the igm . @xcite fit the spectrum of grb050904 and find its best - fit column density to be @xmath131 . to find the expected lyman-@xmath126 absorption in the @xmath128 band , we have convolved the lyman-@xmath126 absorption profile with the filter transmission curve of the @xmath128-band , and obtain the absorption coefficient @xmath132 meaning @xmath133 loss of @xmath128-band flux . also we notice that from figure 6 in @xcite , the absorption around wavelength @xmath134 is negligible , so we take the absorption efficient @xmath135 for the 9500 data . our model fits for the two different scenarios cover a range of bands from the radio , through the ir / optical , to and bat energies . the list of the observational data used in our global fitting is in table [ table : alpha - beta ] . .reverse and forward shock observation data points [ cols="<,^,^,^ , > , < " , ] [ table : parameter_value ] the shift of the afterglow starting time to the epoch @xmath136 seconds modifies the analysis and the interpretation of the early afterglow in model ( b ) , but this should not affect the late - time afterglow light curve evolution , which should be similar to the one for model ( a ) . thus the discussion below will be focused on the light curve of model ( a ) , and where there are differences , these are pointed out . @xcite have fitted their collection of nir data , and find that between 3 hours and 0.5 days after the burst , the fading of the afterglow can be fitted by a power law of index @xmath137 , while after 0.5 days the fading appears to slow down to a temporal index of @xmath138 . at @xmath139 hours ( 0.44 days ) , the spectral index is @xmath140 . a single power - law decay is ruled out at 3.7@xmath141 confidence . @xcite have extended datasets whose observation time reaches up to 7 days after the burst , and fit the lightcurve with a smoothly - broken power law . the fit gives @xmath142 , @xmath143 , and @xmath144 days . in addition , the spectral index at @xmath145 days is calculated to be @xmath146 or @xmath147 by two slightly different fitting codes . thus , based only on the observations in the optical / nir bands , we can divide the light curve into 3 segments : ( d ) @xmath148 : the afterglow decays as a power - law with index of @xmath149 ; ( e ) @xmath150 : the light curve is relatively flat , decaying as a power - law with index @xmath151 ; and ( f ) @xmath152 : the light curve decays with an index of @xmath153 ( see fig . [ fig : all_the_data ] ) . @xcite have argued that the fast decay during stage ( d ) represents the normal afterglow , and the flattening at stage ( e ) is caused by energy injection . then the stage ( f ) would indicate a return to the normal afterglow evolution . here , however , we have presented a different interpretation for the stages ( d ) and ( e ) . the flux at stage ( d ) is considered to be suppressed by the inverse compton interaction between electrons in the forward shock and flare photons , while the flux in stage ( e ) is the normal flux without external inverse compton process . this is motivated by the argument of @xcite suggesting that the flare photons can interact with the electrons in the forward shock regions via inverse compton scattering . the origin of the late - time flares is unknown , although a widely - held view is that they are due to the internal shocks from late time engine activity . since these photons would be coming , in this view , from a region different from ( and behind ) the forward shock , we call this inverse compton ( ic ) process an `` external ic process . '' in this case , the external ic process will contribute significantly to the cooling of the forward shock electrons , since the flare luminosity is much larger than the forward shock ( afterglow ) luminosity , and the compton parameter is determined by the ratio of those two luminosities . if the total radiated energy at a given time is constant , when the energy radiated via the inverse compton process increases , the synchrotron radiation should decrease . in effect , this can be viewed as the synchrotron radiation having been suppressed by the inverse compton process . since the @xmath15-band luminosity at this time is dominated by synchrotron radiation , the observed flux would become smaller in the presence of strong external ic processes . at the end of the flare , the external inverse compton suppression disappears , and the synchrotron radiation can then return to its normal course . if we assume that the averaged luminosity ratio between the flare and the forward shock in the is @xmath154 ( `` fl '' is the flare and `` f '' denotes the forward shock ) , following the similar definition for the compton parameter which is the ratio of the ic luminosity to the synchrotron luminosity @xmath155 where the subscript `` ssc '' indicates the self - compton scattering process @xcite , we can get the new compton parameter , considering the external ic process , as @xmath156=[(k+1)\eta \epsilon_e/\epsilon_b]^{1/2}$ ] . we can see an additional factor of @xmath157 contributes to the compton parameter for the external ic process compared with the compton parameter for the usual ssc case . because normally the parameter @xmath158 , we expect @xmath159 . for the fast - cooling case , @xmath94 , so we have @xmath160^{1/2}$ ] and @xmath161 . the ic process will affect the synchrotron radiation , and change the cooling frequency @xmath80 for synchrotron radiation and the observed flux @xmath162 . for synchrotron radiation , @xmath163 and the external ic process will lead to a much lower cooling frequency @xmath80 due to an increase in the value of the compton parameter . during the time period of stage ( d ) in fig . 1 , the electrons are in the fast - cooling regime , @xmath164 , where @xmath165 is the observing frequency , so that the ratio of the flux without external ic to the flux with external ic is @xmath166 , where @xmath167 is the flux with external ic process considered and the flux without ic is @xmath168 . therefore , the observed flux in the optical / nir band during section ( d ) should be multiplied by a factor @xmath169 $ ] to recover the flux that would be observed without ic suppression effects . we can estimate the required luminosity ratio from the observed suppression factor . taking the power - law index for the electron energy distribution to be @xmath170 , the theoretically expected temporal index is @xmath171 for @xmath172 , the expected regime for the optical afterglow during stage ( e ) . then we can extrapolate the optical lightcurve back from @xmath173s ( where the flux is @xmath174 ) to the observer time @xmath175s , where the observed flux is @xmath176 . this theoretical flux , in the absence of the external ic process , will be @xmath177 . thus , the flux will be suppressed by a factor of @xmath178 . from the observational point of view , we invert this problem and solve for @xmath179 , deriving @xmath180 . apparently , compared to the mean observed luminosity ratio between the flares and the forward shock , @xmath181 ( see fig . [ fig : all_the_data ] ) , this indicates within our picture that only a small fraction of the flare photons have interacted with the electrons in the afterglow . a possible explanation could be that the anisotropic distribution of the incoming flare photons in the comoving frame of the afterglow shock @xcite . this resultsing more head - on scattering which reduces the ic interaction . therefore , the suppression is relatively small and the optical flux is only affected to a reduce degree . grb050904 shows several similarities with grb990123 , including a large isotropic - equivalent gamma - ray energy and a very bright optical flash . in grb990123 , the radio emission was observed to peak at @xmath182 day in the observer frame , and this emission was interpreted as the radio emission from the reverse shock @xcite . given the important role of the reverse shock in our model ( b ) of grb 050904 , it is interesting to consider its radio emission . however , our estimates indicate that the reverse shock radio emission at early time would be suppressed by the large circumburst density in our model , so that we would not expect to observe a bright radio flare . in fig . [ fig : light_curve_flare ] , we have plotted the reverse shock radio emission as a dotted blue line starting from @xmath183s , while the solid blue line shows the combined radio emission from both the forward shock and the reverse shock ; as can be seen , the emission from the reverse shock is negligible . in fig . [ fig : light_curve_flare ] , there is one averaged radio data point centered at @xmath184 seconds and @xmath185 ( indicated with dashed lines ) based on the average of multiple vla observations spanning @xmath186 days ( see fig . 2 of * ? given the large dynamic range in time , we consider this data point as an upper limit of the radio emission during that period . we can see that our best - fit radio light curve for the forward shock in both models ( a ) and ( b ) fits accommodates this upper limit reasonably well , along with the other data points . the density obtained in our fit , @xmath187 for model ( a ) and @xmath188 for model ( b ) , is smaller than the @xmath189 density derived by @xcite . investigating our model fits , we find that the discrepancy arises because their model peaks later in the radio than ours . referring to the posterior distribution for the density parameter in figs . [ fig : for_dis ] and [ fig : for_rev_dis ] , we find that the most likely range ( 90%-confidence ) for density is from 26 to 273.4 for model ( a ) and from 87.8 to 270.6 for model ( b ) . on the other hand , the light curves for different densities demonstrate that there is no big noticeable affect of these density changes on the and optical / nir lightcurves . the radio observations thus provide the best frequency for density constraints , and fail to provide a tight constraint mainly because of the lack of data ( measurements ) late times . with regards to the circumburst density , we note that @xcite observed the spectrum and detected in it fine structure lines including @xmath190 , which were taken to imply an electron density of up to @xmath191 . the density obtained from the spectral lines would thus be consistent with our best - fit value for the density , and suggest that the observed lines formed in region similar to that hosting the grb itself . however , we note that several recent papers @xcite have suggested that these fine - structure transitions in grb afterglows are excited by radiative processes , rather than collisions , which would make the density constraint irrelevant . an important check on our model results , pointed out by @xcite , is to estimate the luminosity from the light curve at some fiducial time ( usually at @xmath192 hours ) . at @xmath193 days , the observed flux over the xrt energy range is @xmath194 . calculating back to @xmath195 hours for the afterglow flux from synchrotron radiation , the predicted flux is @xmath196 . therefore the luminosity should be @xmath197 , and the geometrically - corrected luminosity we find is different from and significantly lower than theirs . the reason for this is that when they calculate the isotropic - equivalent energy , they appear not to have included the @xmath179-correction factor @xmath198 . @xcite also appear not to have included the @xmath179-correction factor when doing statistics on the isotropic - equivalent and geometrically - corrected luminosity . here we do the statistics again after putting the correction , and find that the geometrically - corrected luminosities are still clustered , but the peak value has shifted down to @xmath199 , reduced by a factor of five . once we have obtained the luminosity , we can estimate the kinetic energy with the best fitting parameters and the observed spectral index . in model ( a ) , we have best fitting parameters @xmath200 , @xmath201 , @xmath202 , @xmath203 , and @xmath204 . from the equation [ eqn : kinetic_energy ] in the appendix , we have @xmath205 , which is consistent with the kinetic energy derived from the fit , @xmath206 . this is not surprising since the formula in the appendix is only a shortcut to obtain the kinetic energy . in model ( b ) , we have the best fitting parameters @xmath207 , @xmath208 , @xmath202 , @xmath209 , and @xmath204 , and similarly we obtain @xmath210 . after considering the @xmath179-correction factor for the luminosity by @xcite , we can re - estimate the kinetic energy as @xmath211 with the parameters @xmath212 , and @xmath213 , which is 3 times larger than their best - fit kinetic energy @xmath214 . considering our estimated opening half - angles of @xmath215 for both models ( a ) and ( b ) , we can calculate the geometrically - corrected luminosity for grb050904 to be @xmath216 , which falls within the corrected luminosity range of low - redshift grbs @xcite . we now calculate the geometrically - corrected kinetic energy . our best - fit opening half - angle is @xmath215 for both models , and the fitted kinetic energy for model ( a ) and ( b ) is @xmath217 ergs and @xmath218 ergs , so the geometrically - corrected kinetic energy will be @xmath219 ergs and @xmath220 ergs for models ( a ) and ( b ) , respectively . broadband modeling of 10 low - redshift bursts indicated that the geometrically - corrected kinetic energies of two were anomalously high , @xmath221 ergs , approximately 10 times higher than for the other eight grbs @xcite . our geometrically - corrected kinetic energy for model ( a ) is comparable to those anomalously - large kinetic energies seen from low - redshift grbs . the kinetic energy of model ( b ) , on the other hand , is significantly larger even than this . both models yield a relatively large kinetic energy has been obtained for grb050904 . grb050904 thus suggests that bursts at high redshift are somehow able to tap into a higher - energy reservoir than the low - redshift events . the radiated isotropic - equivalent gamma - ray energy for grb050904 is @xmath222 ergs @xcite . if the isotropic - equivalent kinetic energy for the afterglow of grb050904 , as we concluded in case ( a ) , is @xmath217 ergs , we can estimate the grb efficiency as @xmath223 @xcite , which is @xmath224 for grb050904 . if the isotropic - equivalent kinetic energy of the afterglow , as in case ( b ) , is @xmath225 ergs , then the corresponding grb efficiency is roughly @xmath226 . in either case , this indicates that grb050904 has a high efficiency ; however , such high efficiencies are not unique . @xcite found that a substantial number of grbs have high efficiency . for some bursts like grb990705 , the inferred efficiency even reaches up to @xmath227 . @xcite have shown that the inferred efficiency can be reduced when inverse compton effects are taken into account ( see also * ? ? ? even so , there are still several bursts which have a high observed efficiency , for example , @xmath228 for grb 050315 and @xmath229 for grb 050416 @xcite . the extreme interest in grb050904 has motivated several groups to analyze the burst data and suggest interpretations . these works fall into two categories : ( 1 ) comparing the properties of grb050904 with other bursts ; and ( 2 ) making fits to the grb050904 data either in the framework of internal shock or external shock models . we give a brief description below of the work of other groups and mention key differences between their work and ours . in category ( 1 ) , @xcite made a composite @xmath15-band light curve starting from @xmath230 days to @xmath231 days . after applying extinction correction , they shifted grb050904 as well as lower - redshift bursts to @xmath232 , and made a comparison of afterglow lightcurves . they found that grb050904 is much brighter than other grbs at early times , but of roughly equal brightness at late times . thus they conclude that grb050904 most likely is still a normal grb . the other analyses are all in category ( 2 ) . @xcite argued that grb050904 is a burst with extremely long central engine activity . they put all the observed data within the framework of the internal shock model . by contrast , we only treat the first several hundred seconds ( bat ) as internal shock activity in our model ( corresponding to the stages ( a ) and ( b ) for model ( a ) , and stage ( a ) for model ( b ) in fig . 1 ) . the late - time flares between @xmath233 and @xmath234 seconds may be due to internal shock , but we have interpreted portions of this emission as being due to the forward - shock afterglow . @xcite argued that the @xmath1s flare is from internal shocks on the basis of its fast decay ( the temporal index is @xmath235 relative to the trigger time ) and also because the optical - to- emission of the flare can not be described by a synchrotron radiation model @xcite . they made fits to all the available @xmath15-band data . they argue that the slow - decay portion of the lightcurve is due to energy injection . in our model , besides fitting over all the additional bands ( and radio ) , we propose a new mechanism for the flattening , namely that it is caused by the suppression of the synchrotron radiation by the interaction between the flare photons and afterglow electrons . separately , in our model ( b ) , we introduced a new reference time @xmath65 which flattens the decay index , and allows us to interpret the @xmath236s optical/ flare as arising in the reverse shock . @xcite made broadband model fits including the , nir / optical , and radio data . the difference with our work is partly that we have used the larger data set that later became available . we have included two data points as early as @xmath237 seconds , and we also included the @xmath238 band data observed @xmath239 days after the burst @xcite . since most of the other data are concentrated around @xmath240 seconds , the introduction of these additional and nir data have some impact on the final fitting result . the other major difference is that we have freed all the possible parameters and applied the mcmc method for the global fitting , making an efficient exploration of the full parameter space , and providing the posterior distributions ( including confidence intervals ) for each parameter . @xcite argued that the power - law - like decay right after the flare @xmath1s should be interpreted as forward shock emission . because the extrapolation of the late time flux to early times is lower than the observed value , they found that a wind - type environment was favored by the closure relationship for this early - time segment ( it should also be noted that their spectral index is smaller than @xcite ) . they propose a density - jump model for the afterglow evolution : before a certain radius , the density goes as @xmath241 and after that , the density is a constant ism model . at the transition point a termination shock is formed which lies around @xmath242pc from the central engine . in our model , we consider the same segment of data , but interpret it differently . we argue that the flux between 600 and 800 seconds arises from the combination of three sources : high - latitude emission of the prompt emission , flux from the flare , and forward shock emission . reviewing the data closely , we see that actually there are two other small flares between 800 and 2000 seconds , so the data around @xmath243 seconds has the contribution from the flares and the forward shock ( see fig . once we subtract the flare contribution from the observed data , we argue , the forward shock contribution is what remains . and in fact , we find that an extrapolation of the late - time flux to early times is consistent with this flare - subtracted early - time flux . there remain some substantial differences in parameter values between the two models that we have presented . for example , the most likely value for @xmath244 in model ( a ) is roughly @xmath245 , but the most likely value for @xmath244 in model ( b ) is around @xmath246 . it turns out that the main difference lies in the optical flux at early times . we take as an example the @xmath15-band light curve for the best fitting parameter set in model ( a ) . the flux slowly decays with an index of @xmath247 before @xmath248 and then follows a faster decay with a temporal index of @xmath249 after @xmath250 . the break at @xmath251 seconds is caused by the crossing of the electron s typical frequency @xmath79 through the optical observing frequency . since we have @xmath252 , then the smaller @xmath244 , the smaller the typical frequency , therefore the earlier the crossing time . the fit in model ( b ) requires an earlier break than in model ( a ) , so a smaller value of @xmath244 is expected in model ( b ) . since the observed flux from the forward shock is the same in both models , we expect a higher kinetic energy in model ( b ) . similarly , at the deceleration time , the reverse shock flux is larger than that from the forward shock , so a larger @xmath253 is expected in the reverse shock . we notice that in both our models we find a small value for @xmath88 , e.g. , the best fitting @xmath254 for model ( a ) and @xmath255 for model ( b ) . considering the radiative correction factor @xmath256 ( where @xmath62 is the deceleration time , and we set @xmath257 seconds for model ( a ) and @xmath258 seconds for model ( b ) ) , that is , the amount by which the kinetic energy at time @xmath42 is reduced by comparison to the kinetic energy at time @xmath62 , we find @xmath259 for model ( a ) and @xmath260 for model ( b ) at @xmath261 seconds after the burst . therefore , the radiative losses are mild in either case , consistent with our assumption that the afterglow evolves in an adiabatic fashion . our fitting results also show that the host galaxy dust extinction is quite small , @xmath262 mag or even smaller , consistent with the results from @xcite . they applied several different dust models ( mw , lmc and smc ) over the composite @xmath15-band , and all the models suggested zero or negligible extinction . in the context of grb990123 , it was suggested that the negligible extinction to the burst was the result of dust destruction by the strong burst and early - afterglow emission @xcite . in this paper we have performed the most extensive multiband analysis so far of the grb050904 afterglow . we have considered two scenarios : ( a ) only forward shock emission is considered , and the flares peaking at @xmath263 after the burst are assumed to be due to internal shocks ( or are otherwise independent of the afterglow ) ; and ( b ) the nir and flares at @xmath1s are ascribed to emission from the reverse shock when the ejecta has swept up enough material and starts to decelerate , the synchrotron radiation in the reverse shock produces the optical flare , and the self - compton scattering of synchrotron photons generates the flare observed in the xrt and bat energy bands . combining the early afterglow data with late - time observations in the , optical and radio , and using a markov chain monte carlo method , we present a full characterization of the posterior distributions ( including confidence intervals ) for the various parameters of our model fits . our best - fit parameter values for model ( a ) are @xmath203 , @xmath264 , @xmath201 , @xmath265 , @xmath266 , @xmath267 , and @xmath268mag , with a reduced chi - squared value of @xmath269 . our best - fit parameter values for model ( b ) are @xmath209 , @xmath270 , @xmath271 , @xmath272 , @xmath273 , @xmath274 , @xmath275mag , @xmath276 , and @xmath277 , with a reduced chi - squared value of @xmath278 . note that the subscripts @xmath90 and @xmath89 refer to the reverse shock and forward shock respectively , @xmath279 , and we have assumed @xmath280 . we have compared the density , the geometrically - corrected kinetic energy and the luminosity at @xmath192 hours derived here for our two models of grb050904 against those values for other bursts , as derived from afterglow modeling . the results for both models show that although the luminosity of grb050904 falls within the range for low - redshift grbs , the density and geometrically - corrected kinetic energy are both above the typical values for low - redshift grbs , which suggests that grb050904 may be a member of a distinct population of high - redshift , higher kinetic - energy bursts , whose properties differ from those of low - redshift grbs . a clear preference between our ( a ) and ( b ) models is hard to establish at present , since there is only one high - redshift grb known . one would like access to several more high - redshift grb observational datasets before attempting to discriminate between the two models . it is estimated that @xmath281 grbs are located at @xmath282 @xcite , and detection rate simulations by @xcite indicate that _ swift _ could detect grbs out to redshift @xmath283 , if they are present . @xcite also predict that @xmath284 of the _ swift _ grbs originate at @xmath282 . it appears that one can realistically expect a handful ( 5 to 10 ) of additional high - redshift grb detections with rapid follow - up in the next few years of the _ swift _ mission . in this case , the consistent application of mcmc methods , as used here , will lead efficiently to a set of statistically well - quantified , posterior parameter distributions and confidence intervals . this would enable a statistically meaningful comparison of high - redshift and low - redshift grb parameters , which might well lead us to a definite understanding of the physics and environments of grbs as a function of redshift , up to the highest redshifts detected . this would also have a substantial impact on the study of the large scale structure and star formation processes throughout the universe , and the properties of the cosmic reionization at @xmath285 . l. j. gou thanks g. cusumano for providing the data for this burst , as well as j. cummings , b. zhang , s. kobayashi , zh . wang , and j.f . wang for the helpful discussion . l. j. gou also thanks the support of sigma - xi fellowship . this research has been supported in part through nasa nag5 - 13286 . this research has used the resources of high performance computing ( hpc ) group at the penn state university . once we consider radio emission in our afterglow models , the self - absorption frequency becomes an important parameter to consider . below we give the expressions for the self - absorption frequency in the different regimes . if we assume that the electron distribution follows a power law @xmath286 where @xmath287 , then the self - absorption coefficient for the various possible regimes is @xcite : @xmath288 where @xmath289 , @xmath290 , @xmath291 , @xmath292 and @xmath293 are the typical synchrotron frequencies of electrons with the lorentz factor @xmath294 and @xmath295 , respectively , and @xmath296 is the gamma function . following the definition of @xmath297 for the self - absorption frequency ( where @xmath298 is the thickness of the shell ) , we can find the self - absorption for the forward shock region . the self - absorption frequency for the reverse shock has a similar form , with the difference that the quantities specific to the forward shock region should be replaced with those specific to the reverse shock . fast cooling : @xmath299 , @xmath300 where the superscript `` @xmath301 '' denotes the different regimes . fast cooling : @xmath302 , @xmath303 fast cooling : @xmath304 , @xmath305^{1/(p+5)}\end{aligned}\ ] ] slow cooling : @xmath122 @xmath306 we notice that @xmath307 . to keep the continuity of the flux while the afterglow transits from the fast - cooling regime to the slow - cooling regime , we divide @xmath308 by a factor of @xmath309 . slow cooling : @xmath310 , @xmath311^{1/(3(p+4))}\end{aligned}\ ] ] slow cooling : @xmath312 , @xmath313^{1/(p+5)}\end{aligned}\ ] ] it can be shown that the self - absorption frequency in the regime @xmath312 has the same form as that for the fast - cooling case , @xmath304 . because the observer time , @xmath314 , at @xmath315 is connected to the time in the source frame , @xmath316 , at redshift @xmath128 by the relation @xmath317 , we have the redshift dependence for main characteristic quantities : the shock radius @xmath318 , the shock lorentz factor @xmath319 , the magnetic field @xmath320 , the typical lorentz factor @xmath321 , the cooling lorentz factor @xmath322 . in addition , we have @xmath323 and @xmath324 ( see eqn . [ eqn : forward ] ) . substituting these dependence into the relations for the self - absorptions above , the redshift dependence for the self - absorptions is @xmath325 } , \nu_a^{(4 ) } \propto ( 1+z)^{-1 } , \nu_a^{(5 ) } \propto ( 1+z)^{(p-6)/[2(p+4)]}$ ] , and @xmath326}$ ] . as described by @xcite , the inverse compton flux can be calculated from the following double integral : @xmath327 where @xmath328 is the electron distribution in the shocked shell , @xmath329 is the seed photon flux , and @xmath330 is defined as @xmath331 where the subscript `` @xmath332 '' denotes the seed photon . the distribution of seed photons is described by the synchrotron spectrum , a broken power law with the characteristic quantities ( sari et al . then the inner integral in eqn . ( [ inverse_compton ] ) gives : @xmath333 the integration over different electron energies again needs to be divided into four different regimes : @xmath334 , & \nu_a^{ic } < \nu < \nu_m^{ic } ; \\ \left[\displaystyle \int_{{\gamma}_c}^{{\gamma}_{cr } ( \nu_m)}{d { \gamma}n ( { \gamma } ) i_3 } + \displaystyle \int_{{\gamma}_{cr } ( \nu_m)}^{{\gamma}_{cr } ( \nu_a)}{d { \gamma}n ( { \gamma } ) i_2 } + \displaystyle \int_{{\gamma}_{cr } ( \nu_a)}^{\infty}{d { \gamma}n ( { \gamma } ) i_1}\right ] , & \nu_m^{ic } < \nu ; \\ \end{array } \right.\end{aligned}\ ] ] evaluating the integrals in eqn . ( [ int2 ] ) , we only keep the dominant terms : @xmath335 , & \nu_a^{ic } < \nu < \sqrt{\nu_m^{ic } \nu_c^{ic } } ; \\ ( \frac{\nu}{\nu_a^{ic}})^{-1/2 } [ \frac{2}{(p+2)(p-1 ) } + \frac{1}{3}\log\left ( \frac{\nu_m}{\nu_a}\right)+\frac{5}{9 } ] , & \sqrt{\nu_m^{ic } \nu_c^{ic } } < \nu < 4{\gamma}_m^2 \nu_a x_0 ; \\ ( \frac{\nu}{\nu_a^{ic}})^{-1/2}[\frac{2}{(p+2)(p-1 ) } + \frac{2}{3(p-1)}+ \frac{1}{3}\log\left(\frac{\nu_m^{ic}}{\nu}\right ) ] , & 4{\gamma}_m^2 \nu_a x_0 < \nu < \nu_m^{ic } ; \\ ( \frac{\nu_m^{ic}}{\nu_a^{ic}})^{-1/2}(\frac{\nu}{\nu_m^{ic}})^{-p/2}[\frac{2}{(p+2)(p-1)}+ \frac{1}{p+2}\log\left(\frac{\nu}{\nu_m^{ic}}\right ) + \frac{2}{3(p-1 ) } ] , & 4{\gamma}_m^2 \nu_a x_0 < \nu < \nu_m^{ic } \end{array } \right.\end{aligned}\ ] ] @xmath336 @xmath337 , & \nu_a^{ic } < \nu < 4 { \gamma}_m^2 \nu_a x_0 ; \\ ( \frac{\nu}{\nu_a^{ic}})^{-p/2 } ( \frac{{\gamma}_m}{{\gamma}_c})^{(p-1 ) } [ \frac{2}{(p+2)(p-1)}+\frac{1}{(p+2)}\log\left(\frac{\nu}{4 { \gamma}_m^2 \nu_a x_0}\right)+\frac{2(p+5)}{3(p+2)^2 } ] , & \nu_c^{ic } < \nu ; \end{array } \right.\end{aligned}\ ] ] @xmath338 @xmath339 , & \nu_a^{ic } < \nu < 4 { \gamma}_c^2 \nu_a x_0 \\ ( \frac{\nu}{\nu_{a}^{ic}})^{-p/2}(\frac{{\gamma}_c}{{\gamma}_m } ) [ \frac{2(p-1)}{(p+2)}+ \frac{(p-1)}{(p+2)}\log\left(\frac{\nu}{4 { \gamma}_c^2\nu_a x_0}\right ) + \frac{2(p+5)(p-1)}{3(p+2)^2 } ] , & \nu > 4{\gamma}_c^2 \nu_a x_0 ; \end{array } \right.\end{aligned}\ ] ] @xmath340 @xmath341 , & \nu_a^{ic } < \nu < \sqrt{\nu_m^{ic } \nu_c^{ic } } ; \\ ( \frac{\nu}{\nu_a^{ic}})^{-(p-1)/2 } [ \frac{2(p-1)}{(p+2)}+ \frac{(p-1)}{(p+1)}\log\left(\frac{\nu_c}{\nu_a}\right)+ \frac{2(p+4)(p-1)}{3(p+1)^2 } ] , & \sqrt{\nu_m^{ic } \nu_c^{ic } } < \nu < 4{\gamma}_c^2 \nu_a x_0 ; \\ ( \frac{\nu}{\nu_a^{ic}})^{-(p-1)/2}[\frac{2(p-1)}{(p+2)}+ \frac{(p-1)}{(p+1 ) } \log\left(\frac{\nu_c^{ic}}{\nu}\right)+ \frac{2(p-1)}{(p+1 ) } ] , & 4{\gamma}_c^2 \nu_a x_0 < \nu < \nu_c^{ic } ; \\ ( \frac{\nu_c^{ic}}{\nu_a^{ic}})^{-(p-1)/2}(\frac{\nu}{\nu_c^{ic}})^{-p/2}[\frac{2(p-1)}{(p+2)}+\frac{(p-1)}{p+2}\log\left(\frac{\nu}{\nu_c^{ic}}\right)+\frac{2(p-1)}{(p+1)}],&\nu > \nu_c^{ic } \end{array } \right.\end{aligned}\ ] ] it is noted that a factor of @xmath342 has been multiplied for the slow - cooling case in order to keep the ic flux continuous while the afterglow changes from the fast - cooling regime to the slow - cooling regime . in the observer frame the energy loss rate is equal to the rate at which energy is supplied to the unshocked matter @xmath343 @xcite , multiplied by @xmath344 , the fraction of energy that each particle has lost ( assuming that the relation between the radius and the lorentz factor is described by @xmath345 where a=2 for the grb prompt phase and @xmath346 $ ] for the afterglow deceleration phase ) . @xmath347 where @xmath121 is the radius of the shocked region from the center in the observer frame , @xmath348 is the pressure in the shocked region in the observer frame , and @xmath349 is the energy density of the shocked region in the comoving frame . normally we assume that all the energy stored in electrons has been radiated during the fast cooling phase , i.e. , @xmath350 . however , @xcite have shown that only a portion of the energy will be lost even if the electrons are in the fast - cooling regime ( refer to their fig . the relation between the radiation factor @xmath344 and the electron equipartition parameter @xmath88 is given by @xmath351 ( their eq . 46 ) where @xmath352 for the extreme relativistic case is the adiabatic index . we can solve the above equation numerically , e.g. , @xmath353 for @xmath354 , @xmath355 for @xmath356 , and @xmath357 for @xmath358 . we find that for the reasonable range of parameter values , we have the following relation , @xmath359 , and substituting it and bm solution back into eq . [ eqn : energy_loss ] above , we obtain : @xmath360 we have the solution as @xmath361 where @xmath62 is the deceleration time . because we have applied the blandford - mckee solution during the whole derivation , the above solution is only valid after the deceleration time . normally during the afterglow deceleration phase , the we set the parameter @xmath362 , and thus we obtain the radiative correction factor after prompt phase @xmath363 . we note that the above correction factor can only be applied to the fast cooling case . taking @xmath364 and @xmath257 s , the correction factor at observer time @xmath192 hours would be @xmath365 . it should be mentioned that if the estimated correction factor is much larger than the unity , one should not estimate the kinetic energy only by introducing the radiative correction factor @xmath121 for the adiabatic solution , and should reconsider the whole evolution including the radiation loss effects . because at this point the radiation is not negligible any more , the afterglow evolution will deviate substantially from the bm solution ( e.g. , * ? ? ? * ; * ? ? ? the radiative correction factor @xmath363 for the afterglow is different from the one , @xmath366 , provided by @xcite ( see also @xcite ) for two reasons : ( 1 ) they have taken the value @xmath367 , while we have used @xmath362 which is believed to be much more reasonable from the detailed studies of the hydrodynamic evolution of the afterglow @xcite . ( 2 ) they have assumed all the electron energy will be radiated away if the afterglow is in the fast - cooling regime , but from the results of @xcite , they showed that roughly half of the electron energy is lost . if the band is above the typical frequency @xmath79 and the cooling frequency @xmath80 , the observed flux at a certain observer time is given by : @xmath368 where @xmath48 is the energy distribution index , @xmath369 is the isotropic - equivalent kinetic energy , @xmath88 is the electron equipartition parameter , @xmath87 is the magnetic field equipartition parameter , @xmath42 is the observer time , @xmath370 is the x - ray observing frequency , and @xmath91 is the compton parameter . we can find the kinetic energy @xmath375 from the above equation reversely , and considering the radiative correction factor , we obtain , @xmath376^{4/(p+2 ) } ( 1+z)^{4(p/4 - 1/2-\alpha + \beta)/(p+2 ) } \\ & & \times \epsilon_{e,-1}^{4(p-1)/(p+2 ) } \epsilon_{b,-2}^{-(p-2)/(p+2 ) } t_{10 { \rm h}}^{(2 - 3p)/(p+2 ) } \nu_{18}^{2(2-p)/(p+2 ) } ( 1+y)^{4/(p+2 ) } \label{eqn : kinetic_energy}\end{aligned}\ ] ] where @xmath377^{(17/48 ) \epsilon_e}$ ] is the radiative correction factor , and the derivation is given in appendix [ app : der_rad ] .

massive young stellar objects ( ysos ) are powerful infrared h i line emitters . it has been suggested that these lines form in a outflow from a disk surrounding the yso . here , new two - dimensional monte carlo radiative transfer calculations are described which test this hypothesis . infrared spectra are synthesised for a yso disk wind model based on earlier hydrodynamical calculations . the model spectra are in qualitative agreement with the observed spectra from massive ysos , and therefore provide support for a disk wind explanation for the h i lines . however , there are some significant differences : the models tend to overpredict the br/br ratio of equivalent - widths and produce line profiles which are slightly too broad and , in contrast to typical observations , are double - peaked . the interpretation of these differences within the context of the disk wind picture and suggestions for their resolution via modifications to the assumed disk and outflow structure are discussed . [ firstpage ] radiative transfer methods : numerical stars : early - type stars : winds , outflows

astro-ph0508103

understanding the formation and pre - main - sequence evolution of massive stars ( @xmath2 m@xmath3 ) is an important goal in modern astrophysics . currently , the formation of low - mass stars via gravitational collapse of a molecular cloud and subsequent disk accretion is , at least conceptually , well understood but it is not clear whether massive stars form through a scaled - up version of the same process or whether more complex , environmental effects are important ( see e.g. bonnell , vine & bate 2004 for a recent discussion ) . since massive stars form inside dense molecular clouds , they are observed primarily at infrared ( ir ) and longer wavelengths . embedded massive ysos are powerful sources of ir h i line emission ( simon et al . 1981 , 1983 ; drew , bunn & hoare 1993 ; bunn , hoare & drew 1995 ; blum et al . 2004 ) with equivalent widths ( ews ) in br@xmath0 of @xmath4 90 ( bunn et al . the lines profiles are complex , displaying both fairly narrow ( full - width - at - half - maximum [ fwhm ] @xmath5 50 100 km s@xmath6 ) line cores and broad wings extending out to @xmath7 km s@xmath6 in extreme cases . in the sample of objects observed by bunn et al . ( 1995 ) , the profiles were always single - peaked but recent observations reported by blum et al . ( 2004 ) include at least one object ( ngc 3576 ) which shows clear evidence of double - peaked profiles . by examining flux ratios across the profiles of lines with differing opacities , bunn et al . ( 1995 ) found evidence that , at least in some sources , the lines form in an accelerating outflow . normal , main - sequence ob stars also show h i emission lines . these are produced in the star s fast ( @xmath8 km s@xmath6 ) spherical wind , but the strength of emission in yso spectra is much greater than in o - stars . if the features in ysos were due to reprocessing of radiation in a spherically symmetric wind , the implied mass - loss rates would be up to @xmath9 m@xmath3 yr@xmath6 ( simon et al . 1983 ) , substantially exceeding those of field stars of comparable spectral type . therefore , to interpret the h i observations it is worthwhile to consider alternatives to formation in a spherically symmetric structure . important observational constraints on the geometry of emitting material in ysos are provided by the first overtone bands of co at 2.3@xmath10 m ( e.g. carr 1989 ; carr et al . 1993 ; chandler et al . 1993 ; chandler , carlstrom & scoville 1995 ) . in particular , both carr et al . ( 1993 ) and chandler et al . ( 1995 ) conclude that models based on an accretion disk surrounding the central object reproduce the co observations of ysos . furthermore , although there are some cases in which co data can be modelled in terms of a stellar wind , in general the disk model encounters fewer difficulties ( see chandler et al . the observations strongly suggest that the co emission originates close to the central star : bik & thi ( 2004 ) and blum et al . ( 2004 ) have recently studied co emission from a range of massive ysos and derived disk radii in the range 0.1 5 au . thus it seems probable that ysos harbour accretion disks containing significant amounts of hot gas fairly close to the central object and it is natural to consider whether the h i emission is associated with such a disk or its environment . hamann & simon ( 1986 ) suggested that powerful recombination line emission of the sort discussed above originates in an outflow associated not with a stellar wind but with mass - loss from a disk . their work focused on mwc 349 , an object whose evolutionary status is unclear but which possesses spectroscopic signatures ( including h i and co emission features ) broadly similar to the embedded massive ysos ( see hamann & simon 1986 ; kraus et al . hollenbach et al . ( 1994 ) developed a model for outflow from young massive stars by photoevaporation of the outer regions of their disks . they were motivated by the need to explain the high frequency of occurrence of ultracompact h ii regions photoevaporation can provide a source of mass input for h ii regions to balance the loss by pressure - driven expansion . although the rate of mass - loss in photoevaporation models is large , it occurs at large radii ( typically @xmath11 au ) and the characteristic flow velocities are low ( comparable to the typical sound speed of 10 50 km s@xmath6 , hollenbach et al . 1994 ) . thus , such a flow does not provide a promising origin for relatively high velocity features , such as the ir h i broad line wings ( bunn et al . 1995 ) . recently , drew , proga & stone ( 1998 ) proposed that the intense radiation field produced by a massive yso may drive mass - loss from the surface of the inner parts of its disk . they performed a hydrodynamical simulation which showed that , in addition to driving a normal hot star wind component , radiation pressure from a massive yso could propel a dense equatorial flow from a surrounding disk with terminal velocity in the same regime as the observed h i linewidths . these results led them to speculate that such a model accounts for the ir h i lines . however , they did not perform the radiative transfer calculations required to confirm this conjecture . to investigate the possibility that massive ysos harbour disk winds which give rise to the observed ir h i line emission , this paper presents the results of new two - dimensional radiative transfer calculations which account for both the complex geometry of a yso disk wind ( starting from , but not limited to , the drew et al . 1998 model ) and the detailed atomic physics of h i line formation . in section 2 , the yso model adopted in this investigation and its parameters are discussed . the radiative transfer calculations have been performed using the monte carlo ( mc ) code described by long & knigge ( 2002 ) after incorporating a sophisticated treatment of h i line formation using the approach described by lucy ( 2002,2003 ) ; the method of calculation and code used are discussed in section 3 . results are presented in section 4 and our conclusions discussed in section 5 . for the radiative transfer calculations ( section 4 ) a simply parameterised model for the yso and its disk is adopted . in choosing the parameters for this model , our starting point has been to construct a reasonable representation of the disk wind obtained in the hydrodynamical simulations presented by drew et al . ( 1998 ) . therefore , before discussing our model in detail , a brief description of this hydrodynamical model is given . following the methods of proga et al . ( 1998 ) , drew et al . ( 1998 ) performed a hydrodynamical simulation of a radiatively driven wind from a disk around a massive yso . they considered an early - b type star ( mass @xmath12 m@xmath3 , luminosity @xmath13 l@xmath3 , radius @xmath14 r@xmath3 ) surrounded by an accretion disk extending from the stellar surface to an outer radius @xmath15 @xmath16 , the outer boundary of their computational domain . noting that , for reasonable accretion rates ( @xmath17 m@xmath3 yr@xmath6 ) , the luminosity of the central star far exceeds the accretion luminosity of the disk , they accounted for reprocessing of stellar light by the disk and then computed the radiation force using the castor , abbott & klein ( 1975 ) parameterisation of the force due to spectral lines . their simulation exhibited a steady state solution with two fairly distinct outflow components : a fast ( @xmath18 km s@xmath6 ) polar wind from the central star , rather similar to a normal hot star wind ; and a slower ( @xmath19 km s@xmath6 ) , denser equatorial outflow from the disk at colatitude @xmath20 with mass - loss rate @xmath21 m@xmath3 yr@xmath6 . between these two outflow components they found a complex transitional zone in which the stellar wind streamlines were compressed due to non - radial radiation force components and the presence of the disk wind . drew et al . ( 1998 ) suggested that the high density and moderate velocity of the equatorial disk - wind component make it a promising site for the formation of the h i lines observed in massive yso spectra our goal is to present radiative transfer calculations to examine this hypothesis . figure 1 illustrates the geometrical construction used to describe the wind , parameters for which are discussed in the next sub - section . in order to capture the essence of the drew et al . ( 1998 ) hydrodynamical disk wind model ( described above ) , the structure consists of the following four basic elements : 1 . a central star . the star is assumed to be spherical with radius @xmath16 . an accretion disk . the disk lies in the @xmath22-plane of the adopted coordinate system and is assumed to extend from the surface of the star to an outer radius @xmath23 . for simplicity , it is assumed to be geometrically thin and flat . 3 . a disk wind . the wind , which is launched from the disk , is described following long & knigge ( 2003 ) and knigge , woods & drew ( 1995 ) : namely the streamlines of the disk wind are assumed to converge at a point which lies on the @xmath24-axis a distance @xmath25 below the coordinate - origin . thus the angular boundaries of the disk wind are determined by the choice of the inner and outer disk radii and the distance @xmath25 ; namely @xmath26 and @xmath27(see figure 1 ) . motivated by the drew et al . ( 1998 ) simulation , the flow is assumed to be stationary and smooth . 4 . a spherical stellar wind component . this is assumed to occupy the entire polar region above the disk wind and to have radial streamlines . the boundary between the stellar wind and disk wind is not treated in detail ; it is assumed to be perfectly sharp thus the transition zone between the disk and stellar winds which exists in the drew et al . ( 1998 ) simulation is neglected here . table 1 gives a list of the wind parameters adopted for the reference model ( hereafter , model a ) which are discussed individually below . of our models , model a and the closely related model b ( see section 4.3 ) are those most closely matching the drew et al . ( 1998 ) simulation . the parameters of the central star are those of a b1-b2 main sequence star . the effective temperature was computed from the luminosity using the stefan - boltzmann equation . the star is assumed to emit radiation as a black body . in our reference model , @xmath15 @xmath16 is adopted the same computational domain considered by drew et al . realistically , the disks around massive ysos are likely to extend to significantly larger radii . therefore , we will also consider disks extending to @xmath28 @xmath16 in section 4 . to consider even larger radii becomes impractical since increasing the physical size of the calculation limits the spatial resolution available in the inner regions computing feasibility prevents us from having arbitrarily many grid cells . it is noted that , at least for the treatment of disk radiation adopted in our reference model , considering an even larger outer radius will have negligible effect the disk temperature at such large radii will be too low for its thermal radiation to contribute significantly to the ir spectral range under consideration . given that it lies considerably beyond the scope of this investigation to study radiative transfer within the accretion disk in detail , approximations must be made regarding the treatment of the disk and its emission . in the calculations presented in section 4 , results using two different approaches are considered . it is known from observations ( henning , pfau & altenhoff 1990 ) that there are significant differences in the ir seds of massive ysos some have large near - ir excesses while other have little or no excess . therefore , we have chosen our two models for disk emission to approximately bracket the range of plausible ir seds for massive ysos . + + in some cases , referred to as a `` reflecting '' disk , it is assumed that the disk effective temperature is determined as a function of radius by the accretion rate ( exactly as discussed by long & knigge 2002 ) and that annuli of the disk emit as black bodies at the local temperature . during the mc simulations , it is assumed that any radiation which strikes the disk from above is reflected by a hot disk atmosphere rather than absorbed and reprocessed . since the computational domain is symmetric about the disk plane , this is closely equivalent to the assumption that the disk is completely optically thin ( transparent ) . in model a , the disk is assumed to be `` reflecting '' . alternatively , as a simple `` reprocessing '' disk model , it is assumed that stellar radiation falling on the disk is absorbed and the energy re - radiated . in this case , the disk is divided into concentric annuli and the energy falling on each annulus from the star computed following the discussion of proga et al . ( 1999 ) . by assuming that all this energy is re - radiated locally , and that the annuli radiate as black bodies , the disk temperatures are then obtained . the accretion luminosity is still included in this model but is small relative to the reprocessed luminosity . at the start of the mc simulation , the disk annuli emit black - body radiation according to their effective temperatures thereby accounting for the luminosity due to reprocessing of stellar light and during the subsequent propagation of mc quanta , any radiation which strikes the disk is simply absorbed . computed disk temperature for both reflecting and reprocessing disks are shown , as a function of radius , in figure 2 . spectral energy distributions ( sed ) for radiant energy emitted from models with both reflecting and reprocessing disks are shown in figure 3 . comparing the seds for reflecting and reprocessing disks with @xmath15 @xmath16 ( left - hand panels in figure 3 ) shows that the treatment of the disk makes a significant difference to the continuum brightness in the ir region of interest . if the disk only reflects ( or transmits ) light which strikes it , the @xmath5 2 4 @xmath10 m continuum is a combination of stellar radiation and accretion luminosity released in the disk . ( figure 3 , upper - left panel ) . however , if the disk thermally reprocesses all the stellar light which strikes it , the reprocessed light is dominant ( lower - left panel ) . increasing the outer radius of the disk to 100 @xmath16 increases its brightness . with a reflecting disk , this effect is fairly small below about 4 @xmath10 m ( upper - right panel ) because the temperatures of the outer parts of the disk are very low . however , since the reprocessing disk has higher temperatures ( figure 2 ) , changing its outer radius has a greater effect in the waveband of interest ( figure 3 , lower - right panel ) . we note that since the treatment of the radiation force is grey in the drew et al . ( 1998 ) simulation and that in both our approaches to the disk sed all the radiative flux which falls on the disk is re - emitted locally , both our disk sed models are equally valid in the context of the hydrodynamical calculations . the geometry of the disk wind is specified by the distance between the origin and the focus point , @xmath25 ( see figure 1 ) . here , @xmath29 @xmath16 is adopted leading to @xmath30 and @xmath31 ( see above ) . the dense disk wind component present in the hydrodynamic model of drew et al . ( 1998 ) occupied a rather wider angular range ( @xmath32 ) . however , proga et al . ( 1999 ) showed that improved treatment of the radiative line force leads to significantly more swept back equatorial disk winds . in particular , for a model in which the luminosity of the central object substantially exceeds that of the disk ( their model e ) they found a disk wind with opening angle 8 . thus the @xmath25-value adopted here is chosen to reflect the narrow disk winds obtained by proga et al . ( 1999 ) . the density and velocity structure of the disk wind closely follows that adopted by long & knigge ( 2002 ) . following equation ( 7 ) of long & knigge ( 2002 ) , it is assumed that the mass loss rate per unit area from the disk can be described by @xmath33 on the assumption that the mass - loading is proportional to the local luminous flux , @xmath34 is adopted here . the total mass - loss rate in the disk wind is fixed at the mass - loss rate obtained by drew et al . ( 1998 ) : @xmath35 m@xmath3 yr@xmath6 . note that the improved treatment of the line force ( i.e. proga et al . 1999 ) did not significantly change the total mass - loss rates from those of the earlier treatment . the poloidal velocity on a streamline in the disk wind is described using equation ( 8) of long & knigge ( 2002 ) , namely @xmath36 where @xmath37 is the sound speed on the disk surface at the base of the streamline , @xmath38 is the terminal velocity of the streamline , @xmath39 is the poloidal distance , @xmath40 is a velocity - law scale length for the disk wind and @xmath41 is an exponent which determines the acceleration of the wind . for simplicity , @xmath42 is adopted where @xmath43 is the escape velocity on the disk surface at the base of the streamline . figure 4 shows @xmath38 as a function of @xmath44 these values compare favourably with the terminal velocities found by drew et al . ( 1998 , their figure 3 ) . to describe the acceleration of the wind , a velocity scale length for the disk - wind , @xmath45 and exponent @xmath46 are used . the density of the stellar wind component is chosen such that it would have mass - loss rate @xmath47 m@xmath3 yr@xmath6 if it were spherically symmetric . following e.g. lucy & abbott ( 1993 ) , the velocity in this component is assumed to be radial and given by @xmath48 based on the drew et al . ( 1998 ) simulations , @xmath49 km s@xmath6 is adopted . in addition , @xmath50 is assumed and @xmath51 a standard value in modelling hot star winds is chosen . llc parameter & value & varied + + mass , @xmath52 & 10 m@xmath3 + radius , @xmath16 & 5.5 r@xmath3 + luminosity , @xmath53 & 8500 l@xmath3 + temperature , @xmath54 & @xmath55 k + + inner radius , @xmath56 & @xmath16 & + outer radius , @xmath23 & 10 @xmath16 & + accretion rate , @xmath57 & @xmath9 m@xmath3 yr@xmath6 + + focus point , @xmath25 & 0.14 @xmath16 & + mass - loss rate , @xmath58 & @xmath59 m@xmath3 yr@xmath6 & + mass - loading exponent , @xmath0 & 1 + terminal velocity , @xmath38 & @xmath43 + acceleration length , @xmath40 & @xmath16 + acceleration exponent , @xmath41 & 1.5 + + mass - loss rate , @xmath60 & @xmath61 m@xmath3 yr@xmath6 + terminal velocity , @xmath38 & 2000 km s@xmath6 + launch velocity , @xmath62 & @xmath63 + acceleration exponent , @xmath64 & 1 + in section 4 , the dependence of the results on several of the important model parameters listed in table 1 will be investigated . the calculation is , of course , particularly sensitive to the choice of density in the wind . the density in turn is sensitive to many of the model parameters most fundamentally , the mass - loss rates which determine the mass - loading of the stellar and disk winds . however , the velocity - law parameters ( e.g. @xmath38 , @xmath40 and @xmath41 ) also play a significant role , as does the geometry of the disk wind ( determined by @xmath25 ) . in section 4 , models with density higher than that in model a will be considered : these are created by increasing the mass - loss rate only physically this is the parameter which leads to a simple increase in density everywhere in the wind but it should be noted that there is a degree of degeneracy between this and the other parameters mentioned above . the second departure from the reference model ( model a ) that will be discussed in section 4 is an increase in the outer disk radius , @xmath23 . the relevance of this parameter to the determination of the disk sed has already been discussed ( see section 2.3.2 and figure 3 ) . finally , a model in which an inner hole is introduced to the accretion disk will be considered . in addition to increasing @xmath56 for this model , the @xmath25-value is changed in order to preserve the same opening angles ( @xmath65 and @xmath66 ) for the disk wind . departures of the other parameters from their reference values will not be considered here . for this investigation we wish to restrict ourselves to central objects with parameters suitable for early - type near - main - sequence stars thus we do not consider models with large values of @xmath67 . an increase in the stellar radius , @xmath67 would mimic many of the effects associated with the introduction of an inner hole to the disk since both push the wind out to regions where the rotational velocity is lower . thus , our model with an inner hole can be regarded as a proxy for some of the most important consequences of an enlarged central star . an increase in the stellar luminosity would clearly affect the sed , but could also change the mass - loss rate and possibly the wind geometry further hydrodynamical calculations would be required to fully investigate results for a range of luminosities . the mass - accretion rate ( @xmath57 ) is relevant only to calculations with reflecting disks in such cases , @xmath57 plays an important role in determining the disk temperature and ir sed . for reprocessing disks , these are controlled by the stellar radiation field and @xmath57 plays only a minor role . it is very difficult to reliably obtain @xmath57 by observation or theory and so we restrict ourselves to considering only one value , that used by drew et al . ( 1998 ) . for simplicity , it is assumed that both the disk wind and stellar wind consist entirely of hydrogen . an atomic model with energy levels for principle quantum number @xmath68 to 20 and the continuum state is used in the calculations . bound - bound oscillator strengths are taken from menzel & pekeris ( 1935 ) . bound - bound collision rates are derived from the oscillator strengths using the van regemorter ( 1962 ) formula . photoionization rates are taken from topbase ( cunto et al . 1993 ) and collisional ionization rates are computed using equation 5 - 79 from mihalas ( 1978 ) . in addition to these processes associated with atomic hydrogen , free - free absorption and emission by h ii ions and thompson scattering by free - electrons are included in the calculations . all other processes are neglected . in the disk wind model , the ir h i line formation is expected to be driven by recombination of ionised hydrogen gas fairly close to the central star . the observed line ratios ( e.g. bunn el al . 1995 ) indicate , however , that the line emission is not optically thin making it necessary to perform detailed non - lte radiative transfer calculations to reliably model the line formation process . the radiative transfer calculations discussed in the next section were performed using a modified version of the mc code written by long & knigge ( 2002 ) . since the code has been described in detail elsewhere , only a brief overview and discussion of the important modifications made for this investigation are presented here ( see long & knigge 2002 for further details of the code ) . to synthesise spectra , the code performs a sequence of mc radiative transfer calculations . the code uses indivisible energy packets as the elementary mc quanta , assumes radiative , statistical and thermal equilibrium in the wind , and utilises a sobolev treatment of bound - bound transitions . the code presented by long & knigge ( 2002 ) adopted a two - level approximation in the treatment of line scattering and used approximate excitation and ionization formulae for the computation of level emissivities . these approximations made the code efficient and thus able to handle a large set of atomic data for many chemical elements . however , the two - level approximation also meant that it was not suitable for modelling lines formed by non - resonance scattering or recombination . since the ir h i lines form primarily by recombination , the code has been substantially modified to incorporate recently developed mc radiative transfer techniques which allow the formation of such lines to be modelled . in particular , _ macro atoms _ , as devised and tested by lucy ( 2002,2003 ) , are used . this approach allows the radiative equilibrium constraint to be rigorously enforced at all times without approximation in the treatment of interactions between radiation and matter . initially , several mc simulations are performed to compute the temperature and degrees of excitation and ionization throughout the wind . for these `` ionization cycles '' the mc quanta are launched with frequencies spanning a wide enough range to simulate all the important radiative energy inputs to the system . during each of these mc simulations , estimators for the radiative heating rate ( e.g. due to photoionization ) are recorded in each grid cell . at the end of the simulation , these rates are balanced against cooling rates to determine the local electron temperature which is then adopted in the next simulation . this process is repeated until essentially all grid cells pass a chosen convergence threshold in electron temperature . for this work , the code has been modified to also record high precision mc estimators for the individual radiative rates ( photoionization and bound - bound excitation ) following lucy ( 2003 ) . at the end of each ionization cycle , these are used , together with the various collision rates and radiative decay rates , to compute level populations and the ionization fraction in each grid cell . these populations are used in the next iteration , alleviating the need to use the approximate analytic formulae for ionization and excitation employed by long & knigge ( 2002 ) . no convergence criterion is enforced for the level populations at present . lucy ( 2002 ) has shown that , provided the populations of the lowest levels are estimated sensibly , the indivisible - energy - packet mc method produces highly accurate emissivities even if the populations for the emitting upper levels are relatively poorly known . this insensitivity motivates our use of the mc method since it substantially reduces our reliance on detailed calculation of excited level populations for an adequate non - lte investigation of recombination line formation . we note that our convergence criterion on the electron temperature implies a corresponding convergence in the energy flow between the radiation field and the thermal energy pool . since this energy flow is a summation of all the radiative heating processes in the hydrogen model , its behaviour implies good convergence amongst all the processes which are energetically important . after the `` ionization cycles '' , a set of additional mc simulations are performed to compute the spectrum ( the `` spectral cycles '' ) . during these cycles the mc quanta are created in only a fixed frequency interval while the thermal , ionization and excitation states of each grid cell are fixed to the values previously determined as described above . by tracking the quanta , we compute the spectrum as seen by observers at different inclination angles in the frequency range of interest . @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 .the observed range of properties of massive yso h i ir lines in the sample discussed by bunn et al . note that in several of the objects , br@xmath1 and/or pf@xmath1 are not observed and so the range of ew for these lines is biased towards the brighter sources compared with br@xmath0 . [ cols="<,^,^,^ " , ] lcccccccccccc + & & & & & & + & @xmath75ew & fwhm & @xmath75ew & fwhm & @xmath75ew & fwhm & @xmath75ew & fwhm & @xmath75ew & fwhm & @xmath75ew & fwhm + line & ( ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) + + br@xmath0 & 44 & 300 & 56 & 425 & 70 & 550 & 78 & 610 & 77 & 610 & 68 & 550 + br@xmath1 & 4.0 & 290 & 4.9 & 420 & 5.6 & 520 & 5.8 & 520 & 5.8 & 550 & 5.4 & 550 + pf@xmath1 & 7.2 & 340 & 9.3 & 450 & 12 & 560 & 13 & 560 & 14 & 560 & 13 & 560 + + br@xmath0 & 3.1 & 430 & 4.2 & 610 & 6.1 & 730 & 8.5 & 790 & 14 & 790 & 19 & 790 + br@xmath1@xmath76 & 0.17 & & 0.36 & & 0.41 & & 0.57 & & 0.83 & & 1.2 & + pf@xmath1@xmath76 & 0.76 & & 0.80 & & 1.3 & & 1.7 & & 2.8 & & 3.8 & + + br@xmath0 & 27 & 300 & 38 & 425 & 56 & 550 & 74 & 550 & 100 & 670 & 120 & 670 + br@xmath1 & 3.4 & 290 & 4.8 & 420 & 7.1 & 520 & 9.3 & 550 & 11 & 580 & 12 & 620 + pf@xmath1 & 6.6 & 340 & 9.6 & 450 & 13 & 560 & 17 & 560 & 21 & 560 & 22 & 670 + + br@xmath0 & 24 & 300 & 32 & 425 & 43 & 550 & 53 & 550 & 62 & 610 & 63 & 550 + br@xmath1 & 3.4 & 360 & 3.9 & 420 & 4.5 & 550 & 4.7 & 590 & 4.6 & 590 & 4.2 & 590 + pf@xmath1 & 4.5 & 340 & 5.7 & 450 & 7.6 & 500 & 9.5 & 560 & 11 & 560 & 12 & 560 + + br@xmath0 & 4.6 & 300 & 6.6 & 430 & 10 & 550 & 16 & 610 & 29 & 670 & 49 & 670 + br@xmath1 & 0.80 & 390 & 1.1 & 550 & 1.6 & 650 & 2.3 & 750 & 3.8 & 720 & 5.4 & 720 + pf@xmath1 & 1.2 & 340 & 1.6 & 620 & 2.3 & 730 & 3.1 & 730 & 5.1 & 730 & 8.1 & 730 + + br@xmath0 & 22 & 300 & 28 & 300 & 36 & 425 & 43 & 425 & 51 & 425 & 52 & 490 + br@xmath1 & 2.2 & 260 & 2.4 & 360 & 2.6 & 460 & 2.7 & 490 & 2.7 & 490 & 2.5 & 490 + pf@xmath1 & 2.8 & 220 & 3.8 & 340 & 5.0 & 450 & 6.2 & 450 & 7.5 & 450 & 7.9 & 450 + @xmath76 for model b , the br@xmath1 and pf@xmath1 lines are very weak and only appear marginally above the mc noise therefore , the entries for these lines give only upper limits on their ew . lcccccc + & @xmath77 & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 + + br@xmath0 flux & 5.1 & 5.3 & 4.9 & 4.3 & 3.3 & 2.6 + br@xmath0/br@xmath1 & 1.72 & 1.70 & 1.66 & 1.59 & 1.48 & 1.36 + br@xmath0/pf@xmath1 & 4.86 & 4.74 & 4.67 & 4.48 & 4.13 & 3.81 + + br@xmath0 flux & 4.8 & 5.0 & 4.5 & 3.9 & 3.1 & 2.6 + br@xmath0/br@xmath1@xmath76 & 1.99 & 1.37 & 1.73 & 1.72 & 1.85 & 1.72 + br@xmath0/pf@xmath1@xmath76 & 3.07 & 4.09 & 3.34 & 3.84 & 3.65 & 3.77 + + br@xmath0 flux & 43 & 47 & 43 & 37 & 27 & 20 + br@xmath0/br@xmath1 & 0.91 & 0.93 & 0.93 & 0.95 & 1.05 & 1.19 + br@xmath0/pf@xmath1 & 3.05 & 3.01 & 3.16 & 3.27 & 3.69 & 4.09 + + br@xmath0 flux & 4.0 & 4.1 & 4.0 & 3.6 & 3.0 & 2.5 + br@xmath0/br@xmath1 & 1.54 & 1.59 & 1.60 & 1.59 & 1.56 & 1.52 + br@xmath0/pf@xmath1 & 4.50 & 4.61 & 4.66 & 4.49 & 4.33 & 4.18 + + br@xmath0 flux & 25 & 26 & 25 & 23 & 19 & 16 + br@xmath0/br@xmath1 & 1.32 & 1.40 & 1.49 & 1.51 & 1.56 & 1.64 + br@xmath0/pf@xmath1 & 3.07 & 3.44 & 3.66 & 4.11 & 4.57 & 4.90 + + br@xmath0 flux & 19 & 19 & 19 & 17 & 16 & 14 + br@xmath0/br@xmath1 & 2.37 & 2.37 & 2.29 & 2.18 & 2.06 & 2.03 + br@xmath0/pf@xmath1 & 6.94 & 6.20 & 5.82 & 5.60 & 5.17 & 4.93 + @xmath76 for model b , the br@xmath1 and pf@xmath1 lines are very weak and only appear marginally above the mc noise - therefore , the entries for these lines give only lower limits on the appropriate flux ratios . in all the calculations described below , the model structure and radiation field properties are discretized onto a two - dimension ( @xmath83-@xmath24 ) grid consisting of 3000 cells . during each mc calculation , the input radiative energy ( which arises from the star and accretion disk ) is divided into @xmath84 indivisible quanta . we have explicitly verified that our approach is adequate for the calculation of the spectrum by comparing results for models with identical input parameters using smaller and larger numbers of grid cells . these calculations differed by at most a few per cent in the computed h i line strengths from the standard ( 3000 cell ) case . this implies that the number of packets , number of grid cells and the quality of the mc estimators are all sufficient for calculations to this degree of accuracy . early observations of ir h i emission lines in massive ysos ( e.g. simon et al . 1981 ) had insufficient signal - to - noise ratio to provide detailed information on the line shapes , thus subsequent theoretical work ( e.g. hflich & wehrse 1987 ) was concerned primarily with modelling the observed line fluxes and flux ratios . more recently , higher quality data ( e.g. drew et al . 1993 , bunn et al . 1995 ) has allowed the line shapes to be studied in greater detail in several massive ysos . in common with earlier work ( e.g. simon et al . 1981,1983 ; drew et al . 1993 ) , bunn et al . ( 1995 ) observed a moderately strong , pure - emission line of br@xmath0 in all of their targets . in most cases , they also reported weaker emission in br@xmath1 and pf@xmath1 . their observed line profiles are single peaked and not shifted from the expected local rest velocity by more than a few km s@xmath6 . many of the profiles show line wings which extend to several hundred km s@xmath6 from line centre and in some cases these wings are clearly asymmetric . by considering wavelength - dependent line ratios ( following drew et al . 1993 ) , bunn et al . ( 1995 ) found evidence , in at least some cases , of hybrid profiles consisting of a narrow line core ( which may have a nebular origin ) and broad wings in which the line ratio is consistent with formation in an accelerating outflow . high resolution observations of several massive ysos have very recently been presented by blum et al . interestingly , in their sample , at least one object ( ngc 3576 ) has double - peaked , narrow line profiles . however , it is difficult to study the broad line wings in the blum et al . ( 2004 ) data since the spectral coverage is too narrow to establish the underlying continuum with certainty . in the following sub - sections , results of our models will be compared primarily with the observations presented by bunn et al . ( 1995 ) since their spectra provide good constraints on both line strengths and shapes for a significant sample of objects . unfortunately , owing to the great difficulty in establishing the extinction along the line of sight to massive ysos , interpreting line fluxes and flux ratios for lines at different wavelengths is very difficult ( it is known from br@xmath0/pf@xmath1 flux ratios that the h i lines are not well described by either the lte or the baker menzel case b recombination assumptions thus the br@xmath0/br@xmath1 ratio can not be trusted as an indicator of the reddening [ hflich & wehrse 1987 ] ) . therefore , in our discussion , we prefer to focus on ews , ew ratios ( ewrs ) and line shapes since these are not affected by reddening . for quantitative comparison , table 2 gives a summary of the range of observed ew , fwhm and half - width - at - zero - intensity ( hwzi ) from bunn et al . the table also gives the observed range of ewrs for br@xmath0/br@xmath1 and br@xmath0/pf@xmath1/br@xmath1 ] and ewr[br@xmath0/pf@xmath1 ] respectively . ] . since our model parameters are not fine - tuned to one particular object , these values should be regarded only as indicative of the appropriate regime in which acceptable model predictions should lie . spectra have been computed in the 2 5 @xmath10 m range using the reference wind model ( model a ) which was described in section 2 . in this wavelength range , the model predicts moderately strong emission in the first three lines of the brackett series and weaker emission in lines of the pfund series ( pf@xmath86 and pf@xmath1 ) . this is in qualitative agreement with the observed ir properties of massive ysos ( e.g. simon et al . 1981,1983 ; drew et al . 1993 ; bunn et al . 1995 ) . figure 5 shows the predicted model a line profiles of br@xmath0 , br@xmath1 and pf@xmath1 . spectra are shown for a grid of viewing angles ranging from 30 to 85 ( measured relative to the polar axis ) . for quantitative comparison with observations , the computed ew and fwhm of these lines are tabulated in table 3 . for completeness , computed line fluxes and flux ratios are given in table 4 . for all three lines , the ew is largest for viewing angles ( @xmath87 ) of 70 80 and becomes noticeably smaller at lower @xmath87 . the computed br@xmath0 ews are consistent with the high end of the range of measured ew ( table 2 ) obtained by bunn et al . similar agreement between observations and the model is also found for pf@xmath1 making the modelled ewr[br@xmath0/pf@xmath1 ] similar to the observed ratio . for br@xmath1 , the model a computed ews lie in the mid - range of the observed values ( see table 2 ) . the relative weakness of the computed br@xmath1 line means that the model ewr[br@xmath0/br@xmath1 ] is systematically too high in the observations the ewr for these lines ranges from 1.1 to 4.4 but the computed model a ratio is somewhat higher , ranging from 11 to 13 for the viewing angles considered . it is probable that part of this disagreement is the result of opacities being too low in the models but given the good agreement with observations for ewr[br@xmath0/pf@xmath1 ] , it seems unlikely that a simple underestimate of the opacity in br@xmath0 is solely responsible . perhaps more importantly , the ewr[br@xmath0/br@xmath1 ] will be affected by inaccuracies in the model continuum shape since br@xmath0 and pf@xmath1 lie at similar wavelengths , the choice of continuum shape will have little effect on the computed ratio of their ews but because br@xmath1 lies at a significantly shorter wavelength , the ewr[br@xmath0/br@xmath1 ] is much more sensitive . an alternative treatment of the continuum that invoking a reprocessing disk is considered in section 4.2 . it is also noted that , compared with br@xmath0 and pf@xmath1 , observations of br@xmath1 for heavily embedded targets are more likely to be contaminated by emission from other sources ( e.g. larger scale nebular emission ) owing to the wavelength dependence of the source extinction . the computed line profiles are complex , usually exhibiting double - peaks which become more pronounced at larger @xmath87 . this is in contrast to the observations where only single - peaked profiles are usually observed ( see section 4.1 ) . also , the computed profiles are too broad ( fwhm ranging from 300 to 600 km s@xmath6 as a function of @xmath87 ) . part of this discrepancy between model and observations may be explained by contamination of the observed profiles by narrow nebular emission there is some evidence for hybrid line profiles with narrow cores ( see section 4.1 ) however , the half - width - zero - intensity ( hwzi ) measurements also suggest that the observed line profiles do not extend more than 200 to 300 km s@xmath6 from line - centre in most cases . the large line widths and double - peaked profiles arise due to the rotational velocities in the region of line formation . figure 6 identifies the region of line formation for the br@xmath0 , br@xmath1 and pf@xmath1 lines in model a ; specifically , the plots show the amount of energy that escapes in each spectral line from each grid cell in the model . note that mc noise is responsible for the graininess of the figures . it can be seen that in all three cases , the region of line formation is close to the disk surface , within the first few velocity law scale - lengths along the wind streamlines . br@xmath0 tends to form slightly further out than either br@xmath1 or pf@xmath1 this is to be expected owing to the higher opacity in the br@xmath0 line . to obtain models which produce narrower lines , either more low - velocity gas must be added , or high - velocity gas must be removed . the simplest way in which to introduce low - velocity gas is by increasing the outer disk radius ( @xmath23 ) models with wider disks are presented in section 4.3 . to remove high - velocity gas requires a reduction in the wind density at small values of @xmath83 . given that the inner disk is brighter than the outer disk , it seems unlikely that a radiatively driven flow will not have greatest mass - loading at small radii . therefore , the most plausible way in which to eliminate high - velocity gas is not to change the prescription of mass - loading but rather to consider models in which either the stellar radius is larger than that adopted in the reference model or the disk inner radius is several times greater than the adopted stellar radius . an example of the latter , a model with an empty inner cavity in the disk , is discussed in section 4.5 . to assess the relative importance of the disk wind and stellar wind component , results from a calculation considering only the disk wind component were compared with those which include the stellar wind . these two calculations give virtually identical results this is not unexpected since it is already known that a spherical wind would require a mass - loss rate at least an order of magnitude greater than that in the stellar wind component here ( simon et al . 1981 , see section 1 ) . in view of this , in all the calculations discussed below , the stellar wind component is dropped for computational efficiency . in model a ( discussed above ) it was assumed that the photons striking the accretion disk were reflected rather than absorbed . to investigate an alternative simple hypothesis , namely that the disk absorbs and thermally re - emits the radiation which strikes it ( see section 2.2.2 ) , a second model ( model b ) is now considered . this is the extreme , optically thick disk case . all the system parameters ( table 1 ) for model b are identical to those of model a , saving that the stellar wind component is omitted ( see above ) . the simple treatment of reprocessing adopted ( see section 2.2.2 ) means that the disk is much brighter at ir wavelengths in model b than model a. thus , although the wind parameters in model b are unchanged , the ews of all the emission lines are much smaller ( the ews are given in table 3 ) . at moderate to high @xmath87 , the model b ews for br@xmath0 and pf@xmath1 are comparable to the observed ews in the weaker sources reported by bunn et al . however , the computed br@xmath1 line is very weak . to increase the line ews in the presence of a reprocessing disk the wind density ( and hence emission measure ) needs to be increased . therefore , a further model ( model c ) has been constructed it is identical to model b saving that the disk - wind mass - loss rate has been increased by a factor of four . a higher mass - loss rate may be feasible even under the assumption of driving purely by radiation since the absolute mass - loss rates provided by the hydrodynamical model of drew et al . ( 1998 ) may be uncertain by a factor of a few owing to the parameterisation of the line - driving force . ews and fwhm for model c are given in table 3 and the line profiles computed from this model are shown in figure 7 . the model c ews are slightly greater than even the largest observed values from the bunn et al . ( 1995 ) data , suggesting that the factor of four mass - loss rate increase between models b and c comfortably brackets the range of values that are likely to be consistent with the data for particular objects . the ewr[br@xmath0/pf@xmath1 ] is smaller for model c than model a , lying close to the mid - range of observed values . the ewr[br@xmath0/br@xmath1 ] is also smaller in model c it is still greater than the observed ratio in any of the objects observed by bunn et al . ( 1995 ) but the discrepancy is much less . this change in the line ratios is mostly due to the increased br@xmath0 opacity resulting from the higher wind density . this result suggests that by further increasing the mass - loss rate an ewr[br@xmath0/br@xmath1 ] close to that observed could be achieved . however , it is noted that a further increase in the wind density will lead to br@xmath0 ews which are uncomfortably large . the model c line profiles are generally less double - peaked and more square - topped than the model a profiles . their fwhm are very similar to those in model a and the line wings still extend to several hundred km s@xmath6 at the base . in general , model c appears to suggest that the combination of a reprocessing disk and a higher disk - wind mass - loss rate is in closer agreement with the observations than the reference model ( model a ) . but further hydrodynamical modelling is required to examine whether so high a mass - loss rate is feasible . the outer disk radius paralleling that of drew et al . ( 1998 ) and adopted in models a c is , arguably , rather small . in principle , line formation in a wind rising from the outer parts of a larger disk could eliminate the double - peaked line shape obtained with models a c by adding low - velocity gas . in this section two more models ( models d and e ) are considered : these models are identical , respectively , to models a and c saving that the outer disk radius is set to 100 @xmath16 rather than 10 @xmath16 . the computed line profiles for model d are shows in figure 8 and the ews for models d and e are given in table 3 . the differences between the profile shapes shown in figures 5 and 8 are generally fairly small . the double - peaked shape is slightly filled in when the larger disk is considered ; however , the assumption that the mass - loading of streamlines is proportional to the luminous flux ( see section 2.2.3 ) means that the majority of the mass - loss still originates from the inner disk where velocities are high . the most important consequence of the larger disk is the increase in the ir continuum level ( see figure 3 and the discussion in section 2.3.2 ) . the brighter continuum means that , when the disk is bigger , the line ews are all smaller . also , the ewr[br@xmath0/br@xmath1 ] is smaller the larger disk radius increases the continuum around 4 @xmath10 m by more than at 2 @xmath10 m ( see figure 3 ) . for model d , the br@xmath0 ews lie comfortably in the mid - range of the observed values and the ewr[br@xmath0/pf@xmath1 ] is similar to that found in model a. the ewr[br@xmath0/br@xmath1 ] is smaller in model d than a ( as required ) but the improvement is less significant than obtained with model c ( see section 4.2 ) . model e predicts smaller ews , but still within the range observed . the ewr[br@xmath0/br@xmath1 ] in model e is closer to that observed than in any of the other models this follows since model e benefits from the improvement in this ratio resulting from both a higher line opacity ( density ) and stronger disk continuum . @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 it is apparent that a failing of models a e in comparison with the data is that the computed line profiles are too broad . one of two possible solutions to this problem is to invoke an inner hole in the accretion disk , thereby removing the highest velocity material from the region of line formation . optically thin inner cavities in accretion disks around young luminous stars have previously been suggested in various contexts ( e.g. herbig ae / be stars dullemond et al . 2001 ; tuthill , monnier & danshi 2001 ; natta et al . 2001 ; and see monnier et al . 2005 for a recent discussion ) . interestingly , a disk with an inner hole might also help explain the residual discrepancy between the observed ewr[br@xmath0/br@xmath1 ] and the models discussed above . since the inner parts of the accretion disk are hottest , removing them will push the sed of the disk to the red . this will increase the continuum level at br@xmath0 relative to br@xmath1 and therefore reduce the ewr[br@xmath0/br@xmath1 ] as required . however , a quantitative investigation of this possibility goes beyond the scope of this paper since more sophisticated models for the disk emission than used here would be required . consideration of models in which the disk has a central hole represents a significant departure from the drew et al . ( 1998 ) model and therefore , until further hydrodynamical calculations are performed , it is difficult to place useful constraints on the likely geometry and parameters for mass - loss from such a disk . nevertheless , for illustrative purposes , a model adopting the same wind opening angles ( @xmath65 , @xmath66 ) and mass - loss rate as the reference model ( model a ; section 4.1 ) but with an inner disk radius of @xmath88 @xmath16 and outer disk radius of @xmath89 @xmath16 has been constructed ( model f ) this corresponds to a disk with a circular inner hole of @xmath90 au . it is noted that , conceptually , invoking a wind launched significantly further out than the stellar surface is reminiscent of the photoevaporation model ( e.g. hollenbach et al . 1994 ) however , the distance from the central star at which the mass - loss occurs here is still much smaller than expected from photoevaporation . it is , however , comparable to the smallest disk radii inferred from co observations by bik & thi ( 2004 ) . the computed fwhm and ews for model f are tabulated in table 3 and the profiles are shown in figure 9 . the ews are smaller than those for model a because the density in the wind is lower , a result of the radial dependence of the volume element in a cylindrical polar system . the lower density also leads to lower opacity and hence larger ratios of br@xmath0 to br@xmath1 and pf@xmath1 . however , as anticipated , the model f lines are narrower than in model a , by up to 100 km s@xmath6 in fwhm for moderate inclination angles . the profile widths at base are also smaller in model f the profiles cut - off at @xmath91 km s@xmath92 from line centre . thus holes of the size considered here , or a little larger , could explain why the hwzi measurements of bunn et al . ( 1995 ) are rather smaller than would be expected based on models a e. given that there are no strong constraints on the hole size , geometry or mass - loss rate appropriate for the model here it is not possible to drawn definitive conclusions . however , it seems that models with an inner disk cavity have the potential to resolve the discrepancies between the computed disk wind line profiles and the observations . model f also suggests that the second possibility mentioned in section 4.2 that of an enlarged stellar radius can also provide a possible solution since increasing the stellar radius will affect the line widths in a similar way to introducing an inner gap in the disk . constraints on the size of any inner hole , the stellar radius and the geometry of the disk wind are needed in order to proceed . these could be provided via sophisticated observational techniques such as interferometry ( which has already been applied to herbig ae / be stars by e.g. monnier et al . 2005 ) or spectropolarimetry ( as discussed by vink , harries & drew 2005 ) . as mentioned in section 4 , direct confrontation of theoretical line fluxes with observations is difficult owing to the uncertainty in the extinction along lines - of - sight to massive ysos . but on comparing ews the quantities most readily extracted from observations the radiative transfer calculations presented above lead us to conclude that model disk winds , such as that obtained by drew et al . ( 1998 ) , do predict h i line strengths consistent with observations of massive ysos . this matching is achieved for total mass - loss rates between 0.3 and 1.2 @xmath93 m@xmath3 yr@xmath6 , in contrast with previously calculated spherical models for early b stars which require mass loss rates up to @xmath9 m@xmath3 yr@xmath6 ( simon et al 1983 , nisini et al 1995 ) . there are , however , some difficulties when the models are considered in detail . although the computed ewr[br@xmath0/pf@xmath1 ] agrees well with observations , the ewr[br@xmath0/br@xmath1 ] tends to be overpredicted by the models . we have investigated the effect of a higher density on this ratio by considering models with greater mass - loss rate ( e.g. model c ; recall that the model density is sensitive to several parameters in addition to the mass - loss rate and thus there is a degree of degeneracy among these quantities see section 2.3.5 ) . as expected , increasing the wind density decreases the ewr[br@xmath0/br@xmath1 ] owing to the greater optical depth . however , to fully explain the ewr discrepancy solely by such a modification is difficult since the absolute values of the ews grow rapidly if the density is raised . at the same time , this ewr is very sensitive to the assumed underlying sed , owing to the significant wavelength difference between the br@xmath0 and br@xmath1 lines ( unlike br@xmath0 and pf@xmath1 which lie at similar wavelengths ) . observationally it is clear that there is genuine diversity in the ir seds of massive ysos ( e.g. henning et al . 1990 ) . in our models we have considered only two limiting cases for the treatment of the sed and have not concerned ourselves with trying to fit specific objects . to determine whether the ewr[br@xmath0/br@xmath1 discrepancy is unavoidable , there is now a need to move on to fitting observations of individual well - observed sources with reliable reddening estimates . important differences are also evident in comparisons of the line profiles obtained from our models and those typically observed in massive ysos . to illustrate this , figure 10 compares the observed br@xmath0 profile for gl989 ( taken from bunn et al . 1995 ) with computed profiles from two of the models . gl989 is chosen here since the line profiles for this object have strong extended wings and , in contrast to several of the other objects in the bunn et al . ( 1995 ) sample ( e.g. s106ir ) , have simple profiles suggesting a single dominant region of line formation . the comparison models and viewing angles ( model b with @xmath79 and model e with @xmath78 ) were selected since they predict br@xmath0 ews close to the observed value , thereby allowing a direct comparison of profile shape . the figure shows that the model b br@xmath0 profile at @xmath79 is too broad and , in contrast to the observations , is double - peaked . the model e profile ( @xmath78 ) is in better agreement with the observation the combination of higher optical depth in the model ( owing to the increased mass - loss rate ) and lower viewing angle reduce the profile width and suppress the double - peaks but a significant discrepancy remains . to resolve this difference in line shape between the models and typical observations , one could invoke even smaller inclination angles . although viewing angles are not known for most luminous ysos , this is not an attractive solution since small viewing angles ( close to pole - on ) are statistically disfavoured in an unbiased sample . furthermore , at least one of the objects discussed by bunn et al . ( 1995 ) is known to be viewed almost edge - on ( s106ir ; solf & carsenty 1982 ) . thus we prefer to pursue resolutions to the two styles of line shape discrepancies which do not rely on orientation effects . increasing the physical extent of the disk from which the wind is launched does not readily eliminate the double - peaked line shape from our models ( models d and e ) although some of the wind does occupy regions with lower rotational velocities when the disk is made larger , our assumption that the mass - loading is proportional to the luminous flux of the disk means that most of the mass in the wind is still launched at small radii . the problem of reconciling observed single - peaked line profiles with theoretical predictions of double peaks has arisen elsewhere , particularly in studies of nova - like variables ( see e.g. horne 1997 ) and active galactic nuclei ( agn ) . in the context of agn , murray & chiang ( 1997 ) have argued that substantial radial velocity shear in a disk wind can suppress the formation of double - peaked profiles as required . for this to occur , it is necessary that line optical depths are high such that photon escape is strongly favoured in the poloidal direction where accelerating outflow ensures steep velocity gradients . it is this difference between model c , with enhanced mass loss , and model a that explains the relatively flat - topped profiles in fig . 7 , compared to fig . 5 . hence there is scope to re - organise model parameters in order to exploit this mechanism to suppress double peaks . in particular , a way of raising line opacity without also raising br@xmath0 ews is needed . this will warrant further investigation in future work , focusing on fitting observations of specific objects . some of the discrepancies , in particular the large linewidths predicted by the models , may be explained by departures from the particular disk wind geometry adopted ( which was motivated by the drew et al . 1998 model ) . two interesting possibilities are those of an enlarged radius for the central star , relative to a main - sequence radius , or of an inner gap in the yso disk . these possibilities have rather similar consequences by moving the disk wind out to larger radial distances from the central object , the correspondingly lower rotation velocities reduce the extent of the line wings and make the double - peaks less prominent ( this is illustrated by our model with a cavity in the inner disk , model f ) . also , the sed for a disk with a hole will be redder than that for a complete disk , helping to explain the ewr[br@xmath0/br@xmath1 ] discrepancy . these alternatives might be distinguished by observational techniques that probe the inner disk geometry ( e.g. interferometry and spectropolarimetry ) while the influence of a gap on the sed needs to be investigated fully in subsequent work incorporating more sophisticated treatments of the disk sed than used here . a third possibility is that of disk winds in which the streamlines diverge less at large distances . in our models , the divergence of the streamlines is imposed by our adopted `` split dipole '' geometry . in a geometry with less divergence of the flow , the density would be higher in the outer parts of the wind for a given mass - loss rate and velocity law . the increased emission measure at large radii would help to wash out the double - peaked profiles which form when the emission is dominated by gas in rotation close to the central object . for appropriate viewing angles , line - of - sight optical depths would also be larger for a less divergent outflow . as has already been noted , higher optical depths present a promising solution to part of the ewr[br@xmath0/br@xmath1 ] discrepancy . thus , in the context of studying particular objects in detail , considering a range of different disk wind geometries is likely to prove fruitful in future work . these issues aside , the major success of our radiative transfer simulations is their support of the hypothesis that disk winds can play a significant role in creating the h i lines of massive ysos . in particular , the departure from one - dimensional wind models has alleviated the need to invoke the uncomfortably large mass - loss rates required in earlier spherical models . further investigations of the applicability of the disk wind model to particular objects are now warranted . sas thanks l. lucy for several useful discussions relating to this work and l. mendes for technical support in use of the imperial college astrophysics beowulf cluster . this work was undertaken while sas was a pparc supported pdra at imperial college london ( ppa / g / s/2000/00032 ) . bik a. , thi w. f. , 2004 , a&a , 427 , l13 + blum r. d. , barbosa c. l. , damineli a. , conti p. s. , ridgway s. , + 2004 , apj , 617 , 1167 + bonnell i. a. , vine s. g. , bate m. r. , 2004 , mnras , 349 , 735 + bunn j. c. , hoare m. g. , drew j. e. , 1995 , mnras , 272 , 346 + carr j. s. , 1989 , apj , 324 , 522 + carr j. s. , tokunaga a. t. , najita j. , shu f. h. , glassgold a. e. , + 1993 , apjl , 411 , l37 + castor j. i. , abbott d. c. , klein r. i. , 1975 , apj , 409 , 429 + chandler c. j. , carlstrom j. e. , scoville n. z. , dent w. r. f. , + geballe t. r. , 1993 , apj , 412 , l71 + chandler c. j. , carlstrom j. e. , scoville n. z. , 1995 , apj , 446 , 793 + cunto w. , mendoza c. , ochsenbein f. , zeippen c. j. , 1993 , a&a , + 275 , l5 + drew j. e. , bunn j. c. , hoare m. g. , 1993 , mnras , 265 , 12 + drew j. e. , proga d. , stone j. m. , 1998 , mnras , 296 , l6 + dullemond c. p. , dominik c. , natta a. , 2001 , apj , 560 , 957 + hamann f. , simon m. , 1986 , apj , 311 , 909 + henning th . , pfau w. , altenhoff w. j. , 1990 , a&a , 227 , 542 + hflich p. , wehrse r. , 1987 , a&a , 185 , 107 + hollenbach d. , johnstone d. , lizano s. , shu f. , 1994 , apj , 428 , + 654 + horne k. , 1997 , in wickramasinghe d. t. , bicknell g. v. , ferrario + l. , eds , iau colloquium 163 , asp conference series , vol . + 121 , p.14 + knigge c. , woods j. a. , drew j. e. , 1995 , mnras , 273 , 225 + kraus m. , krgel e. , thum c. , geballe t. r. , 2000 , a&a , 362 , + 158 + long k. s. , knigge c. , 2002 , apj , 579 , 725 + lucy l. b. , abbott d. c. , 1993 , apj , 405 , 738 + lucy l. b. , 2002 , a&a , 384 , 725 + lucy l. b. , 2003 , a&a , 403 , 261 + menzel d. h. , pekeris c. l. , 1935 , mnras , 96 , 77 + mihalas d. , 1978 , stellar atmospheres ( 2nd edition ; san + francisco : freeman ) + monnier j. d. et al . , 2005 , astro - ph/0502252 + murray n. , chiang j. , 1997 , apj , 474 , 91 + natta a. , prusti t. , neri r. , wooden d. , grimm v. p. , mannings + v. , 2001 , a&a , 371 , 186 + proga d. , stone j. m. , drew j. e. , 1998 , mnras , 295 , 595 + proga d. , stone j. m. , drew j. e. , 1999 , mnras , 310 , 476 + simon m. , righini - cohen g. , fischer j. , cassar l. , 1981 , apj , + 251 , 552 + simon m. , felli m. , cassar l. , fischer j. , massi m. , 1983 , apj , + 266 , 623 + solf j. , carsenty u. , 1982 , a&a , 113 , 142 + tuthill p. g. , monnier j. d. , danchi w. c. , 2001 , nature , 409 , + 1012 + van regemorter h. , 1962 , apj , 136 , 906 + vink j. s. , harries t. j. , drew j. e. , 2005 , a&a , 430 , 213 + [ lastpage ]

we develop a first - principles electron - transport simulator based on the lippmann schwinger ( ls ) equation within the framework of the real - space finite - difference scheme . in our fully real - space based ls ( grid ls ) method , the ratio expression technique for the scattering wave functions and the green s function elements of the reference system is employed to avoid numerical collapse . furthermore , we present analytical expressions and/or prominent calculation procedures for the retarded green s function , which are utilized in the grid ls approach . in order to demonstrate the performance of the grid ls method , we simulate the electron - transport properties of the semiconductor / oxide interfaces sandwiched between semi - infinite metal electrodes . the results confirm that the leakage current through the ( 001)si / sio model becomes much larger when the dangling - bond ( db ) state is induced by a defect in the oxygen layer while that through the ( 001)ge / geo model is insensitive to the db state .

1505.06535

electron - transport calculations are important tools to investigate and develop materials for new electronic devices . recently , to obtain more practical knowledge on electron - transport properties of nanoscale structures , long - range and large - scale transport simulations have attracted much interest . however , such simulations are very hard task since huge computational costs growing with a system size are required . therefore , it is important to develop an efficient electron - transport simulator . the lippmann schwinger ( ls ) equation method proposed by lang _ et al . _ @xcite is one of popular methods , which enable us to obtain the scattering wave functions of nanoscale structures sandwiched between electrodes by solving the integral equation of the second - kind fredholm equation . when the reference system consists of only bare left and right electrodes with the empty transition region , scattering wave functions can be efficiently evaluated for a variety of structures of nanoscale junctions set up in the transition region by using the same reference green s function of the bare electrode system , where the computation of the reference green s function has only to be performed once . moreover , for a similar reason , the ls equation is utilized in the implementation of self - consistent calculations for the convergence of electronic states in infinitely open systems @xcite . in the conventional ls equation method , scattering wave functions are expressed in the laue representation , that is , the ls equation is solved by using a 2-dimensional plane - wave expansion in the directions parallel to the electrode surface ( lateral directions ) and a real - space discretization of the coordinate in the direction perpendicular to that ( longitudinal direction ) . in the ls equation method , however , one may frequently encounter a numerical difficulty such that a part of the green s function expressed in a variable - separable form drastically varies due to the appearance of evanescent waves exponentially growing and decaying in the longitudinal direction . to overcome this issue , in the previous study @xcite , we proposed the procedure of the ratio expression for the green s function matrix elements in the laue representation as a remedy for avoiding the numerical collapse . so far , we developed the several simulators to elucidate the electronic properties of nanostructures based on the real - space finite - difference ( rsfd ) approach @xcite , in which the system is divided by equally spaced grid points , within the framework of the density functional theory @xcite . for electron - transport simulations , the rsfd method has several advantages compared with the method of the laue representation from the fundamental and practical points of view . firstly , the finite differentiation for the kinetic - energy operator is treated on the _ equal footing _ in all three directions . this avoids numerical errors due to the artificial anisotropy between the lateral and longitudinal directions at any grid spacing . secondly , the computational accuracy can be improved by employing a higher - order finite - difference formula . thirdly , in the lateral directions , isolated boundary conditions are available as well as periodic ones , which enable us to treat electrodes as leads . furthermore , the algorithm of the rsfd method is suitable for massively parallel computing@xcite . in this paper , we present the fully real - space based ls method and the ratio expression technique for the green s function of the reference system within the approach of the rsfd . this method is referred to the grid ls method . to demonstrate the performance of the grid ls method , we use it to investigate the electron - transport properties of the ( 001)si / sio@xmath0 and ( 001)ge / geo@xmath0 models connected to semi - infinite electrodes . we also estimate how the dangling bond ( db ) caused by an oxygen vacancy contributes to leakage currents across the interface between the semiconductor and oxide . the results indicate that the leakage current attributed to the db state in the si / sio@xmath0 model is much larger than that in the ge / geo@xmath0 model . in the followings of this paper , section [ sec : level2 ] gives details of the computational scheme used to develop the grid ls method . section [ sec : level3 ] presents a demonstration of our method , in which we use it to examine transport properties of si / sio@xmath0 and ge / geo@xmath0 models and to reveal how the leakage current is influenced by the db state that arises due to an oxygen vacancy . conclusions are given in section [ sec : level4 ] and mathematical details are described in appendices a and b. we propose an efficient procedure to obtain the solution of kohn sham equation for a system where the nanoscale junction is sandwiched between semi - infinite electrodes within the framework of the rsfd scheme . the effective potential is close to periodic bulk potentials as it goes deeply inside the left and right electrodes , so that the whole infinite system can be appropriately divided into three parts : the left electrode , the transition region , and the right electrode . the hamiltonian of the system , @xmath1 , is defined by @xmath2 with @xmath3 where @xmath4 and @xmath5 are the hartree and exchange - correlation potentials , respectively , and @xmath6 and @xmath7 are local and nonlocal parts of atomic pseudopotentials , respectively . assuming that the hamiltonian in the transition region can be decomposed into an unperturbed part @xmath8 and a perturbation @xmath9 , we rewrite the kohn sham equation as @xmath10 where @xmath11 is the scattering wave function for an incident wave coming from the left or right electrode with the energy @xmath12 . the subscript 0 on the variables indicates that they are evaluated in the unperturbed reference system . here , for convenience , we assumed @xmath8 not to contain the nonlocal parts of the pseudopotentials . once the retarded green s function @xmath13 in the transition region associated with the unperturbed part @xmath8 is known , eq . ( [ eqn : ls001 ] ) is put into the ls equation in a form of the integral equation , i.e. , @xmath14 with the unperturbed wave function @xmath15 . equation ( [ eqn : ls002 ] ) provides a unified treatment of the kohn sham equation and the boundary conditions@xcite . in the case where the incident bloch wave @xmath16 propagates from deep inside the left electrode , the boundary condition is @xmath17 here , @xmath18 is a reflected wave that propagates and decays into the left electrode , @xmath19 within the right electrode is a transmitted wave , and @xmath20 and @xmath21 are unknown reflection and transmission coefficients , respectively . the lateral ( @xmath22 and @xmath23 ) and @xmath24 directions are set to be parallel and perpendicular to the electrode surface , respectively . the system is assumed to be periodic in the lateral direction and infinite in the @xmath24 direction . the case of incident electrons coming from the right electrode can be considered in the same manner . in this paper , the ls equation is solved within the rsfd scheme @xcite . the rsfd approach enables us to treat arbitrary boundary conditions and to calculate the atomic and electronic structures to high accuracy . the whole system is composed by the transition region sandwiched between semi - infinite left and right electrodes and is divided by grid points with an equal spacing of @xmath25 , where @xmath26 and @xmath27 are the length and the number of grid points in the @xmath28 direction ( @xmath29 , @xmath23 , and @xmath24 ) of the transition region , respectively . here , we assume a 2-dimensional periodicity in the lateral directions and employ a generalized z coordinate @xmath30 instead of @xmath31 , which stands for the group index of @xmath24 coordinates within the closed interval @xmath32 $ ] , where @xmath33 is the number of @xmath22@xmath23 grid planes involved in @xmath30 ( see fig . [ fig : zeta_coord ] ) ; @xmath33 corresponds to the order of the finite - difference approximation for the kinetic energy operator in the kohn sham equation@xcite and is chosen so as to include the nonlocal region of pseudopotentials to obtain highly accurate results . the kohn sham equation is written in a discretized matrix form @xcite as @xmath34\psi(\zeta_{k};{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/ } } ) -b(\zeta_{k+1};{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/ } } ) \psi(\zeta_{k+1};{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})= 0 & , & \nonumber \\ ( k=-\infty, ... ,-1,0,1, ... ,\infty ) & & \label{eqn : ls01}\end{aligned}\ ] ] where @xmath35 and @xmath36 denoting the @xmath37-dimensional block matrices ( @xmath38 ) are the diagonal and off - diagonal elements of the hamiltonian block - tridiagonal matrix @xmath39 , respectively . @xmath35 includes the potential on the @xmath22@xmath23 planes at @xmath40 , @xmath41 is a set of the @xmath37 values of the wave functions on the @xmath22@xmath23 planes at @xmath40 , and @xmath42 is the lateral bloch wave vector within the first brillouin zone . hereafter , for simplicity , @xmath43 is ignored throughout . we assume that the hamiltonian in the transition region can be decomposed into an unperturbed part @xmath44 and a perturbation @xmath45 as well as in the case of the non - discretized treatment mentioned above . when the electrodes in the unperturbed reference system are adopted to be exactly those in the perturbed system described by @xmath46 , the perturbation @xmath45 has nonzero elements only in the transition region @xmath47 as @xmath48 now , by using the discretized retarded green s function @xmath49 in the transition region associated with the unperturbed part @xmath44 , the ls equation is expressed in the discretized form as @xmath50 which is referred to as the grid ls equation . this discretized form within the framework of the rsfd approach unifies eq . ( [ eqn : ls01 ] ) and the scattering boundary conditions . the boundary condition eq . ( [ eqn : ls003 ] ) now reads as @xmath51 as the hamiltonian matrix of the unperturbed reference system , @xmath52 , is a block - tridiagonal form , the @xmath37-dimensional block matrix @xmath49 , which is a component of the retarded green s function matrix @xmath53 , is expressed in terms of the scattering wave functions in a variable - separable form as ( see appendix a ) @xmath54 here , @xmath55 ( @xmath56 ) is the @xmath37-dimensional matrix made of the solutions of the kohn sham equation in the case of electrons coming from the right ( left ) electrode in the reference system , that is @xmath57 where the @xmath37-dimensional columnar vector @xmath58 @xmath59 denotes the scattering wave functions at @xmath30 for the @xmath60th incident wave @xmath61 @xmath62 incoming from deep inside the right ( left ) electrode , where the incident wave is considered to include an evanescent wave as well as an ordinary propagating wave ; more precisely , @xmath63 @xmath64 is taken to be a set of the @xmath37 generalized bloch states consisting of leftward ( rightward ) propagating bloch waves and decaying evanescent waves toward the left ( right ) side , which are the solutions of the @xmath65-dimensional generalized eigenvalue equation @xcite . the matrix @xmath66 stands for the diagonal block - matrix element of the retarded green s function matrix , @xmath67 , the representation of which is derived in appendix [ sec : appa ] as ( see eq . ( [ eqn : ls75 ] ) ) @xmath68^{-1}\end{aligned}\ ] ] with @xmath69 @xmath70 being the diagonal ( off - diagonal ) block - matrix element of @xmath71 . since @xmath72 includes the exponentially growing or decaying evanescent waves , the calculation using eq . ( [ eqn : ls12 ] ) frequently gives rise to the serious numerical errors @xcite . we provide a remedy for this problem as follows . introducing the _ ratio matrices _ @xmath73 and @xmath74 at _ two successive _ @xmath30 points , which are defined as @xmath75 respectively , we obtain the following @xmath76-dimensional block - matrix expression for the retarded green s function eq . ( [ eqn : ls12 ] ) : @xmath77 , \end{aligned}\ ] ] that is , we rewrite the block - matrix element of @xmath78 in eq . ( [ eqn : ls12 ] ) as @xmath79 and from eqs . ( [ eqn : ls14])([eqn : ls10 - 02 ] ) the diagonal block - matrix element @xmath66 reads as @xmath80^{-1}. \label{eqn : ls555}\end{aligned}\ ] ] in the following subsections [ ssec : jelele ] and [ ssec : cryele ] , we will give efficient numerical calculation techniques for the ratio matrices @xmath81 and @xmath82 without employing the matrices @xmath83 and @xmath84 which include evanescent waves explicitly . our previous study @xcite verified that the introduction of the ratio expression such as eqs . ( [ eqn : ls10 - 01])([eqn : ls99 ] ) into the retarded green s function enables us to avoid the numerical collapse originated from the appearance of the rapidly growing and decaying evanescent waves . by contrast , in ls simulations of electron transport through _ long _ conductor systems using the conventional green s function in a variable - separable form , the numerical collapse is inevitable . in the solving of eq . ( [ eqn : ls05 ] ) by using the iterative method such as the conjugate gradient method , the operation of @xmath85 in eq . ( [ eqn : ls05 ] ) is carried out as follows : @xmath86 where @xmath87 it is easily shown that the sequences @xmath88 and @xmath89 satisfy the following recursive relations : @xmath90 & \quad ( k=2,3,\cdots , m+1 ) \end{array } \right . , \\ \label{eqn : ls19 } p_r(\zeta_k)&= & \left\ { \begin{array}{ll } 0 & \quad ( k={m+1},m ) \\ x^0_{k+1}[p_r(\zeta_{k+1})+\psi^\prime(\zeta_{k+1 } ) ] & \quad ( k = m-1,\cdots,1,0 ) \end{array } \right . .\end{aligned}\ ] ] here , we used the fact that @xmath91 outside the transition region of @xmath92 . it should be emphasized that since the elements of @xmath78 in eq . ( [ eqn : ls99 ] ) are no longer in a variable - separable form , the amount of @xmath93 ^ 2 $ ] for each multiplication is expected to be required ; nevertheless , it is reduced to the order of @xmath94 by virtue of eqs . ( [ eqn : ls18 ] ) and ( [ eqn : ls19 ] ) , which means that the present method does not suffer from the numerical collapse without increasing the computational cost . the case in which electrodes are approximated by structureless jellium models is treated . the jellium electrode approximation has been successfully applied to the interpretation of electron - transport properties with less computational load @xcite . a free electron system is chosen as the unperturbed one with the hamiltonian @xmath8 where a completely flat potential is assumed , for simplicity . the green s function in the free - electron system is more conveniently described by using @xmath31 instead of @xmath30 . in appendix [ sec : appb ] , we discuss the analytical expression of the green s function in terms of @xmath31 in a general @xmath33 case . we here give details on the implementation of the analytically expressed retarded green s function in the 3-dimensional central finite - difference ( @xmath33=1 ) case , for example , which is written by @xmath95 here , @xmath96 @xmath97 are the lateral coordinates with @xmath98 [ @xmath99 is chosen an odd integer for convenience ] , and @xmath100 } & \cdots & e_{n_x , n_y}^{(\mathcal{n}_f=1 ) } + \displaystyle{\frac{2}{h_z^2 } } \le e \end{array } \right . .\end{aligned}\ ] ] with @xmath101 in the derivation of eqs . ( [ eqn : ls07])([eqn : ls09 - 02 ] ) , we used eqs . ( [ eqn : lsb-21 ] ) and ( [ eqn : lsb-22 ] ) and the extension of eq . ( [ eqn : lsb-26 ] ) to the case of the 3-dimensional space . since @xmath66 defined by eq . ( [ eqn : ls99 ] ) is the diagonal block - matrix element of the retarded green s function @xmath102 , the @xmath60th row and @xmath103th column element @xmath104 is expressed as @xmath105 one can see from eq . ( [ eqn : ls21 ] ) that @xmath104 , and thus @xmath66 , is @xmath106-independent owing to the translation invariance in the @xmath24 direction . on the other hand , by eq . ( [ eqn : ls99 ] ) , @xmath73 and @xmath74 are given by @xmath107 and from eq . ( [ eqn : ls07 ] ) , the @xmath60th row and @xmath103th column matrix element of @xmath108 are described as @xmath109 this implies that @xmath73 and @xmath74 are also @xmath106-independent and @xmath110 after some calculations , @xmath111_{j , j^\prime } \nonumber \\ & = & \frac{1}{n } \!\!\ ! \sum^{\frac{n_x-1}{2}}_{n_x=-\frac{n_x-1}{2 } } \sum^{\frac{n_y-1}{2}}_{n_y=-\frac{n_y-1}{2 } } \!\!\ ! \exp\bigl(i({\mbox{\boldmath$g$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},n}+{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})\cdot({\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j}-{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j^\prime})\bigr ) \times \exp\bigl(i\mathcal{k}_1 h_z\bigr)\end{aligned}\ ] ] is obtained @xcite . hereafter , @xmath66 and @xmath73 are denoted by @xmath112 and @xmath113 , respectively , since they are @xmath106-independent . the products of the matrix @xmath112 @xmath114 and vectors @xmath115 as required in the computations of eqs . ( [ eqn : ls16 ] ) , ( [ eqn : ls18 ] ) and ( [ eqn : ls19 ] ) can be easily carried out in the momentum space , since they are written in the convolution form of the 2-dimensional discrete fourier transform . owing to the orthogonality of the plane waves , the fourier transformed @xmath112 and @xmath113 are represented as the diagonalized matrices , i.e. , @xmath116_{n , n^\prime } & \equiv & \frac{1}{n } \sum_{{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j } } \sum_{{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j^\prime } } \exp\bigl(-i({\mbox{\boldmath$g$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},n}+{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})\cdot{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j}+i({\mbox{\boldmath$g$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},n^\prime}+{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})\cdot{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j^\prime}\bigr ) ( d^0)_{j , j^\prime } \nonumber \\ & = & -ih_z^2 \delta_{nn^\prime } \frac{1}{\sin(\mathcal{k}_1h_z)},\\ \label{eqn18 - 2 - 2 } \mathcal{f}\left [ x^0 \right]_{n , n^\prime } & \equiv & \frac{1}{n } \sum_{{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j } } \sum_{{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j^\prime } } \exp\bigl(-i({\mbox{\boldmath$g$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},n}+{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})\cdot{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j}+i({\mbox{\boldmath$g$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},n^\prime}+{\mbox{\boldmath$k$}}^b_{{\hspace{-0.5mm}/\hspace{-0.5mm}/}})\cdot{\mbox{\boldmath$r$}}_{{\hspace{-0.5mm}/\hspace{-0.5mm}/},j^\prime}\bigr ) ( x^0)_{j , j^\prime } \nonumber \\ & = & \delta_{nn^\prime } \exp(i\mathcal{k}_1h_z),\end{aligned}\ ] ] respectively . finally , one can obtain the matrix elements of the fourier transform of the terms shown in eqs . ( [ eqn : ls16 ] ) , ( [ eqn : ls18 ] ) and ( [ eqn : ls19 ] ) as @xmath117_{n } \!&=&\ ! \sum_{n^\prime } \mathcal{f}\left [ d^0 \right]_{n , n^\prime } \mathcal{f } \left [ { \displaystyle{\sum_l}}\delta v(z_k , z_l)\psi(z_l ) \right]_{n^\prime } , \\ \label{eqn : ls25 } \mathcal{f } \bigl [ p_l(z_k ) \bigr]_{n } \!&=&\ ! \left\ { \begin{array}{ll } 0 & \quad ( k\!=\!0,1 ) \\ \displaystyle { \sum_{n^\prime } } \mathcal{f}\left[x^0\right]_{n , n^\prime } \bigl ( \mathcal{f } \bigl [ p_l(z_{k-1})\bigr]_{n^\prime}+\mathcal{f } \bigl [ \psi^\prime(z_{k-1 } ) \bigr]_{n^\prime } \bigr ) & \quad ( k\!=\!2,3,\cdots , m\!+\!1 ) \end{array } \right . , \nonumber \\ \\ \label{eqn : ls27 } \mathcal{f } \bigl [ p_r(z_k ) \bigr]_{n } \!&=&\ ! \left\ { \begin{array}{ll } 0 & \quad ( k\!=\!m\!+\!1,m ) \\ \displaystyle { \sum_{n^\prime } } \mathcal{f}\left[x^0\right]_{n , n^\prime } \bigl ( \mathcal{f } \bigl [ p_r(z_{k+1 } ) \bigr]_{n^\prime}+\mathcal{f } \bigl [ \psi^\prime(z_{k+1 } ) \bigr]_{n^\prime } \bigr ) & \quad ( k\!=\!m\!-\!1,\cdots,1,0 ) \end{array } \right . , \nonumber \\ \end{aligned}\ ] ] respectively . for calculating the product in eq . ( [ eqn : ls05 ] ) , the computational cost of @xmath118 is required . however , by introducing the 2-dimensional discrete fast fourier transform ( fft ) algorithm , the cost of the product shown in eqs . ( [ eqn : ls24])([eqn : ls27 ] ) decreases to @xmath119 , since the off - diagonal elements of the fourier transformed matrices @xmath120 $ ] and @xmath121 $ ] are zero as seen in eqs . ( [ eqn18 - 2 ] ) and ( [ eqn18 - 2 - 2 ] ) . the fourier transform of a columnar vector and the inverse fourier transform of @xmath120 \times \mathcal{f}\left [ f_k \right]$ ] and @xmath122 \times \mathcal{f}\left [ f_k \right]$ ] are carried out at each @xmath31 point using fft algorithm . here , @xmath123 $ ] represents the fourier transformed vector of @xmath115 . thus , the maximum order of the calculations is improved from @xmath118 to @xmath124 . the above mentioned discussion on the central finite - difference approximation can be straightforwardly extended to the cases of the higher - order finite - difference approach . a general case is discussed where a system with atomistic crystalline electrodes is chosen as the unperturbed reference system ; one electrode is confronted with the other across the empty transition region . we present efficient procedures for calculating the ratio matrices @xmath73 and @xmath74 in this case . the matrices @xmath125 and @xmath126 defined by eqs . ( [ eqn : ls10 - 01 ] ) and ( [ eqn : ls10 - 02 ] ) are described as @xmath127 where @xmath128 is the self - energy term defined on the left- ( right-)electrode surface and can be calculated by using the continued - fraction equation ; for the details of the derivation of eqs . ( [ egami : eqn9 - 122 ] ) and ( [ egami : eqn9 - 123 ] ) and the computation of the self - energy terms , see refs . 10 and 18 . for the sake of comparison , we note that @xmath129 defined by eqs . ( [ eqn : ls13 ] ) and ( [ eqn : ls13 - 2 ] ) is identical to @xmath130 of eq . ( 15 ) in ref . 18 . we also emphasize that the accuracy of @xmath131 is enhanced by making use of the continued - fraction equation in a self - consistent manner , as shown in eqs . ( 16)(18 ) in ref . it should be noticed that the terms @xmath132 can be sequentially computed as @xmath133 which are easily derived from eq . ( [ eqn : ls671 ] ) , and similarly , the iterative series of @xmath134 are obtainable from eq . ( [ eqn : ls672 ] ) as @xmath135 the recursive relations eqs . ( [ egami : eqn9 - 134 ] ) and ( [ egami : eqn9 - 135 ] ) allow us to calculate all the matrix elements by a linear scaling operation ( order-@xmath37 calculation procedure ) at a limited computational cost . it is also noted that using eqs . ( [ egami : eqn9 - 134 ] ) and ( [ egami : eqn9 - 135 ] ) , @xmath73 and @xmath74 are stably computed without involving error accumulation since the errors due to the appearance of evanescent waves are eliminated by introducing the ratios of these waves at two successive grid points . finally , the diagonal block - matrix element @xmath136 is given by eq . ( [ eqn : ls555 ] ) . once @xmath66 , @xmath73 and @xmath74 for any @xmath106 ( @xmath137 ) are known , all of the matrix elements of @xmath78 in eq . ( [ eqn : ls99 ] ) are determined , and the algorithm of eqs . ( [ eqn : ls18 ] ) and ( [ eqn : ls19 ] ) can be utilized . to demonstrate the performance of the grid ls method , we examine the electron - transport properties of models of semiconductor / insulator interfaces sandwiched between semi - infinite electrodes . recently , the germanium - based metal - oxide - semiconductor field - effect transistor has attracted significant attention because the electronic band gap of germanium ( @xmath138 ev ) is lower than that of silicon ( @xmath139 ev ) , which allows for reduced operating voltages . in a highly integrated circuit , it is known that a large leakage current is induced by defects such as impurities and oxygen vacancies in the thin gate oxide layer . so , the relationship between the db introduced by defects and leakage current in si / sio@xmath0 interfaces has been extensively investigated @xcite , while the role of the db state in ge / geo@xmath0 interfaces is controversial . one of the present authors ( t. o. ) has performed several investigations on ge / geo@xmath0 interfaces @xcite . in recent work , the relationship between atomic configurations and electronic structures of ( 001)si / sio@xmath0 and ( 001)ge / geo@xmath0 models with dbs was explored using first - principles simulations within the framework of the local density approximation ( lda)@xcite . it was found that the si - db state is located near the midgap of the si substrate corresponding to the fermi level , while the ge - db state lies near the top of the valence band which is 0.3 ev below the fermi level@xcite . to examine how db states with different characteristics affect leakage currents , we performed transport simulations of electrons flowing across the ( 001)si / sio@xmath0 and ( 001)ge / geo@xmath0 models . the magnitude of the leakage current flowing through insulators is so small that it can be easily affected by interactions between electrodes and interface models and by the value of the energy band gap , which is underestimated by the lda calculation . therefore , in this paper , we discuss qualitatively the ratio of the leakage current between models with and without a defect . figure [ fig : sige_models ] illustrates a unit cell of each interface model . in these models , the side lengths of the cell in the lateral direction parallel to the interface for the si / sio@xmath0 ( ge / geo@xmath0 ) model were taken to be the experimental lattice constant of bulk si ( ge ) , @xmath140 . the thicknesses of the sio@xmath0 ( geo@xmath0 ) layer and the si ( ge ) substrate were 7.34 ( 7.25 ) and 7.18 ( 7.37 ) , respectively . in calculations for the models with an oxygen vacancy , we introduced the defect into a supercell comprising @xmath141 unit cells in the lateral direction ( fig . [ fig : sige_model ] ) ; this is large enough to avoid interactions between defects in neighboring cells . two si ( ge ) dbs were generated near the interface between the si ( ge ) substrate and the oxide layer in one unit cell by removing a bridging oxygen atom in a manner similar to that used in the previous study@xcite . one of the dbs is passivated by a hydrogen atom , while the other remains with the si ( ge ) atom of the center back - bonded to two neighboring si ( ge ) atoms and an oxygen atom @xmath142si@xmath143si@xmath0o ( @xmath144ge@xmath143ge@xmath0o)@xmath145 . for the no - defect models , the unit cell of each model , depicted in fig . [ fig : sige_models ] , was employed with @xmath141 sampling @xmath106-points in the 2-dimensional brillouin zone for comparison with the models having defects . we first optimized the atomic and electronic structures of the models . first - principles calculations based on the rsfd approach were performed in the manner described in ref . [ 42 ] with a grid spacing of 0.15 . the size of the supercell in the [ 001 ] ( @xmath24 ) direction was taken to be @xmath146 , including a large enough vacuum region , and the top and bottom layers of the models were terminated by hydrogen atoms . as shown in fig . [ fig : sige_model ] , which illustrates the relaxed configurations , the si atom with the db is pulled down to the si substrate whereas the ge atom with the db is slightly raised toward the oxide layer . next , we examined the leakage currents caused by introducing the oxygen vacancies into the models . employing the optimized effective kohn sham potential , we used the grid ls method to evaluate the scattering wave functions for electrons incident from the bottom - side electrode . the conductance at the limits of zero temperature and bias are described by the landauer bttiker formula @xcite . in the transport calculation , the top and bottom sides of each model were connected to aluminum jellium electrodes without terminating hydrogen atoms . the wigner seitz radius @xmath147 was @xmath148 , which corresponds to the valence electron density of bulk aluminum . figure [ fig : sige_conds ] shows the computed conductance spectra for the no - defect si / sio@xmath0 and ge / geo@xmath0 models as functions of incident electron energy measured from the valence band maximum ( vbm ) of the substrate . although some small peaks derived from the bulk states in the valence band of the si and ge substrates appear in fig . [ fig : sige_conds ] , both models exhibit highly suppressed conductivities in the band - gap region between the vbm and conduction band minimum ( cbm ) of the substrates . figure [ fig : sige_cond ] represents the conductance spectra for the si / sio@xmath0 and ge / geo@xmath0 models with the oxygen vacancy . no remarkable peaks appear in the spectrum of the ge / geo@xmath0 model ; however , for the si / sio@xmath0 model , a peak with high transmission occurs around @xmath149 ev , where electrons flow through the oxide layer via the si - db state as shown in the charge density distribution of the scattering electron ( fig . [ fig : si_cd ] ) . in addition , fig . [ fig : si_ldos ] exhibits contour plots of the density of states ( dos ) integrated in the lateral directions ( left panels ) and the charge density distributions of the db states ( right panels ) for the defect - introduced models without electrodes . in fig . [ fig : si_ldos](a ) , the white arrow identifies the peak in the dos derived from the db state between the vbm and cbm of the si substrate . in fig . [ fig : si_ldos](b ) , the ge - db state is coupled to the states in the valence band of the ge substrate . the relative positions of the vbm , cbm , and db state are modulated when each model is connected to electrodes . when a db appears in diamond - structured semiconductors , there are two possibilities : the db state tends to become either more @xmath150 type or more @xmath151 type @xcite . in the @xmath151 type db state , the three remaining bonds tend to become @xmath152 hybridized and , to reduce the strain , prefer to be in a plane . this occurs in fig . [ fig : si_ldos](a ) wherein the si atom with the db is pulled down and the db state spatially extends in the [ 001 ] ( z ) direction . this behavior degrades the insulating properties of the si / sio@xmath0 model . in contrast , when the db state is inclined to be an @xmath150 type , the three remaining bonds tend to become @xmath151 types . in this case , the angular separation of these bonds is reduced from that in the tetrahedral structure where the separation angle is 109.5@xmath153 . as a result , the atom with the db moves away from the three bonded atoms . therefore , for the db state of the raised ge atom , the charge density of the state is distributed in the lateral directions compared with that of the si - db state , and the ge - db state is coupled with interface states of the ge substrate ( fig . [ fig : si_ldos](b ) ) . this behavior barely contributes to electron transport across the model . consequently , by introducing the oxygen vacancy , the leakage current in the si / sio@xmath0 model increases by a factor of 162.9 , while that in the ge / geo@xmath0 model increases by a factor of 11.8 @xcite . we have presented the grid ls equation method based on the fully real - space algorithms to elucidate the scattering wave functions in nanoscale structures sandwiched between semi - infinite electrodes . it is shown that the numerical collapse due to the exponentially growing and decaying evanescent waves and the computational costs can be restrained by using the ratio expression of the retarded green s function obtained analytically ( jellium electrode case ) and by incorporating the self - energy matrices and applying the recursive formulas to the ratio matrices ( crystalline electrode case ) . to demonstrate the performance of our method , we used it to calculate the transport properties of ( 001)si / sio@xmath0 and ( 001)ge / geo@xmath0 models attached to semi - infinite electrodes . the results show that the db state in the ge / geo@xmath0 model gives a much smaller contribution to leakage current than that in the si / sio@xmath0 model . our procedure can precisely and efficiently extend knowledge of the physics underlying the transport of electrons through nanoscale structures . this research was supported by a grant - in - aid for young scientists ( b ) grant no . 24710152 from the ministry of education , culture , sports , science and technology . the numerical calculation was carried out with the computer facilities at the institute for solid state physics at the university of tokyo , the information synergy center at tohoku university . in this appendix , the subscript @xmath154 denoting the reference system is omitted , for simplicity . let us consider the product of the matrix @xmath155 and the @xmath156th columnar vector of @xmath157 , @xmath158 ( @xmath159 ) . the retarded green s function is constructed from _ outwardly _ propagating and decreasing waves , and then , taking it into account , we assume this columnar vector to be represented by @xmath160^t & , & \end{aligned}\ ] ] where @xmath161 and @xmath162 @xmath163 are unknown block matrices , and @xmath164 is defined by eqs . ( [ eqn : ls13 ] ) and ( [ eqn : ls13 - 2 ] ) . hereafter , @xmath164 , @xmath161 and @xmath162 are abbreviated to @xmath165 , @xmath166 and @xmath167 , respectively . by definition , the abovementioned product satisfies @xmath168 where @xmath169 and @xmath170 . since @xmath171 is a set of the solutions of the kohn sham equation , the following equations hold : @xmath172 from eqs . ( [ eqn : ls66])([eqn : ls672 ] ) , one sees that the unknown matrices @xmath173 and @xmath167 are required to satisfy the equations @xmath174 and thus , eqs . ( [ eqn : ls68 ] ) and ( [ eqn : ls70 ] ) lead to the relationships between @xmath173 and @xmath167 as @xmath175 and eq . ( [ eqn : ls69 ] ) decides @xmath167 to be @xmath176^{-1}.\end{aligned}\ ] ] in consequence , the retarded green s function @xmath177 can be described in the following separable form : @xmath178 a 1-dimensional system is firstly considered for simplicity . in the rsfd approach , the kinetic - energy operator @xmath179 is represented by the matrix @xmath180 , and the kinetic - energy term in the kohn sham equation is written as @xmath181,\end{aligned}\ ] ] where @xmath33 is the order of the finite - difference approximation , @xmath182 is a grid spacing and the weight coefficients @xmath183 are determined using the taylor expansion @xcite . in the @xmath33th order finite - difference approximation , the green s function matrix @xmath184 is determined as satisfying @xmath185 where @xmath186 is a complex number and @xmath187 is the unit matrix . the @xmath188 th row@xmath189 th column element of the green s function matrix @xmath190 is described in a spectral representation as @xmath191 where @xmath192 and @xmath193 are the eigenvalue and eigenvector of @xmath180 , respectively , obtained by solving the eigenvalue equation @xmath194 , and are given by @xmath195 where @xmath196 , and @xmath197 is normalized , i.e. , @xmath198 substituting eq . ( [ eqn : lsb-04 ] ) into eq . ( [ eqn : lsb-03 ] ) and changing the integration variable from @xmath151 to @xmath199 and subsequently from @xmath200 to @xmath201 , we obtain @xmath202 the integration can be carried out along the unit circle in the complex @xmath203 plane based on the residue theorem . in the following , we introduce a sensible manner of picking up the poles inside the unit circle that contribute to the integration . these poles @xmath203 s are the @xmath33 solutions of the equation @xmath204 we now define a new variable @xmath150 as @xmath205 and rewrite eq . ( [ eqn : lsb-06 ] ) as @xmath206 which is the @xmath33th order algebraic equation with respect to @xmath150 and its solutions are denoted by @xmath207 . resultantly , for each @xmath208 , the poles @xmath209 s are given by the solutions of the quadratic equation @xmath210 as @xmath211 when the imaginary part of @xmath186 is non - zero , either of @xmath212 or @xmath213 is inside the unit circle , while the other is outside it . hereafter , the inside pole is defined as @xmath209 , and the other is obtained by @xmath214 . using the residue theorem , the integration of eq . ( [ eqn : lsb-05 ] ) is carried out to yield @xmath215 finally , introducing @xmath216 by @xmath217 , hence , @xmath218 , we obtain @xmath219 the retarded green s function is given by @xmath220 it is noted that @xmath216 is a multivalued function since it is defined as @xmath221 . the branch of @xmath222 should be chosen to satisfy the requirement of @xmath223 , which guarantees that @xmath209 exists inside the unit circle in the complex @xmath203 plane . thus , the following relationship is established : @xmath224 * proof : * consider complex numbers @xmath225 and @xmath226 with the relationship of @xmath227 in the interval of @xmath228 . it is readily shown that when @xmath229 and @xmath230 , there exists the one - to - one correspondence between @xmath231 and @xmath232 . in this case , @xmath233 is uniquely so defined as to satisfy @xmath234 and the relationship @xmath235 where @xmath236 is defined as @xmath237 and @xmath238 in eq . ( [ eqn : lsb-14 ] ) is the principal value of the inverse trigonometric ( hyperbolic ) cosine function . thus , @xmath106 varies as @xmath239 in the derivation of eqs . ( [ eqn : lsb-14])([eqn : lsb-16 ] ) , we used well - known formulas , @xmath240 it is straightforward to extend the above argument to the 3-dimensional case . we deal with a free electron system in which the discretized space is infinite in the @xmath24 direction and periodic in the @xmath22 and @xmath23 directions . the green s function in this case is described in a spectral representation by @xmath245 where @xmath246 and @xmath247 are the eigenvalue and eigenvector of the 3-dimensional kinetic - energy matrix , respectively , i.e. , @xmath248 here , @xmath249 are the lateral coordinates with @xmath98 [ for convenience , @xmath99 is chosen an odd integer ] , and the definition of @xmath250 is same as shown in eq . ( [ eqn : ls11 ] ) . now , the green s function represented by eq . ( [ eqn : lsb-11 ] ) reads as @xmath251 where @xmath252 under the requirement of @xmath223 , and @xmath208 is the solution of the @xmath33th order algebraic equation with respect to @xmath253 @xmath254 in the following , we present the _ analytic _ representation of the retarded green s functions in @xmath255 cases in the 1-dimensional free electron system ; there exists no analytic one in a case of @xmath256 , since the algebraic equation ( [ eqn : lsb-08 ] ) with @xmath256 can not be solvable analytically according to galois theory . given the solutions of eq . ( [ eqn : lsb-08 ] ) , @xmath257 , @xmath258 are determined from eqs . ( [ eqn : lsb-13])([eqn : lsb-19 ] ) , and finally we obtain the analytic representation of the green s function eq . ( [ eqn : lsb-11 ] ) . hereafter , we choose @xmath259 ( @xmath260 : an infinitesimal positive number ) in eq . ( [ eqn : lsb-11 ] ) so as to deal with the retarded green s function . substituting @xmath262 and @xmath263 into eq . ( [ eqn : lsb-08 ] ) , we have the equation @xmath264 and its solution @xmath265 since eq . ( [ eqn : lsb-25 ] ) indicates that @xmath266 ( an infinitesimal _ negative _ number ) in the limit of @xmath267 , @xmath268 is determined from eqs . ( [ eqn : lsb-14 ] ) and ( [ eqn : lsb-15 ] ) such that @xmath269 substituting @xmath271 , @xmath272 and @xmath273 into eq . ( [ eqn : lsb-08 ] ) yields the quadratic equation with respect to @xmath150 , @xmath274 this equation has two solutions @xmath275 and @xmath276 given by @xmath277 and @xmath278 subsequently , from eqs . ( [ eqn : lsb-14 ] ) , ( [ eqn : lsb-15 ] ) and ( [ eqn : lsb-18 ] ) , @xmath279 can be described in an analytic form . the substitution of @xmath281 , @xmath282 , @xmath283 and @xmath284 into eq . ( [ eqn : lsb-08 ] ) leads to the cubic equation with respect to @xmath150 , @xmath285 whose solutions are determined according to cardano s formula as @xmath286 here , @xmath287{5 \bigl\ { \sqrt{(144h^2e+155)^2 + 5 \times 19 ^ 3 } \pm ( 144h^2e+155 ) \bigr\}},\end{aligned}\ ] ] and @xmath279 can be analytically represented using eqs . ( [ eqn : lsb-14 ] ) , ( [ eqn : lsb-15 ] ) and ( [ eqn : lsb-18 ] ) . now , substituting @xmath289 , @xmath290 , @xmath291 , @xmath292 and @xmath293 into eq . ( [ eqn : lsb-08 ] ) , we obtain the quartic equation with respect to @xmath150 , @xmath294 ferrari s solutions to eq . ( [ eqn : lsb-36 ] ) are utilized . after tedious but straightforward calculations , we have the following representations : @xmath295 here , @xmath296{\rho + \sqrt{\nu^3 + \rho^2 } } - \sqrt[^{^3}\!]{-\rho + \sqrt{\nu^3 + \rho^2 } } \right ) } } & \cdots & 0 \leq \nu \\ \displaystyle{\sqrt{\alpha \left ( -\frac{1}{3}+\sqrt[^{^3}\!]{\rho + \sqrt{\nu^3 + \rho^2 } } + \sqrt[^{^3}\ ! ] { \rho - \sqrt{\nu^3 + \rho^2 } } \right ) } } & \cdots & \nu < 0 \end{array } \right . , \\ \begin{array}{rclcrclc } \nu & = & \displaystyle{a_0+a_1h^2e } & , & \rho & = & \displaystyle{\frac{1}{2_{}}\nu + a_2 } & , \\ e_0 & = & \displaystyle{\frac{\nu_0 - a_0}{a_1h^2 } } & , & e_0^\prime & = & \displaystyle{\frac{\nu^\prime_0 - a_0}{a_1h^2 _ { } } } & , \\ \end{array } \\ \begin{array}{rclcrclcrcl } a_0 & = & \displaystyle{\frac{3 ^ 2 \cdot 5 \cdot 19}{2 ^ 2 \cdot 31 ^ 2 } } & , & a_1 & = & \displaystyle{\frac{2 \cdot 3 ^ 5 \cdot 5}{7 \cdot 31 ^ 2 } } & , & a_2 & = & \displaystyle{\frac{3 ^ 2 \cdot 5 \cdot 3623}{2 \cdot 7 \cdot 31 ^ 3 } } , \end{array } \end{array } \right.\end{aligned}\ ] ] and @xmath297 is the solution of @xmath298 , which is evaluated as @xmath299 , and @xmath300 with @xmath301 being @xmath302{2\beta^2}/2\alpha$ ] . the usage of eqs . ( [ eqn : lsb-11])([eqn : lsb-19 ] ) leads to the analytic representations of @xmath279 and the retarded green s function . 99 n. d. lang and a. r. williams , phys . rev . b * 18 * , 616 ( 1978 ) . n. d. lang , phys . b * 52 * , 5335 ( 1995 ) . n. d. lang and p. avouris , phys . rev . b * 64 * , 125323 ( 2001 ) . n. kobayashi , m. aono , and m. tsukada , phys . b * 64 * , 121402 ( 2001 ) . s. tsukamoto and k. hirose , phys . b * 66 * , 161402(r ) ( 2002 ) . s. tsukamoto , m. aono , and k. hirose , jpn . j. appl . phys . * 41 * , 7491 ( 2002 ) . j. nara , w. t. geng , h. kino , n. kobayashi , and t. ohno , j. chem . 121 * , 6485 ( 2004 ) . s. tsukamoto , y. egami , k. hirose , and s. blgel , phys . b * 84 * , 115443 ( 2011 ) . y. fujimoto and k. hirose , phys . b * 67 * , 195315 ( 2003 ) . k. hirose , t. ono , y. fujimoto , and s. tsukamoto , _ first - principles calculations in real - space formalism , electronic configurations and transport properties of nanostructures _ ( imperial college press , london , 2005 ) . t. ono and k. hirose , phys . lett . * 82 * , 5016 ( 1999 ) . k. hirose and t. ono , phys . b * 64 * , 085105 ( 2001 ) . t. ono and k. hirose , phys . b * 72 * , 085105 ( 2005 ) . t. ono and k. hirose , phys . b * 72 * , 085115 ( 2005 ) . t. sasaki , t. ono , and k. hirose , phys . e * 74 * , 056704 ( 2006 ) . t. ono , m. heide , n. atodiresei , p. baumeister , s. tsukamoto , and s. blgel , phys . b * 82 * , 205115 ( 2010 ) . y. egami , k. hirose , and t. ono , phys . e * 82 * , 056706 ( 2010 ) . t. ono , y. egami , and k. hirose , phys . b * 86 * , 195406 ( 2012 ) . p. hohenberg and w. kohn , phys . rev . * 136 * , b864 ( 1964 ) . w. kohn and l.j . sham , phys . rev . * 140 * , a1133 ( 1965 ) . y. hasegawa , j .- iwata , m. tsuji , d. takahashi , a. oshiyama , k. minami , t. boku , f. shoji , a. uno , m. kurokawa , h. inoue , i. miyoshi , and m. yokokawa , in _ proceedings of 2011 international conference for high performance computing , networking , storage and analysis _ , sc 11 , acm ( new york , ny , usa 2011 ) , p. 1 ; http://sc11.supercomputing.org/?pg=awards.html ; http://awards.acm.org/award_winners/hasegawa_7239126.cfm j.r . chelikowsky , n. troullier , k. wu , and y. saad , phys . b * 50 * , 11355 ( 1994 ) . chelikowsky , n. troullier , and y. saad , phys . lett . * 72 * , 1240 ( 1994 ) n. d. lang , phys . lett . * 79 * , 1357 ( 1997 ) . n. kobayashi , m. brandbyge , and m. tsukada , surf . * 433 - 435 * , 854 ( 1999 ) ; phys . b * 62 * , 8430 ( 2000 ) . m. okamoto , t. uda , and k. takayanagi , phys . b * 64 * , 033303 ( 2001 ) . p. havu , t. torsti , m. j. puska , and r. m. nieminen , phys . b * 66 * , 075401 ( 2002 ) . s. furuya , y. gohda , and s.watanabe , jpn . j. appl . phys . , part 1 * 45 * , 8991 ( 2006 ) . m. otani , t. ono , and k. hirose , phys . b * 69 * , 121408(r ) ( 2004 ) ; t. ono and k. hirose , phys . rev . lett . * 94 * , 206806 ( 2005 ) ; * 98 * , 026804 ( 2007 ) ; y. egami , s. aiba , k. hirose , and t. ono , j. phys . : * 19 * , 365201 ( 2007 ) . t. ono , j. phys . c * 113 * , 6256 ( 2009 ) . an alternative derivation of eq . ( [ eqn : ls22 - 3 ] ) is given in ref . 17 , in which the retarded self - energy matrix of the jellium electrode , @xmath303 , is analytically calculated . ( [ egami : eqn9 - 122 ] ) for the relationship between @xmath304 and @xmath305 , and note also that @xmath306 in the central finite - difference ( @xmath261 ) case . h. kageshima and k. shiraishi , phys . lett . * 81 * , 5936 ( 1998 ) . m. houssa , g. pourtois , m. caymax , m. meuris , m. m. heyns , v. v. afanas@xmath307ev , and a. stesmans , appl . lett . * 93 * , 161909 ( 2008 ) . m. yang , r. q. wu , q. chen , w. s. deng , y. p. feng , j. w. chai , j. s. pan , and s. j. wang , appl . . lett . * 94 * , 142903 ( 2009 ) . j. f. binder , p. broqvist , and a. pasquarello , microelectron . eng . * 86 * , 1760 ( 2009 ) . p. broqvist , j. f. binder , and a. pasquarello , appl . phys . lett . * 94 * , 141911 ( 2009 ) . m. houssa , e. a. chagarov , and a. c. kummel , mrs bull . * 34 * , 504 ( 2009 ) . l. tsetseris and s. t. pantelides , appl . . lett . * 95 * , 262107 ( 2009 ) . t. ono , phys . b * 79 * , 195326 ( 2009 ) . s. saito and t. ono , jpn . j. appl . phys . * 50 * , 021503 ( 2011 ) . s. saito and t. ono , phys . b * 84 * , 085319 ( 2011 ) . t. ono and s. saito , appl . phys . exp . * 4 * , 021303 ( 2011 ) . j.p . perdew and a. zunger , b * 23 * , 5048 ( 1981 ) . m. bttiker , y. imry , r. landauer , and s. pinhas , phys . b * 31 * , 6207 ( 1985 ) . and @xmath308 in the computational model in the case of @xmath280 . the system consists of the transition region sandwiched between the left and right semi - infinite crystalline electrodes . in the left electrode , the incident wave and the reflected waves consisting of the propagating and evanescent ones are illustrated by @xmath309 and @xmath310 , respectively , while in the right electrode , the transmitted waves composed of the propagating and decaying evanescent ones toward the right side are denoted by @xmath311 . here , @xmath312 and @xmath24 are coordinates perpendicular and parallel to the nanoscale junction , respectively . , width=604 ] and ( b ) ge / geo@xmath0 models with an oxygen vacancy after geometrical optimization . dashed lines indicate the boundaries of supercells . the key to the symbols is the same as in fig . [ fig : sige_models ] . , width=604 ] and ge / geo@xmath0 models as functions of incident electron energy measured from the valence band maximum ( vbm ) of the substrates . red circles and black squares represent the conductance spectra of si / sio@xmath0 and ge / geo@xmath0 models , respectively . vbm and conduction band minimum ( cbm ) of si ( ge ) substrate are indicated by vertical dashed ( dotted ) lines . , width=302 ] and ge / geo@xmath0 models as functions of incident electron energy measured from the valence band maximum ( vbm ) of the substrates . the key to the symbols is the same as in fig . [ fig : sige_conds ] . vbm and conduction band minimum ( cbm ) of si ( ge ) substrate are indicated by vertical dashed ( dotted ) lines . , width=302 ] model with incident energy of @xmath313 ev . the key to the symbols is the same as in fig . [ fig : sige_models](a ) . here the charge density is integrated in the [ 100 ] ( @xmath22 ) direction . each contour represents twice or half the density of adjacent contours ; the lowest contour is @xmath314 electron / ev / @xmath315 . , width=302 ] and ( b ) ge / geo@xmath0 models without electrodes . arrows denote the dos peaks derived from the db states . energies are measured from the fermi level @xmath316 and the @xmath24 coordinate of the atom with db is set to zero . in the right panels , the charge density is integrated in the [ 100 ] ( @xmath22 ) direction . the key to the symbols in the right panels is the same as in fig . [ fig : sige_models ] . the coordinate in the @xmath24 direction corresponds to that of the left panel . each contour represents twice or half the density of adjacent contours ; the lowest contour is @xmath317 electron / ev / ( @xmath318 electron / @xmath315 ) in the left ( right ) panels . , width=604 ]

we investigate whether uncoded schemes are optimal for gaussian sources on multiuser gaussian channels . particularly , we consider two problems : the first is to send correlated gaussian sources on a gaussian broadcast channel where each receiver is interested in reconstructing only one source component ( or one specific linear function of the sources ) under the mean squared error distortion measure ; the second is to send vector gaussian sources on a gaussian multiple - access channel , where each transmitter observes a noisy combination of the source , and the receiver wishes to reconstruct the individual source components ( or individual linear functions ) under the mean squared error distortion measure . it is shown that when the channel parameters match certain general conditions , the induced distortion tuples are on the boundary of the achievable distortion region , and thus optimal . instead of following the conventional approach of attempting to characterize the achievable distortion region , we ask the question whether and how a match can be effectively determined . this decision problem formulation helps to circumvent the difficult optimization problem often embedded in region characterization problems , and also leads us to focus on the critical conditions in the outer bounds that make the inequalities become equalities , which effectively decouples the overall problem into several simpler sub - problems . = 0.2 cm

1502.05428

although the source - channel separation architecture is asymptotically optimal in the point - to - point communication setting @xcite as well as several classes of multiuser communication settings ( see _ e.g. , _ @xcite and references therein ) , uncoded schemes have several particularly attractive properties . firstly , they have very simple encoders and decoders ; secondly , they belong to the so - called zero - delay codes , which can avoid the long delay required to approach the asymptotic performance in the separation - based schemes ; lastly , they are in fact optimal in some settings , while the separation - based schemes are not ( see _ e.g. _ , @xcite ) . it was shown in @xcite that uncoded schemes are optimal when certain matching conditions involving the source probability distribution , the channel transition probability distribution , the channel cost function and the distortion measure function are satisfied . though the focus in @xcite was mainly on the point - to - point setting , recent results @xcite suggest that the concept of matching indeed carries over to the multiuser case . in fact , in multiuser settings , matching may occur naturally when the distortion measure , the channel cost function and source distribution are all fixed , and the channel parameters , which represent physically meaningful quantities , satisfy certain conditions . in this work , we consider such matching , particularly , when the sources and the channels are gaussian , the channel constraints are on the expected average signal power , the distortion measure is the mean squared error ( mse ) , and only the channels parameters , such as the channel amplification factors and the additive noise powers , are allowed to vary . in this context , of interest is whether for a fixed source and fixed coding parameters , the distortion vector such induced is on the boundary of the achievable distortion region ( and thus optimal ) . more specifically , we seek to answer the following questions : * is there a set of ( explicitly ) computable conditions that can be used to certify a fixed uncoded scheme to be optimal for a given source and channel pair ? * if so , is there a non - trivial set of channels that satisfy such conditions for a given source and uncoded scheme pair ? we shall refer to this kind of channels as `` matched channels '' ; a dual question is to ask for `` matched sources '' , however in the context of problems considered here , the dual question is notationally more involved , and thus we choose to investigate the problems from the perspective of `` matched channels '' . one can also ask for `` matched distortion measures '' , similarly as the approach taken in @xcite , however in the gaussian setting we consider here , fixing the mse distortion is practically more important and well - motivated . the set of matched channels should be distinguished from the complete set of channels for which the given uncoded scheme is optimal . the former may be a strict subset of the latter , since the former may be only sufficient for an uncoded scheme to be optimal , which usually depends on the specific outer bounding technique employed . characterizing the latter region is naturally more difficult than answering the questions we posed above . the two questions posed above are in essence the two facets of the same question . since we only provide conditions for matching , or in other words , sufficient conditions for the scheme to be optimal , the set of matched channels may in fact be empty . a trivial condition to answer the first question is simply an impossible one such that we would never be able to certify a channel to be matched . thus the second question is important , and we show indeed for the two problems considered here , there are non - trivial channels that match the source and the uncoded scheme . traditionally , research in information theory asks for characterization of certain achievable region , for which we first derive an expression for an outer bound , and derive an expression for an inner bound , and then make comparison of them . this approach can be challenging because it usually involves optimization over a set of parameters , and solving such an optimization problem explicitly can be difficult . it is not clear whether the obstacle mainly stems from the intractable nature of the underlying communication problem , or it is mainly caused by the embedded optimization problem . the aforementioned difficulty motivates the formulation of the first question , which is a decision problem instead of an optimization problem . an analogy of this situation can be found in computer science algorithm research , where instead of asking whether an optimization problem can be solved in polynomial time , the alternative question is asked whether a decision ( _ e.g. _ , regarding a solution is above a threshold ) can be made in polynomial time . our problem formulation naturally leads to a different approach in the investigation . instead of focusing on comparison of the inner bounds and outer bounds using their expressions , we focus on the necessary conditions that the outer bound becomes tight , _ i.e. , _ the conditions when the information inequalities hold with strict equality . with fixed source and fixed coding parameters , the coding vector can be substituted into the conditions , and the necessary and sufficient conditions for such equality can be derived . the outer bounds naturally provide certain `` decoupled '' conditions , which significantly simplifies the overall task . though this approach may have inherently been used by many researchers in the past , its effectiveness becomes particularly evident in our investigation of uncoded schemes in the joint source channel communication setting . in the rest of the paper , we focus specifically on two joint source channel coding problems using the approach outlined above . the first problem is to send correlated gaussian sources on a gaussian broadcast channel where each receiver is interested in reconstructing only one source component ( or equivalently one specific linear function of the source ) under the mse distortion measure . the second problem is to send vector gaussian sources on a gaussian multiple - access channel , where each transmitter observes a noisy combination of the source , _ i.e. , _ a case of the vector ceo problem , and the receiver wishes to reconstruct the source components ( or equivalently linear functions of the source components ) under the mse distortion measure . general conditions for matching are derived , which either include or generalize well - known existing results on the optimality of uncoded schemes in the multiuser setting . particularly notable are the following cases : * the first problem generalizes the two - user case considered in @xcite and @xcite to the @xmath0-user case , for which we show that the uncoded scheme is optimal for a large set of sources and channels ; our results reveal that uncoded scheme can still be optimal when some source components are negatively correlated . * the results on the second problem includes as specials cases the symmetric scalar gaussian ceo problem @xcite , the problem of sending bivariate gaussian sources on a gaussian multiple - access channel @xcite , and sending remote ( noisy ) bivariate gaussians on a gaussian multiple - access channel @xcite . our results reveal that in addition to the symmetric case considered in @xcite , uncoded scheme is also optimal when the sensor observation quality is proportional to the channel quality . these results also allow the sensor observations to have more general correlation structure and the observations to be noisy , thus extending the results in@xcite and @xcite . when viewed from the perspective of computation , our result also provides new characterizations for the problem of computing linear functions of gaussian random variables on the gaussian multiple - access channels considered in @xcite and @xcite . notationally , we write for a source @xmath1 at time @xmath2 as @xmath3 $ ] , and a length-@xmath4 vector as @xmath5 . for a set of coefficients @xmath6 , we sometimes write it in a ( column ) vector form as @xmath7 . for a real matrix @xmath8 , we write its transpose as @xmath9 . the positive semidefinite order is denoted as @xmath10 . in this section we consider the problem of broadcasting correlated gaussian sources on a gaussian broadcast channel , which can be described as follows ; see also fig . [ fig : bc ] for an illustration . let the zero - mean gaussian source be @xmath11,s_2[n],\ldots , s_m[n])$ ] with covariance matrix @xmath12 , which is assumed to be full rank . the channel is given by @xmath13=x[n]+z_m[n],\end{aligned}\ ] ] where @xmath14 are zero - mean additive noises which are mutually independent , with variances @xmath15 , respectively ; the channel input must satisfy an average power constraint @xmath16 . the transmitter encodes the length-@xmath4 source vector @xmath17 into a length-@xmath4 channel vector @xmath18 , and the @xmath19-th receiver reconstructs from the channel output vector @xmath20 the source vector @xmath21 as @xmath22 , resulting in a distortion @xmath23-\hat{s}_m[n])^2 $ ] . we omit the formal problem definition using encoding and decoding function , which is standard and can be obtained by extending that in , for example , @xcite . the uncoded scheme of interest has the form @xmath24=\sum_{m=1}^m \alpha_m s_m[n],\end{aligned}\ ] ] such that @xmath25=p.\end{aligned}\ ] ] in other words , at each time instance , the channel input is simply a linear combination of the source components with coefficients @xmath6 , such that the resulting signal has an expected variance that is equal to the power constraint @xmath26 . we shall assume @xmath27 . the decoders simply estimate @xmath28 $ ] as @xmath29= { \mathbb{e } } ( s_m[n]|y_m[n])$ ] , at each time instance @xmath30 at decoder @xmath31 . notice that the problem can be equivalently formulated as computation of linear functions of gaussian sources on the broadcast channel , however this alternative formulation is notationally more involved . , width=491 ] define @xmath32 the main result is summarized in the following theorem , which gives a matching condition in a positive semidefinite form . [ theorem : bc ] a gaussian broadcast channel is said to be matched to a given source and an uncoded scheme with non - zero parameters @xmath7 , and the distortion vector induced by the given scheme is on the boundary of the achievable distortion region thus optimal , if @xmath33 where the entries of the symmetric matrix @xmath34 are specified as @xmath35,\quad m=1,2,\ldots , m.\end{aligned}\ ] ] this theorem establishes a condition that is sufficient to guarantee a distortion vector induced by an uncoded scheme to be on the boundary of the achievable distortion region , and thus an optimal solution . the matrix @xmath34 may seems mysterious at the first sight , however , the reason for introducing this matrix will become clear shortly . this theorem clearly answers our first question regarding conditions that can be used to certify whether a given uncoded scheme is optimal . in fact , it also provides clues on the second question regarding whether there exist non - trivial channels that such matching is possible . indeed , in section [ sec : cholesky ] and section [ sec : bcchannels ] we establish several properties of matched channels , through which an answer to the second question is given . before presenting those results , the proof of this theorem is presented next in two parts : the critical conditions in a novel outer bound are outlined in section [ sec : bcouter ] , and then these conditions for the bound to hold with equality in the uncoded scheme are analyzed in section [ sec : bcinner ] . the proof details for the outer bound are relegated to the appendix . in order to obtain the matching condition , we first derive a novel outer bound for this problem . an important technique in the derivation of this outer bound is the introduction of certain appropriate random variables outside of the original problem . this approach is partly motivated by our previous work @xcite , which can further be traced back to ozarow @xcite . consider @xmath0 zero - mean gaussian random variables @xmath36 , independent of everything else , with covariance matrix @xmath37 , and write @xmath38=s_m[n]+w_m[n].\end{aligned}\ ] ] the outer bound will be written as a necessary condition that any distortion vector has to satisfy . for this purpose , we bound the following quantity for any encoding and decoding functions : @xmath39,\end{aligned}\ ] ] where @xmath40 for notational simplicity . an almost identical quantity was used in @xcite to obtain an approximate characterization for the distortion region of the gaussian broadcast problem with bandwidth mismatch . we shall upper - bound this quantity using the channel properties and lower - bound it using the source reconstruction requirements , then combine them to obtain an eventual outer bound . this quantity can be upper - bounded as given in the appendix as @xmath41 with equality holds if and only if @xmath42 and the following condition stemming from the entropy power inequalities hold with equality @xmath43=\exp\left[\frac{2}{n}h(y^n_{m+1}|u^n_m , u^n_{m-1},\ldots , u^n_1)\right]+2\pi e[\sigma^2_{z_m}-\sigma^2_{z_{m+1}}],\nonumber\\ & \qquad\qquad\qquad\qquad\qquad \qquad\qquad\qquad\qquad\qquad \qquad\qquad\qquad\qquad\qquad m=1,2,\ldots , m.\label{eqn : epi}\end{aligned}\ ] ] the conditions in ( [ eqn : epi ] ) are standard , as bergmans @xcite also used the entropy power inequality to establish the gaussian broadcast channel capacity , and in general a gaussian codebook suffices to make them equalities . the condition ( [ eqn : bcpower ] ) intuitively requires that the power is fully utilized . the condition ( [ eqn : degerate ] ) is however rather peculiar , which essentially requires the noisy source @xmath44 to be as useful as the real source @xmath17 in determining the channel output @xmath45 . the quantity @xmath46 can also be lower - bounded as given in the appendix , where its individual summands are bounded as @xmath47\geq \frac{|\sigma_{s_1,s_2,\ldots , s_m}+\sigma_{w_1,w_2,\ldots , w_m}|}{\pi^m_{j=1}(d_j+\sigma^2_{w_j})},\label{eqn : bcsource}\end{aligned}\ ] ] with equality holds if and only if @xmath48,\quad m=1,2\ldots , m,\label{eqn : bccondition1}\\ h(u^n_m|y^n_m , u^n_1,u^n_2,\ldots , u^n_{m-1})&=h(u^n_m|y^n_m),\quad m=2,3,\ldots , m.\label{eqn : bccondition2}\end{aligned}\ ] ] the conditions in ( [ eqn : bccondition1 ] ) are standard which can be viewed as requiring the codes to achieve the given distortions with equality , however the conditions in ( [ eqn : bccondition2 ] ) are peculiar which essentially require all the information @xmath49 on @xmath50 to be from @xmath20 . combining ( [ eqn : bcchannel ] ) and ( [ eqn : bcsource ] ) , we obtain an outer bound , or outer bounds since each set of the auxiliary random variables @xmath36 provides one specific outer bound . in the approach we shall take , the precise form of this outer bound is less important than the extracted matching conditions ( [ eqn : degerate ] ) , ( [ eqn : bcpower ] ) , ( [ eqn : epi ] ) , ( [ eqn : bccondition1 ] ) and ( [ eqn : bccondition2 ] ) . in fact , the conditions ( [ eqn : bcpower ] ) , ( [ eqn : epi ] ) and ( [ eqn : bccondition1 ] ) can be satisfied simply by choosing a proper jointly gaussian coding scheme , yet the conditions ( [ eqn : degerate ] ) and ( [ eqn : bccondition2 ] ) are the effectual non - trivial conditions . note that from the problem setting and taking into consideration the fact that physical degradedness is equivalent to stochastic degradedness in the broadcast setting , we have the markov string @xmath51 where we have utilized the fact that physical degradation is equivalent to stochastic degradation in broadcast channels . this markov string is not sufficient to guarantee ( [ eqn : degerate ] ) and ( [ eqn : bccondition2 ] ) , and thus they require special attention . we first introduce some additional notation and make a few observations . notice that due to the power constraint , the coefficient vector @xmath52 should satisfy @xmath53 and it follows that @xmath54 due to the jointly gaussian distribution of the uncoded scheme , we can write @xmath55 where the three components are mutually independent , since @xmath56 $ ] ; we have also omitted the time index @xmath57 $ ] to simplify the notation . it follows that the covariance matrix of @xmath58 given @xmath59 can be decomposed as follows @xmath60 where @xmath61 let @xmath62 for @xmath31 . with the above observations , we now return to the derivation of the forward matching conditions . as mentioned earlier , we need to substitute the random vectors specified by the uncoded scheme , _ i.e. , _ assigning @xmath63=\sum_{m=1}^m \alpha_m s_m[n]$ ] , into the critical conditions ( [ eqn : degerate ] ) , ( [ eqn : bcpower ] ) , ( [ eqn : epi ] ) , ( [ eqn : bccondition1 ] ) and ( [ eqn : bccondition2 ] ) in order to identify the matching conditions . it is straightforward to see that ( [ eqn : bcpower ] ) , ( [ eqn : epi ] ) and ( [ eqn : bccondition1 ] ) indeed hold with equality due to the jointly gaussian distribution of the uncoded scheme , and the chosen coefficients . thus we only need to focus on ( [ eqn : degerate ] ) and ( [ eqn : bccondition2 ] ) , which in the context of the uncoded scheme are equivalent to the following single - letter forms @xmath64 to satisfy the condition ( [ eqn : bcindep ] ) with the jointly gaussian uncoded scheme , for any @xmath65 , we must have @xmath66+\beta_m\beta_j\sigma^2_{x|y_m}=0 $ ] for @xmath67 . this specifies all the off - diagonal terms of @xmath68 , as @xmath69=\gamma_{m , j}=-\beta_m\beta_j\sigma^2_{x|y_m},\quad 1\leq j < m,\quad m=2,3,\ldots , m . \label{eqn : crossterms}\end{aligned}\ ] ] it remains to determine the diagonal entries of @xmath68 . since @xmath70 in order to satisfy the condition ( [ eqn : bcasifnonoise ] ) with equality , we must have @xmath71 where we have taken into account of the fact that @xmath12 is full rank . thus for any @xmath31 , @xmath72= { \mathbb{e } } [ v_m\sum_{j=1}^m\alpha_jv_j]=0.\end{aligned}\ ] ] it follows that @xmath73 can be determined from @xmath74-\sum_{j = m+1}^m\alpha_j { \mathbb{e } } [ v_mv_j]=\beta_m\sigma^2_{x|y_m}\sum_{j=1}^{m-1}\alpha_j\beta_j+\beta_m\sum_{j = m+1}^m\alpha_j\beta_j\sigma^2_{x|y_j},\end{aligned}\ ] ] since @xmath27 . thus the conditions ( [ eqn : bcasifnonoise ] ) and ( [ eqn : bcindep ] ) being equalities uniquely specify the matrix @xmath68 . conversely , as long as the matrix @xmath75 is positive semidefinite , the conditions ( [ eqn : bcasifnonoise ] ) and ( [ eqn : bcindep ] ) hold with equality and the corresponding auxiliary random variables @xmath36 can be found , and the outer bound derived previously is thus tight . this is exactly the matching condition given in theorem [ theorem : bc ] . * remark : * the outer bounds conditions ( [ eqn : degerate ] ) and ( [ eqn : bccondition2 ] ) in the context of the uncoded scheme provide two constraints on the matrix @xmath68 . their effects on the matrix @xmath68 are largely decoupled : the condition required by ( [ eqn : bccondition2 ] ) being equal determines the off - diagonal entries of @xmath68 , while the condition ( [ eqn : degerate ] ) determines its diagonal entries . this decoupling effect is particularly helpful in deriving the matching conditions . in the second problem we consider in the next section , _ i.e. , _ the multiple access channel problem , this decoupling effect is even more pronounced . the condition given theorem [ theorem : bc ] is in a positive semidefinite form , however , due to the specific problem structure , it can also be represented as a set of recursive conditions , which is discussed in this section . this alternative representation also leads to a necessary condition for matching to hold , which plays an instrumental role for several results given in section [ sec : bcchannels ] , where we answer the second question regarding the existence of non - trivial set of matched channels . determining whether a matrix is positive semidefinite is equivalent to computing the ldl decomposition , and checking whether the resultant diagonal matrix in the decomposition has non - negative entries . the matrix @xmath75 is positive semidefinite if and only if the diagonal matrix in the ldl decomposition has only non - negative entries . computationally this can be accomplished with the cholesky factorization @xcite on the matrix @xmath75 . here we provide an intuitive description of the cholesky factorization in the context of the problem being considered , and its conceptual interpretation as the recursive thresholding determination for the channel to yield a matching . in the first step of the cholesky factorization , we use symmetric column and row gaussian elimination to eliminate all the entries of the @xmath0-th column and the @xmath0-th row , except the diagonal entry . denote the resulting upper - left @xmath76 matrix after this first step as @xmath77 . a necessary condition for the matrix @xmath75 to be positive definite is that the lower right entry of the matrix @xmath75 is strictly positive , or all the entries on the last column are zero . notice that the condition only involves @xmath78 or the channel noise power @xmath79 , which yields a necessary condition on @xmath79 in the form of @xmath80 . continue the cholesky factorization on @xmath77 , and a similar necessary condition is its lower right entry is strictly positive , or the entries on the @xmath81-th row of @xmath77 are zero . similarly as the previous step , the condition on @xmath82 is found to be in the form that @xmath83 . continuing this process will yield a set of conditions in the form of @xmath84 the matrix @xmath75 is positive definite if and only if all such threshold conditions are satisfied . notice that the threshold function @xmath85 for @xmath86 depends on the channel noise power values @xmath87 , but not on @xmath88 . thus these functions @xmath89 , @xmath90 can be viewed as a recursive threshold checking ( or determination ) procedure , and the channel noise power @xmath91 needs to be chosen to be larger than the threshold determined by @xmath87 in every step to yield a matching . given the above observation , it is natural to speculate that if a channel is matched , then any more noisy channel also induces a match . this intuition is in fact correct , and the statement is made more rigorous in the next section as corollary [ coro1 ] . one necessary condition for a matching to exist is that the matrix @xmath34 is positive semidefinite . we can thus apply the cholesky factorization technique on this particular matrix to obtain a necessary condition for matching to exist . for the matrix @xmath34 constructed previously to be positive semidefinite ( with @xmath92 ) , it must be true that @xmath93 , @xmath94 . note that this condition is essentially independent of the channel , as long as the channel is not perfect . this lemma is proved in the appendix . with lemma 1 , we can establish several properties of the set of matched channels , given next as corollaries to theorem [ theorem : bc ] . their proofs are provided in the appendix . these properties essentially provide an answer to the second question posed earlier , and we shall also further illustrate such sources and channels using an example . [ coro1 ] if an uncoded scheme is matched on @xmath95 , then it is matched and thus optimal on any channel with noise powers @xmath96 where @xmath97 , @xmath31 . the corollary reveals a property of matched channels : once a channel is matched , any channel with more noise is also a matched channel and thus the uncoded scheme is optimal . the next corollary states , from the perspective of only the source and the uncoded scheme parameters , the necessary and sufficient condition for matching to exist . [ coro2 ] matching ( on a broadcast channel with finite noise powers ) exists , if and only if @xmath98 and the matrix @xmath99 has its largest eigenvalue being 1 with multiplicity 1 , where @xmath100 is a diagonal matrix with diagonal entries being @xmath101 moreover , if the above condition holds , then any channel with @xmath102 is a matched channel , where @xmath103 , and @xmath104 is the second largest eigenvalue of the matrix @xmath99 . * remark : * if the entries of @xmath105 are strictly positive , then matching is always possible . this follows from the fact that the matrix @xmath99 has positive entries , and @xmath106 is its positive eigenvector , such that @xmath107 is its largest eigenvalue with multiplicity 1 ( by perron - frobenius theorem @xcite ) . different from the case discussed in the previous remark , the next corollary gives another sufficient condition for matching to occur when the sources and the coding parameters satisfy the same positive correlation condition . [ coro3 ] let the entries of the matrix @xmath108 be strictly positive . denote the entries of @xmath12 as @xmath109 , and define @xmath110 any channel with @xmath111 such that @xmath112 for @xmath65 is a matched channel . * remark : * @xmath113 as defined above may in fact be negative . however this does not cause any discrepancy due to the requirement @xmath111 . as a reality check , notice that @xmath114 but @xmath115 unless @xmath116 , which however would contradict our assumption . it thus follows @xmath117 and thus @xmath118 always holds under the condition in the corollary . * remark : * for the symmetric case where @xmath119 , @xmath120 , @xmath121 and @xmath122=\beta x$ ] , for @xmath94 . a necessary and sufficient condition for matching is simply @xmath123 to see this , notice that @xmath124.\end{aligned}\ ] ] and @xmath125 can be computed as @xmath126 checking the first condition in the cholesky factorization , it is easily verified that ( [ eqn : symmetric ] ) is a necessary condition for matching . however , from corollary [ coro3 ] , it is seen that it is sufficient to choose any @xmath112 , where @xmath127 , @xmath65 . this is exactly condition ( [ eqn : symmetric ] ) . pairs for which matching is possible , given in shade . [ fig : rho1rho2],width=302 ] let us consider a source with three components whose covariance matrix is either @xmath128 or @xmath129 and further assume that the coefficients are chosen @xmath130 in the uncoded scheme . in addition to the constraint that the matrix @xmath131 must be positive definite , for a matching to exist , the condition in corollary [ coro2 ] must be satisfied . it can be shown that the eigenvalues of @xmath132 are @xmath133 and we must have @xmath134 and @xmath135 . in the appendix , we show that the valid choices are the @xmath136 pairs such that @xmath137 the corresponding region is plotted in fig [ fig : rho1rho2 ] . notice that the two matrices are equivalent for the purpose of determining whether matching is possible , thus the region in fig [ fig : rho1rho2 ] is valid for both cases . next let us fix a @xmath136 pair , and consider the region of @xmath138 pairs such that matching occurs . the tradeoffs can be computed explicitly , and are illustrated in fig . [ fig : n1n2 ] for @xmath139 . the circles in the plots give the channels specified by corollary [ coro2 ] . the channels given by corollary [ coro3 ] can be computed directly ( given as the dots ) , which is loose in the first case , but on the lower boundary ( and it is an extreme point ) for the second case . since @xmath140 , we also include this boundary in the plot . for the first case , the boundary @xmath141 is also shown , while for the second , the lower bound @xmath142 required by the function @xmath143 in the first step of the cholesky factorization is shown . the corresponding channels that matching occurs are those inside the fan regions . note that there is a tension between the noise powers @xmath144 and @xmath145 for matching to occur with for the fixed source and uncoded scheme . for the two covariance matrices ( [ eqn : cov1 ] ) and ( [ eqn : cov2 ] ) , respectively . [ fig : n1n2],width=604 ] in this section we consider the problem of sending correlated gaussian sources on a gaussian multiple - access channel , where the transmitters observe noise linear combinations of the source components ; see also fig . [ fig : mac ] for an illustration . a zero - mean vector gaussian source @xmath11,s_2[n],\ldots , s_m[n])$ ] has a covariance matrix @xmath12 . there are a total of @xmath146 sensors , whose observations are @xmath147,t_2[n],\ldots , t_l[n])$ ] , respectively , with covariance matrix @xmath148 . the sources and observations are jointly gaussian . each sensor observes @xmath149 , encodes it under an average transmission power constraint @xmath150 , @xmath151 . the channel output is given as @xmath152=z[n]+\sum_{\ell=1}^l\delta_\ell x_\ell[n],\end{aligned}\ ] ] where @xmath153 . the receiver wishes to reconstruct @xmath17 using channel output @xmath154 to minimize the individual mse measure , which achieves mse distortion @xmath155 for @xmath156 , _ i.e. _ , @xmath23-\hat{s}_m[n])^2 $ ] . notice that due to the jointly gaussian distribution , we can write @xmath157\triangleq \tilde{s}_m=\sum_{\ell=1}^l\gamma_{m,\ell}t_\ell,\quad m=1,2,\ldots , m.\end{aligned}\ ] ] the parameters @xmath158 can be conveniently written in a matrix form @xmath159 , and computed as @xmath160 where @xmath161 is the cross - covariance matrix between vectors @xmath162 and @xmath163 . notice that the problem can be equivalently formulated as computation of linear functions of gaussian sources on the multiple - access channel . in this alternative setting , the functions to be computed are @xmath162 , which can be represented as noisy linear functions of the sensor observations @xmath163 . this alternative formulation is notationally more involved in the current problem setting , but we shall explore this connection in a separate work . we assume @xmath164 , and consider the case that the matrices @xmath12 , @xmath148 , @xmath165 and @xmath161 all have full ( row ) rank , which hold in general except certain degenerate cases . denote the entries of @xmath148 as @xmath166 . the uncoded scheme we consider is @xmath167=\eta_\ell\sqrt{\frac{p_\ell}{\psi_{\ell,\ell}}}t_\ell[n],\quad \ell=1,2,\ldots , l,\label{eqn : xform}\end{aligned}\ ] ] where @xmath168 is either @xmath169 or @xmath170 to be specified next . in other words , each sensor sends its noisy observation directly using the full power , but it can choose whether to negate its observations . the @xmath19-th receiver estimates @xmath28 $ ] as @xmath29= { \mathbb{e } } [ s_m[n]|y[n]]$ ] . define @xmath171^{-1}\nonumber\\ & \qquad\cdot\sigma_{(s_1,s_2,\ldots , s_m),(t_1,t_2,\ldots , t_l)}\sigma_{t_1,t_2,\ldots , t_l}\left(\delta_1 \eta_1\sqrt{\frac{p_1}{\psi_{1,1}}},\delta_2 \eta_2\sqrt{\frac{p_2}{\psi_{2,2}}},\ldots,\delta_l \eta_l\sqrt{\frac{p_l}{\psi_{l , l}}}\right)^t,\label{eqn : alphas}\end{aligned}\ ] ] and we assume @xmath27 , @xmath31 , which is true in general except certain degenerate cases . our main result on this problem is summarized in the following theorem . , width=604 ] [ theorem : mac ] a gaussian multiple - access channel is said to be matched to a given gaussian source and an uncoded scheme with parameters @xmath172 , and the distortion vector induced by the given scheme is on the boundary of the achievable distortion region and thus optimal , if 1 . @xmath173 ; 2 . the vector @xmath174 is in the row space of the matrix @xmath161 ; 3 . @xmath175 , where @xmath104 is the second largest eigenvalue of the matrix @xmath176 , where @xmath100 is a diagonal matrix with diagonal entries @xmath177 and @xmath178 s are the entries of the matrix @xmath179 these conditions can be intuitively explained as follows : condition one guarantees that the channel inputs from all transmitters coherently add up ; condition two stems from the requirement that the noise observations should serve the same role as the underlying source for the chosen power constraints and amplification factors , _ i.e. , _ as if the observation noise does not exist ; condition three is similar to the effect in the previous problem where once a channel is matched , a more noisy channel will also induce a match . when all @xmath180 , we can simply choose @xmath181 ( or @xmath170 ) for all @xmath182 to satisfy the first condition . however , when some of the terms @xmath183 are negative , a simple algorithmic approach can be used to determine whether there exists a valid assignment of @xmath168 . in fact this condition is completely source dependent , and the choice of @xmath184 is unique up to a negation ( assuming any component @xmath185 is not completely independent of the others ) , and thus can be considered fixed for a given source observation covariance matrix . the proof of this theorem also has two parts given in section [ sec : outermac ] and section [ sec : innermac ] . this theorem answers the first question regarding the conditions to certify whether an uncoded scheme is optimal in this communication problem . the answer to the second question for this problem turns out to be simpler than that in the broadcast case , and we discuss in section [ sec : macexample ] as special case examples several problems previously considered in the literature . define @xmath186 and thus @xmath187 the reason to introduce @xmath188 s is that in the remote coding setting , the distortion can be decomposed into two independent parts : the first part is due to encoding the observable part of the underlying sources , which are @xmath189 random variables , with a distortion , and the second is due to the inherent noisy nature of the observations which induces a fixed distortion @xmath190 . thus encoding the source @xmath156 to distortion @xmath155 is equivalent to encoding the equivalent source @xmath188 to distortion @xmath191 . we can now derive an outer bound by combining the approach used in the broadcast problem with a technique based on witsenhausen s bound @xcite . again consider @xmath0 auxiliary zero - mean gaussian random variables @xmath36 with covariance matrix @xmath37 , which are independent of everything else , and write @xmath38=\tilde{s}_m[n]+w_m[n],\quad m=1,2,\ldots , m.\end{aligned}\ ] ] notice the markov string @xmath192 and we can write using data processing inequality that @xmath193 where equality holds if and only if @xmath194 following the exact steps as in @xcite ( see also @xcite ) and applying witsenhausen s bound @xcite , we can obtain @xmath195 where @xmath196 and @xmath197 . this inequality intuitively says that the mutual information between the channel inputs and the output is upper bounded by the capacity of a point - to - point channel , whose power constraint is equal to the resultant signal power when all the inputs on the multiple - access channel are coherently added . we will not attempt to further simplify this condition at this point , since in the context of the uncoded scheme , it has a particularly simple form . the right hand side of ( [ eqn : dataprocessing ] ) can be bounded similarly as in the broadcast problem . here the equivalent source is @xmath198 , and the distortion vectors are @xmath199 , and moreover , @xmath200 for @xmath31 . we thus arrive at @xmath201 where equality holds if and only if @xmath202,\quad m=1,2,\ldots , m,\label{eqn : maccondition1}\\ h(u^n_m|y^n , u^n_1,u^n_2,\ldots , u^n_{m-1})&=h(u^n_m|y^n),\quad m=2,3,\ldots , m.\label{eqn : maccondition2}\end{aligned}\ ] ] an outer bound is then obtained by combining ( [ eqn : dataprocessing ] ) , ( [ eqn : channelcondition ] ) and ( [ eqn : sameasbc ] ) . again the precise form of this outer bound is less important than the extracted matching conditions ( [ eqn : macindep ] ) , ( [ eqn : channelcondition ] ) being equality , ( [ eqn : maccondition1 ] ) and ( [ eqn : maccondition2 ] ) . the condition ( [ eqn : channelcondition ] ) being equality and the condition ( [ eqn : maccondition1 ] ) can be satisfied simply by choosing a proper jointly gaussian coding scheme , and the conditions ( [ eqn : macindep ] ) and ( [ eqn : maccondition2 ] ) are almost identical to ( [ eqn : degerate ] ) and ( [ eqn : bccondition2 ] ) in the broadcast case . since the uncoded scheme takes single letter encoding function , ( [ eqn : channelcondition ] ) being equality is equivalent to @xmath203 because in the uncoded scheme the channel input @xmath204 is given in ( [ eqn : xform ] ) , the equality holds as long as @xmath205 this yields the first condition stated in theorem [ theorem : mac ] . the conditions ( [ eqn : macindep ] ) and ( [ eqn : maccondition2 ] ) in the context of uncoded scheme are equivalent to @xmath206 denote @xmath207 for ( [ eqn : macasinnonoise ] ) to hold with equality , two conditions must hold @xmath208=\tilde{x},\label{eqn : dataprocessing1}\end{aligned}\ ] ] and @xmath209=\tilde{x}.\label{eqn : dataprocessing2}\end{aligned}\ ] ] let us consider the first condition ( [ eqn : dataprocessing1 ] ) . due to the jointly gaussian distribution , there exists a set of coefficients @xmath6 such that @xmath208=\sum_{m=1}^m\alpha_m\tilde{s}_m=\sum_{m=1}^m\alpha_m\sum_{\ell=1}^l\gamma_{m,\ell}t_\ell.\end{aligned}\ ] ] however notice that @xmath210 thus the condition ( [ eqn : dataprocessing1 ] ) is equivalent to the fact that the vector @xmath211 is in the row space of the matrix @xmath212 . equivalently , the vector @xmath213 needs to be in the row space of the matrix @xmath161 . this leads to the second condition stated in theorem [ theorem : mac ] . when this condition is satisfied , the coefficients @xmath7 can be determined exactly as in ( [ eqn : alphas ] ) . the conditions ( [ eqn : maccondition2singleletter ] ) and ( [ eqn : dataprocessing2 ] ) are now identical to the broadcast case with @xmath214 being the sources and @xmath215 being the channel input . by corollary 2 , such a channel is matched when the second largest eigenvalue of the matrix @xmath176 is less than @xmath216 , or in other words , the noise power must be above or equal to the given threshold stated in theorem [ theorem : mac ] . * remark : * the first condition in theorem [ theorem : mac ] generally has a unique solution if it can be satisfied , up to a negation of all the signs of the channel input signals . the second condition can almost always be satisfied by choosing appropriate a @xmath217 vector , except a few special cases where an all positive solution does not exist ( recall we have assumed @xmath153 , and thus only all positive solutions are valid ) . if the third condition is satisfied for certain source - channel - code triple , then it is satisfied for any more noisy channels . it is seen that the critical conditions in the outer bound derivation essentially decouples the matching problem into several simpler ones , leading to the three largely independent conditions given in theorem [ theorem : mac ] . in the multiple - access setting , the conditions for matching in theorem [ theorem : mac ] are already rather simple , and there is no need to further investigate the properties of matched channels as in the broadcast case . next we consider two special cases in the general problem setting which extend those considered in @xcite and @xcite , respectively . , width=529 ] consider a zero - mean scalar gaussian source @xmath3 $ ] with variance @xmath218 . there are a total of @xmath146 sensors , whose observations are @xmath219=d_\ell s[n]+z\rq{}_\ell[n ] , \ell=1,2,\ldots , l,\end{aligned}\ ] ] where @xmath220 ( without loss of generality ) and @xmath221$]s are the zero - mean independent additive noise with variance @xmath222 . this special case is depicted in fig . [ fig : ceo ] . it is clear that the first condition in theorem [ theorem : mac ] is satisfied by @xmath223 for all @xmath151 . the second condition for this case is equivalent to @xmath224 where @xmath225 here means a component - wise proportional relation . in other words , the uncoded scheme is optimal if @xmath226 however , the lhs of the above condition can be simplified to @xmath227 where the second term is proportional to @xmath228 , and the first term is proportional to @xmath228 if and only if @xmath229 it remains to check the third condition , however in this case @xmath230 , and the second eigenvalue of the matrix @xmath176 can be viewed as zero , thus any noise power @xmath231 will allow a matching . summarizing the above analysis , it is seen that for the scalar ceo problem on a gaussian multiple - access channel , as long as the condition ( [ eqn : proportional ] ) holds , the uncoded scheme is optimal . conversely , for any noisy observation qualities , there always exists a matched channel by choosing the values of @xmath232 properly . the condition ( [ eqn : proportional ] ) corresponds to a proportional quality requirement : the quality of the observations need to match the transmission powers and the transmission amplification factors . gastpar @xcite showed that when all the sensors have the same observation quality , the same power and the same amplification factor , the uncoded scheme is optimal . our result thus generalizes it to the proportional case . consider the case when @xmath233 , and we shall assume that the first condition in theorem [ theorem : mac ] can be satisfied . the second condition is also satisfied trivially since the matrix @xmath161 is full rank in our problem setting . thus only the last condition needs to be checked in this case . equivalently , when @xmath104 is strictly less than @xmath107 , there always exists a noise power @xmath231 such that the channel is matched and thus the uncoded scheme is optimal . , width=604 ] lapidoth and tinguely @xcite previously considered the case that @xmath234 , @xmath31 ; see fig . [ fig : local ] . it was shown that for covariance matrix @xmath12 with strictly positive entries , there always exists a noise power @xmath231 such that the uncoded scheme is optimal . our result generalizes theirs to the case that the observations can be noisy linear combinations , and the covariance matrix @xmath12 does not necessarily all have strictly positive entries . we considered the problem of determining whether a given uncoded scheme is optimal for multiuser joint source - channel coding . it was shown that for both broadcast and multiple - access in the gaussian setting , matching occurs naturally under certain general conditions . our approach differs from the more conventional approach in that instead of attempting to find explicit outer bound and inner bound then compare them , our focus is on the critical conditions that make the outer bound hold with equality . this approach has a decoupling effect which tremendously simplifies the overall task . as future work , we plan to extend and generalize this approach to explore matching in other channel networks . to upper - bound @xmath46 , first recall the markov string @xmath235 we start by writing the following : @xmath236\nonumber\\ & = \sum_{j=1}^m \left[h(y^n_j|u^n_1,u^n_2,\ldots , u^n_{j-1})-h(y^n_j|u^n_1,u^n_2,\ldots , u^n_j)\right]\nonumber\\ & = \sum_{j=1}^m h(y^n_j|u^n_1,u^n_2,\ldots , u^n_{j-1})-\sum_{j=1}^m h(y^n_j|u^n_1,u^n_2,\ldots , u^n_j).\end{aligned}\ ] ] since physical degradedness is equivalent to stochastic degradedness in the broadcast setting , _ i.e. , _ @xmath237 can be assumed to be decomposable into two independent components as @xmath238 , we can apply the entropy power inequality @xcite for @xmath239 , @xmath240\nonumber\\ & \geq \exp\left[\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^n_j)\right]+\exp\left[\log(2\pi e ( \sigma^2_{z_j}-\sigma^2_{z_{j+1}}))\right]\nonumber\\ & = \exp\left[\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^m_j)\right]+2 \pi e(\sigma^2_{z_j}-\sigma^2_{z_{j+1 } } ) . \label{eqn : applyentropypower}\end{aligned}\ ] ] for @xmath241 , it is clear that @xmath242\geq \exp\left[\frac{2}{n}h(y^n_m|s^n_1,s^n_2,\ldots , s^n_m)\right]= 2 \pi e\sigma^2_{z_m},\end{aligned}\ ] ] with equality if and only if @xmath243 it now follows that @xmath244\nonumber\\ & \leq \sum_{m=1}^m ( \sigma^2_{z_{m}}-\sigma^2_{z_{m+1 } } ) \frac{\exp\left[\frac{2}{n}\sum_{j=1}^m h(y^n_j|u^n_1,u^n_2,\ldots , u^n_{j-1})\right]}{\prod_{j=1}^m\left[\exp\left(\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^n_j)\right)+2 \pi e ( \sigma^2_{z_j}-\sigma^2_{z_{j+1}})\right]},\end{aligned}\ ] ] where for convenience @xmath245\triangleq 0 $ ] . we upper - bound this summation , by considering the summands in the reversed order , _ i.e. _ , @xmath90 . starting with the summands when @xmath246 and @xmath247 , we have @xmath248}{\prod_{j=1}^{m-1}\left[\exp\left(\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^n_j)\right)+2 \pi e(\sigma^2_{z_j}-\sigma^2_{z_{j+1}})\right]}\nonumber\\ & \qquad\qquad+\sigma^2_{z_{m } } \frac{\exp\left[\frac{2}{n}\sum_{j=1}^{m } h(y^n_j|u^n_1,u^n_2,\ldots , u^n_{j-1})\right]}{\prod_{j=1}^{m}\left[\exp\left(\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^n_j)\right)+2 \pi e(\sigma^2_{z_j}-\sigma^2_{z_{j+1}})\right]}\nonumber\\ & = \frac{\exp\left[\frac{2}{n}\sum_{j=1}^{m-1 } h(y^n_j|u^n_1,u^n_2,\ldots , u^n_{j-1})\right]}{\prod_{j=1}^{m-1}\left[\exp\left(\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^m_j)\right)+2 \pi e ( \sigma^2_{z_j}-\sigma^2_{z_{j+1}})\right]}\nonumber\\ & \qquad\qquad\cdot\left[(\sigma^2_{z_{m-1}}-\sigma^2_{z_{m } } ) + \sigma^2_{z_{m}}\frac{\exp\left[\frac{2}{n}h(y^n_m|u^n_{m-1})\right]}{2 \pi e\sigma^2_{z_{m}}}\right]\nonumber\\ & = \frac{1}{2 \pi e}\frac{\exp\left[\frac{2}{n}\sum_{j=1}^{m-1 } h(y^n_j|u^n_1,u^n_2,\ldots , u^m_{j-1})\right]}{\pi_{j=1}^{m-2}\left[\exp\left(\frac{2}{n}h(y^n_{j+1}|u^n_1,u^n_2,\ldots , u^n_j)\right)+2 \pi e(\sigma^2_{z_{m-1}}-\sigma^2_{z_{m}})\right]}.\label{eqn : firststep}\end{aligned}\ ] ] continuing this line of reduction , we finally arrive at when @xmath249 @xmath250}{\exp\left(\frac{2}{n}h(y^n_{2}|u^n_1)\right)+2 \pi e(\sigma^2_{z_{2}}-\sigma^2_{z_{1}})}+\frac{1}{2 \pi e}\frac{\exp\left[\frac{2}{n}\sum_{j=1}^{2 } h(y^n_j|u^n_1,\ldots , u^n_{j-1})\right]}{\exp\left(\frac{2}{n}h(y^n_{2}|u^n_1)\right)+2 \pi e(\sigma^2_{z_{2}}-\sigma^2_{z_{1}})}\nonumber\\ & = \frac{\exp\left[\frac{2}{n } h(y^n_1)\right]}{\exp\left(\frac{2}{n}h(y^n_{2}|u^n_1))\right)+2 \pi e(\sigma^2_{z_{2}}-\sigma^2_{z_{1}})}\left[(\sigma^2_{z_{2}}-\sigma^2_{z_{1}})+\frac{\exp \left[\frac{2}{n}h(y^n_2|u^n_1)\right]}{2\pi e}\right]\nonumber\\ & = \frac{\exp\left[\frac{2}{n } h(y^n_1)\right]}{2\pi e}\leq p+\sigma^2_{z_1},\label{eqn : bcchannela}\end{aligned}\ ] ] where the last inequality is by the concavity of the @xmath251 function and the given power constraint . the chain of inequalities in ( [ eqn : bcchannela ] ) holds with equality holds if and only if @xmath252 as well as ( [ eqn : bcasifnonoisea ] ) and the entropy power inequalities hold with equality . we next lower bound @xmath46 . by the rate - distortion theorem @xcite @xmath253 with equality holds if and only if @xmath254.\end{aligned}\ ] ] furthermore , @xmath255\label{eqn : bccondition3a}\\ & = \frac{n}{2}\log\frac{|\sigma_{s_1,s_2,\ldots , s_m}+\sigma_{w_1,w_2,\ldots , w_m}|}{|(\sigma_{s_1,s_2,\ldots , s_{m-1}}+\sigma_{w_1,w_2,\ldots , w_{m-1}})|[d_m+\sigma^2_{w_m}]},\nonumber\end{aligned}\ ] ] where ( [ eqn : bccondition2a ] ) is because conditioning reduces entropy , and ( [ eqn : bccondition3a ] ) is because gaussian distribution maximizes the entropy for random variables with the same variance , together with the concavity of the @xmath256 function . for ( [ eqn : bccondition2a ] ) to hold with equality , we must have @xmath257 and for ( [ eqn : bccondition3a ] ) to hold with equality requires @xmath258,\quad m=2,3,\ldots , m.\end{aligned}\ ] ] recall @xmath27 . in the @xmath263-th step of the cholesky decomposition @xmath264 , we claim that @xmath265 and @xmath266 . moreover , we claim the matrix partially diagonalized , denoted as @xmath267 , has entries in the following form : * @xmath268 , @xmath269 and @xmath270 ; by symmetry , @xmath271 , @xmath272 and @xmath270 . * @xmath273 , @xmath274 ; * @xmath275 , @xmath276 and @xmath277 ; by symmetry @xmath278 , @xmath279 and @xmath280 ; * @xmath281 $ ] , @xmath282 . where the terms @xmath283 are determined recursively as @xmath284 and @xmath285 for which @xmath286 clearly the claim is true when @xmath287 . suppose it is also true for @xmath288 , and we wish to prove the claim for @xmath289 . it is clear that due to the positive semidefinite requirement for the degenerate case when @xmath290 we must have for @xmath291 @xmath292 and this cholesky step can essentially be skipped , and @xmath283 does not need to be updated . it is easy to check the recursive formula @xmath293 for @xmath294 is indeed valid for this case . if @xmath295 , then clearly @xmath296 due to the assumption in the induction . if @xmath297 then @xmath298 for @xmath291 , and thus @xmath299 . thus the induction is valid in this case . if @xmath300 , then due to the assumption in the induction we have @xmath301 first observe that due to the assumption in the induction , we have @xmath302 using the cholesky factorization , we have for any @xmath303 and @xmath277 @xmath304\nonumber\\ & = -\beta_i\beta_jb^{(k^*+1)}_j.\end{aligned}\ ] ] similarly for @xmath294 @xmath305-\beta^2_m\beta_{m - k^*}b^{(k^*)}_{m - k^*}\frac{\alpha_{m - k^*}}{\sum_{j=1}^{{m - k^*-1}}\alpha_j\beta_j}\nonumber\\ & = \frac{\beta_m}{\alpha_m}b^{(k^*)}_m\sum_{j=1}^{m-1}\alpha_j\beta_j+\frac{\beta_m}{\alpha_m}\sum_{j = m+1}^{m - k^*-1}\alpha_j\beta_jb^{(k^*)}_j\nonumber\\ & \qquad+\alpha_{m - k^*}\beta_{m - k^*}\frac{\beta_m}{\alpha_m}\frac{\sum_{j=1}^{{m - k^*-1}}\alpha_j\beta_j-\alpha_m\beta_m}{\sum_{j=1}^{{m - k^*-1}}\alpha_j\beta_j}b^{(k^*)}_{m - k^*}\nonumber\\ & = \frac{\beta_m}{\alpha_m}\left[b^{(k^*)}_m+\frac{\sum_{j=1}^{m-1}\alpha_{m - k^*}\beta_{m - k^*}}{\sum_{t=1}^{{m - k^*-1}}\alpha_t\beta_t}b^{(k^*)}_{m - k^*}\right]\sum_{j=1}^{m-1}\alpha_j\beta_j\nonumber\\ & \qquad+\frac{\beta_m}{\alpha_m}\sum_{j = m+1}^{m - k^*-1}\alpha_j\beta_j\left[b^{(k^*)}_j+\frac{\sum_{j=1}^{m-1}\alpha_{m - k^*}\beta_{m - k^*}}{\sum_{t=1}^{{m - k^*-1}}\alpha_t\beta_t}b^{(k^*)}_{m - k^*}\right]\nonumber\\ & = \frac{\beta_m}{\alpha_m}\left[b^{(k^*+1)}_m\sum_{j=1}^{m-1}\alpha_j\beta_j+\sum_{j = m+1}^{m - k^*-1}\alpha_j\beta_jb^{(k^*+1)}_j\right].\end{aligned}\ ] ] suppose @xmath306 , this implies that @xmath307 which however contradicts with the positive semidefinite requirement that @xmath308 if @xmath309 , then from the assumption in the induction , we have @xmath310 thus this case does not cause any problem . if @xmath311 , then it also follows that @xmath312 . the lemma is proved . it suffices to consider the case that @xmath313 , @xmath314 , and @xmath315 . denote @xmath316 , and matrices constructed for the two channels as @xmath34 and @xmath317 , respectively . it is clear that @xmath318 however , it is easily seen that this matrix is positive semidefinite since the first @xmath319 diagonal terms are non - negative , and we can remove all the other terms through symmetric elimination , _ i.e. , _ the cholesky factorization . it follows that @xmath320 + \left[\sigma_{v_1,v_2,\ldots , v_m}-\sigma_{s_1,s_2,\ldots , s_m}+p\bar{\beta}\bar{\beta}^t \right]\end{aligned}\ ] ] is positive semidefinite since it is a summation of two positive semidefinite matrices . for the `` only if '' direction , it follows from corollary 1 that matching must hold for the degraded channel with noise power @xmath322 . the requirement ( [ eqn : semidefinite ] ) implies @xmath323 this implies that @xmath324 , since otherwise the left hand side is rank deficient . together with lemma 1 , it follows that @xmath98 . next multiply both sides from left and from right by @xmath100 , it follows that @xmath325 notice that @xmath106 is in fact an eigenvector corresponding to eigenvalue @xmath107 for the matrix @xmath99 , easily verified using ( [ eqn : alphatobeta ] ) . we can write the eigen decomposition of the matrix @xmath99 as @xmath326 where @xmath327 are the other eigenvalues of @xmath99 , and @xmath328 are the corresponding eigenvectors . it follows that @xmath329 which implies @xmath330 , @xmath331 . for the `` if '' direction , we choose a @xmath332 such that ( [ eqn : nlarge ] ) holds , which is always possible when @xmath332 is sufficiently large . this implies that for the channel @xmath333 , condition ( [ eqn : matchingspecial ] ) holds , and thus it is a matched channel . the entries of matrix @xmath108 , denoted as @xmath334 , is given ( by symmetry only the upper - triangle entries need to be specified ) @xmath335 and @xmath336 a necessary and sufficient condition for matching is that the matrix @xmath108 is positive semidefinite . observe that @xmath337 if all the off - diagonal entries of @xmath108 are non - positive , then the matrix is diagonally dominant , and the diagonal entries are all positive , which implies that it is a positive semidefinite matrix . thus as long as @xmath338 and @xmath339 the positive semidefinite condition is satisfied . it is thus sufficient to have @xmath340 together with corollary 1 , this implies the statement given in the corollary is indeed true . since the matrix @xmath131 is positive definite , we have @xmath341 and @xmath342 . since @xmath343 , we must also have @xmath344 , @xmath345 and @xmath346 for matching to occur . the first condition gives that @xmath347 but the latter two require a few more steps . notice that the condition @xmath346 implies that @xmath348 if @xmath349 , then @xmath345 implies @xmath350 if @xmath351 , this yields a condition already implied by @xmath352 ; on the other hand , @xmath353 is an impossible case . it can be verified that @xmath354 is also an impossible case . thus we must have @xmath349 and @xmath351 simultaneously , from which we obtained the set of conditions given in ( [ eqn : source2 ] ) .

an extended bose - einstein condensate ( bec ) in an optical lattice provides a kind of periodic dielectric and causes band gaps to occur in the spectrum of light propagating through it . we examine the question whether these band gaps can modify the spontaneous emission rate of atoms excited from the bec , and whether they can lead to a self - stabilization of the bec against spontaneous emission . we find that self - stabilization is not possible for becs with a density in the order of . however , the corresponding non - markovian behavior produces significant effects in the decay of excited atoms even for a homogeneous bec interacting with a weak laser beam . these effects are caused by the occurrence of an avoided crossing in the photon ( or rather polariton ) spectrum . we also predict a new channel for spontaneous decay which arises from an interference between periodically excited atoms and periodic photon modes . this new channel should also occur in ordinary periodic dielectrics . + pacs : 03.75.fi , 32.80.-t , 42.70.qs + shortened title : spontaneous emission in bose - einstein condensates

cond-mat0004326

it is well - known that the radiation properties of atoms can dramatically be manipulated by changing the environment where atoms emit photons . for micro - cavities it has been demonstrated @xcite and for periodic dielectric media predicted @xcite that a suppression of spontaneous emission ( se ) can be achieved . in the case of a micro - cavity this happens because its geometry reduces the radiation - mode density , whereas in a periodic dielectric medium se is suppressed due to the formation of photonic band gaps ( pbg ) . the recent achievement of bose - einstein condensation in magnetic traps @xcite has provided a new state of matter where all atoms share a single macroscopic quantum state . such a state of matter offers great opportunities to explore and test new phenomena related to macroscopic quantum coherence . recently several authors have theoretically studied spontaneous emission in a trapped bec . in this case the continuous center - of - mass momentum distribution leads to an increase of se @xcite . in addition , the stimulated emission can be increased by the bose enhancement in a bec @xcite . in the case of two becs , interference effects can be important @xcite . the present work is focused on the case of an extended bec and was motivated by the following idea . if an extended bec is placed in an optical lattice it will become periodic . since such a bec does provide a ( quantum ) dielectric it affects the properties of photons propagating through it and phenomena similar to pbgs do occur @xcite . since then the photon mode density around the resonance frequency is reduced one can expect that se is suppressed by non - markovian effects . we thus are led to the following question : can a bec in an optical lattice stabilize itself against spontaneous emission ? a large part of this paper is devoted to the answer of this question . however , we also have studied what will happen for a homogeneous extended bec interacting with a weak running laser beam . surprisingly , non - markovian effects similar to that in a pbg do occur even in this non - periodic situation . this happens because in the presence of such a bec photons and excited atoms do form superpositions called polaritons @xcite . the spectrum of these polaritons contains an avoided crossing which has a similar effect on the se rate as a pbg . the paper is organized as follows . in sec . [ tmodel ] we will present the theoretical model on which our calculations are based . the general derivation of the se rates in laser fields will be done in sec . [ generalresult ] . the results for the case of a bec in a traveling wave laser beam or in a 1d optical lattice beam are discussed in secs . [ resultsrun ] and [ resultsstand ] respectively , and are summarized in sec . [ conclusion ] . the details of the calculations are given in two appendices . we consider a bec composed of two - level atoms which is coupled to the electromagnetic field . the interaction is described by using minimal coupling in rotating - wave approximation under neglection of the term quadratic in the electromagnetic field . the interaction hamiltonian is then given by @xmath3 with @xmath4^{1/2}$ ] for an electromagnetic mode with frequency @xmath5 and polarization vector @xmath6 . the vector @xmath7 denotes the atomic dipole moment and @xmath8 is the resonance frequency . the heisenberg equations of motion for the photon annihilation operators @xmath9 and the field operators @xmath10 and @xmath11 for excited and ground - state atoms can be derived easily and are given by @xmath12 we have neglected the interatomic interaction terms . to address the question of self - stabilization consider the following situation : the atoms in the internal ground - state have formed a bec which is described by a macroscopically occupied coherent collective wavefunction @xmath13 . they interact with a traveling wave or standing wave laser which is described by a coherent c - number field @xmath14 . due to this interaction a part of the bec is coherently excited . we denote the wavefunction for coherently excited atoms by @xmath15 . since both the ground - state bec and the laser beam are described by c - number fields it is easy to see from eq . ( [ edgl ] ) that @xmath15 must be a c - number field , too . it is only through the spontaneous decay of these coherently excited atoms that q - number deviations from c - number solutions to eqs . ( [ edgl ] ) to ( [ adgl ] ) can appear . the corresponding se rate determines the stability of the macroscopic solution . let us start with the assumption that the bec can indeed stabilize itself against se . in that case a stationary macroscopic solution @xmath16 of eqs . ( [ edgl ] ) to ( [ adgl ] ) should exist . the problem then can be divided into two separate parts . we first search for the stationary macroscopic coherent solution which includes all interaction effects between atoms and photons beside se . having found this solution we can perform a stability analysis to analyze the quantum fluctuations ( se ) around it . spontaneous decay will make the coherent solution unstable and the corresponding quantum corrections will become important on a time scale comparable to the atomic lifetime ( which is to be calculated ) . for times shorter than this lifetime the deviations from the coherent solution will be small ( i.e. , there are only few non - condensed atoms and non - laser photons ) . given a stationary macroscopic solution of the heisenberg equations of motion the stability analysis can be performed by applying bogoliubov s method . this is done by writing the quantum field operators in the form @xmath17 \ { \psi_g^{\mbox{{\scriptsize coh } } } ( \vec{k } ) + \delta \psi_g(\vec{k } ) \ } \label{fluctg } \\ \psi_e(\vec{k } ) & = & \exp[-i ( \mu+\omega_l ) t ] \ { \psi_e^{\mbox{{\scriptsize coh}}}(\vec{k } ) + \delta \psi_e(\vec{k } ) \ } \label{flucte } \\ a_\sigma(\vec{k } ) & = & \exp[-i \omega_l t ] \ { a_\sigma^{\mbox{{\scriptsize coh}}}(\vec{k } ) + \delta a_\sigma(\vec{k})\ } \label{flucta } \end{aligned}\ ] ] and retaining in eqs . ( [ edgl ] ) to ( [ adgl ] ) only terms linear in @xmath18 and @xmath19 , which describe the quantum fluctuations around the coherent solution . the resulting linearized equations of motions are given by @xmath20 here @xmath21 is the laser s frequency and @xmath22 its detuning . the linearized equations are valid as long as the photon - atom quantum fluctuations remain small enough , i.e. , there are only few non - condensed atoms and non - laser photons . this is certainly the the case for short times . furthermore several other approximations have been made . first , it is not difficult to see that for a bec of density @xmath0 @xmath1 the chemical potential @xmath23 , the kinetic energy @xmath24 of an atom , and the laser s rabi frequency @xmath25 are typically much smaller than the interaction energy @xmath26 if @xmath27 is of the order of @xmath28 . we thus have neglected all terms in which these quantities do appear . in addition , we have introduced two specific polarization vectors for the electromagnetic field . a `` non - coupled '' polarization vector @xmath29 , which is perpendicular to the photon momentum @xmath30 and the atomic dipole moment @xmath7 , and a coupled polarization vector @xmath31 , which is perpendicular to @xmath30 and @xmath32 ( see fig . [ polvecs ] ) . since the interaction is proportional to the scalar product of the polarization vector and @xmath7 only electromagnetic modes with polarization @xmath33 do interact with the atoms . we associate with these modes the quantum fluctuation operator @xmath34 . it is easy to see that the scalar product @xmath35 , which appears in the definition of @xmath36 , is given by @xmath37 , where @xmath38 is the angle between @xmath30 and @xmath7 . for notational convenience we have defined @xmath39 . it is possible to derive eqs . ( [ fludgle ] ) to ( [ fludgla ] ) from an effective hamiltonian for the quantum fluctuations , @xmath40 the first part , @xmath41 conserves the number of photons plus excited atoms , @xmath42 the first integral in @xmath43 describes the energy of free incoherent photons and atoms . the second integral represents the excitation of atoms from the ground - state bec and the reabsorption of incoherent photons by the bec . its eigenmodes @xmath44 are generally superpositions of photons and excited atoms , i.e. , polaritons @xcite . they are characterized by a continuous , momentum - like quantum number @xmath45 and discrete quantum numbers @xmath46 ( see below ) and can generally be written as @xmath47 the form of the expansion coefficients @xmath48 depends on the particular physical situation and is derived for a traveling and standing - wave laser in appendix [ polaritoneigenmodesrun ] and [ polaritoneigenmodesstand ] , respectively . the second part of the effective hamiltonian is given by @xmath49 it does not conserve @xmath50 and describes the spontaneous decay of coherently excited atoms . if this term would vanish the macroscopic coherent state would be stable against spontaneous decay . the stability analysis essentially comprises to solve the time evolution of the polariton modes for relatively short times during which the occupation of the macroscopic coherent solution does not change very much . this will allow us to derive the initial se rate of coherently excited atoms . we assume that initially all atoms and photons are in the state determined by the macroscopic coherent solution , or in other words , the quantized polariton field ( photon - atom quantum fluctuations ) is initially in the vacuum @xmath51 . this state then evolves under the action of the fluctuation hamiltonian ( [ hfluct ] ) into the time dependent state @xmath52 . to describe this time evolution we rewrite the polariton hamiltonian ( [ hpol ] ) in the convenient form @xmath53 where @xmath54 are the eigenfrequencies of @xmath43 . using eqs . ( [ linsupe ] ) and ( [ linsupa ] ) one also can derive @xmath55 with @xmath56 to describe the evolution of the state @xmath52 we make the following ansatz , which corresponds to the one - photon approximation , @xmath57 the schrdinger equation @xmath58 then can be solved by using the laplace transform @xmath59 r(t ) dt$ ] and similarly for @xmath60 . the resulting equations , @xmath61 have the solution @xmath62 the dependence on the particular physical situation is completely determined by the integral @xmath63 for notational convenience we have defined the complex variable @xmath64 the most important aspect of @xmath65 is its complex analytical structure . this is because the inverse laplace transform is defined by @xmath66 where @xmath67 is chosen so that the path of integration lies to the right of any branch cuts and poles of @xmath68 . from eq . ( [ lapltransf ] ) it becomes clear that the branch cuts of @xmath68 are those of @xmath65 and that the poles of @xmath68 essentially depend on the form of @xmath65 . assuming that all poles of @xmath68 are simple poles we then find for @xmath69 @xmath70 where @xmath71 denote the poles of @xmath68 and @xmath72 the branch cuts . each pole corresponds to a fraction of the coherently excited atoms which decays ( or increases exponentially ) with a se rate of @xmath73 the integration contours around the branch cuts corresponds to a fraction of coherently excited atoms with a non - exponential time evolution . with eq . ( [ poles ] ) we have found a general expression for the se rates that can appear in the presence of a bec . we now want to study the different physical situations of a bec in a traveling or standing wave laser and to derive the corresponding values of @xmath74 . to do so we have to find closed expressions for the polariton eigenfrequencies @xmath75 and the functions @xmath76 in order to derive @xmath77 . these quantities in turn require the knowledge of both the polariton eigenmodes and the fields @xmath78 comprising the macroscopic coherent solution . since the calculations leading to a closed expression for @xmath65 are quite involved we present them in the appendices . in the next two sections we analyse the results and give physical interpretations of the effects involved in the case of a bec interacting with a traveling wave laser , the polariton dispersion relation derived in appendix [ polaritoneigenmodesrun ] does contain an avoided crossing around the resonance frequency of the atoms ( see fig . [ avoidedcrossing ] ) . since nearly resonant photons provide the dominant contribution to se , it is physically evident that this avoided crossing in the dispersion spectrum will produce an effect on the se rate which is similar to what a pbg can do . this effect can be studied by analysing the closed expression for the renormalized value of @xmath77 which we have derived using the generalized wigner - weisskopf approximation presented in appendix [ derivationofirun ] . we find @xmath79 in this expression @xmath80 denotes the se rate in free space and @xmath81 the number of coherently excited atoms ( @xmath82 is the quantization volume ) . the frequency @xmath83 with @xmath84 being the density of atoms in the ground - state , determines the strength of the interaction between photons and excited atoms mediated by the bec . typically we have @xmath85 ( see appendix [ polaritoneigenmodesrun ] ) . as already mentioned in sec . [ generalresult ] the time evolution of the macroscopic coherent solution depends essentially on the analytical structure of @xmath68 of eq . ( [ lapltransf ] ) . in general , @xmath68 has several poles and a branch cut originating from the term including the arcoth in @xmath86 . this cut lies between @xmath87 and @xmath88 . another important property of @xmath86 is that it depends on @xmath89 only through the ratio @xmath90 so that the magnitude of the se modification depends on this ratio , too . in addition , a numerical evaluation of eq . ( [ iren ] ) shows that it is a slowly varying function of the order @xmath91 unless @xmath90 is close to zero . this has the following consequences . in free space @xmath89 can be taken to be close to the pole of @xmath68 , i.e. , of the order of @xmath91 . hence , the magnitude of non - markovian effects is essentially determined by the ratio @xmath92 this ratio is proportional to the total number of excited atoms divided by the number of condensed ground - state atoms per optical wavelength @xmath93 . the request that this ratio should be small has important consequences when applied to becs of a density in the order of @xmath0 @xmath1 . in this case the bec - induced effects can only be relevant if there are very few coherently excited atoms ( in the order of one ) . however , for higher densities significant effects can occur also for a higher number of excited atoms . the request that @xmath94 is of order one also implies @xmath95 since otherwise the rabi frequency would be large enough to excite many atoms . as discussed in appendix [ classicalsolutionrun ] , the macroscopic coherent solution implies in this case for @xmath96 the additional constraint @xmath97 . in fig . [ resultsfigrun ] the real part of the two dominating poles @xmath98 of @xmath77 is shown as a function of @xmath99 for the case @xmath100 and @xmath101 ( corresponding to an atom density @xmath84 of @xmath102 @xmath1 ) . for @xmath96 a third pole appears with a very small negative decay rate ( @xmath103 ) . the occurrence of negative decay rates is consistent within the range of validity of the linearized equations for the quantum fluctuations and may indicate the formation of an atom - photon bound state @xcite . obviously the change in the se rate can be quite large for small @xmath104 . according to eq . ( [ atsol ] ) the fraction of atoms belonging to the poles can be easily calculated by determining the residue at the poles . it turns out that the pole whose real part asymptotically approaches @xmath105 always dominates and that the fraction of atoms belonging to other poles is significant only for small @xmath104 . the same holds for the fraction corresponding to the branch cut . if @xmath106 holds the dominant pole @xmath107 can be calculated by perturbation theory . its real part ( the decay rate ) is then given by @xmath108 we see that the se rate is altered by a factor of @xmath109 . it depends on the sign of @xmath99 whether se is increased or decreased . we remark that the reason why se depends on the detuning is that the coherently excited atoms are driven by the laser field and thus oscillate at the laser frequency @xmath21 instead of the resonance frequency @xmath110 ( see eq . ( [ flucte ] ) ) . we shortly summarize the results that we have found for a bec interacting with a running laser beam . for evolution times smaller than the atomic lifetime and for a weak laser beam which only excites a number @xmath111 of excited atoms ( for bec densities of @xmath0 @xmath1 ) , the spontaneous emission rate is significantly modified by non - markovian effects . these effects result from an avoided crossing in the polariton spectrum caused by the extended homogeneous bec in the internal ground state . in the next section we will examine the corresponding results for a spatially periodic ( lattice ) bec in a standing wave laser beam . a bec in an optical lattice formed by a standing wave laser beam becomes spatially periodic and thus provides a kind of periodic dielectric . the polariton spectrum will therefore contain band gaps . we have derived the corresponding dispersion relation in appendix [ polaritoneigenmodesstand ] . as is well known from pbgs a band gap around the resonance frequency will lead to non - markovian effects in the spontaneous emission of an atom . the examination of these effects will answer the question whether a periodic bec in an optical lattice can stabilize itself against se . to determine the se rate of a bec in a 1d optical lattice we have again to evaluate the integral @xmath65 of eq . ( [ integral ] ) . this task is quite involved and is presented in appendix [ derivationofistand ] . our final analytical form of the renormalized integral @xmath86 is given by the somewhat lengthy expression @xmath112 + \right . \nonumber \\ & & \hspace{1.2 cm } \left . \int_0 ^ 1 v dv \ln \left [ \frac{(f_0(z_s)-v)(\sqrt{v^2 + 1}-f_0(z_s))+f_1(z_s ) } { ( f_0(z_s)-v)(1-v - f_0(z_s))+f_1(z_s ) } \right ] \right \ } + \nonumber \\ & & \frac{i\gamma_{\mbox{{\scriptsize vac } } } \tilde{n}_e}{\pi } { \tilde{\nu}_g\over z_s } \int _ 1^\infty \frac{\sqrt{v } dv}{f_0(z_s)-v } \bigg \ { ( v^2 + 1)^{1/4 } -(v^2 - 1)^{1/4 } - \nonumber \\ & & \hspace{2 cm } h(v)\mbox{arctanh } \left[\frac{(v^2 + 1)^{1/4}}{h(v ) } \right ] + h(v)\mbox{arctanh } \left[\frac{(v^2 - 1)^{1/4}}{h(v ) } \right ] \bigg \}+ \nonumber \\ & & \frac{i\gamma_{\mbox{{\scriptsize vac } } } \tilde{n}_e}{\pi } { \tilde{\nu}_g\over z_s } \int_0 ^ 1 \frac{\sqrt{v } dv}{f_0(z_s)-v } \bigg \ { ( v^2 + 1)^{1/4 } -(1-v)^{1/2 } - \nonumber \\ & & \hspace{2 cm } h(v)\mbox{arctanh}\left[\frac{(v^2 + 1)^{1/4}}{h(v ) } \right ] + h(v)\mbox{arctanh}\left[\frac{\sqrt{1-v}}{h(v ) } \right ] \bigg \ } \ ; . \nonumber \end{aligned}\ ] ] in this result we have introduced a couple of new notations . for notational convenience we have defined @xmath113 as well as the abbreviation @xmath114 . we also introduced a cut - off @xmath115 to regularize the integral ( @xmath116 is the electron s mass ) . two important physical quantities are given by @xmath117 where @xmath82 denotes the quantization volume and the sum runs over the ( real ) momentum components @xmath118 of coherently excited atoms . @xmath119 is simply the total number of excited atoms in the macroscopic coherent field , and @xmath120 describes how these atoms are distributed in momentum space and is always smaller than @xmath119 . it is a measure for the degree of periodicity of the density of excited atoms , very roughly we have @xmath121 . the influence of the bec in a standing wave laser on the se rate is determined by the frequencies @xmath122 in the polariton dispersion relation @xmath123 produces a contribution similar to that of @xmath124 in the case of a bec in a traveling wave laser beam ( avoided crossing ) . @xmath125 produces a real pbg close to the resonance frequency due to the spatial periodicity of a bec in a 1d optical lattice . the two frequencies define the strength of the interaction mediated by the mean density @xmath126 and the periodic part @xmath127 of the ground - state bec , @xmath128 these densities play a similar role to what @xmath129 and @xmath130 do for coherently excited atoms . for a mean density of @xmath131 @xmath1 we find @xmath132 hz . the magnitude of @xmath125 can vary between @xmath123 for a very strong optical potential and 0 if the laser beam is switched off . although eq . ( [ analyticalresult ] ) has a complicated structure it allows to analyze the main features of @xmath86 and hence of the time evolution of the macroscopic coherent solution in the presence of ( small ) quantum fluctuations . it is even possible to estimate the influence of the band gap with some simple arguments . a very important feature of the integral ( [ analyticalresult ] ) is that all parts of @xmath86 are proportional to @xmath133 or @xmath134 . in addition , it becomes obvious that @xmath86 depends on the ground - state bec and on the complex variable @xmath89 essentially through the ratios @xmath135 and @xmath136 . the only exception to this is the first term containing @xmath89 in eq . ( [ f0 ] ) , but this term is negligible compared to @xmath137 and does only serve to keep book on which side of the branch cut @xmath89 is placed ( see remarks below eq . ( [ freecontrib ] ) ) . these facts can be exploited to estimate under which circumstances the influence of the bec on the se rate is significant . since for a bec in a traveling laser beam the contribution of the poles of @xmath68 usually dominates ( see sec . [ resultsrun ] ) , we will focus on this part . the denominator of @xmath68 is of the form @xmath138 , see eq.([lapltransf ] ) . since a numerical analysis of eq . ( [ analyticalresult ] ) shows that @xmath139 is of the order of its pre - factors @xmath133 or @xmath140 unless @xmath141 or @xmath142 are small , a pole @xmath71 must be of the order of these pre - factors ( if the detuning @xmath99 is not very large ) . in analogy to the case studied in sec . [ resultsrun ] one can again infer that the magnitude of the bec - induced effects essentially depends on ratios of the form @xmath143 , for instance . as in sec . [ resultsrun ] this allows the conclusion that for a bec with a density in the order of @xmath0 @xmath1 non - markovian effects can only be relevant if there are very few coherently excited atoms ( in the order of one ) . in the case of a bec in an optical lattice this restriction has additional implications : the rabi frequency @xmath144 of the coherent standing wave laser has to be very small since otherwise too many atoms would be excited . the number of excited atoms is approximately given by @xmath145 , where @xmath146 denotes the total number of condensed ground - state atoms . since @xmath146 is very large the ratio @xmath147 must be very small in order to achieve @xmath148 . this , in turn , means that the optical potential ( @xmath149 ) provided by the standing laser beam is very weak and thus @xmath150 is much smaller than @xmath151 . a small value of @xmath150 simply means that the polaritonic band gap that is formed in the presence of a bec will be small and therefore will not have a significant effect on the se rate . another observation deals with the dependence of @xmath86 on the wavefunction @xmath15 of coherently excited atoms . it is known that in free space the shape of the spatial wavefunction of an excited atom does only have a tiny influence on its se rate @xcite . these small corrections are mainly due to the atomic kinetic energy which we have neglected in the hamiltonian ( [ hfluct ] ) for the quantum fluctuations . in this sense , one would expect that the se rate in eq . ( [ analyticalresult ] ) does also not depend on the shape of the wavefunction for coherently excited atoms and is proportional to total number @xmath119 of excited atoms in which this shape does not enter . however , eq . ( [ analyticalresult ] ) does also include the terms proportional to the quantity @xmath120 which clearly depend on the shape ( for instance , @xmath120 vanishes for a spatially homogeneous wavefunction @xmath15 ) . principally , this contribution can be as large as that depending on @xmath119 . this new dependence on the shape of @xmath15 is an additional effect of the bec on the se rate and no consequence of the polaritonic dispersion relation . to understand its origin it is useful to look at eq . ( [ betasum ] ) where @xmath120 appears first . this contribution obviously does vanish if @xmath152 or @xmath153 , which are momentum - components of the photon modes belonging to momentum @xmath45 and @xmath154 , is zero . this is the case for photons interacting with a homogeneous bec , for instance . the new effect therefore can be considered as arising from the interference between different momentum - components of the photon modes and the wavefunction of coherently excited atoms . we want to emphasize that this effect is not tied to the presence of a bec . the only conditions for its existence are the periodicity of both the wavefunction @xmath15 for excited atoms and the eigenmodes for the photons . since in an ordinary pbg material the photon eigenmodes are periodic , this new contribution to the se rate can be present in ordinary pbg materials , too . in an ordinary periodic dielectric the new interference channel for se even could produce large contributions since the periodicity of the dielectric is produced by , e.g. , mechanical forces but not by the light that is used to excite the atoms . only in the case of a bec do the standing laser beams play a double role , excitation of atoms and production of a periodicity in the bec , which results in a suppressed influence of both the polaritonic band gap and the interference channel on the se rate . we conclude this paper by summarizing the results that we have found . we have examined the self - stabilization of a bec against se by performing a stability analysis of a macroscopically occupied state for photons and two - level atoms , which describes a bec that is coherently coupled to a laser beam . the presence of the ground - state bec thereby leads to the formation of polaritons and introduces non - markovian effects in the spontaneous decay of excited atoms . in the case of a bec in a traveling - wave laser , the polariton spectrum displays an avoided crossing around the resonance frequency which causes similar changes in the se rate as a pbg in periodic dielectrics . its magnitude depends on the ratio @xmath155 between the total number of excited atoms @xmath94 and the number of bec - atoms inside a cube of the size of an optical wavelength @xmath156 . if this ratio is much larger than 1 the se rate will essentially remain unchanged . otherwise the change can be significant as the numerical examples shown in fig . [ resultsfigrun ] demonstrate . the change of the se rate displayed in fig . [ resultsfigrun ] depends on the detuning of the laser because the coherently excited atoms are driven at the laser s frequency @xmath21 . for a bec in a 1d optical lattice two new effects do appear . being a kind of periodic dielectric the bec then produces a real polaritonic band gap . the size of this band gap is determined by @xmath125 . as in the case of a traveling wave laser , se is only significantly altered if there are very few excited atoms . this in turn does imply that the optical lattice must be very weak and therefore produces only a small band gap which has only a very small influence on the se rate . the second new effect in a periodic bec is the appearance of a new channel for se which arises from the interference between different momentum components of the excited - state wavefunction and the photon modes . though its effect in a bec is as small as that of the band gap it should also be present in the case of a pbg in an ordinary periodic dielectric where it can be large . it should be pointed out clearly where exactly the difference between an ordinary periodic dielectric and a bec in an optical lattice comes into play . in an ordinary dielectric medium the periodicity is produced by whatever forces determine the stability of the medium . the excitation of an atom inside such a medium is done by a light beam , i.e. , a completely different physical system . in the case of a bec in an optical lattice , however , the excitation of the atoms and the potential that produces the periodicity of the bec both are provided by the same device : the laser beams of the optical lattice . these lattice beams have to achieve two competing goals : to provide a strong periodic potential ( to produce a large band gap ) and to cause only a weak excitation ( to have few excited atoms ) . as the achievement of both goals is impossible the periodicity of the bec will only have a tiny influence on the se rate and it will essentially cause the same effect as a homogeneous bec in a running laser wave . this argument also provides the answer to the question whether self - stabilization of a bec against se is possible . since the se rate is only significantly changed if there are very few excited atoms , and since a large pbg does only form for strong laser beams , a self - stabilization is not possible for becs with a density in the order of @xmath0 @xmath1 . we finally remark that our results are not applicable to becs confined in a micrometer - sized trap , a case discussed in the literature @xcite . our work is concerned with becs which are extended enough to allow the formation of polaritons . the necessary extension of the bec can be estimated by considering the typical interaction energy for the formation of polaritons which is given by @xmath157 ( see appendix [ polaritoneigenmodesrun ] ) . for a bec with a density of @xmath0 @xmath1 this energy is in the order of @xmath158 hz . for the formation of polaritons a photon must therefore be inside the bec longer than @xmath159 seconds . since it travels at the speed of light the bec must therefore be larger than about 3 mm . * acknowledgment * : this work has been supported by the australian research council and the optik zentrum konstanz . we are interested in finding a particular solution @xmath160 of macroscopically occupied fields to eqs . ( [ edgl])-([adgl ] ) which describes a bec coherently coupled to a running laser wave . we thus make the ansatz @xmath161 \label{cohg}\\ a_\sigma^{\mbox{{\scriptsize coh}}}(\vec{k } ) & = & \exp [ -i \omega_l t ] \delta(\vec{k}-\vec{k}_l ) \delta_{\sigma,\sigma_l } \omega_l [ 2(2\pi)^3 \hbar \varepsilon_0 \omega_{k_l}]^{1/2}/(|\vec{d}| \omega_{\mbox{{\scriptsize res } } } ) \label{coha}\\ \psi_e^{\mbox{{\scriptsize coh } } } ( \vec{k } ) & = & ( 2\pi)^{3/2 } \sqrt{\rho_e } \delta(\vec{k}-\vec{k}_l ) \exp [ -i ( \mu+\omega_l ) t ] \label{cohe } \end{aligned}\ ] ] which corresponds to a homogeneous ground - state bec of density @xmath84 , a laser beam with frequency @xmath21 , rabi frequency @xmath162 , polarization @xmath163 , and wave - vector @xmath164 ( _ inside _ the bec ) , and coherently excited atoms of density @xmath165 and of momentum @xmath166 . inserting these expressions into the heisenberg equations of motions leads to a set of algebraical conditions which fix the chemical potential @xmath23 , the laser wavenumber @xmath167 , and the density of coherently excited atoms @xmath165 which we assume to be smaller than @xmath84 . if we neglect the kinetic energy the density of excited atoms is given by @xmath168 . the wavenumber @xmath167 is fixed by @xmath169 , where @xmath124 is defined in eq . ( [ nug ] ) . note that @xmath167 generally is different from the free - space value @xmath170 . for @xmath171 ( @xmath172 ) the chemical potential is given by @xmath173 which implies @xmath174 ( @xmath175 ) , respectively . note that for @xmath96 the expression for @xmath176 implies the additional constraint @xmath177 . having found the macroscopic coherent solution we are in the position to to derive the polariton eigenmodes because of the delta distribution appearing in the macroscopic solution ( [ cohg ] ) the hamiltonian ( [ hpol ] ) reduces to a sum of two - level systems so that its eigenmodes are quite easy to find . they consist of polaritons with momentum @xmath178 and frequency spectrum @xmath179 where we have defined @xmath180 and @xmath181 . the coefficients of the polariton creation operator in eq . ( [ linsup ] ) are given by @xmath182 so that we find for eq . ( [ gqr ] ) @xmath183 the polariton spectrum @xmath184 clearly exhibits an avoided crossing around @xmath185 of width @xmath186 ( see fig . [ avoidedcrossing ] ) . it also contains a small gap whose edge is reached in the limit @xmath187 and @xmath188 @xcite . this gap will play no role for the se rate since its edges are far away from the resonance frequency . this is physically reasonable since far away from resonance the two - level approximation for the atomic internal structure ceases to be valid . with the help of eq . ( [ gqrrun ] ) the integral @xmath77 of eq . ( [ integral ] ) can be written in the form @xmath189 where @xmath82 denotes the quantization volume . this integral agrees with the one found in absence of a bec ( which describes se in free space ) by setting @xmath190 . we denote this free space integral by @xmath191 . both integrals are linearly divergent and can be treated in the way pointed out by bethe ( see , e.g. , ref . @xcite ) , i.e. , we renormalize the integrals by subtracting the free - electron contribution , @xmath192 and @xmath193 . these renormalized integrals are only logarithmically divergent . at this point it is customary in the calculation of the free - space se to perform the wigner - weisskopf approximation by neglecting the dependence of @xmath194 on @xmath195 . in the presence of a band gap this is inappropriate due to the strong variation of the mode density around the gap @xcite . nevertheless , we can perform a generalized wigner - weisskopf approximation in the following way . we expect that the typical timescale on which se happens is much larger than the optical cycle timescale @xmath196 . from the definition of the inverse laplace transform ( [ inverselapl ] ) it is clear that the variable @xmath195 plays more or less the role of a fourier - transformed time . we thus expect that only values of @xmath195 much smaller than @xmath8 contribute significantly to the se . this implies that we can neglect ( the imaginary part of ) @xmath195 wherever it appears together with @xmath8 or @xmath21 . thus , we are allowed to set @xmath197 in the denominator of @xmath77 while retaining the term depending on @xmath198 . in the case of @xmath199 this procedure immediately reproduces the wigner - weisskopf result @xmath200 , where @xmath105 and @xmath94 are defined in sec . [ resultsrun ] . to fix @xmath201 we follow the theory of bethe ( see , e.g. , @xcite ) and introduce a cut - off frequency of @xmath202 in @xmath199 , where @xmath116 is the electron s mass . calculating the principal value of the integral then leads to @xmath203 . in contrast to free space the se rate in a bec depends on @xmath201 since such a radiative frequency correction shifts the center of the avoided crossing ( or of a band gap @xcite ) . it remains to calculate a renormalized expression of the integral @xmath86 in the presence of a bec . fortunately , this task reduces to integrals proportional to @xmath199 and a couple of convergent integrals and leads to eq . ( [ iren ] ) since in a running laser wave the bec density is not periodic no pbgs can be formed . it is therefore of interest to study a bec interacting with a standing laser wave so that the formation of photonic , or rather polaritonic band gaps @xcite , is possible . the coherent laser field describing a standing wave is given by @xmath204 \ ; . \label{classem}\ ] ] we assume that the amplitude @xmath205 is real and that the polarization @xmath206 of the laser beam is parallel to the dipole moment @xmath7 of the atoms . since the laser field provides a periodic potential for the atoms it is reasonable to assume that the macroscopic atomic fields are periodic , too ( at least for the ground - state of the system ) . one also can make the ansatz that @xmath13 has period @xmath207 so that the coherent solutions can be written as @xmath208 inserting this into eqs . ( [ edgl ] ) to ( [ adgl ] ) leads to the matrix equations @xmath209 we have assumed that @xmath210 is real and does not depend on the sign of @xmath30 and introduced the real rabi frequency @xmath211 . for consistency with the assumption that @xmath205 is a real quantity the coefficients @xmath212 and @xmath213 must be real , too . the system ( [ classa])-([classg ] ) of algebraic equations can easily be solved numerically . to do so we assume that the rabi frequency of the laser beam is a given quantity . for a given value of @xmath167 the two equations ( [ classe ] ) and ( [ classg ] ) then just describe the well - known problem of a two - level atom moving in a standing laser wave . this is a simple system of linear equations and can be solved in a standard manner . the solution then can be inserted into eq . ( [ classa ] ) which then , because @xmath21 and @xmath205 are fixed , determines the value of @xmath167 . we then have reinserted the new value for @xmath167 into the system ( [ classa])-([classg ] ) and iterated the procedure until @xmath167 did not change significantly anymore . in this case the periodic structure of the macroscopic solution ( [ becg ] ) leads to a more complicated structure of the eigenmodes of @xmath214 than in the case of a traveling laser wave . to find these modes we make in eq . ( [ linsup ] ) the ansatz @xmath215 and correspondingly for @xmath216 . now @xmath45 denotes the quasi - momentum of the polariton . for a single standing laser wave along the z - axis @xmath217 is confined to @xmath218 $ ] whereas @xmath219 represent the real momentum of the polariton perpendicular to the laser beam . the index @xmath46 is a collective notation for discrete quantum numbers which include an internal quantum number taking two values ( since two quantum fields @xmath220 and @xmath221 are involved ) and the band index . this ansatz leads to @xmath222 and results in the eigenvalue equations @xmath223 these equations can be substantially simplified by noting that the frequency difference @xmath224 is huge compared to all other energy scales involved unless @xmath225 and @xmath45 is close to @xmath226 . we thus can approximate the photon - part of all modes with @xmath227 as free photons and need only to retain the coefficients @xmath152 and @xmath153 for @xmath228 $ ] and the coefficients @xmath152 and @xmath229 for @xmath230 $ ] , respectively . we will focus here on the case @xmath228 $ ] since the second case can be treated analogously . to solve the resulting equations we introduce the two quantities @xmath231 and @xmath232 and make the approximation @xmath233 so that the problem is reduced to the simple matrix eigenvalue equation @xmath234 here we have introduced @xmath235 and the quantities @xmath126 and @xmath127 which are defined in eqs . ( [ barrhog ] ) and ( [ tilderhog ] ) , respectively . the problem of finding the eigenvalues and eigenvectors of a @xmath236 matrix is a basic one . the eigenvalues @xmath237 fulfill the relation @xmath238 , where @xmath239 is the characteristic polynomial of the matrix and the frequencies @xmath240 and @xmath241 are defined in eqs . ( [ barnug ] ) and ( [ tildenug ] ) . due to the periodicity of the ground - state bec @xmath75 as a function of the quasi - momentum @xmath45 exhibits the phenomenon of band gaps @xcite , see fig . [ bandgap ] . though closed expressions for the eigenvalues @xmath75 do exist they are rather cumbersome and not of much use for our problem . we instead will use a theorem on the eigenvalues to derive the physical quantities of interest . to do so we assume that we already know the eigenvalues . for a given eigenvalue @xmath75 it is easy to solve for the eigenvectors . for the relevant components we find @xmath242 the normalization has been done by requiring the particle number ( [ npol ] ) to be pseudo - normalized , i.e. , @xmath243 the analogous equation for @xmath244 $ ] is obtained if @xmath153 is replaced by @xmath229 and @xmath245 by @xmath246 . to calculate the integral @xmath77 of eq . ( [ integral ] ) we first need to calculate the integral over @xmath30 in the nominator which is an easy task . with eq . ( [ gqrstand ] ) we find for @xmath228 $ ] @xmath247 if @xmath46 is in the lowest two energy bands . the numbers @xmath119 and @xmath120 are defined in eqs . ( [ barne ] ) and ( [ tildene ] ) , respectively . for @xmath230 $ ] and the two lowest energy bands one has to replace @xmath153 by @xmath229 and @xmath248 by @xmath249 . for all higher bands , which according to our approximation just describe free photons , the integral ( [ betasum ] ) has the simple value @xmath250 . from eq . ( [ betasum ] ) it becomes clear that to calculate @xmath65 of eq . ( [ integral ] ) we have to find closed expressions for terms like @xmath251 . for the lowest two energy bands , the sum over @xmath46 now runs over the four eigenvectors of the matrix ( [ 44matrix ] ) . it would be extremely tedious if not practically impossible to derive these sums by simply inserting the complicated closed expressions for @xmath252 into them . instead , we start with the observation that the polynomial appearing in the denominator of eqs . ( [ beta0 ] ) and ( [ beta1 ] ) can be written as @xmath253 where @xmath254 denotes the derivative of the characteristic polynomial ( [ pch ] ) . this enables us to write the sum under consideration in the form @xmath255 this can be further simplified by noting that the characteristic polynomial can also be written in the form @xmath256 so that we have @xmath257 . using this expression it is straightforward if still tedious to find @xmath258 we thus have been able to calculate this sum without explicit knowledge of the eigenvalues of eq . ( [ 44matrix ] ) . the remaining sums which result from the insertion of eq . ( [ betasum ] ) into eq . ( [ integral ] ) can be treated in a similar way and are given by @xmath259 for the lowest two energy bands and @xmath260 $ ] . again the corresponding expressions for @xmath244 $ ] are obtained by replacing @xmath261 by @xmath262 and @xmath245 by @xmath263 . taking everything together the use of eq . ( [ betasum ] ) and the three sums ( [ sum01 ] ) , ( [ sum00 ] ) , and ( [ sum11 ] ) allows us to bring the integral ( [ integral ] ) into the form @xmath264 \displaystyle % \nonumber \\ & & \frac{\bar{n}_e}{i\hbar^2 } \int_{|q_z|>2k_l}dq_z \int_{-\infty}^\infty dq_x dq_y \frac{\zeta(\vec{q})^2 } { z_s - c|\vec{q}| + \omega_{\mbox{{\scriptsize res } } } } + \\[5 mm ] \displaystyle % \label{integ1 } \\ & & \frac{z_s}{i\hbar^2 } \int_{v_- } \frac { \bar{n}_e\zeta(\vec{q})^2 [ z_s^2-z_s \delta_{1}-\bar{\nu}_g \omega_{\mbox{{\scriptsize res } } } ] + \bar{n}_e\zeta(\vec{q}+2\vec{k}_l)^2 [ z_s^2-z_s \delta_{0}-\bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] + 2\tilde{n}_e \tilde{\nu}_g\omega_{\mbox{{\scriptsize res } } } \zeta(\vec{q})\zeta(\vec{q}+2\vec{k}_l ) } { z_s^4-z_s^3(\delta_0+\delta_{1})+z_s^2(\delta_0\delta_{1}-2 \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ) + z_s \bar{\nu}_g \omega_{\mbox{{\scriptsize res}}}(\delta_0+\delta_{1 } ) + \bar{\nu}_g^2\omega_{\mbox{{\scriptsize res}}}^2 -\tilde{\nu}_g^2 \omega_{\mbox{{\scriptsize res}}}^2 } + \\[5 mm ] \displaystyle % \nonumber \\ & & \frac{z_s}{i\hbar^2 } \int_{v_+}d^3q \frac { \bar{n}_e\zeta(\vec{q})^2 [ z_s^2-z_s \delta_{-1}-\bar{\nu}_g \omega_{\mbox{{\scriptsize res } } } ] + \bar{n}_e\zeta(\vec{q}-2\vec{k}_l)^2 [ z_s^2-z_s \delta_{0}-\bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] + 2\tilde{n}_e \tilde{\nu}_g\omega_{\mbox{{\scriptsize res } } } \zeta(\vec{q})\zeta(\vec{q}-2\vec{k}_l ) } { z_s^4-z_s^3(\delta_0+\delta_{-1})+z_s^2(\delta_0\delta_{-1}-2 \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ) + z_s \bar{\nu}_g \omega_{\mbox{{\scriptsize res}}}(\delta_0+\delta_{-1 } ) + \bar{\nu}_g^2\omega_{\mbox{{\scriptsize res}}}^2 -\tilde{\nu}_g^2 \omega_{\mbox{{\scriptsize res}}}^2 } \end{array } % \nonumber \label{integ1}\ ] ] where the areas of integration @xmath265 are given by @xmath266 and @xmath228 $ ] for @xmath267 as well as @xmath244 $ ] for @xmath268 . the first integral in eq . ( [ integ1 ] ) represents the contribution from the higher polaritonic energy bands where the polaritons can be considered as free photons . apart from the restriction @xmath269 , which essentially means that the photons are far off - resonant , it has the same form as the integral @xmath270 which appears in the calculation of the free - space se rate . the second contribution arises from the two lowest energy bands for negative quasi - momentum @xmath271 , and the third integral represents the corresponding contribution for positive quasi momentum @xmath217 . it is not hard to see that eq . ( [ integ1 ] ) reduces to the free integral ( [ i0 ] ) in absence of a ground - state bec , i.e. , for @xmath272 and to eq . ( [ ihomog ] ) if the bec is homogeneous ( for @xmath273 but @xmath274 ) . we now return to the evaluation of eq . ( [ integ1 ] ) . shifting the integration variable @xmath217 and exploiting the symmetries @xmath275 and @xmath276 allows us to combine the last two integrals into a more convenient form . to simplify the process of renormalization it is also advantageous to calculate @xmath277 instead of @xmath65 alone since in this difference the divergence appearing in @xmath278 is canceled . we then find @xmath279 + 2\tilde{n}_e \tilde{\nu}_g\omega_{\mbox{{\scriptsize res}}}\zeta(\vec{q})\zeta(\vec{q}+2\vec{k}_l ) } { [ z_s^2 -z_s \delta_0 - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] [ z_s^2 -z_s \delta_{-1 } - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] - \tilde{\nu}_g^2\omega_{\mbox{{\scriptsize res}}}^2 } \nonumber \end{aligned}\ ] ] to calculate these expressions we have to make one further approximation by neglecting the angular dependence in @xmath280^{1/2}$ ] , where @xmath281 is the angle between @xmath45 and the atomic dipole moment @xmath7 . replacing @xmath282 by @xmath283 for all values of @xmath45 leads to the correct result in absence of a bec and should produce qualitatively correct results for the case under consideration . doing this approximation in the case of a homogeneous bec just amounts in replacing a factor of @xmath284 by @xmath285 in the modifications of the se rate ( [ homogexpan ] ) . we remark that neglecting the dependence of @xmath286 on @xmath282 does only symmetrize the integrand in the @xmath287 plane . this does not correspond to an isotropic band model because the asymmetry between @xmath217 and @xmath287 still persists . the calculation of the first integral of eq . ( [ integ2 ] ) , which roughly corresponds to the contribution of @xmath288 , is quickly done and results in @xmath289 here @xmath290 is a cut - off which usually is taken to be @xmath291 , where @xmath292 is the electron s mass ( see , e.g. , ref . @xcite ) . the dependence on the sign of im(@xmath89 ) originally comes from a logarithm of the form @xmath293 . this expression can be reduced to the one presented in eq . ( [ freecontrib ] ) by doing a generalized wigner - weisskopf approximation as it was introduced above in the case of a running laser wave . though im(@xmath89 ) is also much smaller than any other quantity in the above logarithm , it determines the sign of the imaginary part of the logarithm s argument . since the logarithm has a branch cut along the negative real axis , this sign determines on which side of the cut we are . it is also worth remarking that eq . ( [ freecontrib ] ) is logarithmically divergent with @xmath290 although we did not subtract the free - electron part , a step which in the free - space calculation is done to remove a linearly divergent contribution ( see , e.g. , ref . this is because the integration over @xmath217 does not extend to infinity , thus reducing the degree of divergence by one . as has been already mentioned , eq . ( [ freecontrib ] ) very roughly corresponds to the negative of the free - space integral @xmath278 . as a consequence , its contribution will be mostly canceled after the renormalization of @xmath65 . this renormalization is easily done by noting that @xmath294 so that @xmath295 , where the superscript `` ren '' denotes the renormalized integrals and the renormalized free - space integral is approximately given by @xmath296 this allows us to derive from eq . ( [ integ2 ] ) the expression @xmath297 } { |\vec{q}| \left \ { [ z_s^2 -z_s \delta_0 - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] [ z_s^2 -z_s \delta_{-1 } - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] - \tilde{\nu}_g^2 \omega_{\mbox{{\scriptsize res}}}^2 \right \ } } + \nonumber \\ & & \frac{z_s\tilde{n}_e \tilde{\nu}_g \gamma_{\mbox{{\scriptsize vac}}}c^3 } { 2i\pi^2 } \int_0^{2k_l}dq_z \int dq_xdq_y \frac{|\vec{q}|^{-1/4 } |\vec{q}+2 \vec{k}_l|^{-1/4 } } { [ z_s^2 -z_s \delta_0 - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] [ z_s^2 -z_s \delta_{-1 } - \bar{\nu}_g\omega_{\mbox{{\scriptsize res } } } ] - \tilde{\nu}_g^2\omega_{\mbox{{\scriptsize res}}}^2 } \nonumber \end{aligned}\ ] ] to reduce eq . ( [ mainresult ] ) it is convenient to introduce the scaled variables of integration @xmath298 and @xmath299 and the abbreviations ( [ f0 ] ) and ( [ f1 ] ) for the evaluation . ( [ mainresult ] ) then becomes @xmath300 } { \sqrt{u^2+v } \{[f_0(z_s)-\sqrt{u^2+v } ] [ f_0(z_s ) -\sqrt{(u-1)^2+v } ] -f_1(z_s ) \ } } + \nonumber \\ & & \frac{\gamma_{\mbox{{\scriptsize vac}}}\tilde{n}_e}{4\pi i } { \tilde{\nu}_g\over z_s } \int_0 ^ 1 du \int_0^\infty dv \frac{(u^2+v)^{-1/4 } ( ( u-1)^2+v)^{-1/4 } } { [ f_0(z_s)-\sqrt{u^2+v } ] [ f_0(z_s ) -\sqrt{(u-1)^2+v } ] -f_1(z_s ) } \nonumber \end{aligned}\ ] ] this expression can be further reduced by switching to the integration variable @xmath301 and exchanging the sequence of integration so that one first integrates over @xmath302 . this allows us to reduce the integral @xmath65 to a number of one - dimensional integrals and leads us to our final analytical result ( [ analyticalresult ] ) . 99 p. goy , j.m . raimond , m. gross , and s. haroche , phys . lett . * 50 * , 1903 ( 1983 ) . s. john and t. quang , a * 50 * , 1764 ( 1994 ) . kofman , g. kurizki , and b. sherman , j. mod . opt . * 41 * , 353 ( 1994 ) . m. anderson , j.r . ensher , m.r . matthews , c.e . wieman , and e.a . cornell , science * 269 * , 198 ( 1995 ) ; c.c . bradley , c.a . sackett , j.j . tollet and r. hulet , phys . lett . * 75 * , 1687 ( 1995 ) ; m .- o . mewes , m.r . andrews , n.j . van druten , d.m . kurn , d.s . durfee , c.g . townsend and w. ketterle , phys . lett . * 77 * , 416 ( 1996 ) . b.v . svistunov and g.v . shlyapnikov , sov . jetp * 71 * , 71 ( 1990 ) ; h.d . politzer , phys . a * 43 * , 6444 ( 1991 ) . j. javanainen , phys . lett . * 72 * , 2375 ( 1994 ) . l. you , m. lewenstein , r.j . glauber , and j. cooper , phys . rev . a * 53 * , 329 ( 1996 ) . j.j . hope and c.m . savage , phys . a * 54 * , 3177 ( 1996 ) . savage , j. ruostekoski , and d.f . walls , phys . a * 56 * , 2046 ( 1997 ) . marzlin and w. zhang , phys . a. * 59 * , 2982 ( 1999 ) . k. rzazewski and w. zakowicz , journ . b * 25 * , l319 ( 1992 ) . milonni , _ the quantum vacuum _ , academic press , san diego 1994 . s. john and j. wang , phys . lett . * 64 * , 2418 ( 1990 ) .

we present a study of the variations in the absorbing column density of 25 x - ray defined seyfert 2 galaxies , as inferred from hard x - ray observations , on timescales from months to several years . we show that a significant variation of n ( from 20% to 80% ) is observed in almost all ( 22/25 ) the sources with multiple x - ray observations , although x - ray absorption never vanishes . for a subsample of 11 sources observed at least five times the typical variation time , as defined by a structure function , is less than one year for both heavily absorbed ( n ) and moderately absorbed ( n ) sources . these variations rule out the simplest version of the unified models , based on a homogeneous obscuring torus , and suggest the presence of clumpy circumnuclear material on a scale well below a parsec . we propose a modification of the torus model in which an overabundance of slightly dusty belr clouds obscures the belr . the belr needs , like the torus , to have an axisymmetric structure . this model is closely related to that of elvis ( 2000 ) for type 1 agn . for lightly obscured agn ( n ) the structure function shows an increase at a timescale of yr , indicating a second absorber , most probably on a 5 - 10 pc scale associated with the host galaxy .

astro-ph0107510

much observational evidence suggests that strong obscuration absorbs the strong continuum source in active galactic nuclei over a significant fraction of the solid angle ( lawrence & elvis 1982 , antonucci & miller 1985 , maiolino et al . the effects of this absorption are clearly visible in the x - ray spectra of many agns , where a photoelectric cutoff at energies of a kev or so ( depending on the column density of the gas ) is observed , and in the mid - infrared , where we observe the thermal re - emission of the absorbed uv radiation from the dust associated with the absorbing gas.this circumnuclear material obscures the optical emission of the broad line region , but not that of the more distant narrow line region . as a consequence , objects observed along a line of sight covered by this material appear in the optical as type 2 agns . the simplest geometry for this cold gas surrounding the nucleus , is that of a torus covering @xmath5% of the solid angle ( antonucci & miller 1985 , krolik & begelman 1989 ) . axial symmetry in the material in the central @xmath4 100 pc of nearby agns is suggested by the biconical shapes ( pogge 1989 ) seen well in high resolution hst images ( tadhunter & tsvetanov 1989 , malkan et al . 1998 ) , and polarization requires a highly non - spherical shape ( antonucci & miller 1985 ) . one of the unsolved questions about the structure of this putative torus is its typical dimensions . detailed models of homogeneous circumnuclear tori , both on the 1 pc and 100 pc scale , reproduce the observed infrared sed ( pier & krolik 1992 , granato & danese 1994 ) . hard x - ray observations show that at least 50% of nearby seyfert 2s are obscured by a column density higher than 10@xmath6 @xmath2 ( risaliti et al . 1999 ) . in these cases it is unlikely that the typical dimensions of a homogeneous torus exceed a few parsecs , since otherwise the dynamical mass of the obscuring material would be too large ( risaliti et al . 1999 ) . however , detailed observations both in the x - rays and in the near ir / optical band suggest that the structure of the circumnuclear material is more complex than a simple homogeneous torus . large variations in the cold obscuring column density have been known in a few objects for many years ( ives et al . 1976 , warwick et al . 1988 ) . a more systematic study by malizia et al . ( 1997 ) , shows that variations of several 10@xmath7 @xmath2 are rather common in both type 1 and type 2 seyferts ( 17 out of 23 sources observed at least 3 times shows n@xmath0 variations ) . changes in months or less suggest much smaller sizes , whether the observed change is due to motions across the line of sight , or to a varying ionization state . the broad emission line region ( belr ) has been suggested as a site for this variable absorption in type 1 agns ( ives et al . 1976 ) . since the belr is either hidden or not present in type 2 agn this explanation appears implausible for type 2 objects , which comprise 80% of all agn ( maiolino & rieke 1995 ) . in this paper we investigate the variability of the x - ray absorbing column density in x - ray - defined seyfert 2s having column densities higher than @xmath8 @xmath2 , but less than 10@xmath6 @xmath2 , i.e. compton thin . this includes optically defined both types 1.8 and 1.9 objects ( which show broad lines only in long wavelength lines , e.g. h@xmath9 , pa@xmath9 , osterbrock 1989 ) as well as type 2 seyferts , and all show evidence for cool absorbing material . we collected all the data available in the literature for seyfert 2s and we complemented them with the analysis of unpublished data in the asca and bepposax public archives . we found that a sample of 25 sources were observed at least twice in the hard x - rays . 11 objects out of these 25 have 5 or more hard x - ray observations in a time interval of several years , and we can therefore obtain for them an `` n@xmath0 light curve '' . in the next section we show that variations in n@xmath0 are almost universally present . we analyze in greater detail the well - observed 11 sources , dividing them in two subsamples with n@xmath10 @xmath2 ( 6 objects ) and n@xmath3 @xmath2 ( 5 objects ) . previous x - ray studies show that sources with 10@xmath11 @xmath12n@xmath13 @xmath2 are mostly type 1.8 and 1.9 ( risaliti et al . 1999 ) while those with n@xmath14 @xmath2 are mostly pure seyfert 2s . this dichotomy is also present in our sources : 3 out of 5 objects in the low - n@xmath0 subsample are intermediate type , while 5 out of 6 in the high - n@xmath0 subsample are type 2s ( table 1 ) . in section three we analyze the n@xmath0 variability of the well studied subsample using a `` structure function '' similar to that used in studies of brightness variability of quasars ( fiore et al . 1998 , di clemente et al . 1996 ) . in section four we discuss our results and we compare them with current models of the circumnuclear medium of agns . our conclusions are summarized in section 5 . in table 1 we list all 139 measurements of n@xmath0 ( and 2 - 10 kev flux ) in the literature ( with references ) for all the 25 seyfert 2s with n@xmath15 @xmath2 that have been observed at least twice . quite a fraction ( 17 % ) of the observations have not previously been published , and their inclusion strengthens our results significantly . where bepposax or asca measurements are not published , we measured the n@xmath0 values through a standard analysis of the data from the bepposax sdc public archive and the asca public archive at heasarc . more details of this analysis will be given in risaliti ( 2001 , in preparation , see also appendix a ) . for this and subsequent analysis , we used the relevant data from every past x - ray mission . we note that this should not give problems of intercalibration , since the column density is evaluated from the energy of the steep photoelectric cutoff in an x - ray spectrum . indeed , while the flux calibration between x - ray instruments is notoriously uncertain , the calibration in energy is quite good , especially at energies higher than 1 - 2 kev , where the instrumental responses do not vary rapidly , and where cut - offs for the n@xmath0 range of these seyfert 2s are found . figure 1 shows the factor _ f _ by which the column density changes in the sample objects . _ f _ is defined as the ratio between the measured variation of n@xmath0 and the mean of the two values . when more than two observations are available , we chose according to both the size of the variation and the statistical significance : if at least two measurements were different at a confidence level higher than 90% , we chose those with the largest factor variation . otherwise , we selected the measurements with the most statistically significant variation . the main result of our analysis is clear from fig . 1 : a significant column density variation on a timescale of years is not a peculiar property of a few sources , but is virtually ubiquitous in seyfert 2s : only 3 of the 25 seyfert 2s are consistent with no variation . these have only 2 or 3 observations in the hard x - rays so that variations typical of well - observed sources could easily be missed . in figure 2 we plot the relative n@xmath0 and relative 2 - 10 intrinsic kev flux for each observation of the well observed subsample . both the measurements are normalized to the average values for each source . the intrinsic flux has been obtained using the best fit values for each observation , and assuming no absorption . interestingly , no correlation between variations of n@xmath0 and flux appear . if photoionization plays an important role in n@xmath0 variations , we would expect that n@xmath0 values below the average of a given source ( values @xmath16 1 in fig . 4 ) are associated to flux measurements above the average ( values @xmath17 1 ) , and vice versa . therefore , the regions with @xmath18 , @xmath19 and @xmath20 , @xmath21 ( top - right and bottom - left quadrants ) should be more populated than the regions @xmath18 , @xmath21 and @xmath20 , @xmath19 ( top - left and bottom - right quadrants ) . instead , the density of points in the four regions is about the same , as expected if flux and n@xmath0 variations are uncorrelated . moreover , if photoionization effects were important for an object , we would expect a correlation among the points relative to this object . instead , all the 11 groups of points in fig . 4 ( each of which is marked with a different symbol ) appear to be randomly distributed in the n@xmath0-flux plane . as a further confirmation , we quantitatively estimated the effects of photoionization assuming two different values of the ionization parameter , u=0.1 and u=0.5 , at the inner edge of the absorber . we assumed that the absorber has a total column density n@xmath22 @xmath2 , distributed on a region @xmath23 times thicker that the distance of the inner edge from the center . these parameters are quite extreme for a `` standard '' absorber on the parsec scale . they are reasonable if we assume that the absorber is located in the broad emission line region ( belr ) . we believe this is a likely possibility , as we show in the following sections . in any case , this scenario can be regarded as extreme in the sense that within it photoionization plays a more important role than in standard torus models . the result we obtained is that fitting the photoionized gas with a cold absorption model , the equivalent cold n@xmath0 changes only of @xmath23% if the ionization parameter changes from u=0.1 to u=0.5 ( a strong continuum variation ) . therefore , we conclude that even strong variations in the continuum are not able to cause apparent variations of cold n@xmath0 higher than a few 10@xmath7 @xmath2 . this further rules out photoionization as a cause of the observed absorption variations . in order to analyze in greater detail the timescale and the nature of the observed n@xmath0 variations , we focused on a subsample of 11 bright seyfert 2s with multiple observations in the hard ( 2 - 10 kev ) x - ray band . conveniently these 11 sources are divided into two groups : 5 ( marked ` a ' in table 1 ) have n@xmath24 @xmath2 , while the other 6 ( marked ` b ' ) have n@xmath1 @xmath2 . this division is in rough agreement with the optical classification of type 1.8/1.9 for low - n@xmath0 objects and type 2s for high - n@xmath0 objects ( table 1 , risaliti et al . 1999 ) the `` light curves '' of n@xmath0 are shown for samples a and b in figures 3 and 4 , respectively . the increasing or decreasing of n@xmath0 is not correlated with the instruments used for the measurements , reinforcing the conclusion that the effect is not due to problems in instrumental calibration . figures 3 and 4 seem to show variations on a variety of timescales from months to years . in table 2 we list the fastest change in n@xmath0 for each object in the full sample , detected at 90% significance or greater . the objects are sorted according to the observed variation timescale . we note that the shortest variation times are observed in the most extensively studied objects : only 3 out of the 11 sources with 5 or more x - ray observations show variations only on timescales of several years . therefore , the observed n@xmath0 variability timescales in many objects are only a conservative upper limit : the fastest variations are not detected only because the time delays between the existing observations are too long . the data for the two well observed subsamples can be used to obtain a `` column density structure function '' , defined by analogy with the intensity structure function used to study optical and x - ray variability of qsos ( di clemente et al . 1996 , fiore et al . we considered 10 time intervals , @xmath25 , for the low n@xmath0 subsample `` a '' and 11 for the high n@xmath0 subsample `` b '' , from 1 to 20 years . we defined a _ structure function _ , @xmath26 , as the average ratio of all the n@xmath0 measurements separated by a time between @xmath25 and t@xmath27 for each set of 5 sources : @xmath28 where _ i _ goes from 1 to 5 in subsample a and from 1 to 6 in subsample b , and labels the source ; _ j _ labels the n@xmath0 measurements for each source , while the sum over _ l _ is extended to data n@xmath0(l ) for which the time delay from n@xmath0(j ) is between @xmath25 and @xmath29 . @xmath30 is the total number of pairs of measurements separated by a time between @xmath25 and @xmath29 . the width of the time intervals was chosen in order to have at least 15 data points per bin . since the ratio in the above equation is always greater than 1 , the value is biased positive . following di clemente et al . ( 1996 ) we remove this bias by defining a modified structure function as follows : @xmath31 where @xmath32 is the average of the statistical errors on the ratios in eq the function @xmath33 is plotted in fig . 5a for the low - n@xmath0 subsample ( a ) and in fig . 5b for the high - n@xmath0 sample ( b ) . we removed the points of signal - to - noise less than 3@xmath34 to avoid an unphysical decrease of @xmath33 due to noise . we also used only one of the high 5 signal - to - noise points provided by bepposax observations of the bright source centaurus a. these observations show a low n@xmath0 variation ( although significant at a level higher than 90% ) , but lower than that measured in average in the other observations . including all the 5 observations would bias our results since the statistics would be dominated by the single source centaurus a. from fig . 5a and 5b and fig . 2 we can draw the following conclusions : 1 . the variability is statistically highly significant for both the low - n@xmath0 and the high - n@xmath0 objects at all the timescales considered ( 90 days - 20 years ) . 2 . for both samples the variability is already significant at the shortest timescale that can be investigated with our data ( several months ) . this result is also confirmed by the data in table 2 for to the whole sample : sources with measured variations in a time shorter than 1 year are @xmath4 50% of the whole sample , rising to @xmath470% of the subsample of sources for which we actually have observations within a time interval of less than 1 year . the modified structure function is constant ( within statistical errors ) for the high n@xmath0 sample , while for the low n@xmath0 sources it seems to have a significant increase for @xmath35 5 years . this suggests a characteristic timescale of @xmath4 5 years for variations in the order of 10@xmath11 @xmath2 , since the structure function is expected to increase with time when a slower variation is convolved ( randomly ) with faster ones . 4 . despite the large n@xmath0 variability observed , no object has been observed to change from type 1 to type 2 ( i.e. n@xmath0 never drops below several 10@xmath7 @xmath2 ) 5 . variations of n@xmath0 are not correlated with x - ray flux variations ( fig . in the previous section we demonstrated that n@xmath0 variability in a timescale from months to years is present in the great majority , and possibly in all , seyfert 2 galaxies . the structure function described in the previous section is useful to quantify the general n@xmath0 variability properties of our sources . a physical interpretation of our results must explain both the shape of this function and the different properties of the individual sources . there are two physical reasons that can explain the variability of the absorbing column density : variations in the ionization state of the absorber , due to variations in the ionizing radiation , and variations in the amount of absorbing gas along the line of sight . in the first case , the absorber can be homogeneous , and the variation in n@xmath0 should be correlated with intrinsic flux variations . in the second case , the absorber must be clumpy , and the variation timescales will be correlated with the typical crossing time of an absorbing cloud along the line of sight . however we already showed that a change in ionization does not fit the data ( section 2 , fig . 2 ) we can therefore adopt the second scenario -motions in a clumpy medium- and use the information collected in the previous sections to estimate the distance from the center of the obscuring gas . we can idealize the situation by assuming the typical timescale of variation , _ t _ , to be the crossing time of a discrete cloud across the line of sight . assuming that the absorption is due to spherical clouds moving with keplerian velocities , the distance from the central black hole of mass _ m _ is given by : @xmath36 where @xmath37 is the density of the cloud . the black hole mass and the cloud density have been normalized to extreme values for a putative torus ( krolik & begelman 1989 ) in order to obtain the greatest distance . the black hole mass m@xmath38 m@xmath39 is obtained from mass measurements of central black holes in nearby galaxies ( kormendy & richstone 1998 ) . the typical average density @xmath40 in a standard torus can be evaluated as the ratio between the column density and the thickness of the torus . objects of subsample b have n@xmath41 several 10@xmath42 @xmath2 , then @xmath43 @xmath44 . given the clumpiness of the torus , the actual density of a cloud can be higher , however the value @xmath45 @xmath44 can be regarded as an upper limit . even though most of the physical parameters in the equation above are poorly constrained , several conclusions can nevertheless be drawn . first , the shorter timescale n@xmath0 variations ( @xmath4 2 month or less ) must be due to material that is nearer to the center than about 10@xmath46 cm , while the radius of the standard model torus is 1 - 3 pc ( krolik & begelman 1992 ) . the only physical parameter that can reasonably be larger , assuming different physical conditions , is the cloud density : for example , photoionization models suggest that belr clouds have @xmath47 @xmath44 ( netzer 1990 ) . even in this case though it is hard to imagine a reasonable geometry with distances larger than @xmath48 cm , since this would imply that the absorbing material is confined in a thin slab whose depth is less than 10@xmath49 of the distance from the center , but covering a large solid angle . these parameters are even better constrained for two of the objects in our sample for which a measurement of the black hole mass is available : centaurus a ( m@xmath50 m@xmath39 , marconi et al . 2001 ) and ngc 4258 ( m@xmath51 m@xmath39 , miyoshi et al . 1995 ) . using the fastest n@xmath0 variations observed ( which are only upper limits to the actual variation timescales ) we obtain : @xmath52 and @xmath53 these results pose severe problems to the standard torus model , according to which the x - ray absorption is due to cold gas distributed in a toroidal geometry on the parsec scale . an alternative possibility , within the standard agn model ( antonucci 1993 ) is that the x - ray absorber is located in the broad emission line region , much nearer to the central black hole than the `` standard '' torus . an indication for an absorber on the belr scale is also provided by recent high resolution soft x - ray spectra obtained with xmm - newton ( sako et al . if we assume that the broad line clouds are responsible for the absorption in the x - rays , we can find a consistent combination of the parameters in eq . 1 . for example , assuming @xmath54n@xmath1 @xmath2 , m=10@xmath55 m@xmath39 , @xmath56 @xmath44 , typical for seyfert galaxies ( netzer 1990 ) , and t=5 days we obtain r@xmath57 cm , @xmath4 3 light days , a distance typical for the belr based on reverberation mapping ( peterson et al . our point can be graphically illustrated as in fig . 6 , where we show that our observed variability is not compatible with a parsec - scale torus , while a variability of @xmath4 1 day could be in agreement with an absorber located in the belr . the typical column density of the obscuring clouds also lies in the regime of belr clouds . this variability consequently requires that the circumnuclear absorber is not homogeneous , but clumped into several clouds . assuming gaussian fluctuations , the average number of clouds along the line of sight , n@xmath58 , should be of the order @xmath59 . the right hand axis of fig . 1 shows this number to be of order a few for most of the sample . in no case though , even for n@xmath60 , did n@xmath0 drop below 5@xmath61 @xmath2 , well above the lowest detectable value for these observations therefore , despite the large variability in n@xmath0 , there are no changes from type 2 to type 1 . from fig . 1 we estimate that the average column density of the single clouds , n@xmath62 , range from a few 10@xmath7 @xmath2 in low - n@xmath0 objects , up to several 10@xmath42 @xmath2 in high n@xmath0 objects . we report the estimates of n@xmath63 in the last column of table 2 . the n@xmath63 distribution is plotted the histogram in fig . however , the standard model for agns fails to explain x - ray absorption through the broad line clouds because it would predict several unseen features . photoionization models require that in type 1 agns the covering factor of the broad line clouds ( assuming an isotropic distribution around the central black hole ) is typically 10% and certainly much lower than 100% . we would then expect that ( 1 ) 10% of type 1 agns would be x - ray absorbed , and ( 2 ) , repeated observations of the same object would reveal , in 10% of the cases , an x - ray absorbed spectrum . at present about 20 bright seyfert 1s have been extensively studied in the x - rays for several years , by means of several x - ray observatories ( turner & pounds 1989 , nandra & pounds 1989 , nandra et al . 1997 ) , and these effects have been observed only in one case ( ngc 3516 , costantini et al . 2000 ) . in the optical , where many more objects have been extensively observed , there are only few known cases of transition from type 1 to type 2 or vice versa ( ngc 7603 , tohline & osterbrock 1976 ; mkn 1018 , cohen et al . 1986 ; ngc 7582 , aretxaga et al . 1999 ; for other cases see references in aretxaga et al . we conclude that the standard agn model is not able to reproduce the observed variability of x - ray absorption in seyfert 2 galaxies . from the discussion above we conclude that the observed x - ray variability must be explained by an absorber with the following properties : ( 1 ) close to the central engine ( unless moving faster than keplerian ) , ( 2 ) clumped ( to explain the n@xmath0 variability , with n@xmath64 ) , ( 3 ) covers a significant fraction , @xmath65 , of the solid angle ( in order to reproduce the 4:1 ratio between seyfert 2s and seyfert 1s ) , and ( 4 ) since agn rarely change from type 2 to type 1 , the absorber can not be spherically symmetric so that the 80% of lines of sight covered must always remain the same . a straightforward extension of the standard unified model that can reproduce the right x - ray absorption properties is to suppose that in seyfert 2s the belcs are , somewhat paradoxically , _ much more numerous _ than in seyfert 1s , so that the covering factor is 100% and on average several clouds cover the line of sight . we can repeat the argument used above to rule out the parsec - scale torus , to constrain the variability timescales within this model : assuming that the absorber is made by belcs ( @xmath66 @xmath44 ) moving with keplerian velocity , we can obtain a consistent distance from the center only assuming variations on timescales of days . assuming a typical velocity of 5000 km s@xmath67 for the belcs , the crossing time of a cloud is @xmath68 days , and the distance from the center is r@xmath69 cm . rewriting equation 3 for a belr absorber , and using a typical n@xmath0 variation for the high - n@xmath0 subsample , we have : @xmath70 such a model is novel because would ascribe the differences between type 1 and type 2 agns to physical properties , rather than to orientation effects . orientation is reintroduced because the x - ray absorber can not be spherically distributed around the center ( because of the shortage of objects changing from type 1 to type 2 and vice versa , as outlined in the previous section ) . the simplest possible geometry for a non - spherically symmetric absorber is axisymmetric : a bi - cylinder or a bi - cone . this geometry is appealing because the same structure has been recently proposed by elvis ( 2000 ) in order to explain the absorption and scattering properties of type 1 agns . according to this model , some disk instability generates a two - phase wind from a narrow range of disk radii : the warm phase , with temperatures of the order of @xmath71 k , is responsible for most of the scattering phenomena observed in agns , and for both the uv narrow absorption lines and the x - ray warm absorbers seen in a significant fraction of type 1 agns by the hubble space telescope ( crenshaw et al . 1999 ) and asca ( reynolds 1998 ) . the cold phase of the wind is formed by the belcs ( figure 8) . a simple extension of this model could be that in type 2 agns the belcs confined the warm wind are far more common and completely cover all lines of sight , i.e. on average more than one cloud is present along each line of sight . the fraction of matter in the different phases of multi - phase media , e.g. the ism , is not predictable with current theory , and may be history dependent . some alteration in the seed medium produced by a change in the disk instability at least provides a plausible site for such a bi - modal state to originate . a variation of this is that a larger radial thickness of the wind might produce the dichotomy between type 1 and type 2 agn . in this case , the average distance of the center of the x - ray absorber could be significantly larger than the distance of the belr , since the belcs will be only those located close to the inner edge of the wind ( fig . indeed , indications of a large thickness of the belr come also from reverberation mapping studies , according to which low ionization lines , as mgii @xmath722800 , are emitted by a region several times farther from the center than that emitting high ionization lines like heii @xmath721640 and nv @xmath721240 ( peterson 1993 ) . an interesting prediction of this model is that the change from type 2 to type 1 could occasionally happen , when the line of sight is freed from clouds , because of random motions . in these cases , broad lines in the optical and soft x - rays outburst would be observable . the duration of these breaks in the clouds is of the same order of the crossing time of a cloud along the line of sight . assuming the parameters used in eq . 2 , and a column density for a single cloud of @xmath8 @xmath2 , we can estimate this time to be t@xmath4 1 day , too short to be caught with anything but the most intense monitoring . the probability @xmath73 that this happens depends on the average number of clouds n@xmath58 along the line of sight . assuming n@xmath58=4 and n@xmath58=10 ( two plausible values , see fig . 1 ) we have p@xmath41% and p@xmath4 0.01% , respectively . a detailed study of this case will be presented in nicastro et al . 2001 ( in preparation ) . a problem with this `` many clouds '' model comes from to the optical emission of seyfert 2s . agns that are obscured only by the broad line cloud gas should not be obscured in the optical continuum , but only in the resonant lines themselves . but seyfert 2s continua are weak ( koski 1978 ) . instead seyfert 2s clearly suffer extinction due to dust ( peterson 1997 ) . some dust must then be present in the belcs of type 2 agn . certainly the outer layers of clouds are heavily shielded from the ionizing continuum and could be the locus of dust formation , although the timescales required in our case are probably too short . another possibility is that the material inflowing from the host galaxy and feeding the agn is dusty , and part of the dust is transferred to the external part of the wind , that is much thicker and farther from the center than in the `` optically thin '' model of elvis 2000 . therefore , part of the dust could survive until it reaches the wind . the amount of inflowing material could be the physical quantity that tunes the thickness of the wind , thus determining the type 1 or type 2 classification . this possibility is interesting , given the growing evidence suggesting a low dust to gas ratio in the absorbing medium of many agns ( maiolino et al . 2000 , risaliti et al . a dust poor absorber , located close to the central black hole , has already been suggested by granato et al . these authors show that the infrared emission of seyfert 2s is best reproduced assuming a low dust - to - gas ratio , and suggest that most of the x - ray absorber is located inside the agn sublimation radius . our `` intrinsic '' model for the difference between type 1 and type 2 agn also explains the otherwise puzzling lack of strong infrared dust emission observed in most seyfert 1s . edelson et al . ( 1987 ) show that only about 1/3 of seyfert 1s show strong dust contribution to infrared emission in iras . this observational fact is hard to include in the standard unified models : seyfert 1s should have the same obscuring tori as seyfert 2s , and therefore their infrared emission should be of the same order of that of seyfert 1s , with respect to the bolometric intensity . the model we proposed predicts that orientation apply only to objects with `` many '' clouds , that are type 2 if seen through the wind and type 1 ( with high ir dust emission ) if seen pole - on . objects with `` few '' clouds are type 1 along every line of sight , and are not expected to be strong ir emitters . finally , this scenario is appealing because it predicts a link between the absorbing column density and the amount of material inflowing from the host galaxy , in agreement with the finding that heavily absorbed agns are preferentially hosted in strongly barred galaxies ( maiolino et al . 1999 ) . within this model we can estimate the fraction @xmath74 of objects with `` many '' clouds and the average opening angle , @xmath9 , of the wind ( fig . 8) . using the ratio 4:1 between type 2s and type 1s ( maiolino et al . 1995 ) and the fraction 1/3 for the dust emitting type 1s ( edelson et al . 1987 ) we obtain @xmath75 and @xmath76% . we note that the opening angle of the wind in the model of elvis 2000 is higher ( @xmath77 ) . however , the presence of a second absorber ( see next section ) can reduce this discrepancy . the increase in the n@xmath0 structure function for low n@xmath0 objects at t@xmath4 5 years is not explained by a `` many clouds '' , belr , origin for the cold x - ray absorber . instead an additional , second , absorber is needed , located much farther from the center . inhomogeneities in this absorber could be related to the long term variability on timescales of years revealed in our sample of objects with n@xmath3 @xmath2 . assuming a column density variation of 5@xmath78 @xmath2 ( typical for the long - timescale variations observed in the low - n@xmath0 subsample ) , we can parametrize the distance from the center of this second absorber as follows : @xmath79 this absorber can be compatible with the standard torus model ( r@xmath80 cm ) , in objects with a lower m@xmath81 or with a lower cloud density . the existence of a second absorber is needed by the belc - origin model , that , in the original version of elvis ( 2000 ) implies a covering factor of @xmath4 0.5 and so can not reproduce the high ratio between seyfert 2s and seyfert 1s , which is known to be @xmath82 for nearby objects ( maiolino & rieke 1995 ) . moreover , strong suggestions of the existence of two distinct absorbers come from the analysis of the x - ray spectra of several nearby agns ( malaguti et al . 1999 , turner et al . 2000 , vignali et al . 1998 ) the orientations of the two absorbers are likely unrelated , since neither radio jet axes nor emission line bi - cones ( pogge 1989 ) align with galaxy minor axes ( ulvestad & wilson 1984 ) . there will then be objects obscured only by one of the two absorbers . this obviously would increase the fraction of obscured lines of sight . objects obscured only by the more distant medium can not have column density variations in timescales of days . this is in agreement with the measurements described in the previous sections . a consequence of this second absorber is that the ratio between free and wind - covered lines of sight can be higher than predicted in the previous section , since some fraction of the objects are type 2 because they are covered by the farther absorber . this would reduce the opening angle discrepancy between our model and that of elvis ( 2000 ) . we can give a quantitative estimate of this effect if we assume that intermediate type objects ( @xmath4 sample a ) are those observed through the outer torus , but not through the wind . we therefore have a new , lower , ratio between type 1.8 - 2 and type 1 - 1.5 objects , and an estimate of the covering factor of the outer obscurer from the ratio between intermediate and type 1 seyferts . with these number we calculate an half - opening angle of 35@xmath83 and a fraction of objects with `` many clouds '' @xmath74=75% . the half - opening angle is still significantly lower that in the model by elvis ( 2000 ) . however , we note that in the elvis model the wind is turned by radiation pressure . we expect that the wind in our model , more massive that that of elvis ( 2000 ) , is harder to turn by radiation pressure . in this paper we have shown that large amplitude ( 20 - 100% ) , rapid ( @xmath84 months ) variability of x - ray absorbing column density is an almost ubiquitous property of seyfert 2 galaxies . from our compilation of 25 objects with at least 2 observations in the hard x - rays ( 2 - 10 kev ) we found that in 21 cases the n@xmath0 varies at a confidence level greater than 90% . we demonstrated that the variations of n@xmath0 are not related to the underlying x - ray flux and so are not due to variations in the ionization state of the absorber . variations of the amount of absorbing gas along the line of sight is the only obvious alternative . our result requires the absorbing gas to be clumpy ( in order to reproduce the observed variability ) and its distribution around the black hole can not be spherically symmetric ( since changes from type 2 to type 1 are not observed in the x - rays ) . for a well observed subsample of 11 sources with at least 5 observations in the hard x - rays , the structure function shows that the typical variation timescales are shorter than several months . the data do not probe shorter timescales well , so this is an upper bound . the shortest measured variation is less than 1 year for 70% of the sources for which this can be measured . from the structure function of the low n@xmath85 @xmath2 ) sample we also find an indication of the presence of a larger scale absorber , responsible for n@xmath0 variations on timescales of the order of @xmath4 5 years . the fast variability is incompatible with the standard parsec - scale torus of unified models . in order to reproduce the observed x - ray properties , the absorbers have to be clumpy and close to central black hole ( distance r@xmath86 m@xmath81/m@xmath39 cm ) . we propose a model in which a superabundance of broad emission line region clouds produces the absorption . the bi - cylindrical geometry of elvis ( 2000 ) , in which the absorber is the cold phase of a wind arising from the accretion disk reintroduces orientation as a factor in whether we see a type 1 or a type 2 agn . the model we proposed also predicts that a change from type 2 to type 1 could occasionally be observable in the x - rays ( further details will be in nicastro et al . 2001 , in preparation ) . we found significant variability of n@xmath0 on all the timescales that we could investigate . moreover , the model we propose predicts n@xmath0 variability in timescales of days . therefore , this work can be usefully complemented by an analysis of the n@xmath0 variations at shorter timescales . several observations made with asca and bepposax have a sufficient statistics to look for n@xmath0 variations of timescales from @xmath87 seconds to 2 days . a more detailed study on this issue will be the subject of a forthcoming paper ( risaliti et al . 2001 , in preparation ) . this work has made use of data obtained through the science data center of the italian space agency and the high energy astrophysics science archive center ( heasarc ) on - line archive , provided by nasa - goddard space flight center . this work was supported in part by nasa grant nag5 - 4808 . . 22 measurements listed in table 1 ( 4 with asca and 18 with bepposax ) are previously unpublished . these data permit the inclusion in the sample of two heavily obscured objects ( ngc 4941 and ngc 3081 ) and are also important for the statistical significance of the structure function , since almost all the spectra have an high signal to noise , allowing n@xmath0 measurements among the most precise in all sample . - _ asca data _ : we only analyzed the two gis ( 0.7 - 10 kev ) observations , extracting a spectrum from a circular region of radius 4 centered on the source . the background was obtained from a free region in the same field . the data were rebinned in order to have at least 20 counts per channel . - _ bepposax data _ : we analyzed the data from the lecs ( 0.1 - 10 kev ) , mecs ( 1.65 - 10.5 kev ) and pds ( 20 - 200 kev ) instruments . we extracted a spectrum from the lecs and mecs data using a region equal to that used for the asca data . we used the pds spectrum provided by the sax science data center ( sdc ) . - _ models _ : we fitted the data with a baseline model composed by an absorbed power law plus a thermal component . in many cases , given the high signal to noise of the spectra , several featured are clearly non fitted . in order to obtain a good fit , we added extra components to the baseline model ( a gaussian emission line to fit the iron k@xmath9 emission at [email protected] kev , a cold reflection component , an high energy cutoff for sources observed with bepposax ) . we give a detailed description of these fits in another paper ( risaliti 2001 , in preparation ) . the important point for the measures used in this work is that in all cases we finally obtained a statistically acceptable fit , and the n@xmath0 measurements are not significantly affected by the details of the fit of the other spectral components . akylas , a. , georgantopoulos i. , & comastri a. 2000 , mnras , 324 , 521 aretxaga , i. , joguet , b. , kunth , d. , melnick , j. , & terlevich , r.j . 1999 , apj , 519 , 123 antonucci , r.r.j . , & miller , j. 1985 , apj , 297 , 621 antonucci , r.r.j . 1993 , ara&a , 31 , 473 awaki , h. , koyama , k. 1993 , adv . space res , 13 , 221 awaki , h. , ueno , s. , taniguchi , y. , weaver , k.a . 2000 , apj , 542 , 175 baity , w.a . , 1981 , apj , 244 , 429 bassani , l. , dadina , m. , maiolino , r. , salvati , m. , risaliti , g. , della ceca , r. , matt , g. , zamorani , g. 1999 , apjs , 121 , 473 cohen , r.d . , puetter , r.c . , rudy , r.j . , ake , t.b . , & foltz , c.b . 1986 , apj , 311 , 135 costantini , e. , et al . 2000 , apj , 544 , 283 crenshaw , d.m . , kraemer , s.b . , boggess , a. , maran , s.p . , mushotzky , r.f . , & wu , c.c . 1999 , apj , 516 , 750 lawrence , a. , & elvis , m. 1982 , apj , 256 , 410 di clemente , a. , giallongo , e. , natali , g. , trevese , d. , & vagnetti , f. 1996 , apj , 463 , 466 edelson , r.a . , malkan , m.a . , & rieke , g.h . 1987 , apj , 321 , 233 elvis , m. 2000 , apj , 545 , 63 fiore , f. , laor , a. , elvis , m. , nicastro , f. , & giallongo , e. 1998 , apj , 503 , 607 forster , k. , leighly , k.m . , & kay , l.e . 1999 , apj , 523 , 521 georgantopoulos , i. , & papadakis , i.e. 2000 , mnras , 322 , 218 gilli , r. , maiolino , r. , marconi , a. , risaliti , g. , dadina , m. , weaver , k.a . , & colbert , e.j.m . 2000 , a&a 355 , 485 granato , g.l . , danese , l. 1994 , mnras , 268 , 235 granato , g.l . , danese , l. , & franceschini , a. 1997 , apj , 486 , 147 halpern , j.p . 1981 , ph.d thesis , harvard university ives , j.c . , sanford , p.w . , & penston , m.v . 1976 , apj , 207 , l159 iwasawa , k. , fabian , a.c . , mushotzky , r.f . , brandt , w.n . , awaki , h. , & kunieda , h. 1996 , mnras , 279 , 837 iyomoto , n. , makishima , k. , & fukazawa , y. 1997 , pasj , 48 , 425 krolik , j.h . 1999 , `` active galactic nuclei : from the central black hole to the galactic environment '' , princeton university press krolik , j.h . , & begelman , m.c . 1988 , apj , 329 , 702 malizia , a. , bassani , l. , stephen , j.b . , malaguti , g. , & palumbo , g.g.c . 1997 , apj , 113 , 311 maiolino , r. , & rieke , g.h . 1995 , apj , 454 , 95 maiolino , r. , salvati , m. , bassani , l. , dadina , m. , della ceca , r. , matt , g. , risaliti , g. , & zamorani , g. 1998 , a&a , 338 , 781 maiolino , r. , risaliti , g. , & salvati , m. 1999 , a&a , 241 , l35 maiolino , r. , marconi , a. , salvati , m. , risaliti , g. , severgnini , p. , oliva , e. , la franca , f. , vanzi , l. 2001 , a&a , 365 , 28 malaguti , g. , bassani , l. , & caroli , e. 1994 , apjs , 94 , 517 malaguti , g. , , et al . 1999 , a&a , 344 , 857 malkan , m.a . , gorjian , m. , & tam , r. 1998 , apjs , 117 , 25 marconi , a. , capetti , a. , axon , d. , koekemoer , a. , macchetto , f.d . , & schreier , e.j . 2001 , apj , 549 , 915 miyoshi , m. , moran , j. , herrnstein , j. , greenhill , l. , nakai , n. , diamond , p. , & inoue , m. 1995 , nature , 373 , 127 morini , a. , f. , & molteni , d. 1989 , apj , 347 , 750 mulchaey , j.s . , colbert , e. , wilson , a.s . , mushotzky , r.f . , & weaver , k.a . 1993 , apj , 414 , 144 mushotzky , r.f . , serlemitsos , p.j . , becker , r.h . , boldt , e.a . , & holt , s.s . 1978 , apj , 220 , 790 nandra , k. , george , i.m . , mushotsky , r.f . , turner , t.j . , & yaqoob , t. 1997 , apj , 476 , 70 peterson , b.m . 1993 , pasp , 105 , 247 peterson , b.m . 1997 , `` an introduction to active galactic nuclei '' , cambridge university press . pier , e.a . , & krolik , j.h . 1992 , apj , 401 , 99 pogge , r.w . 1989 , apj , 345 , 730 reynolds , c.s . 1997 , mnras , 286 , 513 reynolds , c.s . , nowak , m.a . , & maloney , p.r . 1998 , apj , 540 , 143 risaliti , g. , maiolino , r. , & salvati , m. 1999 , apj , 522 , 157 risaliti , g. , maiolino , r. , & bassani , l. 2000 , a&a , 356 , 33 risaliti , g. , marconi , a. , maiolino , r. , salvati , m. , & severgnini , p. 2001 , a&a , 371 , 37 sambruna , r.m . , eracleous , m. , & mushotzky , r.f . 1999 , apj , 526 , 60 severgnini p. , risaliti g. , marconi a. , maiolino r. , & salvati m. 2000 , a&a , 368 , 44 singh , k.p . , garmire , g.p . , & nousek , j. 1985 , apj , 297 , 633 smith , d.a . , & done , c. 1996 , mnras , 280 , 355 stark , j.p.w . , a. , 366 , 435 tadhunter , c. , & tsvetanov , z. 1989 , nature , 341 , 422 tohline , j.e . , & osterbrock , d.e . 1976 , apjl , 210 , l17 turner , t.j . , & pounds , k.a . 1989 , mnras , 240 , 833 turner , t.j . , george , i.m . , nandra , k. , & mushotsky , r.f . 1997 , apjs , 113 , 23 turner , t.j . , perola , g.c . , fiore , f. , matt , g. , george , i.m . , piro , l. , & bassani , l. 2000 , apj , 531 , 245 ulvestad , j.s . , & wilson , a.s . 1984 , apj , 277 , 92 vignali , c. , comastri , a. , stirpe , g.m . , cappi , m. , palumbo , g.g.c . , matsuoka , m. , malaguti , g. , & bassani , l. 1998 , a&a , 331 , 411 warwick , r.s . , pounds , k.a . , & turner , t.j . 1988 , mnras , 231 , 1145 warwick , r.s . , pounds , k.a . , sembay , s. , yaqoob , t. , makishima , k. , ohashi , t. , tashiro , m. , & komhura , y. 1993 , mnras , 265 , 412 weaver , k.a : , ph.d thesis , university of maryland weaver , k.a . , arnaud , k.a . , & mushotsky , r.f . 1995 , apj , 447 , 121 wilkes , b.j . , mathur , s. , fiore , f. , antonelli , a. , & nicastro , f. 2000 , apj , 549 , 248 xue , s. , otani , c. , mihara , t. , cappi , m. , & matsuoka , m. 1998 , pasj , 50 , 519 1 & mkn 348 & 13.06.1987 & ginga & 106@xmath88 & 2.2 & 2 + & sy 2 & 04.08.1995 & asca & 160@xmath89 & 0.8 & 24 + 2 & ngc 526a@xmath90 & 23.05.1978 & heao 1 & 15@xmath91 & 1.5 & 6 + & sy 2/1.5 & 09.12.1978 & heao 2 & 17.5@xmath92 & 6.2 & 9 + & & 05.08.1983 & exosat & 21@xmath93 & 3.1 & 3 + & & 28.06.1985 & exosat & 13@xmath94 & 0.8 & 3 + & & 23.08.1985 & exosat & 25@xmath95 & 0.8 & 3 + & & 15.12.1988 & ginga & 7.6@xmath96 & 1.2 & 2 + & & 30.11.1995 & asca & 15@xmath97 & 7.1 & 4 + & & 01.01.1999 & sax & 16@xmath98 & 1.9 & 1 + 3 & ngc 1365 & 25.01.1995 & asca & @xmath99 & & 12 + & sy 1.8 & 12.08.1997 & sax & 400@xmath100 & 2.1 & 13 + 4 & ngc 1386 & 26.01.1995 & asca & 280@xmath101 & 0.065 & 12 + & sy 2 & 10.12.1996 & sax & @xmath102 & & 22 + 5 & iras 04575 - 7537 & 04.10.1990 & ginga & 12.3@xmath103 & 1.6 & 2 + & sy 2 & 04.11.1996 & asca & 10.5@xmath104 & 2.0 & 23 + 6 & ngc 1808 & 15.10.1990 & ginga & 105@xmath97 & 0.3 & 25 + & sy 2 & 26.02.1994 & asca & @xmath105 & 0.09 & 4 + 7 & iras 05189 - 2524 & 15.02.1995 & asca & 44@xmath106 & 0.6 & 10 + & sy 2 & 03.10.1999 & sax & 85@xmath107 & 0.7 & 10 + 8 & ngc 2110@xmath90 & 09.10.1978 & heao1 & 67@xmath108 & 3.3 & 6 + & sy 2 & 26.09.1989 & ginga & 22.8@xmath98 & 4.0 & 2 + & & 06.12.1990 & bbrxt & 25@xmath109 & 3.5 & 7 + & & 12.03.1990 & asca & 26.5@xmath110 & 3.2 & 4 + & & 14.10.1997 & sax & 34@xmath111 & 3.9 & 8 + 9 & ngc 2992@xmath90 & 24.05.1978 & heao 1 & 4.0@xmath112 & 9.4 & 32 + & sy 1.9 & 22.11.1978 & heao 2 & 6.4@xmath113 & 10.0 & 9 + & & 02.06.1979 & heao 2 & 14.5@xmath114 & 2.8 & 9 + & & 06.06.1980 & heao 2 & 10@xmath115 & 6.8 & 9 + & & 09.06.1980 & heao 2 & 14.5@xmath116 & 6.1 & 9 + & & 18.12.1983 & exosat & 7@xmath117 & 1.3 & 9 + & & 06.04.1984 & exosat & 7.8@xmath118 & 4.5 & 9 + & & 09.11.1984 & exosat & 7@xmath119 & 3.7 & 3 + & & 30.04.1990 & ginga & 16.1@xmath120 & 1.57 & 2 + & & 30.06.1996 & asca & 7@xmath121 & 0.35 & 4 + & & 30.11.1997 & sax & 14@xmath122 & 0.7 & 14 + & & 30.11.1998 & sax & 9@xmath123 & 7.8 & 14 + 10 & mcg-5 - 23 - 16@xmath90 & 12.05.1978 & heao 1 & 50@xmath124 & 11.7 & 6 + & sy 2 & 23.06.1979 & heao 2 & 18@xmath125 & 10.2 & 9 + & & 31.12.1980 & heao 2 & 14.5@xmath122 & 9.6 & 9 + & & 04.01.1981 & heao 2 & 25@xmath126 & 7.4 & 9 + & & 13.12.1983 & exosat & 15@xmath106 & 7.5 & 3 + & & 23.04.1984 & exosat & 16@xmath106 & 5.9 & 3 + & & 30.11.1988 & ginga & 19.4@xmath127 & 4.7 & 2 + & & 05.12.1988 & ginga & 18.1@xmath128 & 2.2 & 2 + & & 27.11.1991 & rosat & 13.1@xmath129 & 5.6 & 15 + & & 11.05.1994 & asca & 16@xmath130 & 8.5 & 4 + & & 25.04.1998 & sax & 15.9@xmath131 & 10.5 & 1 + 11 & ngc 3081 & 13.05.1996 & asca & 500@xmath132 & 2.6 & 1 + & sy 2 & 20.12.1996 & sax & 640@xmath133 & 0.8 & 22 + 12 & ngc 4258@xmath134 & 15.05.1993 & asca & 136@xmath135 & 1.4 & 34 + & sy 1.9 & 23.05.1996 & asca & 92@xmath136 & 2.0 & 34 + & & 06.06.1996 & asca & 88@xmath137 & 2.2 & 34 + & & 28.12.1996 & asca & 97@xmath107 & 2.5 & 34 + & & 19.12.1998 & sax & 121@xmath137 & 1.4 & 1 + & & 17.05.1999 & asca & 95@xmath138 & 1.1 & 34 + 13 & ngc 4388 & 04.07.1993 & asca & 315@xmath139 & 3.2 & 20 + & sy 2 & 21.06.1995 & asca & 334@xmath140 & 1.5 & 20 + & & 09.01.1999 & sax & 380@xmath141 & 8.2 & 1 + & & 03.01.2000 & sax & 480@xmath142 & 2.8 & 1 + 14 & ngc 4507@xmath134 & 07.07.1990 & ginga & 490@xmath143 & 5.1 & 2 + & sy 2 & 12.02.1994 & asca & 343@xmath144 & 4.9 & 4 + & & 09.04.1997 & sax & 590@xmath145 & 5.7 & 1 + & & 19.02.1998 & sax & 540@xmath146 & 4.5 & 1 + & & 04.01.1999 & sax & 710@xmath147 & 1.6 & 1 + 15 & ngc 4941 & 19.07.1996 & asca & @xmath102 & & 1 + & sy 2 & 22.01.1997 & sax & 450@xmath148 & 0.3 & 22 + 16 & iras 13197 - 164 & 18.07.1995 & asca & 760@xmath149 & 0.6 & 11 + & sy 1.8 & 22.07.1998 & sax & 330 @xmath150 & 1.9 & 1 + 17 & centaurusa@xmath134 & 27.07.1975 & oso 8 & 139 @xmath136 & 148 & 27 + & sy 2 & 19.01.1976 & ariel 5 & 110@xmath151 & 112 & 28 + & & 01.08.1976 & oso 8 & 144@xmath152 & 79 & 27 + & & 15.01.1978 & heao 1 & 99@xmath109 & 78 & 29 + & & 15.03.1978 & ariel 5 & 160@xmath153 & 49 & 28 + & & 15.07.1978 & heao 1 & 95@xmath109 & 45 & 29 + & & 15.01.1979 & heao 2 & 126@xmath154 & 163 & 9 + & & 03.08.1979 & heao 2 & 213@xmath155 & 57 & 9 + & & 13.02.1984 & exosat & 164@xmath107 & 35 & 30 + & & 08.06.1984 & exosat & 140@xmath156 & 20 & 30 + & & 30.07.1984 & exosat & 170@xmath156 & 25 & 30 + & & 29.05.1985 & exosat & 144 @xmath157 & 108 & 30 + & & 14.08.1993 & asca & 100@xmath156 & 43 & 31 + & & 20.02.1997 & sax & 100@xmath121 & 34 & 1 + & & 06.01.1998 & sax & 93.0@xmath158 & 43 & 1 + & & 10.07.1999 & sax & 94.3@xmath159 & 40 & 1 + & & 02.08.1999 & sax & 93.6@xmath160 & 40 & 1 + & & 08.01.2000 & sax & 94.3@xmath161 & 39 & 1 + 18 & ngc 5252 & 28.01.1994 & asca & 43@xmath137 & 0.75 & 4 + & sy 1.9 & 20.01.1998 & sax & 68@xmath162 & 3.8 & 1 + 19 & ngc 5506@xmath90 & 20.01.1978 & heao 1 & 41@xmath163 & 6.8 & 6 + & sy 1.9 & 13.07.1979 & heao 2 & 38@xmath152 & 8.8 & 9 + & & 21.07.1979 & heao 2 & 48@xmath164 & 6.9 & 9 + & & 30.07.1979 & heao 2 & 43@xmath165 & 8.7 & 9 + & & 09.08.1979 & heao 2 & 43@xmath166 & 9.2 & 9 + & & 06.01.1981 & heao 2 & 43@xmath167 & 8.2 & 9 + & & 24.01.1986 & exosat & 28@xmath109 & 9.0 & 3 + & & 19.07.1988 & ginga & 29@xmath121 & 1.0 & 2 + & & 06.01.1992 & rosat & 21.9@xmath168 & 8.0 & 15 + & & 30.01.1997 & sax & 37@xmath169 & 7.5 & 1 + & & 14.01.1998 & sax & 39@xmath170 & 9.1 & 1 + 20 & iras 18325 - 5926 & 13.05.1989 & ginga & 16.3@xmath171 & 12.8 & 2 + & sy 1.8 & 11.09.1993 & asca & 13.2@xmath104 & 1 & 26 + & & 27.03.1997 & asca & 10.4@xmath131 & 2.5 & 1 + 21 & eso 103-g35@xmath134 & 04.09.1983 & exosat & 228@xmath172 & 1.8 & 18 + & sy 2 & 19.04.1984 & exosat & 247@xmath173 & 2.2 & 18 + & & 04.05.1985 & exosat & 237@xmath174 & 2.5 & 18 + & & 02.08.1985 & exosat & 71@xmath175 & 2.7 & 18 + & & 13.08.1985 & exosat & 81@xmath176 & 1.9 & 18 + & & 04.09.1985 & exosat & 167@xmath177 & 1.2 & 18 + & & 23.09.1988 & ginga & 176@xmath178 & 5.7 & 19 + & & 12.04.1991 & ginga & 292@xmath179 & 4.6 & 2 + & & 03.09.1994 & asca & 216@xmath180 & 3.5 & 20 + & & 26.09.1995 & asca & 168@xmath181 & 4.1 & 20 + & & 18.03.1996 & asca & 216@xmath182 & 6.2 & 20 + & & 03.10.1996 & sax & 179@xmath136 & 5.8 & 21 + & & 14.10.1997 & sax & 202@xmath183 & 2.4 & 21 + 22 & ic 5063 & 02.10.1990 & ginga & 252@xmath184 & 2.3 & 2 + & sy 2 & 25.04.1994 & asca & 256@xmath185 & 2.6 & 4 + & & 27.04.1994 & asca & 218@xmath186 & 2.7 & 4 + 23 & ngc 7172@xmath134 & 26.10.1989 & ginga & 112@xmath187 & 7.7 & 2 + & sy 2 & 28.10.1985 & exosat & 90@xmath188 & 3.2 & 3 + & & 11.05.1995 & asca & 82@xmath106 & 6.0 & 4 + & & 17.05.1996 & asca & 77@xmath166 & 2.1 & 1 + & & 14.10.1996 & sax & 90@xmath189 & 1.8 & 5 + & & 06.11.1997 & sax & 83@xmath109 & 0.9 & 5 + 24 & ngc 7314 & 05.09.1983 & exosat & 13@xmath190 & 1.3 & 3 + & sy 1.9 & 25.03.1984 & exosat & 8.2@xmath191 & 3.2 & 3 + & & 20.11.1994 & asca & 8.9@xmath192 & 3.6 & 4 + & & 08.06.1999 & sax & 12.2@xmath193 & 2.6 & 1 + 25 & ngc 7582@xmath134 & 14.12.1978 & heao 2 & 80@xmath194 & 9.6 & 9 + & sy 2 & 26.05.1979 & heao 2 & 97@xmath195 & 5.7 & 9 + & & 02.06.1979 & heao 2 & 126@xmath196 & 4.2 & 9 + & & 21.11.1979 & heao 2 & 199@xmath197 & 11.2 & 9 + & & 15.05.1980 & heao 2 & 48@xmath198 & 4.5 & 9 + & & 09.06.1984 & exosat & 149@xmath199 & 3.0 & 16 + & & 25.10.1988 & ginga & 484@xmath200 & 9.8 & 19 + & & 14.11.1994 & asca & 120@xmath201 & 2.3 & 33 + & & 21.11.1996 & asca & 75@xmath122 & 2.8 & 33 + & & 10.11.1998 & sax & 144@xmath202 & 4.3 & 17 + ccccc centaurus a & 2 months & 46% & 18 & 2.3 + ngc 4258 & 5 months & 10% & 6 & 1.6 + ngc 2110 & 6 months & 15% & 5 & 2.2 + ngc 4941 & 6 months & @xmath17 81% & 2 & 27.5 + ngc 7582 & 6 months & 122% & 10 & 3.5 + ngc 2992 & 6.5 months & 40% & 12 & 0.4 + eso 103-g35 & 6.5 months & 19% & 13 & 5.5 + ngc 4507 & 11 months & 25% & 5 & 19 + ngc 4388 & 1 year & 50% & 4 & 16.5 + mcg-5 - 23 - 16 & 1.1 years & 94% & 11 & 0.6 + ngc 1386 & 2 years & @xmath17112% & 2 & 36 + ngc 1365 & 2.5 years & @xmath17 85% & 2 & 30 + iras 13197 - 164 & 3 years & 44% & 2 & 22 + ngc 1808 & 3.3 years & 200% & 2 & 45 + ngc 5506 & 3.5 years & 28% & 11 & 1.1 + iras 18325 - 5926 & 3.5 years & 24% & 3 & 0.2 + ngc 7172 & 3.5 years & 31% & 6 & 1.5 + ngc 5252 & 4 years & 36% & 2 & 1.3 + ngc 7314 & 4.5 years & 31% & 4 & 0.2 + iras 05189 - 2524 & 4.6 years & 64% & 2 & 2.0 + ngc 256a & 5 years & 94% & 8 & 0.4 + mkn 348 & 8 years & 20% & 2 & 2.5 + iras 04575 - 7537 & & & 2 & + ngc 3081 & & & 2 & + ic 5063 & & & 3 &

we present a catalog of 110 variable stars within of sgr a * based on image subtraction of near - infrared ( and ) photometry . our images were obtained over 133 nights from 2000 to 2002 in-band and over 134 nights from 2001 to 2002 in-band ; the typical fwhm is . we match the catalog to other near - infrared , x - ray , and radio ( i.e. , maser ) data , and we discuss some of the more interesting objects . the catalog includes 14 sources with measurable periods , several known long - period variables and three new lpv candidates . we associate irs 10 * with oh , sio , and ho masers and a bright x - ray point source ; this analysis suggests irs 10 * is an agb star with an accreting companion . among the newly discovered sources are a probable cataclysmic variable , a potential edge - on contact 84 day period eclipsing binary , and a possible 41 day period pulsating variable

astro-ph0703769

the galactic center offers a unique opportunity for studying star formation and stellar populations near a supermassive black hole ( smbh ) , namely , sgr a*. active galactic nuclei and supernovae feedback have been evoked to explain many of the observed inhomogeneities in different galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but it is only in our own galaxy that we are able to resolve individual stars within the circumnuclear ring surrounding the central smbh @xcite . within the milky way , the cluster of stars at the galactic center is also unique in that it is comprised of both old and young stars ( see e.g. , * ? ? ? * ; * ? ? ? it is believed that the young stellar population is the result of a burst of star formation at the galactic center a few megayears ago @xcite , and that perhaps sgr a * was more actively accreting mass then than it is observed to be today . furthermore , there is evidence that the stellar initial mass function near a smbh is top heavy @xcite , implying that the galactic center is a viable location for finding highly massive stars . photometric variability is one way of probing the mixed galactic center stellar population on a large scale , as both evolved low mass and high mass stars are prone to photometric variations . there have been several photometric variability studies of the galactic center , some of which we summarize in table [ tbl : surveys ] . while some of these studies have covered a much longer baseline than ours does ( e.g. , * ? ? ? * ; * ? ? ? * ) , with more than 130 epochs in each band , our light curves have a much higher sampling . these data also represent the first major multi - wavelength photometric variability study of the galactic center since @xcite . we present here our catalog of 110 variable stars within @xmath6 pc of sgr a * , including 14 periodic sources . we outline our observations in [ sec : obs ] . in [ sec : analysis ] we discuss the reduction of data with image subtraction . we discuss our variable selection in [ sec : selection ] , and we present the catalog of variables and their light curves along with a discussion of some of the more well - known and interesting objects in [ sec : catalog ] . we adopt a galactocentric distance of 7.6 kpc @xcite . the observations from which irs 16sw was studied by @xcite and @xcite were of a field of view of @xmath7 arcseconds ( 4.12 pc projected ) approximately centered on the galactic center ; we describe here the stellar photometric variability in these data . the observations of the galactic center were made from 20002002 at the cerro - tololo inter - american observatory ( ctio)/yale 1-m telescope using the facility optical / infrared imager ( andicam ; see @xcite for details ) . andicam has a pixel scale of @xmath8 pix@xmath9 on a @xmath10 array . both @xmath1- and @xmath2-band ( centered at 1.6@xmath11 m and 2.2@xmath11 m respectively ) images were taken in the 2001 and 2002 observing seasons ; @xmath1-band data were also obtained in 2000 . the observing campaign consists of every usable night from utc 2000 august 13 through utc 2000 october 14 , utc 2001 may 20 through utc 2001 november 3 , and utc 2002 june 9 through utc 2002 september 25 . each night , a set of seven slightly offset images were obtained and then combined and trimmed ( decreasing the field of view ) to form a final nightly image . the @xmath1-band images consist of 30 s exposures , and it took about four minutes to obtain the group of seven images ; the @xmath2-band images consist of 10 s exposures and took about two minutes to obtain . after image quality cuts were made , there are a total of 133 @xmath1-band images spanning 774 nights and 134 @xmath2-band images spanning 495 nights . the full - width half - maximum ( fwhm ) ranges from @xmath12 to @xmath13 ; in general , the @xmath1-band images are of somewhat higher quality ( typical fwhm @xmath14 ) than the @xmath2-band ( with typical fwhm @xmath15 ) . the final @xmath1- and @xmath2-band images are also offset relative to one another by @xmath16 ; most of the stars detected in only one of the two bands are due to this offset . because the field is crowded , we reduced the data with the isis image subtraction package @xcite following the procedures outlined by @xcite . isis measures the change in flux for an object relative to a reference image , after correcting for the relative change in the point spread function ( psf ) . our reference images ( and thus `` fiducial '' magnitudes ) for both bands are based on images from the 2001 observing season . the reference @xmath1-band image is presented in figure [ fig : himage ] . we identified the stars running daophot / allstar progam @xcite on the astrometric reference image for each band . the final list consists of 1665 sources in @xmath1-band , with @xmath17 to @xmath18 , and 1785 sources in @xmath2-band , with @xmath19 to @xmath20 . we based the astrometric and photometric calibration on the 2mass catalog @xcite . for the astrometry , we obtained an astrometric solution by matching the 2mass @xmath2-band catalog to our @xmath2-band data . we similarly solved for the coordinate transformation between the @xmath1- and @xmath2-band data sets , which enabled us to find the coordinates of sources detected only in @xmath1 band . the astrometric solution is good to @xmath21 in ra and @xmath22 in dec ( about @xmath23 pixel in both coordinates ) . we calibrated our absolute photometry using this same set of stars matched to the 2mass catalog ; the photometric calibration error is @xmath24 mag . we define the limiting magnitude for each band to be where the magnitude distribution peaks , as denoted by the vertical dashed lines in figure [ fig : mags ] ; this is approximately the magnitude at which the sky level starts to dominate the signal . the limiting @xmath1-band magnitude is 15.1 mag , and the limiting @xmath2-band magnitude is 12.6 mag ; because these magnitude determinations are based on composite images ( as opposed to individual images ) , the actual limiting magnitudes for individual observations of variable sources are significantly lower . the color - magnitude diagram is presented in figure [ fig : cmd ] . the easiest to identify variable stars are those which vary about a mean magnitude more than typical stars of the same average mangitude . we therefore flagged potential variables independently in both @xmath1- and @xmath2-bands according to a root - mean - squared ( rms ) cut , as shown in figure [ fig : rms ] . the cutoffs are given by @xmath25 this led to @xmath26 of @xmath1-band sources and @xmath27 of @xmath2-band sources being flagged as potential variables . each potential variable was visually inspected by two of us ( m.s.p . and k.z.s . ) , and flagged as `` periodic '' ( if an aov test yielded a believable period ) , `` long term , '' `` suspected , '' or not variable . we also performed two period searches ( aov and ls ; @xcite ) of all of the sources . these searches revealed only two periodic sources not already flagged by the rms cut as well as three other clearly variable , low - amplitude non - periodic sources . the final sample includes 93 sources with both @xmath1- and @xmath2-band light curves , 14 of which are periodic , as well as 10 stars identified only in @xmath1-band and 7 stars identified only in @xmath2 . the distribution of magnitudes for all of the variable sources are shown as the shaded regions of the histograms in figure [ fig : mags ] . several of the sources flagged as variables were clearly blended , i.e. , nearby stars show similar light - curve shapes . as described in 4.6 of @xcite , the true variable is the one which displays the greatest amplitude of variability in flux . thus , while in most cases we have not attempted to correct for blending , we are able to identify which sources are the true variables . the positions and magnitudes of our variable sources without clear period determinations are given in table [ tbl : vary ] , and the light curves are shown in figure [ fig : lcs ] , [ fig : lch ] , and [ fig : lck ] . the periodic sources are summarized in table [ tbl : periodic ] and figure [ fig : periodic ] . in this section , we discuss indvidual interesting variable sources . for previously identified sources , we compare our data to other studies . table [ tbl : surveys ] summarizes some of the main near - infrared variability surveys of the galactic center to date . at higher energies , @xcite used the _ chandra x - ray observatory _ to study x - ray sources in the @xmath28@xmath29 kev band in the central @xmath30 of the galaxy from september 1999 to june 2002 . of the more than 2000 point sources studied , 77 sources at the galactic center displayed long - term variability . on the other end of the spectrum , several masers have been identified at the galactic center . specifically , @xcite studied 25 22 ghz h@xmath4o masers within 2@xmath31 and 18 43 ghz sio masers within @xmath32 of sgr a * ; these masers are associated with oh stellar masers found by @xcite . a summary of the sources discussed in this section is presented in table [ tbl : discuss ] ; we first discuss previously identified sources , starting with the irs sources ( sorted numerically ) , followed by several sources also found by @xcite ( sorted by their @xcite variable catalog name ) . finally , we discuss some of the more interesting new variables in our catalog ; these subsections are sorted by ra , as in tables [ tbl : vary ] & [ tbl : periodic ] . irs 1ne ( psd j174540.58 - 290026.7 ) passes the rms cut described in [ sec : selection ] , but visually does not appear to vary over the timescales to which we are sensitive in this study . @xcite similarly found irs 1ne to pass their @xmath33 variability cut , but this source did not pass their visual inspection test either . it is possible that irs 1ne is in fact variable , but over shorter timescales to which either @xcite s or our survey was sensitive , i.e. , @xmath34 day . irs 7 ( psd j174540.04 - 290022.7 ) is the brightest source in the field , and the only one to saturate in our @xmath2-band images . @xcite find this m2 supergiant long - period variable ( lpv ) to have a @xmath35 . we find a @xmath36 ( from @xmath37 to 9.33 ) . irs 9 ( psd j174540.47 - 290034.6 ) is a large - amplitude variable , with @xmath38 mag and @xmath39 mag ( though most of the @xmath1-band large amplitude variability is seen in the year 2000 , when no @xmath2-band data were taken ) . in our data , irs 9 is blended with two stars detected in only @xmath1-band . @xcite found the @xmath40 color of this m7 lpv to change by @xmath41 mag from september 1989 to july 1993 . @xcite find irs 9 to have brightened by @xmath42 mag in @xmath2-band over about two years ( from 1991 to 1993 ) . @xcite find irs 9 ( their 3 - 2753 , also known as v4920 sgr ) to have @xmath43 mag over a period of 463 days ; while our data do not rule out such a period , the light curve appears more consistent with that of a large - amplitude irregular or semi - regular variable . @xcite found a variable source between irs 10e and irs 10w . dubbed irs 10 * , its brightness increased by at least one magnitude between 1991 and 1992 . @xcite found similar long - period brightness variations on the order of 1.4 mag . like @xcite , we do not separately resolve irs 10e , 10w , and 10*. in the subtracted images from isis , there appears to be only one point - source of variability in the irs 10 region ( i.e. , irs 10e and 10w are presumably non - varying ) . we associate this variability seen in the irs 10 complex with irs 10 * , even though daophot detects two sources in @xmath1-band ( apparently the brighter nearby irs 10e and 10w ) and only one source in @xmath2-band . because of this , the position we state for irs 10 * ( psd j174540.64 - 290023.6 ) in table [ tbl : vary ] is from @xcite ; the stated @xmath2-magnitude ( 10.2 ) is probably largely due to irs 10e . we find the @xmath2-magnitude of irs 10 * to be relatively constant in both the 2001 and 2002 observing seasons , with the 2002 value being @xmath44 mag brighter . furthermore , irs 10 * was much redder in 2002 than it was in 2001 . both @xcite and @xcite associated irs 10 * with the oh maser oh 359.939 - 0.052 @xcite , but @xcite associated it with the @xcite maser oh 359.946 - 0.047 further from the irs 10 complex than oh 359.946 - 0.047 . ] . @xcite found an sio maser and @xcite found a h@xmath4o maser coincident with oh 359.946 - 0.047 ; @xcite and @xcite further found the sio maser and the h@xmath4o maser to be variable , respectively . @xcite additionally identify an x - ray source , cxogc j174540.7 - 290024 , within @xmath45 of irs 10 * with an ( unobscured ) x - ray luminosity @xmath46 erg s@xmath9 . typically oh / ir stars are asymptotic giant branch ( agb ) stars ; these pulsating stars are experiencing mass loss , which accounts for both the nir variability and the presence of masers . @xcite spectroscopically confirm that irs 10 * ( their irs 10ee ) is a late - type ( i.e. , consistent with being an agb ) star . while it is possible that the x - ray flux is due to some other nearby source ( or collection of sources ) , under the assumption that all three masers and the x - ray source are associated with irs 10 * ( and not irs 10e or 10w ) , this might be the largest x - ray luminosity associated with an agb star to date @xcite . it has been proposed that agb stars should be bright x - ray sources ( @xmath47 erg s@xmath9 ) due to strong magnetic fields , but various x - ray studies have not confirmed this prediction ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; typical x - ray luminosities for , e.g. , mira , are @xmath48 erg s@xmath9 @xcite . @xcite suggest that agb stars are radiating in the x - ray as proposed , but this radiation is not observed due to the high column density of the ejected material surrounding the star ( see also @xcite ) . furthermore , all agb stars with detected x - ray radiation are in binary systems ( e.g. , mira a & b , @xcite ) ; if this trend is in fact a prerequisite for a high agb x - ray luminosity , then irs 10 * should have a companion . because sio , h@xmath4o , and oh masers are believed to be due to different regions of the circumstellar envelope ( sio masers are relatively close to the star , h@xmath4o masers are further out , and oh masers are even further away from the star ; @xcite ) , the presence of all three masers implies that irs 10 * is enshrouded by a substantial circumstellar envelope . yet , as this material does not absorb the x - ray radiation , the x - ray flux is more likely to be due to accretion onto a companion star than the agb star itself . @xcite , @xcite , @xcite , and @xcite found irs 12n ( psd j174539.79 - 290035.2 , spectral type m7iii ) to be variable . @xcite observed an increase of @xmath49 mag over 2 years , and @xcite found similar variations . @xcite found irs 12n ( their 3 - 2753 ) to be a periodic variable of similar magnitude ( @xmath50 ) and amplitude ( @xmath51 mag ) with a period @xmath52 days . we can not deblend irs 12n and irs 12c ; five variable sources are detected within @xmath53 of irs 12n with similar features to one another . we find this source to potentially be periodic , with @xmath54 or @xmath55 days and a @xmath56 and a @xmath57 ; no variations are observed on shorter timescales . these data are in agreement with the classification of irs 12n as an lpv @xcite . irs 14sw ( psd j174540.02 - 290037.2 ) is seen to have variability with amplitude @xmath58 mag in both @xmath1- and @xmath2-bands in the 2001 and 2002 observing seasons . additionally , the @xmath1-band lightcurve is @xmath59 mag lower in the 2000 observing season than in 20012002 . @xcite state that irs 14sw is a `` probable '' variable ; @xcite find that irs 14sw passes their @xmath33 cut , but they do not comment on its variability . it is clear from our data that irs 14sw is , in fact , a variable source . irs 14ne ( psd j174540.11 - 290036.4 ) is a known agb star ( spectral type m7iii ; @xcite ) . @xcite find irs 14ne to have @xmath60 mag and amplitude @xmath61 , though they do not flag it as a potential variable . this is consistent with our findings of @xmath62 and @xmath63 ; we find its variability to have clear structure . irs 15sw ( psd j174539.99 - 290016.5 ) is a known wolf - rayet ( wr ) star ( also known as wr 101i ; @xcite ) with spectral type wn8/wc9 @xcite . we detect small - amplitude variations in irs 15sw , with @xmath64 mag . like many of the stars in the irs 16 cluster , irs 16ne ( psd j174540.25 - 290027.2 ) is a potential luminous blue variable ( lbv ) star @xcite with spectral type ofpe / wn9 @xcite . we observe irs 16ne to have an amplitude of only @xmath65 mag in both @xmath1 and @xmath2 ; furthermore , we do not observe the @xmath40 color to change substantially or systematically . on the other hand , the light curve is potentially consistent with a period of 205 days . irs 16sw ( psd j174540.12 - 290029.6 ) was originally proposed to be a massive eclipsing binary by @xcite . using the same data presented here , @xcite proposed that irs 16sw is instead a new kind of massive pulsating star , but a recent re - reduction of the data by @xcite revealed a sign error in the @xciteanalysis . @xcite presented a radial velocity curve of irs 16sw ; while it appears to be a single - line binary , both their analysis and that of @xcite found the data to be consistent with a contact binary of twin 50@xmath66 stars . while @xcite also support the idea that irs 16sw is an eclipsing binary , their light curve shows an asymmetry in the rise- and fall - times the rise - time appears to be @xmath67 times that of the fall - time which they propose is due to tidal deformations causing asynchronous rotational and orbital periods of the two stars . as discussed by @xcite , our data do not show such a strong asymmetry ; such an asymmetry is not seen by @xcite either . irs 28 ( psd j174540.83 - 290034.0 ) is another source described as a `` probable '' variable by @xcite . @xcite classify irs 28 as an lpv , and @xcite find irs 28 ( their 3 - 72 , also known as v4923 sgr ) to be a periodic source with an amplitude of @xmath68 mag and a period of @xmath69 days . we find this source to _ not _ be clearly periodic ( and a period of @xmath70200 days is ruled out ) . we find @xmath71 mag and @xmath72 mag ; in the two observing seasons with both @xmath1- and @xmath2-band data , the two lightcurves have similar structures , but the @xmath1-band lightcurve clearly has a larger amplitude . @xcite find irs 34w ( psd j174539.76 - 290026.4 ) to be an irregular variable ofpe / wn9 star , suggesting that it is a transitional object between the o supergiant and lbv phases of its evolution . we find an @xmath40 color of 3.5 mag , which is consistent with what @xcite found for a similar epoch . @xcite found v4910 sgr ( their 3 - 270 , our psd j174537.24 - 290045.7 ) to have a @xmath73 and vary with amplitude 0.8 mag over a period of 601 days . we find v4910 sgr to have @xmath74 and @xmath75 , with amplitudes @xmath76 mag and @xmath77 mag . while a period of @xmath78 days is not ruled our by our data , this source appears to be an irregular or semi - regular large - amplitude variable with no convincing signs of periodicity . similarly , @xcite found v4911 sgr ( their 3 - 88 , our psd j174538.02 - 2901002.6 ) to have a @xmath79 with amplitude 0.5 mag over a period of 528 days . we find v4911 sgr to have @xmath80 and @xmath81 , with amplitudes @xmath82 mag and @xmath83 mag ( much of the variation seen in @xmath1-band is from the 2000 observing season , when no @xmath2-band data were taken ) . while a period of @xmath84 days is not ruled out by our data , this source appears to be an irregular or semi - regular large - amplitude variable with no convincing signs of periodicity . @xcite associate v4911 sgr with the sio maser sio 359.930 - 0.045 , though the projected seperation is @xmath85 . psd j174542.72 - 285957.4 is one of the brightest stars in our sample . also known as v4928 sgr , this source is associated with oh , sio , and h@xmath4o masers , oh 359.956 - 0.050 . ] at @xmath86 , @xmath87 @xcite . @xcite find v4928 sgr ( their 3 - 5 ) to have an @xmath88 of 7.89 mag , with amplitude @xmath89 mag and period @xmath90 days . we find amplitudes of @xmath91 mag and @xmath92 mag , but it is clear that the `` true '' amplitude is greater than this , as at the end of the 2001 observing season the light curve shows no signs of flattening . though a period of @xmath78 days does not appear consistent with our data , we can not rule out that v4928 sgr is indeed periodic . the observed large - amplitude variability in the presence of masers flags this source as an lpv candidate . it is possible that v4928 sgr is , in fact , the lpv irs 24 @xcite . the projected distance between irs 24 ( at @xmath93 , @xmath94 j2000.0 ) and v4928 sgr is @xmath95 ; however , when @xcite first discovered an h@xmath4o maser at @xmath96 , @xmath97 j2000.0 , they associated it with the bright near - infrared irs 24 . v4930 sgr ( psd j174543.19 - 290013.0 ) is a known mira ( lpv ) variable @xcite . we observe v4930 sgr , which is one of the brighest variable stars in our sample , to smoothly vary with an amplitude of @xmath98 mag in @xmath1 and @xmath99 mag in @xmath2 . @xcite likewise find v4930 sgr ( their 3 - 16 ) to have an amplitude of @xmath100 mag over a period of 554 days . while our @xmath1-band data do not rule out a period of @xmath101 days , the light curve does not appear to be clearly periodic . psd j174535.60 - 290035.4 and psd j174538.34 - 290036.7 are two of the most enigmatic variables in our sample . for simplicity , in this section alone , we will refer to psd j174535.60 - 290035.4 as psd 35.6 - 35.4 and psd j174538.34 - 290036.7 as psd 38.3 - 36.7 . both of these variables are clearly periodic . psd 35.6 - 35.4 has a period @xmath102 days with amplitude @xmath103 and @xmath104 ; psd 38.3 - 36.7 has a period @xmath105 days with amplitudes @xmath106 and @xmath107 mag ( psd 38.3 - 36.7 is quite faint in @xmath1-band ) . because these periods , magnitudes , and variability amplitudes are so similar , we will discuss these two sources together . one obvious intriguing possibility is that these stars are cepheids . both of these sources have @xmath108 mag ; assuming these are cepheids and taking @xmath109 days , we can calculate the absolute @xmath2-band magnitude using the cepheid period - luminosity relation from @xcite : @xmath110 - 5.71 = -7.78.\ ] ] if we now assume that these stars are at the galactic center , with @xmath111 pc , we can take this absolute magnitude and calculate the @xmath2-band extinction , @xmath112 this @xmath113 , which corresponds to an @xmath114 mag @xcite , is rather large , even for the galactic center . @xcite give @xmath115 near psd 38.3 - 36.7 to be @xmath116 ( derived from pa@xmath117 to h92@xmath117 radio recombination line - emission ratios ) to @xmath118 ( derived from pa@xmath117 to 6 cm radio continuum emission ratios ) ; @xmath115 is unmeasured near psd 35.6 - 35.4 . another way of expressing this discrepancy is to say that if we use the @xcite s average @xmath113 for the field , @xmath119 mag , then equation ( [ eqn : ak ] ) yields an @xmath120 that is @xmath6 mag fainter than is predicted by equation ( [ eqn : cepheidpl ] ) . it is possible , of course , that the intervening dust is patchy on scales smaller than @xcite could resolve , and that one or both of these stars happens to be behind a dense clump of obscuring material . under the assumption that these stars are cepheids , we can constrain the actual @xmath113 for each source from the observed @xmath40 colors ; cepheids have an unreddened @xmath121 mag @xcite . taking @xmath122 and @xmath123 from @xcite , we find @xmath124 for @xmath125 , the observed @xmath40 color of 2.2 for psd 35.6 - 35.4 yields @xmath126 mag and the observed @xmath40 color of 2.6 for psd 38.3 - 36.7 yields @xmath127 mag . these extinction values are substantially less than the @xmath128 mag calculated above ; if either of these stars is in fact a cepheid , some combination of a different reddening law or a different period - luminosity relation than used here would have to come into play . if neither psd 35.6 - 35.4 nor psd 38.3 - 36.7 is a cepheid , then the possible nature of the variability seen in these stars is unclear . the shape of the light curve of psd 35.6 - 35.4 remains reminiscent of a pulsating variable ; a steep brigtening is followed by a longer fall - time . the light curve of psd 38.3 - 36.7 , however , is much more symmetric ; it is possible that it is an eclipsing binary , perhaps similar to those found by @xcite in the large magellenic cloud . if this is the case , then like irs 16sw ( see [ sec:16sw ] ) it is both close to in contact ( because the light curve does not flatten out of eclipse ) , composed of near - equal surface brightness stars ( because the depths of the eclipses are the same ) , and we are viewing it almost edge - on ( because the depths of the eclipses are close to 0.75 mag ) . these conditions are suspicious given that , if psd 38.3 - 36.7 is a binary , then its period is 84.8 days . for comparison , irs 16sw has a period of 19.45 days and is comprised of two 50@xmath66 stars with a separation of @xmath129 au @xcite . spectroscopic monitoring is the only way to determine the nature of the variability of psd 35.6 - 35.4 and to break the mass - radius degeneracy if psd 38.3 - 36.7 is an eclipsing binary . we observe psd j174538.98 - 290007.7 to have a period of @xmath130 days with an amplitude of @xmath131 mag in @xmath1-band and @xmath132 mag in @xmath2-band . based on this large period and these large amplitudes of variation , we propose that psd j174538.98 - 290007.7 is an lpv . we associate psd j174537.98 - 290134.4 and psd j174539.31 - 290016.3 with cxogc j174537.9 - 290134 and cxogc j174539.3 - 290016 respectively @xcite . psd j174537.98 - 290134.4 has marginal variability , but it is a bright source ( @xmath133 and @xmath134 ) . @xcite detected neither short - term nor long - term variability for either x - ray source . psd j174540.16 - 290055.7 is , within uncertainties , coincident with the long - term x - ray variable cxogc j174540.1 - 290055 ( @xmath135 erg s@xmath9 , @xcite ) . we find this source to potentially be periodic , with a period of @xmath136 days and an amplitude of @xmath137 mag and @xmath138 mag . @xcite observed the x - ray flux , on the other hand , to vary by a factor of five over the course of 17 days . these data are consistent with psd j174540.16 - 290055.7 being a bright cataclysmic variable ; the observed period of @xmath136 days is potentially the orbital period of the evolved star around an accreting white dwarf . psd j174541.39 - 290126.6 is one of the reddest stars in our sample , with an observed @xmath139 . in @xmath1-band we observe an amplitude of @xmath140 mag , though the more coherent @xmath2-band light curve only shows variations of @xmath141 mag . both @xcite and @xcite find this object ( their 61 - 7 and 3 - 205 respectively ) to have an amplitude of @xmath142@xmath143 mag and a period of @xmath144 days , both of which are consistent with our data . furthermore , @xcite associate this large - amplitude variable with the oh / ir star oh 359.932 - 0.059 @xcite . psd j174542.36 - 290011.2 is a periodic source with @xmath145 days and amplitudes @xmath146 mag and @xmath147 mag . psd j174542.36 - 290011.2 shows regular brightness variations over a period of @xmath148 days , but when phased at a period of @xmath149 days , both the maxima and the minima are less extreme in the first half of the oscillation than in the second . while minima alternating between shallow and deep is characteristic of an rv tauri type star , a period of @xmath150 days is too long for psd j174542.36 - 290011.2 to be an rv tauri type star @xcite . we observe psd j174543.14 - 290050.2 to have amplitudes of @xmath151 and @xmath152 . while it does appear to be periodic , the actual period is not clear ; it is possible that the period varies from cycle to cycle . based on these observations , we propose that psd j174543.14 - 290050.2 is an lpv . we present a catalog of variable stars within @xmath6 pc of sgr a * in the near - infrared @xmath1 and @xmath2 bands over a three - year baseline ; @xmath153% of these variables are previously unidentfied . this is the first photometric variability study of the galactic center to use the technique of image subtraction . we find several new periodic sources . of these , psd j174535.60 - 290035.4 and psd j174538.34 - 290036.7 have the shortest periods , with @xmath154 days . we believe psd j174535.60 - 290035.4 is a pulsating variable , though it is unlikely to be a cepheid . psd j174538.34 - 290036.7 appears to be a nearly edge - on contact eclipsing binary system , in which case the period is actually @xmath155 days and the individual components are likely to be quite large . another periodic source , psd j174540.16 - 290055.7 ( @xmath156 days ) , is coincident with the x - ray variable cxogc j174540.1 - 290055 ; we suspect that this source is a cataclysmic variable . among the previously identified sources in the catalog , we associate irs 10 * with a bright x - ray point source and oh , h@xmath4o , and sio masers , suggesting that it is an agb star with an accreting companion . and @xmath2 , sorted by ra as in table [ tbl : vary ] . for each source , the extent of the @xmath1- and @xmath2-band ranges plotted is the same . points lying below the magnitude limit ( see [ sec : analysis ] ) are not included.[fig : lcs ] ] lrrrrl psd j174535.45 - 290007.6 & @xmath157 & @xmath158 & 12.6 & & + psd j174535.89 - 290113.5 & @xmath159 & @xmath160 & 14.2 & 10.9 & + psd j174535.98 - 290011.5 & @xmath161 & @xmath162 & 12.1 & 9.7 & + psd j174536.18 - 290047.6 & @xmath163 & @xmath164 & 12.4 & 9.6 & + psd j174536.53 - 290125.9 & @xmath165 & @xmath166 & 11.9 & 9.5 & + psd j174536.63 - 290015.4 & @xmath167 & @xmath168 & 10.7 & 8.8 & + psd j174536.85 - 290114.5 & @xmath169 & @xmath170 & 13.1 & 10.1 & + psd j174536.97 - 290138.0 & @xmath171 & @xmath172 & 15.0 & 10.2 & + psd j174536.97 - 290019.4 & @xmath173 & @xmath174 & 12.5 & 10.0 & + psd j174537.10 - 290103.9 & @xmath175 & @xmath176 & 13.9 & 10.8 & + psd j174537.24 - 290045.7 & @xmath177 & @xmath178 & 12.6 & 9.9 & v4910 sgr + psd j174537.42 - 290042.2 & @xmath179 & @xmath180 & 13.8 & 10.6 & + psd j174537.52 - 290046.2 & @xmath181 & @xmath182 & 14.5 & 10.6 & + psd j174537.52 - 290041.7 & @xmath183 & @xmath184 & 12.8 & 10.0 & + psd j174537.66 - 285955.2 & @xmath185 & @xmath186 & 12.7 & & + psd j174537.74 - 290009.2 & @xmath187 & @xmath188 & 13.1 & 10.2 & + psd j174537.91 - 290045.9 & @xmath189 & @xmath190 & 12.7 & 10.0 & + psd j174537.98 - 290134.4 & @xmath191 & @xmath192 & 9.5 & 8.6 & cxogc j174537.9 - 290134 + psd j174537.98 - 290004.4 & @xmath191 & @xmath193 & 12.0 & 9.2 & + psd j174538.02 - 290102.6 & @xmath194 & @xmath195 & 12.5 & 9.3 & v4911 sgr ; blended + psd j174538.05 - 290058.1 & @xmath196 & @xmath197 & 12.7 & 10.2 & + psd j174538.10 - 290033.5 & @xmath198 & @xmath199 & 13.3 & 10.3 & + psd j174538.19 - 290005.0 & @xmath200 & @xmath201 & 12.5 & 9.7 & + psd j174538.29 - 290011.1 & @xmath202 & @xmath203 & 13.4 & 9.9 & + psd j174538.61 - 290054.3 & @xmath204 & @xmath205 & 13.0 & 9.5 & + psd j174538.74 - 290012.7 & @xmath206 & @xmath207 & 11.1 & 9.2 & + psd j174538.79 - 290004.6 & @xmath208 & @xmath209 & 11.8 & 9.3 & + psd j174538.95 - 290139.6 & @xmath210 & @xmath211 & & 9.4 & + psd j174538.99 - 290102.2 & @xmath212 & @xmath213 & 14.3 & 12.4 & + psd j174539.09 - 290039.3 & @xmath214 & @xmath215 & 12.7 & 9.8 & blended + psd j174539.23 - 290036.3 & @xmath216 & @xmath217 & 14.1 & 10.8 & + psd j174539.31 - 290016.3 & @xmath218 & @xmath219 & 11.6 & 9.2 & cxogc j174539.3 - 290016 + psd j174539.31 - 290054.1 & @xmath218 & @xmath220 & 12.6 & 10.0 & + psd j174539.36 - 290055.6 & @xmath221 & @xmath222 & 13.0 & 10.5 & + psd j174539.39 - 290014.6 & @xmath223 & @xmath224 & 10.9 & 8.4 & blended + psd j174539.45 - 290056.6 & @xmath225 & @xmath226 & 12.4 & 9.1 & blended + psd j174539.56 - 290057.3 & @xmath227 & @xmath228 & & 12.5 & + psd j174539.58 - 290049.5 & @xmath229 & @xmath230 & 12.4 & 9.6 & + psd j174539.60 - 290107.6 & @xmath231 & @xmath232 & 15.3 & 11.8 & @xmath1 past threshold + psd j174539.73 - 290050.2 & @xmath233 & @xmath234 & 12.5 & 9.9 & + psd j174539.75 - 290055.5 & @xmath235 & @xmath236 & 12.5 & 9.6 & irs 22 + psd j174539.76 - 290026.4 & @xmath237 & @xmath238 & 14.1 & 10.6 & irs 34w + psd j174539.77 - 290043.9 & @xmath239 & @xmath240 & 12.1 & 9.6 & + psd j174539.79 - 290008.5 & @xmath241 & @xmath242 & 11.4 & 8.5 & + psd j174539.79 - 290029.9 & @xmath241 & @xmath243 & 13.2 & 10.6 & irs 13e + psd j174539.83 - 290053.9 & @xmath244 & @xmath245 & 9.6 & 7.6 & blended + psd j174539.86 - 290024.7 & @xmath246 & @xmath247 & & 10.5 & + psd j174539.86 - 290019.7 & @xmath248 & @xmath249 & 12.8 & 10.0 & + psd j174539.90 - 290034.5 & @xmath250 & @xmath251 & 13.3 & 10.6 & blended + psd j174539.91 - 290011.0 & @xmath247 & @xmath252 & 12.5 & 9.9 & cxogc j174539.9 - 290012 + psd j174539.93 - 290024.9 & @xmath253 & @xmath254 & 13.9 & 10.5 & + psd j174539.95 - 285951.1 & @xmath255 & @xmath256 & 12.2 & & + psd j174539.95 - 290110.8 & @xmath257 & @xmath258 & 14.2 & 10.2 & + psd j174539.99 - 290016.5 & @xmath259 & @xmath260 & 12.1 & 9.8 & irs 15sw + psd j174540.02 - 290037.2 & @xmath261 & @xmath262 & 11.8 & 9.4 & irs 14sw + psd j174540.04 - 290018.0 & @xmath263 & @xmath264 & 12.3 & 10.3 & irs 15ne + psd j174540.04 - 290022.7 & @xmath263 & @xmath263 & 9.0 & & irs 7 , saturated in @xmath2 + psd j174540.04 - 290027.0 & @xmath263 & @xmath235 & 11.6 & 9.3 & irs 16nw + psd j174540.04 - 290019.5 & @xmath265 & @xmath266 & 12.6 & & + psd j174540.06 - 290112.1 & @xmath267 & @xmath268 & 12.0 & 9.9 & + psd j174540.11 - 290036.4 & @xmath269 & @xmath270 & 11.8 & 9.1 & irs 14ne + psd j174540.21 - 290043.2 & @xmath271 & @xmath272 & 14.1 & 10.7 & + psd j174540.21 - 290056.3 & @xmath271 & @xmath273 & 15.6 & & + psd j174540.31 - 285953.8 & @xmath274 & @xmath275 & 13.0 & & + psd j174540.31 - 290039.5 & @xmath274 & @xmath276 & 12.3 & 9.4 & + psd j174540.37 - 285954.0 & @xmath277 & @xmath278 & 11.7 & & blended + psd j174540.38 - 290033.6 & @xmath279 & @xmath199 & & 11.4 & + psd j174540.45 - 290036.3 & @xmath260 & @xmath217 & 13.8 & 11.1 & cxogc j174540.4 - 290036 + psd j174540.47 - 290034.6 & @xmath219 & @xmath280 & 11.3 & 8.6 & irs 9 , blended + psd j174540.58 - 290026.7 & @xmath281 & @xmath239 & 11.9 & 9.2 & irs 1ne + psd j174540.64 - 290023.6 & @xmath282 & @xmath255 & & 10.2 & irs 10 * , blended + psd j174540.83 - 290034.0 & @xmath283 & @xmath284 & 12.2 & 9.1 & irs 28 + psd j174540.84 - 290027.0 & @xmath285 & @xmath286 & 12.0 & 9.5 & irs 32 + psd j174541.04 - 290022.7 & @xmath287 & @xmath263 & 11.1 & 8.5 & + psd j174541.09 - 285952.6 & @xmath288 & @xmath289 & 13.2 & & + psd j174541.17 - 290046.9 & @xmath290 & @xmath291 & 10.7 & 7.9 & irs 19 , blended + psd j174541.20 - 290039.0 & @xmath292 & @xmath210 & 11.6 & 9.1 & + psd j174541.27 - 290049.9 & @xmath293 & @xmath294 & 12.8 & 9.6 & + psd j174541.35 - 290033.0 & @xmath295 & @xmath296 & 13.2 & 10.2 & + psd j174541.36 - 290022.7 & @xmath297 & @xmath298 & 11.9 & 9.4 & + psd j174541.39 - 290126.6 & @xmath299 & @xmath300 & 13.9 & 9.6 & + psd j174541.56 - 285949.6 & @xmath301 & @xmath302 & 12.4 & & + psd j174541.59 - 290023.0 & @xmath303 & @xmath304 & 13.7 & 10.8 & + psd j174541.75 - 290004.6 & @xmath305 & @xmath209 & 12.1 & 9.0 & + psd j174541.75 - 290013.0 & @xmath305 & @xmath306 & 12.3 & 9.3 & + psd j174542.01 - 290024.4 & @xmath307 & @xmath308 & 13.2 & 10.0 & + psd j174542.13 - 290045.8 & @xmath309 & @xmath310 & 13.5 & 10.3 & + psd j174542.39 - 285950.7 & @xmath311 & @xmath312 & 10.8 & & + psd j174542.60 - 290103.9 & @xmath313 & @xmath176 & 14.3 & 11.0 & + psd j174542.72 - 285957.4 & @xmath314 & @xmath315 & 10.5 & 7.7 & v4928 sgr , blended + psd j174542.76 - 290125.5 & @xmath316 & @xmath317 & 10.7 & 8.7 & + psd j174542.88 - 285951.3 & @xmath318 & @xmath319 & 12.7 & & + psd j174542.92 - 285958.4 & @xmath320 & @xmath321 & 14.2 & 11.8 & + psd j174543.01 - 290011.9 & @xmath322 & @xmath323 & 10.8 & 8.5 & + psd j174543.19 - 290013.0 & @xmath324 & @xmath306 & 10.9 & 8.2 & v4930 sgr , blended + psd j174543.29 - 290118.8 & @xmath325 & @xmath326 & & 11.9 & + psd j174543.31 - 290014.3 & @xmath327 & @xmath328 & 14.8 & 12.6 & + psd j174543.50 - 290005.3 & @xmath329 & @xmath292 & & 9.0 & + lrrrrrl psd j174535.60 - 290035.4 & @xmath330 & @xmath331 & 14.5 & 12.3 & @xmath332 & cxogc j174535.6 - 290034 + psd j174537.11 - 290033.1 & @xmath333 & @xmath296 & 10.0 & 12.8 & @xmath334 & + psd j174538.28 - 290006.8 & @xmath335 & @xmath336 & 11.9 & 9.3 & @xmath337 & + psd j174538.35 - 290036.7 & @xmath338 & @xmath339 & 14.9 & 12.1 & @xmath340 & + psd j174538.97 - 290123.7 & @xmath341 & @xmath161 & 14.6 & 10.8 & @xmath342 & + psd j174538.98 - 290007.7 & @xmath343 & @xmath287 & 13.0 & 9.9 & @xmath344 & + psd j174539.39 - 290039.5 & @xmath345 & @xmath346 & 11.9 & 9.8 & @xmath347 & + psd j174539.79 - 290035.2 & @xmath348 & @xmath349 & 11.8 & 8.8 & @xmath350 & irs 12n , very blended + psd j174540.12 - 290029.6 & @xmath351 & @xmath352 & 10.8 & 8.3 & @xmath353 & irs 16sw + psd j174540.13 - 290016.8 & @xmath354 & @xmath355 & 12.1 & 9.4 & @xmath356 & irs 15n , blended + psd j174540.16 - 290055.7 & @xmath357 & @xmath358 & 12.1 & 9.3 & @xmath359 & cxogc j174540.1 - 290055 + psd j174542.36 - 290011.2 & @xmath360 & @xmath203 & 13.4 & 10.6 & @xmath361 & + psd j174540.25 - 290027.2 & @xmath266 & @xmath233 & 10.5 & 8.3 & @xmath362 & irs 16ne + psd j174543.14 - 290050.2 & @xmath363 & @xmath234 & 12.0 & 9.1 & @xmath364 & blended + lllcl @xcite & @xmath365 & 1991 , 1992 , 1993 ; & @xmath2 & irs 9 , 10 * , 12n , 14sw , 28 + & & 3 epochs & & + @xcite & central @xmath366 & may 1993april 1995 ; & @xmath367 , @xmath1 , @xmath2 , @xmath368 & irs 7 , 9 , 12n + & & 58 epochs & & + @xcite & @xmath369 & august 1992may 1998 ; & @xmath2 & irs 7 , 9 , 10 * , 12n , 16sw and + & & @xmath370 epochs & & potentially irs 1ne , 6 , 14sw + @xcite & @xmath371 & 19941997 ; @xmath372 epochs & @xmath2 & only large - amplitude , + & & & & periodic variability + @xcite & @xmath373 & 19952005 ; 50 epochs & @xmath2 & irs 16sw , 16nw , 16cc , 29n , + & & & & and 12 s stars lllll irs 1ne & [ fig : lcs](d ) & & & psd j174540.58 - 290026.7 + irs 7 & [ fig : lch ] & & lpv & psd j174540.04 - 290022.7 + irs 9 & [ fig : lcs](d ) & & lpv & psd j174540.47 - 290034.6 + irs 10 * & [ fig : lck ] & & oh / ir , x - ray source & psd j174540.64 - 290023.6 + irs 12n & [ fig : periodic ] & @xmath350 & lpv & psd j174539.79 - 290035.2 + irs 14sw & [ fig : lcs](d ) & & & psd j174540.02 - 290037.2 + irs 14ne & [ fig : lcs](d ) & & agb & psd j174540.11 - 290036.4 + irs 15sw & [ fig : lcs](c ) & & wr & psd j174539.99 - 290016.5 + irs 16ne & [ fig : periodic ] & @xmath362 & lbv ? & psd j174540.25 - 290027.2 + irs 16sw & [ fig : periodic ] & @xmath374 & binary & psd j174540.12 - 290029.6 + irs 28 & [ fig : lcs](d ) & & & psd j174540.83 - 290034.0 + irs 34w & [ fig : lcs](c ) & & lbv ? & psd j174539.76 - 290026.4 + v4910 sgr & [ fig : lcs](a ) & & & psd j174537.24 - 290045.7 + v4911 sgr & [ fig : lcs](b ) & & sio maser ? & psd j174538.02 - 290102.6 + v4928 sgr & [ fig : lcs](e ) & & oh / ir & psd j174542.72 - 285957.4 + v4930 sgr & [ fig : lcs](e ) & & mira & psd j174543.19 - 290013.0 + psd j174535.60 - 290035.4 & [ fig : periodic ] & @xmath332 & & + psd j174538.34 - 290036.7 & [ fig : periodic ] & @xmath340 & binary ? ? & + psd j174538.98 - 290007.7 & [ fig : periodic ] & @xmath344 & lpv ? & + psd j174537.98 - 290134.4 & [ fig : lcs](a ) & & x - ray source & + psd j174539.31 - 290016.3 & [ fig : lcs](b ) & & x - ray source & + psd j174540.16 - 290055.7 & [ fig : periodic ] & @xmath359 & variable x - ray & + psd j174541.39 - 290126.6 & [ fig : lcs](e ) & & oh / ir & + psd j174542.36 - 290011.2 & [ fig : periodic ] & @xmath361 & & + psd j174543.14 - 290050.2 & [ fig : periodic ] & @xmath364 & & + , d. l. , atwood , b. , belville , s. r. , brewer , d. f. , byard , p. l. , gould , a. , mason , j. a. , obrien , t. p. , pappalardo , d. p. , pogge , r. w. , steinbrecher , d. p. , & teiga , e. j. 2003 , in instrument design and performance for optical / infrared ground - based telescopes . edited by iye , masanori ; moorwood , alan f. m. proceedings of the spie , volume 4841 , pp . 827 - 838 ( 2003 ) . , m. iye & a. f. m. moorwood , 827838 , f. , genzel , r. , alexander , t. , abuter , r. , paumard , t. , ott , t. , gilbert , a. , gillessen , s. , horrobin , m. , trippe , s. , bonnet , h. , dumas , c. , hubin , n. , kaufer , a. , kissler - patig , m. , monnet , g. , strbele , s. , szeifert , t. , eckart , a. , schdel , r. , & zucker , s. 2005 , , 628 , 246 , t. , genzel , r. , martins , f. , nayakshin , s. , beloborodov , a. m. , levin , y. , trippe , s. , eisenhauer , f. , ott , t. , gillessen , s. , abuter , r. , cuadra , j. , alexander , t. , & sternberg , a. 2006 , , 643 , 1011 , m. f. , cutri , r. m. , stiening , r. , weinberg , m. d. , schneider , s. , carpenter , j. m. , beichman , c. , capps , r. , chester , t. , elias , j. , huchra , j. , liebert , j. , lonsdale , c. , monet , d. g. , price , s. , seitzer , p. , jarrett , t. , kirkpatrick , j. d. , gizis , j. e. , howard , e. , evans , t. , fowler , j. , fullmer , l. , hurt , r. , light , r. , kopan , e. l. , marsh , k. a. , mccallon , h. l. , tam , r. , van dyk , s. , & wheelock , s. 2006 , , 131 , 1163

we introduce definitions of contact blow up from several perspectives . such different approaches to the contact blow up are related . we prove that the contact topology coincides in the case of blow ups along transverse embedded loops .

1210.1687

in his book _ partial differential relations _ @xcite , m. gromov proposed a definition of the blow up operation in the contact category , see exercise ( c ) on page 343 . this article discusses this definition as well as related constructions . + let @xmath0 be a smooth manifold and @xmath1 an embedded submanifold . the normal bundle of @xmath2 in @xmath0 will be denoted by @xmath3 . recall that it is defined through the short exact sequence of smooth vector bundles over @xmath4 @xmath5 given a complex vector bundle @xmath6 , we denote by @xmath7 the fiberwise projectivization of @xmath8 . + suppose the normal bundle @xmath3 is a complex bundle , then we may produce a manifold @xmath9 , the topological blow up of @xmath0 along @xmath4 . it is defined as the connected sum @xmath10 of the manifolds @xmath0 and @xmath11 with the reversed orientation along @xmath4 . let @xmath12 be the zero section of @xmath3 . the submanifold @xmath4 is embedded in the first factor through @xmath13 and in the second as the section @xmath14 in the category of symplectic manifolds the normal bundle is a complex bundle and the manifold @xmath9 can be endowed with a symplectic structure . in this paper we address the question for contact manifolds . + in the above reference , m. gromov conjectured that a contact blow up construction exists along a contact submanifold @xmath4 embedded in a contact manifold @xmath0 provided a pair of hypotheses are satisfied . these are : + * the contact submanifold @xmath15 is a boothby wang manifold . see definition [ def : bw ] . in particular , the reeb vector field associated to @xmath16 has all its orbits periodic with the same period . let @xmath17 be the quotient space of its orbits and @xmath18 the projection map . + * the normal symplectic bundle @xmath3 is isomorphic to the pull back of a symplectic bundle @xmath19 through @xmath20 . that is , there exists an isomorphism @xmath21 of symplectic bundles . these two hypotheses allow to give a definition . nevertheless the contact blow up will not be a contact structure on the topological blow up @xmath9 of @xmath0 . we will first illustrate a reason for this in a simple example , see section [ sec : pr ] . we provide a definition producing a contact structure on a manifold constructed as a different connected sum with @xmath0 . it has the same geometrical properties as the symplectic blow up . it is this manifold we would rather call the contact blow up . + apart from the construction of contact manifolds , the contact blow up construction is relevant for the existence problem of contact structures on @xmath22manifolds . see @xcite . + the content of the paper is organized as follows . the section [ sec : pr ] provides a brief review of the topological blow up . in section [ sec : bw ] , we introduce the classical boothby wang construction @xcite . it will be described with some concrete examples that shall be used later on . then , three alternative constructions of contact blow up are introduced : + * the contact blow up for embedded transverse loops , produced as a surgery operation . this had been introduced in the article @xcite , it will be reviewed in section [ sec : srgy ] . * the contact blow up defined _ la gromov _ is the content of section [ sec : gromov ] . * the contact blow up as a contact quotient is described in section [ sec : quotient ] . + these three constructions are inspired by the three alternative constructions for the symplectic blow up : the _ ad hoc _ construction with explicit gluings , the description using frame bundles , found in pages 239 and 243 in @xcite respectively , and the symplectic cut procedure discussed in @xcite . finally , section [ sec : un ] relates these constructions in the case of transverse loops . + * acknowledgements . * we want to acknowledge k. niederkrger for useful discussions . in particular for asking us to relate the contact cut and the contact blow up . this project was partially developed during the aim workshop _ contact topology in higher dimensions_. the first and third authors are supported by the spanish national research project mtm201017389 . second author would like to thank ictp for offering a visiting position that allowed him to develop this article . in this section we introduce the basic definitions , explain the topological blow up procedure and discuss an example . a contact structure on a smooth manifold @xmath23 is a maximally non integrable smooth field @xmath24 of tangent hyperplanes . a contact manifold @xmath25 is a choice of a contact structure @xmath24 on @xmath0 . the maximal non integrability can be described in terms of local equations for @xmath24 . a smooth field @xmath24 of tangent hyperplanes is maximally non integrable if and only if for any @xmath26 there exist an open subset @xmath27 containing @xmath28 and a @xmath29form @xmath30 such that @xmath31 and @xmath32 . equivalently , the form @xmath33 is non degenerate when restricted to @xmath34 in case the form @xmath35 can be chosen to be globally defined , i.e. @xmath36 , the contact structure @xmath24 is called coorientable . a contact structure is cooriented if a choice of global contact form has been made . + let @xmath37 be a cooriented contact manifold with fixed global contact form @xmath35 , i.e. @xmath36 satisfies @xmath38 , @xmath39 . a smooth submanifold @xmath40 is called a _ contact submanifold _ if the induced distribution @xmath41 is a contact structure on @xmath4 . + the notion of a blow up has its origins in algebraic geometry . first , we define the concept for a complex vector space . see @xcite for further details . [ def : lnbu ] the blow up @xmath42 of the @xmath43dimensional complex vector space @xmath44 at the origin is the smooth manifold @xmath45,p):p\in l\}\subset{\mathbb{cp}}^n\times{\mathbb{c}}^{n+1}$ ] along with the restriction of the projection onto the second factor @xmath46 . note that @xmath47 restricted to @xmath48,p):p\in l , p\neq0\}$ ] induces a diffeomorphism onto the image @xmath49 . the projective space @xmath50 is called the exceptional divisor . the topological blow up of @xmath0 along @xmath4 defined in the previous section coincides with the previous definition if @xmath51 is the origin in @xmath52 . more generally , from the definition of @xmath9 we can conclude the following [ prop : tpbl ] let @xmath0 be a smooth manifold and @xmath2 a submanifold with complex normal bundle . there exists a smooth submanifold @xmath53 diffeomorphic to the total space of a projective smooth bundle over @xmath4 such that , as smooth manifolds , @xmath54 . the topological blow up can be performed along any complex submanifold @xmath4 of a complex manifold @xmath0 . in this case the blown up manifold @xmath9 inherits a canonical complex structure . analogously , if @xmath55 is a symplectic manifold and @xmath4 a symplectic submanifold , the topological blow up manifold @xmath9 can also be endowed with a symplectic structure . in the symplectic case there is no uniqueness , see @xcite . the topological blow up could also be performed along a contact submanifold of a contact manifold because the normal bundle is symplectic and hence it is also complex . [ rem : key ] 1 . suppose the normal bundle @xmath3 splits as a direct sum of isomorphic complex line bundles @xmath56 : @xmath57 . then there is a second projection map @xmath58 defined as follows . given a point @xmath59 , let @xmath60 be non zero vector in the fiber . then a point @xmath61 \in \nu_m(s)_p$ ] is mapped to @xmath62)=[l_1/s_p , \cdots , l_r / s_p]$ ] . it is simple to verify that the map is well defined , i.e. independent of the choice of vector @xmath63 . the hypothesis above is satisfied in some cases . for instance , let @xmath4 be the base locus of a projective , resp . symplectic , lefschetz pencil . then @xmath4 conforms the hypothesis for @xmath64 . in such case the fibers of @xmath65 are projective , resp . this also occurs with contact pencils , see @xcite . * example * : _ let @xmath66 be a @xmath22dimensional contact manifold and @xmath4 is a @xmath29dimensional contact compact submanifold , i.e. a transverse embedded loop . if we perform a topological blow up along @xmath4 , the exceptional divisor is @xmath67 . we are in the hypothesis of the previous remark : @xmath68 is trivial . therefore we have a projection @xmath69 . in the contact case , if we assume that @xmath8 is a contact submanifold , it is not possible to ensure that the fibers of such projection map are contact : there is no contact distribution on @xmath70 whose fibers are all transverse to the contact structure , see @xcite . _ + in the previous example , the non transversality of the fibers occurs only because we are using the topological blow up as our blown up manifold . we will further argue from different perspectives that the blown up manifold @xmath9 we should consider in contact topology is not the topological blow up discussed above . instead , the correct manifold is obtained through a procedure that substitutes @xmath71 by the standard contact sphere @xmath72 , not @xmath70 . in such a case , the natural projection map @xmath73 is the hopf fibration , whose fibers are transverse to the contact structure . + in this section we explain the construction of a contact manifold from an integral symplectic manifold as developed in @xcite . it will be used in understanding the contact structure on the manifold obtained after a contact blow - up . + a symplectic manifold @xmath74 is called integral if the class @xmath75 $ ] lies in the image of the map @xmath76 , i.e. the periods of @xmath77 are integers . such a form @xmath77 is called integral . for instance , a khler form on a complex compact manifold is integral if and only if the manifold is a smooth projective algebraic variety . in the definition above a circle has length @xmath29 . note that the lift of @xmath75 $ ] to @xmath78 may not be unique if @xmath79 contains torsion elements . + given an integral form @xmath80 there exists a hermitian complex line bundle @xmath81 admitting a compatible connection whose curvature is @xmath82 . see @xcite for the details . this leads to the following [ def : bw ] let @xmath74 be an integral symplectic manifold . the boothby wang manifold @xmath83 is the contact manifold whose total space is the unit circle bundle associated to the line bundle @xmath84 and its contact structure is defined as the restriction of any connection with curvature @xmath85 to the circle bundle . [ rek : uniqueness_bw ] the contact structure is independent of the choice of connection . indeed , the space of choices for a connection as above is an affine space modelled on the vector space of flat connections and hence is contractible . the stability theorem of j. gray applies to ensure the uniqueness up to contactomorphisms of the contact structures . for the case @xmath86 we will sometimes omit the subindex . note that the topology of the total space varies with the parameter @xmath87 . the exact relationship between the topology and the parameter @xmath87 is the content of the following [ lem : cov ] let @xmath74 be a symplectic manifold . then the boothby wang manifold @xmath88 is a @xmath87covering of @xmath83 . we fix a hermitian connection on @xmath56 , this induces a hermitian connection on @xmath89 . define the unitary non linear map between line bundles @xmath90 it preserves the connections on the two bundles . there exists a unitary connection preserving action of @xmath91 , the cyclic group of order @xmath87 , in @xmath56 given by @xmath92 this action induces the trivial action in @xmath89 and thus becomes the deck transformation group of a covering between the total spaces of the associated circle bundles . this map is certainly compatible with the contact structures . * examples * : 1 . let @xmath93 be a lens space , i.e. the orbit space of the action @xmath94 the lens space naturally inherits a contact structure @xmath95 from the standard contact structure of @xmath96 induced by the complex tangencies . lemma [ lem : cov ] provides a contactomorphism between @xmath97 and @xmath98 + 2 . consider the @xmath99torus @xmath100 and @xmath101 an integral area form with total area one . then the boothy wang manifolds @xmath102 associated to @xmath103 give rise to quotients of the heisenberg group by discrete subgroups @xmath104 and thus provide several examples of contact nilmanifolds different from the 3torus . + the construction of the contact blow up will involve the quotient of the product of two boothby wang manifolds . with this in mind , we proceed to describe the boothby wang construction when the base sympletic manifold is a product . we show that the boothby wang construction and the cartesian product _ commute_. in precise terms , let @xmath105 be the boothby wang manifold associated to @xmath106 then we have the following : [ thm : bw ] let @xmath107 and @xmath108 be symplectic manifolds and @xmath109 a pair of coprime integers . consider the product @xmath110 of the boothby wang manifolds and the action @xmath111 then the space of orbits is a manifold diffeomorphic to @xmath105 . this space of orbits carries a contact structure induced by a connection with curvature @xmath112 and hence is contactomorphic to @xmath113 let @xmath114 and @xmath115 be the subgroup defined as the image of the embedding @xmath116 let @xmath117 be the @xmath118principal bundle with base space @xmath119 induced by the @xmath120principal bundles @xmath121 and @xmath122 . our aim is to describe @xmath123 as a bundle over @xmath119 . in general @xmath124 is not a @xmath125principal bundle but it is the case when both @xmath118 and @xmath125 are closed lie groups and @xmath125 is a normal sub - group of @xmath126 actually , they are abelian and since @xmath127 , @xmath123 is also a @xmath128principal bundle over @xmath119 . taking into account the exact group sequence @xmath129 where the second morphism is given by multiplication by @xmath130 , we conclude that the space of orbits @xmath123 is a manifold diffeomorphic to @xmath105 . the claim about the connection and the associated curvature follows from the short exact sequence @xmath131 finally , it follows from remark [ rek : uniqueness_bw ] that the manifolds are , in fact , contactomorphic . there are a few simple cases worth mentioning . + * examples * : 1 . let @xmath132 and @xmath133 arbitrary . then neither the topology of the resulting space nor the contact structure depend on @xmath134 . indeed , @xmath135 is diffeomorphic to @xmath136 analogously , the parameter @xmath137 is vacuous if @xmath138 . in particular , @xmath139 quotiented by any @xmath140 coprime @xmath120action is diffeomorphic to @xmath120 . let @xmath141 be symplectic manifolds with the fubini study form . then the space @xmath142 is diffeomorphic to @xmath143 regardless of the values @xmath144 , see @xcite for a proof of this fact . further , the symplectic structure of the associated line bundle depends only on @xmath145 . note that there is an alternative construction of a contact structure in @xmath143 using an open book decomposition with @xmath146 pages and an even power of a dehn twist as monodromy , however such a procedure may only produce vanishing first chern class and is thus different from @xmath142 if @xmath147 . see @xcite for more . the previous example can be generalized to construct contact structures on @xmath148 . indeed the result implies that the total space of @xmath149 is a @xmath150bundle over @xmath151 . the hopf action is explicit enough for the classifying map to be described as the element @xmath152 consequently the resulting manifold is diffeomorphic @xmath148 if @xmath43 is odd or @xmath87 is even . we would like to remark that it is not known whether the product of any contact manifold with the @xmath99sphere admits a contact structure . it will be essential for the contact blow up construction to be able to extend a connection on a submanifold to a global connection , let us now prove that this is possible under suitable conditions : [ lem : exten_bw ] let @xmath4 be a closed submanifold of @xmath153 , possibly with smooth boundary , and @xmath56 the line bundle associated to @xmath77 . assume that the restriction morphism @xmath154 is surjective and let @xmath155 be a connection over @xmath156 whose curvature is @xmath82 . then , there is a connection @xmath157 on @xmath56 with curvature @xmath82 such that its restriction to @xmath4 is @xmath155 . let @xmath158 be a connection on the line bundle @xmath159 with curvature @xmath82 . denote @xmath160 , then @xmath161 is a closed @xmath29form over @xmath4 . in order to complete our argument we need to extend @xmath162 to a global closed @xmath29form . + by hypothesis the map @xmath163 is a surjection . therefore there exists a cohomology class @xmath164 $ ] on @xmath165 , such that restricted to @xmath4 coincides with @xmath166 $ ] . its difference over @xmath4 will be the trivial class on @xmath167 , so @xmath168 , for some smooth function @xmath169 . we extend @xmath170 to a global smooth function @xmath171 . the form @xmath172 is the required global connection with curvature @xmath82 and extending @xmath155 . let @xmath173 be a contact manifold . in this section we recall the blow up construction from section 5 in @xcite . it is an operation defined in a neighborhood of a transversely embedded loop . topologically it consists of a surgery along the loop : the interior of @xmath174 is removed and a tubular neighbourhood of the @xmath175sphere @xmath176 is glued along the common boundary @xmath177 . the sphere @xmath178 whose neighbourhood is attached is called the exceptional divisor . let us discuss this surgery operation in the contact category . + consider the manifold @xmath179 with spherical coordinates @xmath180 . let @xmath181 be the standard contact form for the contact structure @xmath182 on the sphere @xmath183 . define the following two contact forms in @xmath184 : @xmath185 fix an integer @xmath186 and consider the diffeomorphism @xmath187 it pulls back the contact form @xmath188 to @xmath189 $ ] . + given a subset @xmath190 , let @xmath191 denote a small neighbourhood of @xmath192 in @xmath0 . these ingredients suffice to prove the following : [ thm : blow - up ] @xmath193thm . 5.1 in @xcite@xmath194 let @xmath195 be a contact manifold . let @xmath196 be a smooth transverse loop in @xmath0 . there exists a manifold @xmath197 satisfying the following conditions : * there exists a contact structure @xmath198 on @xmath197 . * there exists a codimension@xmath99 contact submanifold @xmath8 inside @xmath197 with trivial normal bundle . the manifold @xmath199 is contactomorphic to the standard contact sphere @xmath200 . * the manifolds @xmath201 and @xmath202 are contactomorphic . the manifold @xmath203 will be called the contact surgery blow up of @xmath0 along @xmath4 . the contact submanifold @xmath199 is called the exceptional divisor . by gray s stability , we may assume that a tubular neighbourhood of the embedded loop is contactomorphic to @xmath204 with the contact form @xmath188 as in ( [ eq : std ] ) , for some small radius @xmath205 . we enlarge this tubular @xmath206neighbourhood using the squeezing technique from @xcite to obtain a radius 2 neighbourhood . more precisely , we need the following auxiliary lemma : [ lem : ekp ] _ ( _ proposition 1.24 in @xcite _ ) _ [ lem : shrink ] let @xmath207 be a positive integer and @xmath208 a radius . then the following map is a contactomorphism @xmath209 and it restricts to the identity at @xmath210 . consider @xmath211 in the lemma above , then we need @xmath87 large enough to satisfy @xmath212 we may therefore assume that the tubular neighbourhood for which the standard equation ( [ eq : std ] ) holds for @xmath188 has radius @xmath64 . in the annulus corresponding to radius @xmath213 use @xmath214 to induce the contact structure given by @xmath215 . declare @xmath216 to define the contact structure in the radius area @xmath217 $ ] . it is left to find a strictly increasing function interpolating between @xmath218 and @xmath219 in the middle region . this can be done , see figure [ fig : h ] . and @xmath220 . ] the process described in the proof can be modified to include the radius squeezing in the gluing map . it suffices to use @xmath221 as gluing map instead of @xmath214 in the domain . indeed , denote @xmath222 and consider the contact structures @xmath223 define the map @xmath224 where @xmath225 is the obvious radius in the image . then the following diagram is commutative in the contact category : @xmath226^{\varphi}\ar@{->}[r]^{\phi_1 } & ( t_2,\xi_0)\ar@{->}[d]^{\psi_k}\\ ( t_{\varepsilon(k)},\xi_l)\ar@{->}[r]^{\phi_l } & ( t_{\varepsilon(k)},\xi_0 ) } \ ] ] where lemma [ lem : ekp ] is performed with parameter @xmath227 . note that the contactomorphism type of the exceptional divisor is that of the standard sphere , the parameter in the construction allows us to discretely vary the radius of the tubular neighbourhood we are collapsing . the maps @xmath221 and @xmath228 are smoothly isotopic if and only if @xmath229 is even . let @xmath230 be a circle coordinate . consider the morphism @xmath231 if @xmath232 denotes @xmath87times the standard circle action on @xmath183 induced by @xmath233 , then it is clear that @xmath228 is realized as @xmath234 . since @xmath235 and @xmath236 under this identification , @xmath237 if and only if @xmath229 is even . + it is left to prove that @xmath238 and @xmath214 are not isotopic , for @xmath43 even . construct two manifolds @xmath239 and @xmath240 by gluing two copies of the manifold @xmath241 respectively using @xmath238 and @xmath214 along the boundary . these manifolds are not diffeomorphic . a sphere is a spin manifold and the product formula for characteristic classes implies that so is @xmath242 . + the manifold @xmath240 is not spin . this can be seen by using any section @xmath243 of the twisted bundle @xmath244 , such @xmath243 exists because @xmath245 . denote by @xmath246 the normal bundle to the section and let @xmath247 be the complex bundle over @xmath151 such that @xmath248 . then @xmath249 . note that @xmath250 if @xmath43 is odd and @xmath251)=w_2(tx_{|s({\mathbb{s}}^2)})=w_2(\nu(s({\mathbb{s}}^2))=w_2(s^*(\nu(s({\mathbb{s}}^2)\oplus { \mathbb{r}}))= w(e_1).\ ] ] hence @xmath238 and @xmath214 are not isotopic . in particular , the smooth type of the contact blow up manifold will depend on the parity of the positive integer fixed for the construction . as for the contact type , it follows from theorem 1.2 in @xcite that the maps @xmath228 and @xmath221 are not contact compactly supported isotopic if @xmath252 . this does not imply that the contact structures are different , but at least there is no local contactomorphism relating the two contact structures . in this section we develop the contact blow up along a boothby wang submanifold , as suggested in @xcite.the existence of a minimum radius for the tubular neighbourhood of the submanifold along which we will perform the blow up will play an important role . this feature will be revisited in the definition provided in section [ sec : quotient ] . + let us review the definition of the symplectic blow up , see @xcite for more details . let @xmath253 be a symplectic manifold and @xmath4 a symplectic submanifold of codimension @xmath254 . consider the symplectic normal bundle @xmath255 of @xmath4 in @xmath0 and fix a compatible almost complex structure . the choice of a compatible almost complex structure for a symplectic form induces a metric and the equality @xmath256 implies that the structure group of @xmath257 can be considered to be @xmath258 . thus @xmath257 is an associated vector bundle of a @xmath258principal bundle @xmath259 . + the symplectic blow up of @xmath0 along @xmath4 is obtained by the fiberwise symplectic blow up of @xmath257 . hence we require definition [ def : lnbu ] for the case of symplectic vector spaces . let @xmath260 be the standard symplectic vector space and @xmath261 the standard fubini study form on complex projective space . we will use the following [ def : sybu ] a symplectic blow up of @xmath260 at the origin with radius @xmath262 is a symplectic manifold @xmath263 such that : * @xmath264 is a topological blow up of @xmath265 at the origin . the symplectic form induced on the exceptional divisor @xmath266 is @xmath267 . * for any @xmath268 , there exists a symplectomorphism @xmath269 * the unitary group @xmath258 acts hamiltonially in @xmath263 . + the symplectic blow up of @xmath260 at the origin exist for each @xmath262 . note that the definition depends on @xmath262 , this parameter does not appear in definition [ def : lnbu ] since any linear homothety at the origin is a complex isomorphism . let us describe the non linear symplectic blow up of @xmath0 along @xmath4 . property 3 in the above definition allows us to associate to @xmath117 a bundle @xmath270 over @xmath4 with fiber @xmath271 . let @xmath272 be a connection in @xmath117 , there are induced coupling forms @xmath35 and @xmath273 , in @xmath257 and @xmath274 respectively , restricting to the symplectic form on each fiber and coinciding away from the radius @xmath275 , see thm . 6.17 in @xcite . define the forms @xmath276 on the bundles @xmath257 and @xmath277 . these are symplectic forms close to the zero section and to the exceptional divisor respectively . + these forms also coincide away from a neighbourhood of @xmath4 of radius @xmath278 . let @xmath279 be a neighbourhood of the zero section of the symplectic normal bundle . by the symplectic neighbourhood theorem there is a neighbourhood @xmath280 of the symplectic submanifold @xmath4 and a symplectomorphism @xmath281 . thus any fiberwise symplectic blow up on @xmath257 with radius @xmath282 can be glued back to the initial manifold @xmath0 using the symplectomorphism @xmath283 . the resulting manifold is the symplectic blow up of @xmath0 along @xmath4 with radius @xmath262 . + observe that the radius of the tubular neighbourhood of @xmath4 can not be estimated a priori . therefore the symplectic volume of the exceptional divisor can not be assumed to be arbitrarily large . this will be an obstruction to develop the gromov s approach in the contact category . + * example : * let @xmath284 be a rank@xmath285 symplectic vector bundle over a symplectic manifold @xmath74 . then the total space is symplectic as well . thus , we are able to blow up the symplectic manifold @xmath284 along its zero section @xmath17 . in case the symplectic form @xmath77 is of integer class , the symplectic form in the resulting blown up manifold will be of an integer class if the blow up radius is @xmath286 , @xmath287 . we call this a radius @xmath288 blow up . + we now define the contact blow up in terms of the symplectic blow up . this is the second notion listed in section [ sec : intro ] . + let @xmath25 be a contact manifold and @xmath289 a contact submanifold . we assume : * the contact submanifold @xmath4 is contactomorphic to a boothy wang manifold @xmath290 . + * let @xmath291 be the circle bundle projection . there exists a symplectic bundle @xmath284 over @xmath17 such that , as symplectic bundles @xmath292 . the total space of @xmath284 carries a symplectic form @xmath293 in the same cohomology class of @xmath75 $ ] , under the natural identification of @xmath294 with @xmath295 . as previously explained , there exists a symplectic manifold @xmath296 obtained by blowing up @xmath284 along its zero section @xmath17 . suppose that the parameter multiplying the class of the exceptional divisor @xmath8 in the symplectic blow up is a positive integer , i.e. the symplectic form in @xmath297 is integral . + the construction of the contact blow up is based on the following diagram : @xmath298 & \mathbb{s}(v ) \ar@{->}[d ] & \mathbb{s}(\widetilde{v})\supset\mathbb{s}(e)=\mathbb{s}(\widetilde{v})|_e \ar@{->}[d ] \\ & ( s,\xi_s)\cong\mathbb{s}(w ) \ar@{->}_{\pi}[d ] & ( v,\overline{\omega } ) \ar@{->}[ld ] & ( \widetilde{v},{\overline{\omega}_w})\supset e \ar@{->}[l ] \\ & ( w,\omega)}\ ] ] each map is a bundle projection . it is essential to understand the relation between the contact manifolds @xmath299 and @xmath300 . this is the content of the following : in the hypotheses above , @xmath301 is a contact submanifold of @xmath302 . there are contactomorphic neighbourhoods @xmath303 and @xmath280 in @xmath302 and @xmath0 respectively . the choice of symplectic form on @xmath284 implies that there exists a symplectic embedding of @xmath17 in @xmath284 and therefore @xmath301 is contained in @xmath302 as a contact submanifold . the tubular neighbourhood theorem states that the normal bundle @xmath3 is diffeomorphic to a small neighbourhood of @xmath4 in @xmath0 , but @xmath304 so the same situation applies to @xmath301 in @xmath302 . the statements now follow from the contact neighbourhood theorem . in consequence , @xmath305 provides a local model . thus we only need to perform the blow up of @xmath284 along @xmath17 and study whether the boothby wang structures associated to them allow us to glue back the resulting blown up model to @xmath0 . this is the content of the following : + [ propo : bundle_blow ] let @xmath306 be a boothby wang contact submanifold of @xmath307 . suppose we symplectically blow up @xmath308 by collapsing a radius @xmath29 neighbourhood . then , there is a choice of contact form for @xmath309 such that @xmath310 is a contact submanifold of @xmath309 and the complement of an arbitrary small neighbourhood of @xmath310 in @xmath309 is contactomorphic to the complement of some neighbourhood of @xmath301 in @xmath302 . for the sake of a clearer exposition the proof is explained at the end of this subsection . + suppose we can choose a tubular neighbourhood @xmath311 with radius larger than @xmath29 which is contactomorphic to a tubular neighbourhood @xmath312 . then we have the following the contact blow up of @xmath25 along @xmath289 is the contact manifold @xmath313 obtained by removing the neighbourhood @xmath280 and gluing along its boundary a small neighbourhood of @xmath310 in @xmath309 . the contact manifold @xmath313 is contactomorphic to @xmath0 away from small neighbourhoods of @xmath310 and @xmath4 respectively . the _ exceptional divisor _ of the contact blow up is defined to be @xmath310 , where @xmath8 is the exceptional divisor of the symplectic blow up over which it is locally modelled . observe that for the definition to work we need @xmath4 to have a tubular neighbourhood of radius at least @xmath29 inside @xmath0 . + * example * : 1 . the most simple example of contact blow up is the case of a transverse loop @xmath314 in @xmath66 . the loop is contactomorphic to @xmath315 and its normal bundle is the pull back of the trivial bundle over the point . thus h1 and h2 are satisfied . the symplectic model corresponds to the blow up of @xmath316 at the origin , collapsing a neighbourhood of radius @xmath29 , and therefore @xmath317 . hence , @xmath318 , i.e. the standard contact @xmath319sphere . this particular case can be seen , at least topologically , as a surgery along a loop . + 2 . in the previous example we may symplectically blow up with radius @xmath320 . the exceptional divisor is then @xmath321 , i.e. the sphere bundle associated to the polarization @xmath322 of @xmath323 , which is the lens space @xmath324 with its standard contact structure . therefore , even the diffeomorphism type of the blown up contact manifold changes with the blow up radius @xmath320 . + note that there is not natural projection map from @xmath310 , the exceptional divisor , to the blow up locus @xmath301 . in the case of a loop in a @xmath22dimensional manifold , the exceptional divisor for a radius @xmath29 blow up is @xmath72 and the blow up locus is the circle @xmath120 . this is a difference with respect to the symplectic and algebraic cases where the exceptional divisor is a bundle over the submanifold along which the blow up is performed . it is true though that there is a natural projection @xmath325 , but it does not lift to @xmath301 . the assumption of the integer radius can be fulfilled in certain cases . for instance in the blow up along a transverse @xmath120 we can use the lemma [ lem : shrink ] . therefore the construction in this case will have two natural parameters : the integer radius that determines the topology of the exceptional divisor , and the choice of framing in the spirit of the lemma . the above described construction la gromov does not show in general the appearance of this second positive integer , this is a reason to introduce a third way of defining the blow up highlighting these two choices . to conclude this subsection we prove the assertion that allowed us to glue the boothby wang construction over the exceptional divisor in the contact blow up construction . + _ proof of proposition [ propo : bundle_blow]_. we need to find an appropriate connection on the topological boothby wang manifold corresponding to @xmath297 . + notice from the construction of the symplectic blow up as given in @xcite , we know that given a sufficiently small neighbourhood of @xmath8 in @xmath297 it is possible to choose a symplectic form @xmath293 on @xmath297 such that complement of that neighbourhood in @xmath297 is symplectomorphic to a small neighbourhood of @xmath17 in @xmath284 . furthermore , observe that the exceptional divisor is just the inverse image of @xmath17 contained in @xmath284 as the zero section under the blow up projection @xmath326 . + now recall from the definition [ def : bw ] that the contact structure of @xmath309 is determined by the choice of a connection over the associated line bundle whose curvature is @xmath327 . so let @xmath157 be the connection over @xmath56 that determines the contact structure on @xmath302 , and denote by @xmath328 an arbitrarily small neighbourhood of @xmath17 inside @xmath284 . from the construction of the symplectic form @xmath293 on @xmath297 we can assume that the map @xmath329 is a symplectomorphism of @xmath330 and @xmath331 . therefore the connection @xmath332 satisfies the required properties over @xmath331 . it remains for it to be extended to a connection all over @xmath297 with curvature @xmath327 . by lemma [ lem : exten_bw ] such an extension is possible provided that the restriction morphism from @xmath333 is surjective . it is sufficient to show that @xmath334 . + indeed , observe that @xmath297 is homotopic to a @xmath335bundle over @xmath17 , with @xmath336 , and hence @xmath337 holds . note that the manifold @xmath331 is diffeomorphic to @xmath330 and the set @xmath330 is homotopy equivalent to a sphere bundle over @xmath17 with fibers of dimension greater than @xmath99 . from the long exact sequence of homotopy we conclude that @xmath338 it now follows from lemma [ lem : exten_bw ] that on @xmath309 there is a choice of contact form with the required properties . away from a given small neighbourhood of @xmath310 in @xmath309 a contact form can be chosen such that it induces a contactomorphism from the complement of such a neighbourhood to the complement of some neighbourhood of @xmath17 in @xmath339 this is because we can choose the symplectic form on @xmath297 with the required property . @xmath340 in this section we define the contact blow up of a contact manifold @xmath0 along a boothby wang contact submanifold @xmath4 using the notion of contact cuts . given a @xmath120action on a manifold @xmath0 , topologically the cut construction is based on collapsing the boundary of a tubular neighborhood of a given submanifold invariant by the action . basic knowledge on the contact reduction procedure is assumed in the next few paragraphs , see @xcite . let us recall the construction of a contact cut for a contact @xmath120action as developed by e. lerman : [ thm : contcut ] @xmath193thm . 2.11 in @xcite@xmath194 let @xmath341 be a contact manifold with a @xmath120action preserving @xmath35 and let @xmath342 denote its moment map . suppose that @xmath120 acts freely on the zero level set @xmath343 . then the set and @xmath344 for some @xmath345 . ] @xmath346 is naturally a contact manifold . moreover , the natural embedding of the reduced space @xmath347 into @xmath348 is contact and the complement @xmath349 is contactomorphic to the open subset @xmath350 note that the contact reduction requires the regular value to be @xmath351 , whereas in the symplectic reduction any regular value is licit . this is so because in the contact reduction it is imposed that the orbits of the isotropy subgroup are tangent to the contact structure , see @xcite . let @xmath173 be a contact manifold and @xmath352 a codimension@xmath285 contact submanifold . suppose that @xmath353 for some symplectic manifold @xmath74 , @xmath354 , and @xmath355 as complex bundles over @xmath4 . we will define the _ contact blow up of @xmath0 along @xmath4_. any isocontact embedding is isocontact if @xmath356 . ] of a contact @xmath319fold in a sphere has trivial normal bundle . this situation does occur : any closed cooriented @xmath319fold admits an isocontact embedding into the standard contact @xmath357sphere . a tubular neighbourhood of the contact submanifold @xmath4 is contactomorphic to @xmath358 with the contact structure given by @xmath359 , where @xmath360 is the standard contact form in @xmath361 . let @xmath362 and consider the @xmath120action @xmath363 @xmath364 this action is generated by the field @xmath365 where @xmath366 are the reeb vector fields associated to @xmath35 and @xmath360 . the moment map of the above action is @xmath367 @xmath368 the contact cut can only be performed in the pre image of the regular value @xmath369 , it is thus a necessary condition that @xmath370 . this can always be achieved if @xmath134 is large enough . let @xmath371 be a contact submanifold of @xmath25 with fixed trivial normal bundle @xmath372 . let @xmath362 be such that @xmath370 . the @xmath373contact blow up @xmath374 of @xmath0 along @xmath4 is defined to be the contact cut of @xmath0 for the moment map associated to the circle action @xmath375 @xmath376 the collapsed region @xmath377 will be called the _ exceptional divisor _ , it is a contact manifold of dimension @xmath378 . the induced @xmath120action in the level set @xmath379 coincides with the action @xmath380 defined in theorem [ thm : bw ] with @xmath381 and @xmath382 . thus , the orbit space is @xmath383 notice that both the topology and the contact structure of the exceptional divisor strongly depend on the choice of the parameters @xmath137 and @xmath134 . consequently , so does @xmath374 . * example * : 1 . in the case of a contact @xmath22fold , a transverse circle the simplest contact submanifold is replaced by a ( quotient of a ) standard contact @xmath319sphere , as in section [ sec : srgy ] . this new construction of the blow up along a transverse loop will be compared with the previous ones in the next section . + 2 . consider the contact blow up along a contact @xmath319sphere @xmath384 . the topology of the exceptional divisor will depend on the element of the corresponding higher homotopy group , cf . section [ sec : bw ] . if in the previous example @xmath25 is a @xmath22dimensional contact manifold , the exceptional divisor of the @xmath385 blow up is contactomorphic to @xmath72 . in higher dimensions , the exceptional divisor of a @xmath385 blow up along @xmath72 is diffeomorphic to @xmath386 for @xmath387 and @xmath87 even . we define the contact blow up along a contact submanifold with a general normal bundle . the construction will clearly coincide with the previous blow up in the case of a trivial normal bundle . in smooth topology the smooth structure of a neighbourhood of a submanifold is retained by the normal bundle . the contact geometry nearby a contact submanifold @xmath289 is determined by the normal bundle @xmath257 along with a conformally symplectic structure . such a structure exists because @xmath257 can be identified with the symplectic orthogonal @xmath388 . the statement of the contact neighbourhood theorem is as follows : @xmath1932.5.15 in @xcite@xmath194 let @xmath389 and @xmath390 be contact pairs such that @xmath391 is contactomorphic to @xmath392 . if @xmath393 as conformally symplectic bundles , then there exists a contactomorphism between suitable neighbourhoods of @xmath394 and @xmath395 . there exist contact submanifolds with non trivial normal bundle in a closed contact manifold . let us provide some examples . + * examples * : 1 . let @xmath396 be a cooriented contact manifold and @xmath24 itself be non trivial as an abstract vector bundle . the contact form provides a contact embedding @xmath397 such that the normal bundle of the contact submanifold @xmath0 is isomorphic to @xmath24 . let @xmath398 be an integral symplectic manifold and @xmath17 a symplectic submanifold with non trivial normal bundle . for instance , @xmath399 . then the normal bundle of the contact submanifold @xmath400 in @xmath307 is also non trivial . let @xmath173 be a closed cooriented contact manifold . consider an isocontact embedding @xmath401 see @xcite for the existence of such an embedding . since the tangent bundle of the spheres are stable , it is simple to give sufficient conditions for the normal bundle to be non trivial , e.g. @xmath0 not spin . the contact blow up construction has been used in another context . given a complex vector bundle @xmath8 on @xmath0 , the contact submanifold @xmath196 is defined as the vanishing set of a section in @xmath402 . then @xmath403)\neq0 $ ] . this occurs for the base locus of contact lefschetz pencil decompositions of @xmath25 . see @xcite . in the blow up construction for the trivial normal bundle case there are two circle actions . the first one exists on the contact submanifold @xmath4 , since it is a boothby wang manifold , and it is extended to a local neighbourhood . the second circle action is the gauge action provided by the complex structure in the conformally symplectic normal bundle . the latter is still available in the non trivial normal bundle case , the former can _ a priori _ no longer be extended in a neighbourhood . + we hence require a lifting condition for the circle action on @xmath4 : the appropriate set up is depicted as in the diagram 1 in section [ sec : gromov ] : @xmath298 & \mathbb{s}(v ) \ar@{->}[d ] \\ & ( s,\xi_s)\cong\mathbb{s}_a(w ) \ar@{->}_{\pi}[d ] & ( v,\overline{\omega } ) \ar@{->}[ld ] \\ & ( w,\omega)}\ ] ] where @xmath284 is a symplectic bundle over a symplectic manifold @xmath17 . assume @xmath404 for simplicity . + in the hypotheses above , the circle action provided by the boothby wang structure can be naturally extended to a neighbourhood of @xmath4 . since @xmath17 is a symplectically embedded submanifold of @xmath284 , @xmath400 is a contact submanifold of @xmath307 . the tubular neighbourhood theorem tells us that the normal bundle @xmath3 is diffeomorphic to a small neighbourhood of @xmath4 in @xmath0 , but after the smooth isomorphism @xmath405 the same situation applies to @xmath400 in @xmath307 . since the isomorphism holds at the level of symplectic bundles , the contact tubular neighbourhood theorem ensures that there exists a contactomorphism @xmath406 between a contact neighbourhood of the zero section in @xmath407 and a contact neighbourhood of @xmath400 in @xmath307 . consequently , the circle action in @xmath307 can be carried along @xmath406 to a neighbourhood of @xmath4 . let us spell out the moment map of the circle action . we refer to the circle action on the normal bundle induced by its complex structure as the gauge action . this action is the natural @xmath120action when working with a contact pair @xmath408 . further , the radius coordinate @xmath409 is a global coordinate regardless of the non triviality of the normal bundle . the remaining action described above will be referred as the boothby wang action . it is the natural action when identifying a neighbourhood of @xmath4 in @xmath0 with a neighbourhood of @xmath400 in @xmath307 via the map @xmath406 in the proof of the lemma . the moment map of the @xmath120action @xmath410 is @xmath411 . the moment map of the gauge action is @xmath412 . for the boothby wang one , the circle action realizes the reeb vector field and thus its moment map is @xmath29 . we express the @xmath413 coordinate through the contactomorphism @xmath406 as @xmath414 . since we are using the action @xmath410 the statement follows . recall that the contact cut can be performed if @xmath351 lies in the image of the moment map . the same argument using a multiple of the gauge action concludes that we may modify the action in order to ensure this : @xmath406 maps the zero section to @xmath400 and thus the values of @xmath415 form a decreasing sequence in @xmath134 that eventually crosses zero , @xmath416 being bounded below . the boothby wang action may as well be arranged to period @xmath137 : the concatenation action is denoted @xmath380 . we are in position to write the @xmath193contact blow up@xmath194 let @xmath371 be a contact submanifold of @xmath25 . let @xmath144 be such that the origin is contained in the image of the moment map @xmath417 for the action @xmath380 . the @xmath193a , b@xmath194contact blow up @xmath374 of @xmath0 along @xmath4 is defined to be the contact cut of @xmath0 for the action @xmath380 , i.e. @xmath418 . in this section we relate the three constructions of the contact blow up . the construction that can be performed in the most general situation is the one involving the contact cut . it has two degrees of freedom : a pair of positive integers @xmath137 and @xmath134 . these two parameters relate to previous integers appearing in the first two constructions . indeed , the parameter @xmath419 in the contact surgery blow up corresponds to @xmath134 . for gromov s construction , the choice of collapsing radius @xmath420 gives rise , in the case of transverse loops , to the exceptional divisor @xmath421 and it corresponds to the parameter @xmath137 . it is quite obvious that the diffeomorphism type of the blown up manifolds is the same regardless of the chosen construction as soon as the parameters coincide as just mentioned . + let us turn our attention to the contact structure : we restrict ourselves to the case of transverse loops . denote by @xmath422 the surgery contact blow up defined in section [ sec : srgy ] with parameter @xmath134 . the contact blow up as defined in section [ sec : gromov ] with radius @xmath137 is denoted by @xmath423 . and @xmath424 will be the contact cut blow up as defined in section [ sec : quotient ] , performed with parameters @xmath373 . let us show that uniqueness holds in this case , more precisely we prove the following [ thm : unique ] let @xmath25 be a contact manifold . performing the blow up along a fixed transverse loop with the three procedures introduced previously , the resulting blown up manifolds @xmath425 , @xmath426 and @xmath427 endowed with the blown up contact structures are contactomorphic . further , given any pair of integers @xmath373 , the following contactomorphisms hold : @xmath428 the relation between the different constructions is already hinted in section [ sec : srgy ] . since the exceptional contact divisors coincide and the procedure is of a local nature , i.e. the contact manifold is not altered away from a neighbourhood of the embedded transverse loop , the study should focus on the natural annulus contact fibration . let us review a few facts . + a contact fibration is a fibration @xmath429 such that the fibers are contact submanifolds . we consider contact fibrations over the disk @xmath430 . the base being contractible , the fibration is trivial and we also assume it to be trivialized . let us introduce the following let @xmath431 be polar coordinates on the disk @xmath432 . a trivialized contact fibration over the disk @xmath433 is said to be radial if the contact structure admits the following equation @xmath434 where @xmath435 is a smooth function such that @xmath436 . notice that for the total space of a radial contact fibration to have an induced contact structure it is required that @xmath437 it is convenient to extend the previous definition in order to include the general situation , where lens spaces may appear as exceptional contact divisors : the action in the fiber sphere @xmath96 is generated by an @xmath440rotation along the hopf fiber , whereas the action in the base @xmath432 is the standard @xmath440rotation in the disk . they preserve respectively the standard contact structure in @xmath96 and the @xmath29form @xmath441 in the disk . hence , the fibration becomes equivariant if the function @xmath125 is preserved by the action . + topologically it is fairly straightforward that the blow up operations we are performing are tantamount to a priori different fillings of the fibration over an annulus to form a manifold lying over the disk this being always considered up to a finite action @xmath439 , for lens spaces fillings . the transition from @xmath174 to @xmath176 can be understood in these terms : both fibrations over the annulus produced by restricting to @xmath442 are filled in the origin with a circle and @xmath96 respectively . in the transverse loop case it will be enough to use the following [ lem : unique ] let @xmath284 be a manifold with contact structures @xmath443 and @xmath444 . assume that there are two smoothly isotopic diffeomorphisms @xmath445 which are contactomorphisms for @xmath443 and @xmath444 respectively . let the two fibrations be radial contact fibrations with common contact fiber @xmath446 and satisfying that the diffeomorphism @xmath447 is the identity close to the boundary . then , the contact structures @xmath443 are @xmath444 are isotopic . + further , if the fiber is @xmath448 and the contact fibrations are @xmath439equivariant , the contact structures are isotopic through @xmath439equivariant contactomorphisms . this can be reduced to the setup with a fibration @xmath449 with two different radial contact structures @xmath450 such that the hamiltonians @xmath451 and @xmath452 coincide close to the boundary . in that setting , we just need to construct a path of functions @xmath453 connecting them , relative to the boundary , satisfying the contact equation ( [ eq : contact ] ) and the condition @xmath454 . but this is possible since the space of such functions is convex . + the argument still works in the equivariant case : the only sentence to be added is that the space of equivariant hamiltonians is also convex . thus , to conclude uniqueness we study the contact topology of the different blow up constructions and ensure that the lemma applies . + _ proof of theorem [ thm : unique ] _ : let us describe the common model fibration that underlies the three constructions in this case . consider a standard contact neighbourhood @xmath455 of the given fixed loop and the morphism @xmath456 it does generalize the diffeomorphism provided by equation ( [ eq : change ] ) that reflects the case @xmath404 . if @xmath137 is greater than @xmath29 , it becomes a @xmath457 covering . the covering transformation is provided by @xmath439 acting through : @xmath458 which is free as long as @xmath127 . to understand the change in the contact structure , note that the pull back of the standard contact form @xmath459 is given by @xmath460.\ ] ] denote by @xmath461 the critical radius where the distribution becomes horizontal . in these coordinates , for any fixed small @xmath268 , the projection onto the first two factors @xmath462 provides a radial contact fibration on the annulus and since the function @xmath463 is strictly positive in @xmath464 , it can be extended to the interior of the disk to a @xmath439equivariant radial contact fibration . in order to glue back the model to the manifold we should quotient the equivariant contact fibration by @xmath439 , this allows us to use the map @xmath465 to insert the model back into the manifold . + it is thus left to verify that the three blow up procedures provide examples of such an extension for particular values of @xmath373 . then lemma [ lem : unique ] will apply to provide the uniqueness of the constructions . note that the contact surgery blow up construction is by definition a radial contact fibration , with @xmath404 , as shown in section [ sec : srgy ] . let us study the two remaining cases . + to understand the proof in section [ sec : gromov ] , let us proceed backwards and instead of applying the boothby wang construction , we produce a contact structure and then quotient the resulting contact manifold by the reeb @xmath120action to study whether it is the correct object . once the coordinate change @xmath465 is performed , the reeb vector field @xmath466 becomes @xmath467 this vector field extends to the interior of the disk fibration and so we may quotient the resulting manifold @xmath241 . we obtain the blown up symplectic ball @xmath468 as its quotient . we can further quotient by the free @xmath439action to obtain a non trivial fibration over the disk @xmath432 . this proves that a suitable choice of connection leads to an equivariant contact fibration . + there are other choices of connection though . from the principal bundle point of view , a radial contact fibration over the annulus @xmath469 corresponds to a connection on @xmath470 certainly , after proposition [ propo : bundle_blow ] the contact structure is fixed with the choice of a connection . note that the space of connections is affine and thus , after gray s stability theorem , the resulting contact structures are contact isotopic for different choices of connections . in conclusion , this second model also provides an extension of the model fibration . + we describe the third procedure also beginning with the resulting contact manifold and giving the pull back of the action . this contact cut construction is also an equivariant radial contact fibration since the pull back of the vector field generating the @xmath120action , that is @xmath471 is expressed as @xmath472 after the coordinate change . there , the contact cut is just an equivariant radial contact fibration , see the proof of theorem 2.11 in @xcite . @xmath340 + using lemma [ lem : unique ] , we can show that the contact blow up is unique up to the choice of a trivializing chart of the neighbourhood of the transverse loop . in order to prove the uniqueness of the blow up along transverse loops , we would need to study the space of isocontact embeddings of the contact manifold @xmath174 in @xmath0 . it is probably false that it is connected , which is the requirement needed to ensure the uniqueness of the blow up once the parameters @xmath473 are fixed . xxxx r. bott , l. tu , differential forms in algebraic topology , _ gtm _ 82 , springer w. boothby and h. wang , on contact manifolds , _ annals of math . _ 68 ( 1958 ) , 721734 . r. casals , d. pancholi , f. presas , almost contact @xmath22folds are contact , arxiv:1203.2166v2 . y. eliashberg , s. s. kim , l. polterovich , geometry of contact transformations and domains : orderability versus squeezing , _ geom . _ 10 ( 2006 ) , 16351747 . o. van koert , open books on contact five - manifolds , _ annales de linstitut fourier _ 58 ( 2008 ) , 139157 . h. geiges , an introduction to contact topology , _ cambridge studies in advanced mathematics _ 109 . h. geiges , constructions of contact manifolds , _ math . _ 121 ( 1997 ) , 455564 . e. giroux , structures de contact en dimension trois et bifurcations des feuilletages , _ inventiones math . _ 141 ( 2000 ) 615689 . m. gromov , partial differential relations , _ ergeb . math . _ 9 ( 1986 ) , springer r. hartshorne , algebraic geometry , _ gtm _ 52 , springer verlag . e. lerman , contact cuts , _ israel j. math . _ 124 ( 2001 ) , 7792 . e. lerman , symplectic cuts , _ math . _ 2 ( 1995 ) , 247258 . d. mcduff , d. salamon , introduction to symplectic topology , _ oxford math . ( 1998 ) , oxford . f. presas , lefschetz type pencils on contact manifolds , _ asian j. math . _ 6 ( 2002 ) , 277302 . f. presas , a class of non fillable contact structures , _ geometry & topology _ 11 ( 2007 ) 22032225 . y. wang , w. ziller , einstein metrics on principal torus bundles , _ j. differential geom . _ 31 ( 1990 ) , 215248 .

the effects of gd doping and o vacancies on the magnetic interaction and curie temperature of euo are studied using first - principles calculations . linear response calculations in the virtual crystal approximation show a broad maximum in the curie temperature as a function of doping , which results from the combination of the saturating contribution from indirect exchange and a decreasing contribution from the- hopping mechanism . non - heisenberg interaction at low doping levels and its effect on are examined . the electronic structure of a substitutional gd and of an o vacancy in euo are evaluated . when the spins are disordered , the impurity state goes from single to double occupation , but correlated bound magnetic polarons are not ruled out . at higher vacancy concentrations typical for eu - rich euo films , the impurity states broaden into bands and remain partially filled . to go beyond the homogeneous doping picture , magnetostructural cluster expansions are constructed , which describe the modified exchange parameters near gd dopants or o vacancies . thermodynamic properties are studied using monte carlo simulations . the curie temperature for gd - doped euo agrees with the results of the virtual crystal approximation and shows a maximum of about 150 k. at 3.125% vacancy concentration the increases to 120 k , consistent with experimental data for eu - rich film samples .

1308.1223

euo is a ferromagnetic oxide semiconductor @xcite with unusual magnetic and transport properties.@xcite electron doping by trivalent rare - earth elements nearly doubles its curie temperature @xmath0 , while the @xmath4 curve deviates significantly from the brillouin function.@xcite a similar enhancement of @xmath0 is observed in eu - rich euo films.@xcite the magnetic transition in bulk eu - rich euo is accompanied by a metal - insulator transition,@xcite while the @xmath0 is not enhanced . recent interest in euo is due to its potential applications in spintronics as a spin filter in tunnel junctions @xcite in insulating pure form and as a high spin - polarization material @xcite when doped . it was shown that euo can be grown epitaxially on various substrates including si , gan , and gaas.@xcite ferromagnetism in pure euo is due to the @xmath3-@xmath5 hopping mechanism.@xcite first - principles calculations confirm that magnetic interaction is short - range and ferromagnetic.@xcite large enhancement of @xmath0 under doping is induced by the filling of the conduction band and the associated indirect exchange mechanism . this mechanism is physically of the ruderman - kittel - kasuya - yosida ( rkky ) type , @xcite but at low doping levels it deviates from the heisenberg form owing to the half - metallic character of the conduction band in the ferromagnetic state.@xcite this non - heisenberg character may be responsible for the peculiar shoulder observed in the @xmath4 curves for gd - doped euo . @xmath0 goes through a broad maximum as a function of gd concentration.@xcite first - principles calculations have been limited to the virtual crystal approximation and were based on the fitting of the total energies for several magnetic configurations to the heisenberg model.@xcite these calculations yield a broad maximum in @xmath0 in agreement with experiment . it was , however , argued @xcite that this maximum is associated with dynamical effects . on the other hand , mairoser _ _ et al.__@xcite observed that the carrier density appears to fall behind the actual gd concentration and raised the question whether the intrinsic limit of @xmath0 has been achieved . the metal - insulator transition in eu - rich euo involves impurity levels on o vacancies,@xcite which absorb free charge carriers from the conduction band in the paramagnetic state . whether a similar transition exists in gd - doped euo @xcite due to shallow donor levels is not fully clear , as the transition is not observed in all measurements and may be due to the presence of oxygen vacancies.@xcite here we investigate the electronic structure and exchange interaction in gd - doped and eu - rich euo using first - principles electronic structure methods combined with both linear response and total energy calculations . the paper is organized as follows . linear response calculations in the virtual crystal approximation are described in section [ vca ] . the electronic structure of isolated gd dopants and o vacancies , and in particular the impurity levels introduced by o vacancies , are discussed in section [ isolated ] . the magnetostructural cluster expansions for gd - doped and o - deficient euo are constructed in section [ sec : msce ] . thermodynamic properties are evaluated using monte carlo simulations in section [ sec : mc ] . the conclusions are drawn in section [ conclusions ] . the results of sections [ sec : msce ] and [ sec : mc ] for gd - doped euo were , in part , briefly reported in conference proceedings.@xcite in this section we evaluate the effect of electron doping on the exchange interaction and curie temperature of euo using the linear response method for the calculation of exchange parameters @xcite within the density functional theory ( dft ) . we use green s function - based formulation of the tight - binding linear muffin - tin orbital ( tb - lmto ) method @xcite in the atomic sphere approximation ( asa ) . the lattice constant is fixed at the experimental value 5.14 . to improve agreement with the full - potential band structure , two empty spheres ( es ) per formula unit are placed at the centers of all nearest - neighbor eu tetrahedra . the electron doping is introduced within the virtual crystal approximation ( vca ) by adding the electron doping concentration @xmath6 to the nuclear charge of the eu atom . in order to expose the different mechanisms of exchange interaction , we have considered three different models . model 1 is designed to provide the best fit to the quasiparticle self - consistent @xmath7 ( qs__gw _ _ ) calculation.@xcite the basis set includes @xmath8 , @xmath9 , @xmath5 , and @xmath3 states on eu , @xmath10 , @xmath11 , and @xmath12 states on o , and @xmath13 , @xmath14 , and @xmath12 states on the empty spheres . the lmto sphere radii were chosen as 1.79 for eu , 1.16 for o , and 0.74 for the empty spheres . strong correlations within the @xmath3 shell were included using the lda@xmath15 method @xcite with @xmath16 ev and @xmath17 ev . in addition , an auxiliary external potential of @xmath18 ev was applied to the o @xmath11 states in order to place them at the correct binding energy.@xcite the resulting band structure of undoped euo is shown in fig . [ bandeuo ] . the indirect @xmath19-x band gap is 0.76 ev , and the exchange splitting of the conduction band minimum ( cbm ) at the x point is 0.78 ev . the band gap is a little too small ( the experimental optical absorption gap is 0.95 ev at low temperatures@xcite ) , and the cbm splitting is somewhat larger than the experimentally reported 0.6 ev.@xcite in models 2 and 3 the `` open core '' approximation is used for the eu @xmath3 electrons , whereby these states are treated as a polarized core and thereby not allowed to hybridize . in these models the lmto sphere radii were 1.69 , 1.27 , and 0.86 for eu , o , and empty spheres , respectively . in model 2 , a @xmath20 ev external potential is applied to the o @xmath11 states just as in model 1 , but in model 3 this is not done . the difference between these two models highlights the effects of oxygen hybridization with the conduction - band states . table [ vcajij ] lists the important exchange parameters ( see fig . [ scheme]a for an illustration ) calculated for the three models at zero and 5% electron doping levels , as well as the differences between them . first , let us comment on the physical meaning of these parameters . in the linear response method @xcite the exchange parameter for a pair of sites @xmath21 and @xmath22 is defined as the second derivative of the total energy with respect to the angles of rotation of the exchange - correlation fields in the corresponding atomic spheres : @xmath23 . in the lmto setups all sites carry local moments , including o and the empty spheres es . therefore , one can formally define the pair exchange parameters for the eu - o pairs , eu - es pairs , etc . table [ vcajij ] lists all such parameters that are not negligibly small in undoped euo ( others are less than 0.01 mev ) . in the doped case a long - range indirect interaction sets in , which will be discussed separately below . although individual pair parameters for this long - range part are not listed in table [ vcajij ] , it contains the full lattice sum of the eu - eu pair parameters ( @xmath24 ) , as well as the full lattice sum of all pair parameters connecting a given site x to the rest of the lattice ( @xmath25 ) . note that the exchange parameters are first calculated in reciprocal space and then fourier - transformed to real space . the lattice sums are thus calculated exactly , without a real - space cutoff . a sensible effective model of exchange interaction in euo should only include eu spins as dynamical variables . the small magnetic moments on o are induced by exchange interaction ( and hybridization for model 1 ) with the @xmath3 states , while those on es sites simply reflect the extension of the eu and o orbitals beyond their atomic spheres . these local moments do not fluctuate independently , but should be assumed to adiabatically follow the fluctuating @xmath3 moments on eu . we should not , therefore , interpret the exchange parameters connecting o and es to other sites as having direct physical meaning , but these interactions should contribute to the true effective eu - eu exchange parameters . for the estimate of the curie temperature @xmath0 , in the mean - field approximation we should consider the total weiss field acting on a given eu spin . in doing so , we should exclude the part attributable to the tails of the orbitals on the central eu site itself . in other words , we should exclude the spurious effects of the orbital `` tails '' being fixed while its `` head '' is rotated , which is what happens , for example , when a eu - o exchange parameter is evaluated using the linear response technique in lmto . since such parameters ( eu - o and eu - es ) extend only to the nearest - neighbor sites of the given kind , on a crude level we can simply reduce each such parameter by a factor @xmath26 , where @xmath27 is the number of nearest - neighbor eu atoms for site @xmath22 ( which is either o or es ) . for model 1 we can then estimate the `` true '' effective field parameter @xmath28 as @xmath29 mev , which is only slightly smaller than @xmath30 mev . the classical mean - field estimate of @xmath31 for undoped euo then becomes 140 k ( we use classical mean - field estimates of @xmath0 to facilitate comparison with classical monte carlo results in the following ) . this value is about twice as large compared to the experimental value . this overestimation can only partially be attributed to the mean - field approximation , and the remaining error is probably related to the asa . .exchange parameters ( mev ) for pure and electron - doped euo in the virtual crystal approximation . [ cols="<,^,^,^,^,^,^,^,^,^ " , ] [ vcajij ] comparison between the three models provides information about the exchange mechanisms . in model 2 , where the @xmath3 electrons are not allowed to hybridize , @xmath32 and @xmath33 are reduced by an order of magnitude , while @xmath34 becomes negative . the eu - o coupling , on the other hand , is only weakly affected . these features confirm that the dominant part of the ferromagnetic coupling in euo is due to the hybridization of @xmath3 states with the conduction - band eu states . @xcite note that in ref . strong antiferromagnetic coupling was found in the open - core approximation , the origin of which is unclear to us . weak ferromagnetic coupling in our open - core calculations is due to the bloembergen - rowland mechanism,@xcite which involves the hybridization of o states with the empty conduction band . the downward shift of the o @xmath11 states in model 2 reduces this mechanism compared to model 3 , but it is quite small in both models 2 and 3 . let us now focus on the effects of electron doping . it is seen from table [ vcajij ] that the changes in the exchange parameters due to 5% electron doping are quite similar across all three models . the dominant effect of doping is the introduction of an additional indirect contribution to eu - eu coupling , while the parameters connecting eu to o and es sites are only weakly modified . this indirect - exchange enhancement of the eu - eu coupling is not sensitive to the position of the o @xmath11 states ( compare models 2 and 3 ) , as expected for this mechanism . fig . [ tcdop ] shows the enhancement of @xmath0 in euo as a function of electron doping @xmath6 evaluated in the mean - field approximation for models 1 and 3 as @xmath35 . in model 1 the @xmath0 increases with the doping concentration @xmath6 up to an optimal doping level of 10 - 15% and then declines . this behavior agrees with earlier theoretical results @xcite and with experimental data ( see e. g. ref . ) . on the other hand , for model 3 with @xmath3 electrons treated as open core , @xmath36 increases monotonically as a function of doping . ) under electron doping for model 1 ( solid black ) and model 3 ( dashed blue ) calculated in the mean - field approximation . red triangles : @xmath36 for model 3 with a correction for non - heisenberg interaction ( see text).,scaledwidth=45.0% ] takahashi @xcite argued that the maximum in the doping dependence of @xmath0 in euo is due to dynamical effects in the effective interaction between @xmath3 and conduction electrons . in his model @xmath0 saturates as a function of doping unless these dynamical effects are taken into account ; he also suggested that the maximum in @xmath0 found in first - principles calculations of refs . is due to the use of total energy calculations , which partially take into account the effects of spin disorder . however , our linear response calculations applied to the ferromagnetic state still produce a maximum in @xmath0 , but only if the @xmath3 electrons are included in the basis set . these results strongly suggest that the observed maximum in @xmath0 is due to the competition of two exchange mechanisms . the contribution of indirect exchange mediated by conduction electrons increases monotonically with doping , but the rate of its increase drops near the point where the fermi level reaches the bottom of the minority - spin conduction band . on the other hand , the @xmath1-@xmath2 hopping mechanism , which dominates in pure euo , is suppressed by the filling of the conduction band and by the loss of its spin polarization . the combination of these two effects leads to a broad maximum in @xmath0 . as we mentioned above , the strength of the @xmath1-@xmath2 hopping mechanism is overestimated in our calculations , so one may expect that the decline of @xmath0 may also be somewhat overestimated . so far we have assumed that the exchange interaction may be described by the heisenberg model . however , at low doping concentrations the indirect exchange in euo has a non - heisenberg character due to the half - metallicity of the conduction band above the threshold polarization level . @xcite in this half - metallic region the effective weiss field no longer depends on the magnetization , which may explain the pronounced shoulder observed in the @xmath4 curve ( see e. g. refs . for recent data on high - quality samples ) . the effective exchange parameter in the paramagnetic state is larger compared to the ferromagnetic state , and thus the above estimates of @xmath36 should be corrected at low doping levels . to this end , we have calculated the exchange fields as a function of the polarization of the @xmath3 shell in euo at 1% and 3% doping levels . these calculations were performed for models 2 and 3 by varying the polarization of the open - core @xmath3 shell , but this variation also represents the effects of temperature - dependent magnetization . the results for 1% doping are displayed in fig . [ var4f ] . contribution to the local moment on eu atoms ( @xmath37 , blue circles ) and magnetization ( @xmath38 , red squares ) as a function of the polarization @xmath39 of the @xmath3 shell . ( b ) effective exchange field on eu ( @xmath40 , blue circles ) as a function of @xmath39 ( left axis ) . in both panels the straight lines are linear fits to the data below and above the point of half - metallicity . all filled symbols are for model 3 , and empty symbols are for model 2 . red solid line in panel ( b ) : effective indirect exchange parameter @xmath41 ( right axis ) extracted from the fits to @xmath42 for model 3.,title="fig:",scaledwidth=35.0% ] contribution to the local moment on eu atoms ( @xmath37 , blue circles ) and magnetization ( @xmath38 , red squares ) as a function of the polarization @xmath39 of the @xmath3 shell . ( b ) effective exchange field on eu ( @xmath40 , blue circles ) as a function of @xmath39 ( left axis ) . in both panels the straight lines are linear fits to the data below and above the point of half - metallicity . all filled symbols are for model 3 , and empty symbols are for model 2 . red solid line in panel ( b ) : effective indirect exchange parameter @xmath41 ( right axis ) extracted from the fits to @xmath42 for model 3.,title="fig:",scaledwidth=35.0% ] as seen in fig . [ var4f]a , the conduction - band magnetization @xmath38 increases almost linearly as a function of @xmath39 up to approximately @xmath43 . this is the point @xmath44 where the minority - spin conduction - band bottom moves above the fermi level due to the exchange splitting induced by the @xmath3 shell . for @xmath45 the magnetization stays constant at @xmath46 , which is equal to the electron doping level . however , the local moment on the eu atom continues to grow at larger @xmath39 at a smaller rate . a similar change of slope is observed for the effective exchange field ( fig . [ var4f]b ) . we interpret these features as a combination of indirect exchange interaction and other mechanisms . indirect exchange contributes an approximately linear term to @xmath37 and @xmath42 at @xmath47 , but at @xmath48 its contribution stays constant . other mechanisms ( bloembergen - rowland ) contribute a linear term for the entire range of @xmath39 . thus , the difference in slopes below and above @xmath49 gives the indirect exchange contribution to @xmath42 , from which we can recover the effective indirect exchange @xmath41 . the latter is shown in fig . [ var4f]b by a solid ( red ) line . we see that @xmath41 is 2.55 times larger in the paramagnetic state ( @xmath50 ) compared to the ferromagnetic state ( @xmath51 ) . this enhancement decreases with the doping level as the half - metallic region shrinks and eventually disappears . the corrected @xmath36 for 1% and 3% doping is shown in fig . [ tcdop ] by ( red ) triangles . faster increase of @xmath0 with doping is in good agreement with experimental data . following mauger,@xcite we can also estimate the effect of the non - heisenberg character of the indirect exchange on the magnetization curve . assuming that the electron gas remains degenerate , we solve the mean - field equations with magnetization - dependent indirect exchange contribution @xmath41 added to the heisenberg part . instead of using the overestimated calculated value for the heisenberg part of @xmath52 , we normalize it to produce the experimental @xmath0 of 69 k. the resulting @xmath4 curve for 1% doping is shown in fig . ( here we used the brillouin function for quantum spin @xmath53 , which makes the @xmath0 enhancement larger compared to that shown in fig . [ tcdop ] . ) the shape of the @xmath4 curve is similar to earlier calculations@xcite and qualitatively similar to experiment ( see e. g. ref . ) . the cusp , which is a consequence of our fitting procedure , should be smeared if a more accurate fitting is used near the crossover point @xmath49 . for undoped euo.,scaledwidth=45.0% ] in this section we have considered the exchange interaction in electron - doped euo within the virtual crystal approximation . in reality the individual dopants introduce local perturbations in the electronic structure and magnetic interaction due to their chemistry and structural relaxations around them . these effects are considered in the following sections . to gain more insight in the role of different electron dopants in euo , in this section we consider isolated gd impurities and o vacancies . here and in the next section we use the pseudopotential plane - wave method with the projected augmented wave ( paw ) potentials @xcite as implemented in the vasp package.@xcite generalized gradient approximation ( gga ) @xcite is employed in combination with the hubbard @xmath54 correction for eu and gd @xmath3 orbitals . the atomic configurations were optimized in the ferromagnetic configuration using the conjugate - gradient algorithm . all the technical details of calculations were similar to ref . . single impurities were introduced in the @xmath55 cubic supercell containing 64 atoms , producing gdeu@xmath56o@xmath57 and eu@xmath57o@xmath56 structures . as previously discussed , @xcite gd impurities induce fairly large displacements of nearby o atoms towards themselves . thus , in each nearest - neighbor eu - o - gd chain the eu - o bond becomes 0.12 - 0.15 longer than the gd - o bond ( see fig . [ scheme]b for an illustration ) . energy gained in this relaxation is responsible for the strong positive pair interaction for second - nearest neighbors in this system . in the case of an o vacancy , the neighboring eu atoms shift 0.13 away from the vacancy site ( see fig . [ scheme]c ) . figure [ dos ] shows site - projected densities of states ( dos ) for the cations in the gdeu@xmath56o@xmath57 supercell ( panel ( a ) ) and in the eu@xmath57o@xmath56 supercell ( panel ( b ) ) . the sites are labeled in the order of their distance from the impurity site . ( for the supercell with an o vacancy , eu@xmath58 is equivalent to eu@xmath59 . ) o@xmath57 supercell , ferromagnetic state . ( b ) eu@xmath57o@xmath56 supercell , ferromagnetic state . the arrow indicates the filled localized impurity state . ( c ) eu@xmath57o@xmath56 supercell , random collinear spin configuration . the o vacancy has 3 eu neighbors each with spin up and down , which are denoted eu@xmath60 and eu@xmath61 . the cut - off peaks in all panels are the @xmath3 states.,title="fig:",scaledwidth=42.0% ] o@xmath57 supercell , ferromagnetic state . ( b ) eu@xmath57o@xmath56 supercell , ferromagnetic state . the arrow indicates the filled localized impurity state . ( c ) eu@xmath57o@xmath56 supercell , random collinear spin configuration . the o vacancy has 3 eu neighbors each with spin up and down , which are denoted eu@xmath60 and eu@xmath61 . the cut - off peaks in all panels are the @xmath3 states.,title="fig:",scaledwidth=42.0% ] o@xmath57 supercell , ferromagnetic state . ( b ) eu@xmath57o@xmath56 supercell , ferromagnetic state . the arrow indicates the filled localized impurity state . ( c ) eu@xmath57o@xmath56 supercell , random collinear spin configuration . the o vacancy has 3 eu neighbors each with spin up and down , which are denoted eu@xmath60 and eu@xmath61 . the cut - off peaks in all panels are the @xmath3 states.,title="fig:",scaledwidth=42.0% ] each gd atom donates one electron to the conduction band . as seen in fig . [ dos]a , the deviations from homogeneous occupation of the conduction band are relatively small , indicating that the shallow impurity state is fully incorporated in the conduction band at the considered gd concentration . still , the occupied conduction - band partial dos ( fig . [ dos]a ) is somewhat larger for the gd atom and its second - nearest eu neighbors ( eu@xmath62 ) compared to other eu atoms . the increase for gd is due to its larger nuclear charge , but for the eu@xmath62 atoms it is due to the displacement of the o atoms between eu@xmath62 and gd towards gd and away from eu@xmath62 , which reduces charge transfer to o and shifts the electronic states on eu@xmath62 towards lower energies . the downward shift of the majority - spin @xmath3 states on eu@xmath62 has the same origin . in the next section we will see that the associated changes in the local conduction - band magnetic moments significantly modify the exchange interaction near gd impurities . small differences in the conduction - band occupations between gd and eu atoms are in sharp disagreement with ref . arguing that gd @xmath5 states are occupied much more than the eu @xmath5 states . this difference might be due to the use of the hubbard @xmath54 correction for the eu @xmath5 orbitals in ref . , which results in the band structure that disagrees with gw calculations.@xcite a charge - neutral o vacancy contributes two extra electrons . [ dos]b shows that in the ferromagnetic state only one of these electrons is donated to the conduction band , while the second one occupies a majority - spin impurity state at @xmath63 ev , which does not extend beyond the eu atoms adjacent to the vacancy ( this state is indicated by the arrow in the figure ) . the minority - spin counterpart of this impurity state lies about 0.25 ev above the fermi level . for the large vacancy concentration in our supercell , the impurity states are considerably broadened into impurity bands ; as a result the minority - spin impurity band becomes slightly occupied . the energy dependence of dos for the impurity states is typical for a single tight - binding band in the simple cubic lattice , which corresponds to the choice of the cubic supercell with one vacancy . the spatial structure of the occupied impurity state is seen from its charge density shown in fig . the charge is concentrated on the vacancy site as well as on its six nearest eu neighbors , extending further along the [ 001 ] directions towards the o atoms . o@xmath56 supercell ( units of 10@xmath64 @xmath64).,scaledwidth=48.0% ] the exchange splitting of the impurity state on the o vacancy is approximately 1.4 ev , which is significantly larger than the exchange splitting of the empty conduction band in euo . as will be further discussed below , a large part of this exchange splitting is due to the exchange - correlation field generated by the impurity state itself . the binding energy of the impurity states , their exchange splitting , and energy cost of double occupancy are important factors in the theories of the metal - insulator transition in eu - rich euo.@xcite let us , therefore , examine these features more closely . first we discuss the relevant parameters and then turn to their implications for the finite - temperature magnetic and transport properties . sinjukow and nolting proposed a model @xcite in which the impurity states in eu - rich euo are described in the atomic limit of the correlated kondo lattice model.@xcite in their description each impurity state interacts with one eu @xmath3 spin . as a result , there are 4 quantum states for the `` impurity atom '' characterized by quantum numbers @xmath65 and @xmath66 , where @xmath65 is the number of electrons in the impurity orbital , and @xmath66 is the total spin of the impurity atom : @xmath67 , @xmath68 , @xmath69 , and @xmath70 , where @xmath71 is the spin of the half - filled eu @xmath3 shell . the corresponding energies are 0 , @xmath72 , @xmath73 , and @xmath74 , where @xmath75 is the binding energy of the impurity state in the absence of exchange coupling , @xmath66 is its exchange coupling with the @xmath3 spin , and @xmath54 the coulomb energy in the doubly occupied state . the four single - particle excitation energies correspond to the transitions @xmath76 , @xmath77 , @xmath78 , and @xmath79 , where the electron is extracted from the reservoir with the chemical potential @xmath80 . the corresponding transition energies are @xmath81 , @xmath82 , @xmath83 , and @xmath84 . in this model the four transition energies do not depend on temperature , but their spectral weights do . @xcite based on these assumptions , it was argued @xcite that the notion of a temperature - dependent exchange splitting of the impurity state @xcite is unphysical . contrary to the model assumption of ref . , an impurity state on an o vacancy interacts not with one , but with six eu @xmath3 spins . therefore , the atomic limit with four energy levels is irrelevant to this system , while the treatment based on the temperature - dependent exchange splitting is quite appropriate . indeed , consider a cluster of six eu spins interacting with an impurity spin and placed in an effective weiss field @xmath85 to describe the ferromagnetic state . this gives an effective hamiltonian @xmath86 where @xmath87 is the operator of the impurity spin . the empty and doubly occupied impurity states have spin zero and do not interact with the @xmath3 spins at all . for the singly occupied impurity states @xmath87 is the spin @xmath88 operator . the singly occupied eigenstates can then be classified by the total spin @xmath89 of the @xmath3 shells and its projection @xmath90 , plus the total spin of the system , as @xmath91 and @xmath92 . the corresponding energy levels are @xmath93 and @xmath94 . the states with 0 and 2 impurity electrons have energies @xmath95 and @xmath96 . the value @xmath89 lies within the range @xmath97 . thus , there are multiple energy levels with different values of @xmath90 . this situation is quite different from the atomic limit , @xcite in which the @xmath3 spin is always equal to @xmath53 . the statistical probabilities of levels with different @xmath90 are determined by the weiss field . at zero temperature only @xmath98 is occupied , and the impurity levels are fully spin - split . the ground state is @xmath99 and there are two single - particle transitions to empty and doubly occupied impurity orbital with energies @xmath100 and @xmath101 . in the paramagnetic state , the most probable states have small values of @xmath90 , for which the exchange coupling with the impurity spin is weak . for these statistically important configurations the impurity state can be either singly or doubly occupied depending on the sign of @xmath102 . thus , up to statistical fluctuations around the mean - field solution , the situation is adequately captured by the temperature - dependent exchange splitting . in the following we will therefore disregard the unimportant quantum character of the @xmath3 spins and treat them classically . in order to simulate the paramagnetic state , we repeated the calculation for the @xmath55 supercell containing one o vacancy , but now with a random assignment of spin directions ( up or down ) on all eu atoms . the @xmath3 moments add up to zero both in the entire cell and on the six eu neighbors of the o vacancy . the self - consistent dos shown in fig . [ dos]c indicates that the impurity state in this configuration remains spin - split by @xmath103 ev . since the @xmath3 spins are randomized , this splitting is due to the exchange - correlation field intrinsic to the impurity state . [ dos]c also shows that the impurity states have different weights on the eu atoms with different orientations of the @xmath3 spins . specifically , an electron occupying a spin - up impurity state is about twice more likely to be found next to an eu atom with spin up compared to an eu atom with spin down . this feature indicates that the wave functions of the impurity states adjust to the given configuration of the @xmath3 spins by partially localizing towards eu atoms with parallel spins . the degree of this partial localization is controlled by the relation between the exchange splitting of the impurity state and the hybridization of the eu @xmath5 orbitals participating in its formation ; it should not be sensitive to the vacancy concentration . note that since the wave functions are different for different spin directions , the configuration of the doubly occupied impurity state is not a simple spin singlet . we will now estimate the parameters controlling the energies and occupation of the impurity states , including the binding energy @xmath75 , the exchange couplings , and the charging energy @xmath54 . we set the energy zero at the bottom of the unpolarized conduction band in the paramagnetic state . all estimates are intended to be accurate to no better than 0.05 ev . [ [ exchange - energies . ] ] exchange energies . + + + + + + + + + + + + + + + + + + by analyzing the local moments on different sites , we estimate that the impurity orbital has an approximately 38% weight on the vacancy site itself , 34% weight on the six nearest eu sites , and the remaining weight extending to more distant sites . it is easy to see that the dominant part of @xmath104 must be due to the exchange field generated by the `` head '' of the impurity orbital on the vacancy site . indeed , the exchange - correlation kernel for the eu @xmath5 orbital is close to 1 ev , while the local moments on the nearest eu sites are less than @xmath105 . this means that the contribution of eu atoms to @xmath104 is less than 0.02 ev . the minority - spin impurity state is approximately half - filled . therefore , the exchange - correlation energy for this orbital is @xmath106 ev . from the local moment and potential parameter splitting in an lmto calculation we found that the exchange - correlation kernel @xmath107 for the @xmath108 orbital on the vacancy site is approximately 5 ev . this value is consistent with the magnitude of @xmath109 ( @xmath110 ) . in the paramagnetic state , the impurity levels are shifted downward by @xmath111 due to their partial localization . since the weights on the eu sites with parallel spins are about twice larger compared to those with opposite spins ( fig . [ dos]c ) , a simple arithmetic calculation shows that @xmath112 , where @xmath113 is the downward shift of the majority - spin state in the ferromagnetic configuration due to exchange coupling with the @xmath3 states . the full exchange splitting of the impurity state in the ferromagnetic configuration is @xmath114 ev . since @xmath115 , we find @xmath116 ev . [ [ charging - energy - u_imp . ] ] charging energy @xmath117 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in the ferromagnetic state ( fig . [ dos]b ) the fermi level lies close to zero energy ( i. e. half - way between the majority and minority - spin cbm ) . the filled and empty impurity bands are centered approximately at @xmath63 ev and 0.25 ev , respectively . the center of gravity of both bands is therefore at @xmath118 ev . in the paramagnetic state ( fig . [ dos]c ) the center of gravity of the impurity states is approximately at @xmath119 ev ( referenced from the cbm ) . the upward shift of @xmath120 ev is equal to @xmath121 ( the minority - spin impurity band is close to half - filling in fig . [ dos]c ) . thus , we find @xmath122 ev . this value is quite reasonable for an @xmath123 center - like state extending to nearest neighbors of the vacancy . [ [ binding - energy - epsilon_0 . ] ] binding energy @xmath75 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + on the model level , the electron removal energy in the ferromagnetic state is equal to @xmath124 . to estimate the removal energy from our dft calculation , we need to correct the kohn - sham eigenvalue at @xmath63 ev by subtracting @xmath125 ev . since @xmath126 in our ferromagnetic calculation , we find @xmath127 ev . we are now in the position to discuss the temperature - dependent occupation of the impurity states . let us first consider the limit of a very low vacancy concentration , for which the impurity bands do not broaden into bands , while the chemical potential lies close to the cbm . in the ferromagnetic state the cbm is exchange split , and @xmath128 ev according to our calculations or @xmath129 ev if we take the experimental value 0.6 ev for the conduction band splitting.@xcite the impurity orbital is obviously singly occupied by an electron with spin parallel to the magnetization . the excitation to an unoccupied state has energy @xmath130 ev , and the electron addition energy is @xmath131 ev . in the paramagnetic state we should set @xmath132 ( cbm is not polarized ) , @xmath133 , and add the localization shift @xmath134 to the impurity levels . the doubly occupied state now has energy @xmath135 ev with respect to the singly occupied state , which means that the doubly occupied state becomes dominant . as long as there are charge carriers in the conduction band , it is preferable for them to get trapped by the impurity levels . we could arrive at the same result by following slater s transition - state argument , which states that the addition energy can be estimated as a dft eigenvalue half - way between the singly and doubly occupied configurations . in the random spin configuration ( fig . [ dos]c ) , the minority - spin impurity state happens to be very close to half - filling . therefore , the addition energy can be estimated as the center of gravity of both impurity states minus the chemical potential . since we are interested in the dilute limit , the chemical potential should be placed at the cbm . thus , our results suggest that slightly eu - rich euo should be semiconducting with an activation energy of about 0.05 ev . this picture is consistent with the so - called `` helium atom model '' of the metal - insulator transition in eu - rich euo,@xcite in which the ferromagnetic order results in the change from double to single occupation of the impurity states and a subsequent promotion of free charge carriers into the conduction band . eu - rich euo samples undergoing the metal - insulator transition exhibit semiconducting conductivity in the paramagnetic state with an activation energy of about 0.3 ev.@xcite our estimate of 0.05 ev is therefore too small . the difference may be due to the finite size of the supercell , as well as to errors intrinsic to dft . on the other hand , the observed activation energy may actually correspond to a different bound state . a bound magnetic polaron @xcite is a popular candidate , in which the second localized electron aligns its spin parallel to the deeply bound impurity state and polarizes the spins of the eu atoms in its vicinity . we note , however , that double occupation of @xmath123 center - like impurity states in the paramagnetic state is consistent with the experimental observation that @xmath0 in bulk eu - rich euo samples is not increased compared to its value for pure euo , while even small concentrations of gd enhance it notably . if bound magnetic polaron is the dominant defect in eu - rich euo , we would expect a @xmath0 enhancement of a magnitude similar to gd doping at approximately double the concentration of o vacancies . as we argued above , in the dilute limit o vacancies in euo should capture two electrons in the paramagnetic state . the situation changes at larger vacancy concentrations on the order of a few percent , which can be reached in films . @xcite compared to the dilute limit , there are two changes : the impurity levels broaden into impurity bands , and the chemical potential is increased unless there are compensating defects . although the increase of the chemical potential tends to further stabilize the doubly occupied state , the lowest - energy state can now correspond to a fractional occupation of the impurity band . this is the situation seen in fig . [ dos]c , where it corresponds to the vacancy concentration of 3.125% . of course , there should be no metal - insulator transition in this case , and @xmath0 should be significantly increased due both to the filling of the conduction band and to the exchange coupling mediated by the magnetically polarized impurity states , as is usually observed in eu - rich euo films . in principle , the spin - polarized impurity orbitals could interact directly and order ferromagnetically above the @xmath0 of pure euo . however , the following considerations show that this interaction is too weak . in order to interact strongly , two o vacancies must share an eu neighbor , which means they must occupy nearest or next - nearest o sites . the exchange coupling for such pairs can be estimated as the difference in the exchange - correlation energy @xmath136 ( @xmath137 ev ) on the shared eu atom(s ) for parallel and antiparallel orientations of the impurity spins . at a 3% vacancy concentration , the vacancy orbitals carry local moments of about 0.5 @xmath138 ( fig . [ dos]c ) . taking into account that the weight of the impurity orbital on each neighboring eu atom is about 5.6% , the local moment on eu for parallel and antiparallel orientations should be about 0.056 @xmath138 and 0 , respectively , yielding an energy difference of about 1.5 mev . if the vacancy orbitals are fully spin - polarized ( local moment 1 @xmath138 ) , this difference would be 4 times larger , or about 6 mev . at 3% concentration each vacancy has a roughly 50% chance of having another vacancy as its first or second neighbor . thus , even in the mean - field approximation the maximum possible @xmath0 due to this exchange interaction is less than 10 k , which is well below the ordering temperature of pure euo . thus , we expect that the vacancy orbitals should behave as anderson local moments , and their dominant effect on @xmath0 should be through an enhancement of the exchange coupling between the eu @xmath3 spins that they overlap with . the local moment on the impurity orbital varies gradually in the whole range of magnetization from ferromagnetic to paramagnetic ( see fig . [ dos]b and [ dos]c ) , and the impurity band does not become half - metallic at any temperature . therefore , it is likely that the interaction introduced by these orbitals does not deviate strongly from the heisenberg model . the magnitude of this interaction will be evaluated in section [ msce : vac ] . to conclude this section , we note that filling of the impurity states on o vacancies suggests a possible explanation for the results of the arpes measurements on gd - doped euo , @xcite which showed the transfer of the spectral weight from the electron pockets near x points toward bound states of unknown origin as the system is heated beyond the curie temperature . it was suggested @xcite that gd - induced o vacancies could be responsible for a defect band contributing the bound states . our present results not only provide evidence in support of this hypothesis , but also show the possible mechanism of the temperature - dependent spectral weight transfer . the fermi level is shifted upwards in gd - doped euo , tending to stabilize the doubly occupied impurity state . therefore , the impurity bands may get fully filled in the paramagnetic state even at fairly large vacancy concentrations . in this section we describe a magnetostructural cluster expansion ( msce ) approach , which is a combination of the cluster - expansion ( ce ) technique @xcite for substitutional alloys with the spin - cluster expansion @xcite for magnetic systems , enabling the description of the interaction between the configurational and magnetic degrees of freedom in a magnetic alloy . this approach allows one to take into account the dependence of the magnetic interaction parameters on the local chemical environment , as well as the influence of magnetic order on the configurational thermodynamics . our goal here is to evaluate the effects of gd doping and o deficiency on the magnetic properties of euo beyond the virtual crystal approximation , and in the actual calculations we will assume that the dopants are randomly distributed over the lattice . the main idea of this method and its application to gd - doped euo were briefly reported earlier.@xcite here we provide a more general formulation of the method and additional details justifying the choice of parameters for gd - doped euo . a simple msce model for o - deficient euo is also presented . in general , we may consider a magnetic substitutional alloy , whose configuration is described by occupations numbers @xmath139 , where @xmath21 is a lattice site , @xmath140 the component index , and @xmath141 if and only if site @xmath21 is occupied by component @xmath140 . atoms of some components also have a classical spin variable @xmath142 associated with them , which we treat in the adiabatic approximation.@xcite we assume that the total enthalpy of the alloy is a function of all occupation numbers and spin variables : @xmath143 . this total energy can be , quite generally , represented as an expansion in a set of spin - structure clusters @xmath144 here @xmath145 denotes a subset of lattice sites ( of size @xmath146 ) with a mapping to the set of components , and @xmath147 is the corresponding projector : @xmath148 . the index @xmath149 designates a magnetic subset of @xmath150 ( with the inherited mapping ) . the effective magnetostructural cluster interaction @xmath151 is an energy term attached to a particular spin basis function @xcite @xmath152 ( labeled by @xmath153 ) for a magnetic cluster @xmath149 . the angular brackets denote averaging over the entire crystal , and the factor @xmath154 is the multiplicity factor needed to count the clusters correctly depending on their symmetry.@xcite the influence of the alloy configuration on the magnetic interaction is reflected in the dependence of the parameters @xmath151 on @xmath150 . although in general the expansion in eq . ( [ msce ] ) can include arbitrary multispin interactions , here we restrict ourselves to heisenberg exchange . the resulting model is therefore not expected to correctly describe the non - heisenberg indirect exchange at low doping levels , which was discussed in section [ vca ] . the exchange parameters , however , will depend on the chemical environment . in this special case , the magnetic subcluster @xmath149 always contains two sites , the index @xmath153 is superfluous , and the total energy can be written as @xmath155 where @xmath156 , and the dependence on the components residing at sites @xmath21 and @xmath22 , as well as on the environment , is schematically absorbed in the configuration dependence of the exchange parameter @xmath157 . as explained in section [ isolated ] , the shifts of the o atoms toward gd increase the local moments of eu atoms that have gd as their second - nearest neighbors . the same applies to gd atoms that have next - nearest gd neighbors . [ m_of_z ] shows the dependence of gd and eu local moments in two representative gd@xmath158eu@xmath159o@xmath160 structures on the number of first and second - nearest gd neighbors . it is evident that the local moments strongly correlate with the number of next - nearest gd neighbors , but not with the number of nearest gd neighbors . further , fig . [ m_of_z ] shows the dependence of the energy needed to reverse an individual spin of eu or gd on the number of nearest or next - nearest gd atoms . the same trend is observed as for the local moments in fig . [ e_of_z ] . clearly , larger local moments indicate increased conduction - electron density and enhanced indirect exchange interaction . eu@xmath159o@xmath160 structures as a function of the number of their nearest ( left panel ) and next - nearest ( right panel ) gd neighbors.,scaledwidth=45.0% ] but for spin - reversal energies instead of local moments.,scaledwidth=45.0% ] this effect can be explained by the fact that some of the o neighbors of such eu and gd atoms are shifted away from them . as a result , their electronic states are lowered , and the conduction states become more populated . the increased local moments lead to enhanced exchange coupling with the neighboring cations . we used this physical insight to select the magnetostructural cluster set , using the standard cross - validation score @xcite as an indicator of the fit quality . this empirical selection resulted in the choice : @xmath161 , \label{jij}\ ] ] where the dependence of the exchange parameters on the concentration @xmath6 of gd is included explicitly to represent the homogeneous effect of electron doping . in the following , @xmath162 stands for eu and @xmath163 for gd . the parameters @xmath164 are non - zero for first and second neighbors ; their values for @xmath165 and ratio @xmath166 for all @xmath6 are taken from the separate fitting for pure euo ; and the dependence on @xmath6 is represented by multiplying @xmath167 by a quadratic polynomial of @xmath6 with a unit free term . the parameters @xmath168 are non - zero only for nearest neighbors and are represented by a quadratic polynomial . the parameters @xmath169 are non - zero for first and second neighbors and are independent of @xmath6 . the second term in eq . ( [ jij ] ) includes configurational corrections for nearest - neighbor exchange : @xmath170 if @xmath21 , @xmath22 are nearest neighbors and 0 otherwise . the configuration - dependent terms are defined as : @xmath171 where @xmath172 , @xmath173 , @xmath174 , and @xmath175 are the fitting parameters , @xmath176 is the set of 6 second - nearest neighbors of site @xmath21 , @xmath177 is the number of gd atoms occupying the sites in @xmath176 , and @xmath178 only if the directions @xmath179 and @xmath180 are orthogonal ( otherwise @xmath181 ) . all the cluster types in eqs . ( [ jij])-([si ] ) are depicted in fig . [ graphh ] . note that the first term in eq . ( [ jij ] ) corresponds to @xmath182 in the general expansion of eq . ( [ msce ] ) , and its three nonequivalent clusters are shown separately in fig . [ graphh](b ) . the remaining terms with @xmath142 given by eq . ( [ si ] ) correspond to more complicated clusters @xmath150 , @xmath183 . the structure of these clusters is selected so as to incorporate the dominant effects of interactions mediated by the local relaxations of the o atoms . these interactions `` propagate '' between the next - nearest neighbor pairs , and a few clusters of this type are included in eq . ( [ si ] ) . in particular , the cluster corresponding to the last term in eq . ( [ si ] ) is a right triangle of gd atoms , whose two sides connect next - nearest neighbors . each of the other two terms in eq . ( [ si ] ) , in fact , corresponds to more than one nonequivalent type of cluster . the first term defines inequivalent clusters with different energies @xmath184 , while the second term includes two inequivalent clusters with the same energy @xmath174 . this choice reduces the number of fitting parameters and is empirically justified by the quality of the fit . note that the subtraction of 1 in the second term of eq . ( [ si ] ) is purely conventional and is designed to remove the explicit `` self - interaction . '' ) -([si ] ) . red crosses connected by arrows : sites @xmath21 , @xmath22 . green circles : gd atoms ; blue circles : eu atoms ; black circles : either gd or eu . the clusters enter the msce with coefficients : ( a ) @xmath184 ; ( b ) @xmath185 ; ( c ) @xmath175 ; ( d ) and ( e ) @xmath174.,scaledwidth=40.0% ] for the fitting of the msce ( [ jij])-([si ] ) we used 41 supercells containing up to 64 atoms with different configurations of gd dopants with concentrations up to 25% . full structural optimization was performed in the ferromagnetic spin configuration . spin - reversal energies for a subset of individual eu and gd atoms in these supercells were then calculated . the structures used and the calculated spin - reversal energies are listed in the supplementary material . the resulting msce has a cv score of 6.6 mev / spin and the root - mean - square misfit of 5.6 mev / spin , and the fitted parameters are : @xmath186 ; @xmath187 ; @xmath188 ; @xmath189 ; @xmath190 ; @xmath191 ; @xmath192 ; @xmath193 ; @xmath194 ( all in mev ) . the parameters @xmath172 , @xmath173 , @xmath174 , and @xmath175 correspond to 3-site and 4-site clusters of high multiplicity , and their impact on the quality of the fit is , thereby , quite significant . note that the @xmath6-dependent contributions to both @xmath195 and @xmath196 are almost identical , reflecting the same mechanism of doping - induced interaction . the value of @xmath196 is much smaller than @xmath195 at @xmath165 , which is reasonable , because the occupied gd @xmath3 orbitals are strongly bound and do not participate in the @xmath1-@xmath2 hopping mechanism . antiferromagnetic @xmath197 coupling is also reasonable in view of the fact that gds is and gdo is predicted to be antiferromagnetic.@xcite concentration - dependent coupling between two gd atoms was not included , because it did not improve the quality of the fit . the msce ( [ heis ] ) assumes the heisenberg form of magnetic interaction and is fitted to configurations that differ from ferromagnetic by one spin reversal . as we discussed in section [ vca ] , at small doping levels the indirect exchange has a non - heisenberg character , because the conduction band becomes half - metallic at sufficiently large magnetization . therefore , the msce is expected to underestimate @xmath0 at small doping levels . still , we found that the total energy differences between randomly selected magnetic configurations of the gdeu@xmath56o@xmath57 supercell are faithfully reproduced by the heisenberg msce , even though their average energy is underestimated by msce by about 13%.@xcite therefore , we may expect that deviations from heisenberg behavior have a relatively small effect on @xmath0 for concentrations of 3% or more . this is in agreement with vca results shown in fig . [ tcdop ] . for o - deficient euo , we would like to understand the large enhancement of @xmath0 observed in eu - rich euo films , which may contain much higher vacancy concentrations.@xcite such films may be strongly heterogeneous , @xcite although there is also evidence against phase separation.@xcite in order to compare strongly o - deficient with gd - doped euo , we constructed a simple msce for the vacancy concentration of 3.125% under an assumption that the configurations with vacancies appearing next to each other are statistically insignificant . thus , we considered one eu@xmath57o@xmath56 supercell with one o vacancy ( same as in section [ isolated ] ) and calculated the total energies of several magnetic configurations with reversed spins on eu atoms . as in the msce for gd - doped euo , we assume that the magnetic interaction has a heisenberg form , which is likely a fair approximation for the 3% electron doping level ( see the discussion in sections [ vca ] , [ isolated ] and a further comment below ) . there are four inequivalent eu atoms in the chosen supercell . the energy required to reverse the spins on the eu atom adjacent to the vacancy is 119.5 mev , while those for the more distant eu sites range from 63 to 69 mev . since these latter energies are very similar , we assumed that the exchange interaction is homogeneous beyond the nearest neighbors of the vacancy and used the average value 67.6 mev for it . this value includes the contribution of homogeneous electron doping . large enhancement of the spin - reversal energy for the nearest neighbor of the o vacancy is due to the impurity state ( section [ isolated ] ) mediating strong ferromagnetic coupling between the eu atoms with which it overlaps . this is the main feature that we wish to include in the msce . in order to resolve this interaction by pairs of atoms , we also calculated the energy changes needed to reverse two of the six eu neighbors of the vacancy . there are two inequivalent configurations of this kind , and the corresponding energies are 221.2 mev ( when two eu atoms are nearest neighbors ) and 234.4 mev ( when they are not ) . the calculated total energies can be well fitted using eq . ( [ heis ] ) with @xmath198 where @xmath199 is the interaction determined separately for pure euo , @xmath150 is the scaling factor accounting for the effect of homogeneous electron doping , @xmath200 if site @xmath21 has a vacancy at an adjacent o site , @xmath170 if @xmath21 , @xmath22 are nearest neighbors ( same as in eq . ( [ jij ] ) ) , and @xmath201 if @xmath21 , @xmath22 are second - nearest neighbors . the fitted parameters are @xmath150=1.45 , @xmath202 mev and @xmath203 mev . the root - mean - squared error for the predicted magnetic energies is 4.1 mev . this representation is designed for a particular vacancy concentration ( 1 per 32 o sites ) . to quantify the deviations from heisenberg behavior , we have compared the prediction of the msce ( [ smvac ] ) for the eu@xmath57o@xmath56 supercell with randomized directions ( its dos is shown in fig . [ dos]c ) with the calculated total energy . the predicted total energy difference with respect to the ferromagnetic state is underestimated by about 11% , which is similar to the gd - doped system at the same concentration . this underestimation can again be attributed to the half - metallicity of the conduction band in the ferromagnetic state , and its small magnitude supports the use of the heisenberg model at this vacancy concentration . the parameter @xmath150 in ( [ smvac ] ) represents the homogeneous enhancement of the exchange interaction ( relative to pure euo ) due to the introduction of free charge carriers , and the same role is played by the @xmath6-dependent @xmath195 parameter in the case of gd doping . as we ve seen in section [ isolated ] , both gd and o vacancy donate one electron to the conduction band . using the fitted expression for @xmath204 for gd - doped euo from section [ sec : msce ] , we find that at @xmath205 it is enhanced by a factor 1.37 compared to @xmath165 . this value is similar to the parameter @xmath150 for eu - rich euo . on physical grounds we may expect that in a reasonable range of o deficiency the parameter @xmath150 should change similar to @xmath204 in gd - doped euo , while the other parameters ( @xmath206 and @xmath207 ) reflecting the contribution from the localized state should be independent on the concentration . we have also considered the possibility of the formation of complex defects when both gd substitution and o vacancies are simultaneously present . to this end , we have calculated the total energies of a gdeu@xmath56o@xmath56 supercell , which contains one substitutional gd and one o vacancy , for their 4 different mutual arrangements in this @xmath55 supercell . for these arrangements the gd and vacancy sites are connected by vectors @xmath208 , @xmath209 , @xmath210 , and @xmath211 in units of the lattice parameter @xmath212 . the lowest total energy is achieved for the second of these arrangements , relative to which the four configurations are 138 , 0 , 46 , and 19 mev higher , respectively . as gd - doped euo films are usually grown at temperatures of 350@xmath213c or higher,@xcite these results suggest that o vacancies should occupy the sites next to gd atoms less often than in the random alloy , but there is no preference for the formation of complex defects . the thermodynamic properties of gd - doped and eu - rich euo were studied using classical monte carlo simulations for the msce models constructed in section [ sec : msce ] and the metropolis algorithm . the simulation box contained up to 32000 cation atoms , and the dopants were randomly distributed over their sublattice . the equilibration and averaging passes for each temperature were performed with @xmath214 monte carlo steps per site or more . in gd - doped euo from monte carlo simulations for the msce model ( black circles ) and from the mean - field estimate based on linear - response results in vca ( blue squares ) . the latter is shifted by a constant so that @xmath215 coincides with the mc result.,scaledwidth=45.0% ] the @xmath4 curves for a wide range of gd compositions in gd@xmath216eu@xmath217o do not reveal any unusual features.@xcite fig . [ mc ] shows the curie temperature @xmath218 obtained from the inflection point of the @xmath4 curve . for comparison , the plot of @xmath36 obtained in vca for model 1 ( fig . [ tcdop ] ) is added to the monte carlo value of @xmath0 at @xmath165 for the msce model . the curie temperature for pure euo is approximately 72 k in excellent agreement with experiment . the concentration dependence of @xmath0 is similar in vca and in the msce - based mc . thus , the effects of lattice contraction and local environment on the exchange interaction do not appear to modify the doping dependence strongly even at larger doping levels . the maximum @xmath0 of approximately 150 k is reached in a broad range of gd concentrations in good agreement with experiments ( see e. g. ref . ) . it is therefore unlikely that @xmath0 can be further increased in gd - doped euo . similar to vca , the msce - based mc results should underestimate the rate of increase of @xmath0 at low doping levels due to the neglect of the non - heisenberg character of the indirect exchange . in gd - doped euo there are two types of magnetic atoms , while in eu - rich euo the eu neighbors of an o vacancy are strongly coupled by the localized state . it is therefore interesting to examine the contributions to the total magnetization coming from these different types of atoms . [ mcd]a shows the atomic contributions to the total @xmath4 in 15% gd - doped euo for eu and gd atoms . [ mcd]b provides a similar plot for eu - rich euo with a 1/32 vacancy concentration , sorting out the atomic contribution to @xmath4 of eu atoms that have an o vacancy next to them . curves in ( a ) 15% gd - doped , and ( b ) 3.125% eu - rich euo . solid lines show the total magnetization . in panel ( b ) the atomic magnetization of the eu neighbors of o vacancies ( denoted eu@xmath59 ) is displayed separately from that of all other eu atoms ( denoted eu@xmath219).,title="fig:",scaledwidth=45.0% ] curves in ( a ) 15% gd - doped , and ( b ) 3.125% eu - rich euo . solid lines show the total magnetization . in panel ( b ) the atomic magnetization of the eu neighbors of o vacancies ( denoted eu@xmath59 ) is displayed separately from that of all other eu atoms ( denoted eu@xmath219).,title="fig:",scaledwidth=45.0% ] for gd - doped euo we see that the reduced magnetization of the gd atoms is notably suppressed compared to that of the eu atoms . this is because the gd - eu exchange coupling is significantly weaker compared to the eu - eu coupling , as mentioned in section [ sec : msce ] . the difference in the @xmath4 curves for eu and gd atoms could be experimentally verified using an element - specific probe such as nuclear magnetic resonance . as seen in fig . [ mcd]b , in eu - rich euo there is a large enhancement of the magnetization for eu sites adjacent to o vacancies , which is due to strong exchange coupling mediated by the localized states centered at the vacancies . the @xmath0 for the chosen vacancy concentration of 3.125% is close to 120 k , which is only achieved at about twice larger gd concentration in gd - doped euo . thus , the homogeneous enhancement of the exchange interaction due to the filling of the conduction band and inhomogeneous coupling due to localized states at the vacancy sites contribute nearly equally to the enhancement of @xmath0 in eu - rich euo . several authors observed that the deviations of the @xmath4 curves from the brillouin function with a long `` tail '' extending to higher temperatures is likely due , at least in part , due to the finite magnetic field used in the magnetization measurements.@xcite this feature is readily reproduced in mc simulations with a zeeman field added to msce , as shown in fig . [ zeeman ] . curves for gd@xmath220eu@xmath221o in an external magnetic field . the field is indicated next to the curves in mev units of maximum zeeman energy per site . for @xmath222 magnetic moment , 1 mev is equivalent to @xmath223 t.,scaledwidth=45.0% ] overall , our results show that , although on the microscopic level the magnetic interaction in gd - doped and eu - rich euo deviates strongly from the homogeneous electron doping model ( vca ) , this inhomogeneity does not strongly modify the macroscopic magnetic properties such as @xmath218 and @xmath4 . the observed deviations of the @xmath4 curves from the conventional behavior should therefore be due to other mechanisms . at low concentrations , as noted above , these deviations may be explained by the non - heisenberg character of the indirect exchange in the half - metallic regime.@xcite on the other hand , the observed anomalies at large gd concentrations may be induced by finite external fields used in the measurements,@xcite or may indicate phase separation as suggested in ref . . the magnetization curves in eu - rich euo films usually look like a superposition of two separate magnetization curves with different curie temperature , one of which coincides with that of pure euo . this feature suggests phase separation,@xcite but contrary evidence @xcite shows the need for further investigation . we have studied the electronic structure , magnetic interaction , and thermodynamic properties of gd - doped and eu - rich euo using first - principles calculations . homogeneous electron doping of euo , which was modeled in the virtual crystal approximation , leads to a broad maximum in @xmath0 . this maximum is due to the competition between the monotonically increasing indirect exchange contribution and the @xmath1-@xmath2 hopping mechanism , which generates ferromagnetism in pure euo but is suppressed at large doping concentrations . calculations with variable polarization of the @xmath3 shell reveal a pronounced non - heisenberg character of the indirect exchange interaction at small doping concentrations in agreement with the model of mauger,@xcite which leads to strongly enhanced @xmath0 and an anomaly in the @xmath4 curve . we have analyzed the electronic structure of isolated gd impurities and o vacancies in euo . both these defects donate one electron to the conduction band in the ferromagnetic state . an o vacancy introduces an exchange - split @xmath123 center - like level , which is half - filled in the ground state . estimates based on the key parameters extracted from our calculations suggest that in the limit of low vacancy concentrations the impurity state should become doubly occupied , assuming that the @xmath3 spins are fully disordered . this is consistent with the `` he atom model '' of the metal - insulator transition observed in the bulk samples of eu - rich euo @xcite and with no enhancement of @xmath0 in such samples . however , the estimated 0.05 ev activation energy in the paramagnetic state is too low compared to the transport measurements , and we can not rule out the formation of bound magnetic polarons . at higher vacancy concentrations typical for eu - rich euo films , the impurity states broaden into impurity bands , the upper one of which remains partially filled in the paramagnetic state . the impurity states behave as anderson local moments in this regime . this is consistent with the observed strong enhancement of @xmath0 in eu - rich euo films . localized states on o vacancies could also provide the bound states observed in arpes in gd - doped euo . @xcite to go beyond the homogeneous doping model , we employed a magnetostructural cluster expansion ( msce ) approach , which explicitly includes the effect of local chemical environment on the magnetic interaction . for the case of gd - doped euo we constructed an msce based on the input data from first - principles calculations that covers a wide range of gd concentration of up to 25% . for eu - rich euo , we set up a simple msce based on the data for one supercell with a single vacancy . the resulting hamiltonians were used in monte carlo simulations . the curie temperature in gd - doped euo was found to behave similar to the virtual crystal approximation , exhibiting a maximum as a function of gd concentration . the results suggest that the intrinsic limit @xcite of @xmath0 in gd - doped euo is approximately 150 k. oxygen vacancies induce a sharper rise in @xmath0 compared to a similar concentration of gd dopants due to the strong exchange coupling of the neighboring eu atoms mediated by the polarized impurity levels . microscopic inhomogeneity does not lead to any anomalies in the @xmath4 curves , at least in the random alloys that we have considered . we thank s. v. barabash , p. a. dowben , and a. g. petukhov for useful discussions . this work was supported by dtra ( grant no . hdtra1 - 07 - 1 - 0008 ) and nsf ( grant no . dmr-1005642 ) . computations were performed utilizing the holland computing center at the university of nebraska . k. d. b. acknowledges support from the research corporation through the cottrell scholar award .

an important challenge in design of modulation schemes for molecular communication is positivity of the transmission signal ( only a positive concentration of molecules can be released in the environment ) . this restriction makes handling of the intersymbol interference ( isi ) a challenge for molecular communication . previous works have proposed use of chemical reactions to remove molecules from the environment , and to effectively simulate negative signals . however , the differential equation describing a diffusion - reaction process is non - linear . this precludes the possibility of using fourier transform tools . in this paper , a solution for simulating negative signals based on the diffusion - reaction channel model is proposed . while the proposed solution does not exploit the full degrees of freedom available for signaling in a diffusion - reaction process , but its end - to - end system is a linear channel and amenable to fourier transform analysis . based on our solution , a modulation scheme and a precoder are introduced and shown to have a significant reduction in error probability compared to previous modulation schemes such as csk and mcsk . the effects of various imperfections ( such as quantization error ) on the communication system performance are studied . molecular communication , modulation , isi mitigation , precoder , diffusion - reaction .

1603.01819

diffusion - based molecular communication is one of the promising mechanisms for communication among nano - scale machines . in such systems , molecules are used as information carriers instead of the electromagnetic waves . information may be encoded in the number , type or release time of the molecules . released molecules gradually diffuse in the medium , resulting in channel memory and inter - symbol interference ( isi ) . one of the main theoretical challenges of molecular communication is handling of the channel memory . while isi is a common issue in classical communications , it is more challenging and prominent in the context of molecular communication because one can not readily combat isi with classical channel equalization techniques . this is due to the fact that molecular communication systems are _ positive systems _ , i.e. , transmitter can only release a positive amount of concentration of a specific molecule into the medium . as a result , consecutive transmissions at the transmitter side can only contribute positively to the molecules at the receiver side , and can not cancel out the previous transmissions . one approach to handle _ negative _ signals is to add a constant value ( dc term ) to both negative and positive transmission concentrations to make all the transmitted concentrations positive . for instance , to transmit amplitudes @xmath0 and @xmath1 , we can consider a dc term of @xmath1 , i.e. , making transmitted concentrations as @xmath2 and @xmath3 for negative and positive messages , respectively . however , this will increase the number of transmitted molecules ( transmission power ) and may not be appropriate . in @xcite , negative signals are handled by assuming that the input signal is the derivative of molecule concentration . the derivative of a positive signal can be both positive and negative . thus , when we want to send positive signals , it is suggested to increase the current level of diffusion and if we want to send negative signals , we reduce the current level of transmission . however , the fact that the input signal has to be the derivative of some non - negative waveform does impose restrictions on the input signal . how this constraint translates into the fourier domain is a challenge in employing the model of @xcite . to simulate negative signals , another known promising idea exploits a unique feature of molecular communication , namely chemical reactions . the general idea is that when molecules of different type bond together , they produce new molecule types and reduce the concentration of original molecule types . in @xcite , the release of enzymes throughout the environment to put down the remaining molecular concentration from previous transmissions is proposed . however , this would also weaken the direct link between the transmitter and the receiver and would imply higher transmission powers . chemical reactions are also employed in design of molecular logic gates like xor , with two types of molecules that nullify each other @xcite . two works that consider chemical reactions to simulate negative signals are @xcite and @xcite . authors in @xcite propose that one puts information on molecules of type @xmath4 , but also follow it by the release molecules of type @xmath5 ( called a poison signal ) to chemically cancel effect of remaining isi terms ; thus , molecules of type @xmath5 are playing the role of negative signals . in @xcite , the idea of using @xmath6 and @xmath7 ions is proposed wherein transmitting each of these ions reduces the concentration of the other one in the medium . therefore these ions can play roles of positive and negative signals . diffusion - reaction equations are used to describe the variation of the concentration of ions in the medium . the main challenge of this system is that these diffusion - reaction equations are _ nonlinear _ , making its analysis a challenging problem . a different suggestion is given in @xcite to use two molecule types of @xmath4 and @xmath5 as primary and secondary signals to mitigate isi . information is transmitted on the primary signal , via molecules of type @xmath4 . the number of molecules of type @xmath4 at each time slot are either due to the current transmission , or previous transmissions ( isi term ) . to convey the isi term to the receiver , the idea is that after each transmission of primary molecules of type @xmath4 , with a certain delay , a proportional amount of molecules @xmath5 is emitted to the medium . no reaction occurs between molecules of type @xmath4 and @xmath5 , and molecules of type @xmath5 experience a similar channel as molecules of type @xmath4 . observing that molecules of type @xmath5 are released after a delay , the idea is to treat the number of arriving molecules of type @xmath5 as the isi term of molecules of type @xmath4 that we wish to cancel . then , the receiver subtracts the number of received molecules of type @xmath4 and @xmath5 to suppress the isi . a numerical optimization is run to find the best choice of delay and the amount of released molecules of type @xmath5 to minimize the error probability . while this approach is shown to reduce the isi , its performance is obtained numerically . to sum this up , previous works consider four solutions : ( 1 ) adding a dc to make the signal positive , ( 2 ) coding information on the signal derivative , ( 3 ) using chemical reactions , and ( 4 ) using pre - equalization ( with primary and secondary signals ) . our goal is to find a solution to simulate the effect of negative signals such that the resulting system still _ remains linear_. the property of linearity will allow us to apply classical communication approaches , such as spectral transformations , for communication and detection paradigms in the general multiple transmitter and multiple receivers setting . in particular , this scheme will permit us to use a _ precoder _ filter to compensate the adverse effects of the channel . we propose to exploit the fact that any arbitrary real signal is the _ difference _ of two positive signals . the idea is to use two different types of molecules ( denoted by molecules @xmath4 and @xmath5 ) for positive and negative transmissions . in other words , when we would like to send a positive signal , we release molecules of type @xmath4 , and for sending a negative signal , we release molecules of type @xmath5 . that is , if the transmitter wants to transmit signal @xmath8 , if @xmath9 then @xmath8 moles of molecules of type @xmath4 will be released , and if @xmath10 , then @xmath11 moles of molecules of type @xmath5 will be inserted into the medium . [ fig.signal ] demonstrates this scheme for transmitting @xmath12 into the medium . in this transmission strategy , the difference of the concentration of molecules of types @xmath4 and @xmath5 is equal to the signal @xmath13 . we call this signaling the type based sign ( ts ) modulation . our solution has similarities and differences with each of the previous solutions . while our solution is not based on chemical reactions as in @xcite , crucially we see that the resulting system remains linear ; in fact , chemical reactions can be utilized for an improved performance . next , our idea is similar to the one given in @xcite in the sense that we look at the _ difference _ of concentrations of molecules of type @xmath4 and @xmath5 to identify the signal . but our approach differs from the ones given in @xcite as we use molecules @xmath4 and @xmath5 in a _ symmetric way _ for positive and negative amplitudes , whereas in @xcite molecules of type @xmath4 are the primary and carriers of information , while molecules of type @xmath5 follow molecules of type @xmath4 by mirroring them appropriately in order to reduce the channel isi . thus , molecules of type @xmath5 are treated as secondary or poison signals . on the other hand , our approach for canceling the isi term is via a _ precoder _ for an effective channel derived using the diffusion - reaction equations . furthermore , the conceptual reason for taking the difference of concentrations of molecules of type @xmath4 and @xmath5 is not because we view one as a poison for the other , but because the difference of concentrations follows a linear differential equation that is not affected by the reaction rate between molecules . furthermore , there is a unique feature in our approach . to explain this , let us consider an open unbounded medium with no drift and no reaction . then , if we release one molar of molecules at the origin at time zero , the concentration of molecules equals @xmath14}{(4 \pi d t)^{1.5 } } { \mathsf{exp}}(-\frac{\|r\| _ 2 ^2 } { 4dt}),\end{aligned}\ ] ] where @xmath15 is the concentration at time @xmath16 at location @xmath17 . for large values of @xmath16 , @xmath15 vanishes like a constant times @xmath18 , as @xmath19 this inverse polynomial decay is slow and leads to large values of isi . ; for instance , see @xcite for a bounded region with a reflexive boundary condition . however , one should note that the value of @xmath20 is proportional to the diameter of the region ( it depends on the poincar constant of the medium , which is proportional to its diameter @xcite ) . as a result , the decay can be still very slow and lead to large values of isi . ] on the other hand , we get much faster decays if we allow for chemical reactions . for a broad class of chemical - reaction equations , it is shown in @xcite ( using the entropy method ) that the concentration of molecules converges exponentially fast in time to its stationary distribution like @xmath21 . to sum this up , chemical reactions allow for fast exponential decays . this is helpful for combatting the isi , as the effect of previous transmissions dies out faster . but chemical reactions not only decrease the isi terms , but also the direct link . in fact , the approach of @xcite based on the release of enzymes has this drawback . however , the unique feature of our approach is that it allows us to overcome this difficulty because the differential equation that governs the difference of molecules of type @xmath4 and @xmath5 is not affected by any of chemical reaction rates . thus , in an intuitive sense , chemical reactions are utilized in a dimension perpendicular to the direction of the signal propagation . our contribution can be summarized as follows : ( i ) we propose a new type based sign ( ts ) signaling strategy to introduce negative signals for molecular communication ( ii ) the ts signaling is utilized for precoder design . the effect of various imperfections on the error probability of the precoder are studied ; supporting analytical results and numerical simulations are also given . the remainder of this paper is organized as follows . in section [ sec : system - model ] , fick s law of diffusion as well as reaction - diffusion equations are reviewed . section [ sec : tsmodul ] gives the principles of the ts modulation . the scheme is applied to a time slotted communication setting in section [ sec : tssig ] . in section [ sec : model ] , we derive a precoder , utilizing the ts modulation . in section [ section : imperfect - reaction ] , the effect of imperfect chemical reactions on the noise variance at the receiver is studied . section [ numeric ] provides extensive numerical results to verify the various claims given in the paper , and finally section [ sec : conc ] concludes the paper . diffusion of molecules can be studied from either a microscopic or macroscopic perspectives , depending on the number of released molecules . if the number of released molecules is small , the microscopic perspective is relevant and the movement of individual molecules should be studied through brownian motion and stochastic differential equations . if possibility of chemical reaction between molecules of different types is considered , one will have brownian motion with interacting particles . on the other hand , if the number of released molecules is large , the macroscopic perspective is relevant and the variation of the concentration of molecules should be studied through the deterministic fick s laws of diffusion . if chemical reaction between molecules also exists , one should look at reaction - diffusion differential equations ( still a deterministic equation ) . herein , we assume that the number of released molecules is large and utilize the macroscopic perspective . the macroscopic diffusion of a molecules over an medium is described as follows : assume that the medium has a time - invariant flow velocity @xmath22 at location @xmath17 . the fick s second law of diffusion states that @xmath23 where @xmath24 is the concentration of molecule at time @xmath16 and position @xmath17 , @xmath25 is the diffusion coefficients of molecules , and @xmath26 denotes the concentration of molecule production rate at point @xmath17 ( by the transmitter ) , _ i.e. , _ the number of molecules added to the environment between time @xmath27 $ ] in a small volume @xmath28 around @xmath17 is equal to @xmath29 . equation can be understood as follows : any change in @xmath24 is due to the following factors : * the term @xmath30 due to _ medium inhomogeneity _ : molecules flow from area of high concentration to areas of low concentration , and the rate of this flow is proportional to the difference of concentrations . thus , the gradient of @xmath24 gives the flow due to medium inhomogeneity . but the mere existence of a flow does not necessarily imply a change in concentration of molecules at @xmath17 . for instance , in a uniform flow , the number of molecules that enter @xmath17 equals the number of molecules that exit @xmath17 . concentration of molecules at @xmath17 changes if there is a mismatch between the incoming and outgoing flows at @xmath17 , _ i.e. , _ variations in the gradient of @xmath24 , which is @xmath30 . * the term @xmath31 is due to the velocity of the medium . the velocity of the medium can alter concentration of molecules at @xmath17 , but as velocity mixes molecules from a neighborhood of @xmath17 , it should be considered in conjunction with the concentration of molecules in the vicinity of @xmath17 . this is reflected in computing the divergence of the product @xmath32 . * the molecule production rate @xmath26 at @xmath17 directly contributes to a change in @xmath24 . the molecule production rate is due to the transmitter(s ) . let us denote the medium by a bounded region @xmath33 , and its boundary by @xmath34 . to ensure that has a unique solution for @xmath24 , initial , boundary conditions as well as some technical mathematical assumptions about the boundary of @xmath34 and @xmath26 will be imposed @xcite . throughout this paper , we assume _ initial rest and zero boundary condition_. the medium is in initial rest if it is zero and remains zero until a non - zero input is applied to the medium by the transmitter(s ) , _ i.e. , _ if the molecule production rate @xmath35 for all @xmath36 and becomes non - zero for some @xmath37 , we assume that @xmath38 for all @xmath17 . our ts signaling strategy can be utilized with any type of boundary condition as long as a zero boundary condition is assumed , _ e.g. _ for a dirichlet boundary condition , a perfectly absorbing boundary is assumed ; for a neumann boundary condition , a perfectly reflective boundary is assumed . has the desired smoothness properties , _ e.g. , _ for the dirichlet boundary condition , we assume that it has the outside strong sphere property ( for every point @xmath39 on the boundary , one can find a closed ball @xmath5 outside @xmath40 such that @xmath41 ) . ] finally , the molecule production rate @xmath26 and velocity @xmath42 are assumed to be hlder continuous @xcite . is said to be hlder continuous if there are non - negative real constants @xmath43 , @xmath44 , such that @xmath45 for all @xmath8 and @xmath46 in the domain of @xmath47 , and @xmath48 is any arbitrary norm ( all norms are equivalent in a finite dimensional space ) . ] under these conditions , is guaranteed to have a unique solution @xmath49 that is continuously differentiable in @xmath16 and twice continuously differentiable in @xmath17 @xcite . a zero boundary condition in conjunction with initial rest imply that the differential equation describes a linear time invariant ( lti ) system when we view @xmath26 as the input and @xmath24 as the output . * diffusion of two types of molecules : * below , we provide the diffusion equations for two molecule types of @xmath4 and @xmath5 , in the absence or presence of chemical reactions . _ the case of no chemical reaction : _ consider diffusion of molecules of type @xmath4 and @xmath5 over an environment with flow velocity @xmath22 . the fick s second law of diffusion states that @xmath50 where @xmath51 and @xmath52 are the concentration of molecules @xmath4 and @xmath5 , @xmath53 and @xmath54 are the diffusion coefficients , and @xmath55 and @xmath56 are the molecule production rates . observe that equations and are decoupled . _ diffusion with chemical reaction : _ assume that molecules of @xmath4 and @xmath5 make chemical reactions due to the following equation @xmath57 in which @xmath58 is the rate of reaction . note that in , the coefficients of @xmath4 and @xmath5 are assumed to be equal ; hence equal concentrations of molecules @xmath4 and @xmath5 cancel out each other . assuming that the output products of do not include either of molecules of type @xmath4 or @xmath5 , the reaction - diffusion law can be expressed as @xmath59 the new term @xmath60 indicates that when both @xmath61 and @xmath62 are positive , molecules of type @xmath4 and @xmath5 disappear from the environment at a rate @xmath60 . indeed , the larger the values of @xmath61 and @xmath62 , the larger @xmath60 , and the faster they die out in the medium ( this is like a negative feedback that pulls down @xmath61 and @xmath62 once they become positive ) . despite its clear intuitive meaning , unfortunately the product term @xmath60 is a _ non - linear _ term . in appendix [ somecomment ] , we comment on the reaction diffusion equations given in and and the effect of the non - linear term on it . due to the non - linear term , no explicit analytical solution is known for and in general . showing that the above differential equations with proper boundary conditions have a _ unique non - negative _ solution is non - trivial from a mathematical point of view ( _ e.g. _ see the first two chapters of @xcite ) . we assume that the transmitter s action is the molecule production rate @xmath55 and @xmath56 of molecules of type @xmath4 and @xmath5 at origin . the functions @xmath55 and @xmath56 are assumed to be hlder continuous . as mentioned in the introduction , in the ts modulation , we encode information in the difference of concentrations of molecules @xmath4 and @xmath5 . in other words , we only track the difference @xmath63 as a function of time @xmath16 and location @xmath17 . in particular , at the receiver side , concentrations of both molecules @xmath4 and @xmath5 are measured and their difference is considered as the received signal . that is if @xmath64 moles molecules of @xmath4 and @xmath65 moles molecules of @xmath5 are measured at the receiver , receiver will interpret the value of @xmath66 as the received signal . similarly , if the transmitter injects @xmath67 moles molecules of @xmath4 and @xmath68 moles of @xmath5 in the environment , we interpret it as injecting @xmath69 . as a result , we need to find the differential equation that describes @xmath63 in terms of molecule injection difference @xmath70 . _ diffusion with no chemical reaction : _ let us assume that @xmath71 . by subtracting ( [ eq2 ] ) from ( [ eq1 ] ) we have @xmath72 thus , if we define @xmath73 the @xmath74 follows the same _ linear _ differential equation , similar to ( [ eq2 ] ) and ( [ eq1 ] ) , when we interpret @xmath75 as the molecule production rate . assumption of initial rest and zero boundary condition for @xmath61 and @xmath62 implies the same condition for their difference . _ diffusion with chemical reaction : _ assume that molecules of @xmath4 and @xmath5 make chemical reactions according to , and follow the reaction - diffusion equations of and with @xmath71 . observe that by subtracting equations and , the non - linear terms @xmath76 cancel out and we get back equation ( the equation for the no - reaction case ) . ] where outputs " do not involve any molecules of type @xmath4 or @xmath5 . ] _ benefits and drawbacks of ts : _ the benefit of the ts scheme is having allowed for negative input signal while keeping the system linear . this allows for combating isi , as discussed below . drawbacks include the use of two molecule types and the loss of degrees of freedom due to the fact that information is encoded only in the difference of molecule densities . nonetheless , we will see that the ts scheme already outperforms previously proposed schemes using two molecule types . _ observation noise at the receiver _ : while the macroscopic diffusion process is deterministic , ligand receivers have signal dependent noise , proportional to the number of measured concentration . as the concentration of molecules increases , the observation noise ( known as the _ particle counting noise _ @xcite ) increases . more specifically , if the receiver is located at @xmath77 , and we denote the concentration of molecule type @xmath4 around the receiver by @xmath78 , the receiver s measurement of molecules of type @xmath4 at time @xmath79 can be written as @xmath80 where @xmath81 is an additive sampling noise , distributed according to @xmath82 where @xmath83 is the receiver volume @xcite . we adopt the assumption that the measurement noise at different time instances @xmath79 and @xmath84 are independent conditioned on knowing @xmath85 and @xmath86 . is very small . ] the above discussion means that it is possible to have small _ difference _ in concentrations of molecules @xmath4 and @xmath5 at the receiver , however each one may have large concentrations ; this would degrade the detector performance when we measure the concentration of the individual molecules separately ( we incur two separate particle counting noises for measuring molecules of type @xmath4 and @xmath5 ) . to put down this noise , one can utilize one of the following two ideas : * use molecules of type @xmath4 and @xmath5 that react with each other according to . the reaction puts down the concentration of both molecules , while keeping the difference of the two concentrations intact . * we may use a molecules of type @xmath43 at the receptor area which together with molecules of @xmath4 and @xmath5 make a reaction and disappear . that is , the receiver releases molecules of type @xmath43 in its vicinity just before making the measurements . molecules of type of @xmath43 are such that the following chemical reaction holds : @xmath87 but that @xmath4 or @xmath5 alone with @xmath43 do not form any reaction . note that the chemical reaction is occurring only locally around the receiver . further assume that the output products of do not include either of molecules of type @xmath4 or @xmath5 . this would imply that release of @xmath43 causes the concentration of both @xmath4 and @xmath5 to drop by the same amount ; but their concentration difference will be preserved , as one can observe from the following @xmath88 + [ example1n ] an example is the following reaction ( for instance see ( * ? ? ? * ch.17 ) ) . @xmath89 in which molecules acetone ( @xmath90 ) , bromine ( @xmath91 ) and hydrogen ion ( @xmath92 ) play role of molecules a , b and c in ( [ reaction ] ) , respectively . the diffusion coefficient of acetone and bromine are equal to @xmath93 and @xmath94 , respectively ( @xcite ) . observe that the diffusion coefficients are not exactly the same , but simulation results provided later show that diffusion coefficient mismatch of up to 10 percent does not significantly degrade the error performance . in this section , we explicitly work out equations of ts signaling for a time slotted communication in which duration of each slot is @xmath95 . transmission and reception occur at the beginning and end of each time slot respectively . this model is employed in the next section for precoder design . see appendix [ somecomment ] for a discussion on the tradeoff between @xmath95 , diffusion coefficients @xmath53 , @xmath54 and reaction rate @xmath58 for the reaction diffusion equations given in and . _ encoder : _ the encoder has a message @xmath96 consisting of a sequence of information bits . it encodes this information bits into channel inputs , which is a sequence of transmission levels @xmath97 for the @xmath98-th time slot . let us fix a given transmission waveform @xmath99 , which we assume is hlder continuous and @xmath100 for @xmath101 . we also expect @xmath99 to be zero only when @xmath17 is not in the vicinity of origin @xmath102 , which is where the transmitter is located . then , if @xmath103 , we produce a molecule production rate of @xmath104 for molecules of type @xmath4 . on the other hand , if @xmath105 , we produce a molecule production rate of @xmath106 for molecules of type @xmath5 . then , the transmission densities ( of equations , , , ) are equal to @xmath107 @xmath108 the transmitted _ differential concentration _ ( i.e. , the difference of concentration of molecules of type @xmath4 and @xmath5 ) equals @xmath109 _ channel : _ the released molecules propagate through the medium , possibly react with each other , and will be received at the receiver space , complying with the diffusion laws . that is molecular concentration varies according to both distance between transmitter and receiver and the time receiver is measuring the concentration . regardless of whether molecules @xmath4 and @xmath5 react with each other as they propagate , or whether they react with molecules @xmath43 at the receiver side , the system that takes in the differential concentration @xmath110 and outputs the differential concentration at the receiver @xmath111is linear . here @xmath77 is the location of the receiver . if the characteristics of the medium does not change , this system is also time - invariant , and hence lti . the response of this system to input impulse function @xmath112 in a 3-dimensional medium without any reaction and drift velocity is given by the green function @xmath113}{(4 \pi d t)^{1.5 } } { \mathsf{exp}}(-\frac{\|r\| _ 2 ^2 } { 4dt}),\end{aligned}\ ] ] where @xmath114 is the diffusion coefficient of the transmitted molecules and @xmath115 denotes the @xmath116 norm . the differential molecule concentration around the receiver is equal to @xmath117 where @xmath118 is the convolution operator . then , @xmath119 where @xmath120 a typical sample of @xmath121 is depicted in fig . [ fig.channel ] for a smooth pulse - like @xmath99 waveform . the actual shape of @xmath121 depends on our choice of @xmath99 , but generally speaking the intensity of @xmath121 initially increases and then decreases . it is suggested in @xcite to take the sampling time interval @xmath95 , at the time instance where the peak of @xmath121 occurs , i.e. , @xmath122 for example , the value of @xmath95 in fig . [ fig.channel ] is @xmath123 . with this choice and letting @xmath124 we have @xmath125 the value of @xmath126 is taken large enough that we may practically assume that @xmath127 . interestingly , a new reason for making this choice of @xmath95 arises in section [ sec : stability ] . _ receiver : _ the receiver measures the molecule concentration at time instances @xmath128 at location @xmath77 , denoted by @xmath129 which can be expressed as @xmath130 where @xmath131 is the measurement noise . the measurement noise can be calculated as follows : assume that at time @xmath132 , the concentration of molecules of type @xmath4 and @xmath5 are denoted by @xmath133 and @xmath134 , we know that their difference is @xmath135 ; the actual values of @xmath133 and @xmath134 depend on whether the involving molecules react and cancel out each other . the measurement counting noise of molecules of type @xmath4 and @xmath5 are independent conditioned on @xmath133 and @xmath134 , and distributed according to @xmath136 where @xmath83 is the receiver volume . therefore , the total counting noise @xmath131 conditioned on @xmath133 and @xmath134 is distributed as @xmath137 furthermore , the counting noise @xmath131 in different time - slots ( for different values of @xmath138 ) are assumed to be independent of each other , when conditioned on the densities of @xmath4 and @xmath5 in the time slots . equation is not in an explicit form . thus , we first consider the case of fast reactions ( perfect reactions ) , for which an explicit formula can be found . imperfect reaction is considered in section [ section : imperfect - reaction ] . assume a fast reaction , that is , the reaction - diffusion equations given in and or the one given in with large reaction rate @xmath58 . fast reactions have been the subject of many studies in the literature , _ see @xcite . a typical result , shown in different papers for different settings , is that the solutions of the fast reaction differential equations converge as @xmath139 . furthermore , they converge to the solution of the _ instantaneous reaction_. consider the reaction - diffusion equations given in and . as discussed in @xcite , as we increase @xmath58 , we observe a spatial segregation of areas with at which molecules of type @xmath4 and @xmath5 are present ; and the reaction mainly happens in the vicinity of the interface of regions that separate molecules of type @xmath4 and @xmath5 . similarly , with the use of molecule @xmath43 as in , a similar phenomenon occurs in the vicinity of the receiver . thus , if @xmath140 , there are more molecules of type @xmath4 than @xmath5 , molecules of type @xmath5 get canceled out and , @xmath141 and @xmath142 . if @xmath143 after the reaction , @xmath144 and @xmath145 . hence , in both cases @xmath146 thus , we can write @xmath147 where the noise @xmath131 is distributed according to @xmath148 thus , we find that the probability density function ( pdf ) of @xmath149 is given by @xmath150 in other words , @xmath149 is a doubly stochastic random variable ( a normal distribution whose mean and variance are themselves random variables ) . ) for transmission of a rectangular pulse . parameters are @xmath151 , and @xmath152 . , scaledwidth=70.0% ] having allowed for negative input signal while keeping the system linear , our goal is to design a proper precoder to eliminate isi for the channel described in equation and . equation can be written as @xmath153 , where @xmath154 denotes the sequence @xmath155 and @xmath156 denotes the sequence @xmath157 . taking the @xmath158-transform from both sides , we obtain @xmath159 , where @xmath160 . [ fig.block ] ( a ) . to design a precoder for this system , we add a precoder block at the transmission side , as depicted in fig . [ fig.block ] ( b ) . in this scenario the input sequence @xmath161 is first fed into a causal realization of the inverse channel ( precoder ) , i.e. , @xmath162 , to produce the transmission symbols of @xmath163 . because @xmath162 is the inverse of the system @xmath164 , @xmath165 , _ i.e. , _ @xmath166 this is done to cancel the isi at the receiver side . using equation , we can rewrite as follows : @xmath167 where @xmath168 . in other words , @xmath169 with the isi completely canceled on the receiver side . of course , isi cancellation is achieved assuming perfect reaction , _ @xmath170 in . however , as we argue in section [ section : imperfect - reaction ] , an imperfect reaction comes close to a perfect reaction ( in a sense that will be clarified later ) exponentially fast in @xmath171 where @xmath172 is the duration of the reaction . assume a simple bpsk modulation with i.i.d . and uniform @xmath173 . then , the transmitted symbols @xmath163 s are produced by passing @xmath174 s through the precoder . we need to compute @xmath175 for the average consumed transmission power . furthermore , it is important to investigate the stability and causality of the precoder filter . let us begin with showing the stability of the precoder . we need to prove that all of the poles and zeros of @xmath162 are inside the unit circle . observe that @xmath176 where @xmath177 . we have @xmath178 by equation . therefore it is enough to show that all of the roots of @xmath179 lie in the unit disk . to show this , we use the following theorem : ( enestrom - kakeya@xcite ) [ th:1 ] if all coefficients of the polynomial @xmath180 are positive reals , then for any ( complex ) root @xmath181 of this polynomial , we have @xmath182 considering theorem [ th:1 ] , since @xmath183 , all roots of @xmath179 have absolute value of smaller than @xmath184 . therefore all poles of @xmath162 lie inside unit disk , _ i.e. _ , the filter is bounded input / bounded output stable . thus , stability provides a new reason for choosing of the sampling time @xmath95 as in equation . now , let us turn to the average power consumption of the precoder . while finding an analytical expression for @xmath175 seems difficult , one can find an upper bound for @xmath175 . observe that @xmath185 . thus , we need an upper bound on the energy of @xmath163 . the i.i.d . sequence of @xmath174 is a stationary discrete random process , which is passed through the filter @xmath162 . therefore , the autocorrelation function and @xmath186 is finite and can be computed . * diffusion coefficient mismatch : * this occurs when @xmath187 . as mentioned in example [ example1n ] and shown in section [ numeric ] via numerical simulations , coefficient mismatch of up to 10 percent does not significantly degrade the error performance . while no explicit solution is known in general , for the case of fast reaction @xmath139 , the differential equations and reduce to the ones for instantaneous reaction @xcite , with @xmath188 @xmath189 with @xmath24 satisfying @xmath190 where @xmath191 for @xmath192 and @xmath193 for @xmath194 . when @xmath195 , this is a special case of the generalized porous medium equation ( * ? ? * eq . ( 2.10 ) ) ( see also @xcite and the references cited for existence , uniqueness , stability of this equation ) . approximate solutions for this problem has been found @xcite for certain boundary conditions . interestingly , like our simulation result given in section [ numeric ] , it can be observed from ( * ? ? ? * fig 1 . ) that a mismatch of up to 10 percent does not significantly vary the movement of reaction interface . * quantization : * even if the input of the precoder is binary , the output of the precoder will have multiple levels ; this may make the transmission scheme complicated . as a result , we can consider quantization of input @xmath163 before its transmission . this can reduce the complexity of the transmitter . generally , we can define a quantizer consisting of a set of intervals @xmath196 and a set of levels @xmath197 , so that the quantizer @xmath198 ( often called the quantization rule ) can be expressed as @xmath199 where the indicator function is given by @xmath200 the transmitter sends @xmath201_{\mathcal{q}}=\mathcal{q}(x_k)$ ] instead of @xmath163 . thus , @xmath202_{\mathcal{q}}$ ] is the quantization error . then , the concentration of molecules around the receiver equals @xmath203_{\mathcal{q}}=y - p\star e.\ ] ] the second term @xmath204 is the quantization error at the receiver . note that @xmath205 where @xmath206 is the maximum quantization error . thus , the output error term @xmath204 vanishes as we increase the quantization levels and reduce @xmath206 . however , given a fixed number of quantization levels , we can ask for the best possible quantization intervals and levels . while the simplest quantizer is the _ uniform _ quantizer , non - uniform quantizers have better performance in general . the literature on quantization generally measures the quality of a quantizer by the distance between the input and output of the quantizer . an example of a distance function is the mean squared error ( mse ) : @xmath207,\end{aligned}\ ] ] where @xmath208 is the average distortion , @xmath209 $ ] denotes the expectation operator over the density function of the random variable @xmath154 given to the quantizer . the optimum quantizer is then found to minimize the average distortion ( [ distortion ] ) over all valid quantization rules , _ i.e. _ , @xmath210 for the mean squared distance defined in ( [ distortion ] ) , the optimum quantizer of ( [ opt_quan ] ) is called the lloyd quantizer @xcite , and is non - uniform in generals . for the scenario considered in this paper , minimizing ( [ distortion ] ) does not generally lead to minimize symbol error rate ( ser ) for the transceiver link . that is , although lloyd quantizer has the maximum reduction in distortion between the input and output of the quantization block , it does not have the best performance considering ser metric . this effect is investigated in more details in section [ numeric ] via numerical simulations . as mentioned earlier , chemical reaction allows for decreasing the noise variance . below , we consider slow reaction with large or small reaction times separately . thus far , we considered the case of fast reactions . but one may not always need fast reactions . weak reaction rates @xmath211 can be compensated by increasing the reaction time window @xmath95 for it to take effect . here , we provide a non - rigorous argument to illustrate this point . let us consider the use of molecule @xmath43 as in ( local reaction at the receiver ) and one - way imperfect reaction with some finite @xmath212 . for simplicity consider that the environment has no drift . while doing an exact calculation can be difficult , we proceed with an approximate calculation to gain some insight . molecules are released at the beginning of each time slot of duration @xmath95 , for a period of @xmath213 seconds . assume that the receiver samples the environment at the end of time - slot ; @xmath214 seconds prior to that , it releases molecules of type @xmath43 to clean up and prepare the environment for sampling . assume that the inter - transmission interval @xmath95 is large enough for the concentration of molecules of type @xmath4 and @xmath5 to stabilize and reach the steady state values @xmath215 and @xmath216 around the receiver , at the time of the release of molecule @xmath43 . due to this stabilization , the terms @xmath217 and @xmath218 are zero . at this time , the receiver releases molecules of type @xmath43 to clear up its surroundings . if the concentration of @xmath43 is assumed to be kept fixed at some level @xmath219 in its surroundings , equations and in the vicinity of the receiver simplify to @xmath220 where @xmath221 . assuming the initial condition @xmath222 and @xmath223 for any @xmath17 in the vicinity of the receiver , we are interested in computing the concentration of @xmath4 and @xmath5 after passage of @xmath172 seconds . subtracting from , we see that @xmath224 is invariant in @xmath16 , and thus equal to @xmath225 . without loss of generality assume that @xmath226 ( the case @xmath227 is similar ) . then , equation reduces to @xmath228 solving this equation , we get @xmath229 this shows that the concentration of @xmath5 converges to zero exponentially fast in @xmath16 . we verify this exponential behavior in section [ numeric ] via numerical simulations . as a result , as long as @xmath230 we can say that the concentration of molecules of type @xmath5 is small in comparison with the concentration of molecules of type @xmath4 at the time of measurement by the receiver . from here , one can conclude the possibility of compensating for small @xmath58 by increasing @xmath172 ; as long as the product @xmath171 is large , the error probability of the system will be close to that of perfect reaction @xmath170 . we now consider the case that both reaction rate @xmath58 and the time - slot duration @xmath95 are small . let us begin with the extreme case of no reaction amongst molecules . in this case the diffusion system is linear . we can write @xmath231 @xmath232 as a result , @xmath233 hence , from @xmath234 , and @xmath235 for @xmath236 , we obtain @xmath237 comparing this equation with , we see that the absolute value is moved inside the summation on the individual terms . now , consider the case of partial reaction . either of the reactions of the types and , or have a net effect of reducing the concentration of molecules of type @xmath4 and @xmath5 . however , if this reaction is imperfect , the densities of both of molecules of types @xmath4 and @xmath5 can be still positive at the receiver . we need the sum value @xmath238 to be able to compute the measurement noise at the receiver . because diffusion - reaction equations are deterministic , the sum @xmath238 is a ( non - linear ) deterministic function of past transmissions ( or equivalently the past @xmath239 transmissions because the channel memory is @xmath126 ) . we can denote this sum by @xmath240 . then , we will have @xmath241 with this notation , for full reaction " and no reaction " , we have @xmath242 @xmath243 to find @xmath240 one has to solve the non - linear differential reaction - diffusion equations to find the individual concentrations @xmath133 and @xmath134 . for instance , consider equations and . the difference @xmath244 follows a linear differential equation and can be solved for . using the fact that @xmath245 , we get the following differential equation for @xmath61 , @xmath246 there is a rich literature on reaction - diffusion equations . while no explicit analytical solution is known for them , the literature provides thorough studies of specific reaction - diffusion equations ; these special cases provide insights into the complexities of such differential equations . , and the difference of the concentrations is equal to one , @xmath247 for all @xmath248 . then , reduces to @xmath249 this is the fisher s equation and is known to have a _ traveling wave solution_. this highlights how different can @xmath61 and @xmath74 behave ; here the difference of concentrations , @xmath74 , is a constant function while @xmath61 is a wave function . ] despite lack of explicit analytical solution for the non - linear diffusion - reaction equation , it is possible to prove bounds for @xmath250 . it is intuitive that the noise variance should be between that of no - reaction and full reaction , _ i.e. , _ @xmath251 the inequality @xmath252 easily follows from the fact that the sum @xmath238 is always greater than or equal to @xmath253 . to prove the inequality @xmath254 we prove the following result : [ thm2d]in an environment with no flow velocity @xmath255 , the function @xmath250 is subadditive , _ @xmath256 holds for any arbitrary @xmath257 and @xmath258 . the proof is given in appendix a. one can deduce from this theorem as follows : take an arbitrary tuple @xmath257 . let @xmath259 $ ] and @xmath260 $ ] for @xmath261 , where @xmath262 $ ] is the indicator function . then , @xmath263 where follows from the fact that in @xmath264 all @xmath265 s are positive and hence only molecules of type @xmath4 are released ( hence no reaction may occur ) . similarly , in @xmath266 only molecules of type @xmath5 are released , hence there is no possibility of any reactions . finally , follows from @xmath267 for @xmath261 . for numerical results we consider equiprobable binary sequences and three modulations and signaling of csk , mcsk , and the proposed ts signaling with precoder . in csk for macroscale molecular communication ( amplitude modulation ) , only one molecule type is used and information bits are encoded in the concentration of the released molecules into the medium . in order to transmit the bit @xmath184 , the transmitter releases a fixed ( say @xmath268 ) molar of molecules of a give type . to transmit the bit @xmath2 , no molecules are released . in the mcsk scheme , similar to csk , same amount of molecules are released into the medium for each information bit , with the difference that two types of molecules , _ i.e. _ , @xmath4 and @xmath5 are utilized for odd and even time slots , respectively . one can consider the csk and mcsk schemes for two extreme cases : one for _ no - memory _ and the other for _ infinite - memory _ at the receiver @xcite . for the no - memory scheme , receiver has no memory unit and hence can not remember the previous transmissions , _ i.e. _ , no knowledge about the current isi . in the infinite - memory scheme , we suppose that at the end of each time slot , a _ genie _ would inform the receiver about the exact value of the current isi due to the previous transmissions . while exact analytical expressions are not previously reported for a genie - aided csk and mcsk , we used the map rule and derived the optimum detector for each transmission , and used the explicit expressions in our plots . we also consider transmission in a one - dimensional unbounded environment with zero boundary conditions at plus and minus infinity . as we saw earlier , transmissions waveforms should be hlder continuous . for consistency with the existing literature on molecular communication , we assume instantenous release of molecules ( delta function ) ; as commonly mentioned in engineering textbooks , one can find a sequence of hlder continuous functions that approximate the delta function . for fair comparisons between the proposed scheme and the other ones , we keep the expected total sum of the released concentration of molecules of type @xmath4 and @xmath5 in all schemes equal to each other and equal to @xmath268 . the system parameters are @xmath269 , and @xmath270 . in fig . [ fig.comp ] , the probability of error for all modulations are compared with each other versus the concentration of transmitted molecules ( we denote an scale of this value by power ) . as fig . [ fig.comp ] depicts , higher error is for the csk modulation which is a simple modulation scheme that uses one molecule type to transmit information bits . in addition , it can be observed that the proposed ts scheme ( with precoder and assuming @xmath170 ) significantly outperforms other modulation and signaling schemes , even with infinite memory at the receiver . on the other hand , our proposed precoder has a more complicated design compared to the csk or mcsk schemes . furthermore , in csk and mcsk , we have only two transmission levels , but in the one with precoder , transmitter is assumed to be able to diffuse at any arbitrary level ( subject to the average transmission power ) . [ fig.diffa ] depicts the performance when we have a mismatch in diffusion coefficients . as expected , once the value of @xmath271 deviates from @xmath184 , the probability of error increases . however , this increase in error probability is relatively negligible if the mismatch between @xmath53 and @xmath54 is less than 10 percent . to make design of the precoder simpler , we investigate the impact of quantizer in fig . [ fig.quantize ] ( a ) versus the average number of released molecules ( power ) , utilizing lloyd algorithm . it is clear that as the number of quantization levels ( @xmath96 ) increase , the mmse of quantizer decreases . however , this does not necessarily improve the performance of the transceiver link . that is , from fig . [ fig.quantize ] it can be realized that although by increasing the quantization levels the average distortion decreases , the error rate does not decrease . for instance , the error probability for the quantization level of @xmath272 is less than the one for @xmath273 . this phenomenon is not strange as lloyd quantizer just minimizes the mmse and not the end - to - end error probability . for a better comparison , in fig . [ fig.quantize ] ( b ) we have plotted the error probability for different quantization levels for lloyd and uniform qunatization rules . although for large quantization levels lloyd quantizer outperforms uniform quantizer , for some small levels , uniform quantizer can outperform the lloyd quantizer . to evaluate the effect of incomplete reaction on the performance of receiver , we investigate the affection of forward reaction rate on the number of received molecules @xmath4 and @xmath5 at the receiver . to this end , we consider the system of equations given in ( [ eq1av])-([eq2av ] ) for a one - dimensional medium . to solve the system of equations , we employ the finite difference method ( fdm ) @xcite , @xcite . in this method , the x - axis and time are discretized into infinitesimal intervals of @xmath274 and @xmath275 , respectively , and a set of recursive equations are derived accordingly . [ fig.realization ] depicts the concentration of molecules @xmath4 and @xmath5 for two consecutive transmission of @xmath4 and then @xmath5 , for different values of the reaction rate . in particular @xmath276 molecules of type @xmath4 are released at time zero , and then @xmath276 molecules of type @xmath5 are released at time @xmath277 . the receiver volume is equal to @xmath278 and the distance between the transmitter and receiver is @xmath279 . as the reaction rate increases , although the difference between the concentration of molecule @xmath4 and @xmath5 is constant , more molecules make reaction and disappear from the medium . focusing on the second transmission , it can be seen that for the reaction with @xmath280 , the limiting reactant ( molecule @xmath4 ) is almost completely removed from the medium at @xmath281 . this verifies the calculations presented in section [ secslowreact ] that for reaction rate larger than a threshold , the concentration of the smaller reactant degrades exponentially fast . molecules of type @xmath4 from origin at time zero , and then @xmath282 molecules of type @xmath5 at time @xmath283 . here , @xmath284 . , scaledwidth=65.0% ] .,scaledwidth=65.0% ] fig . [ fig.imperfect2 ] verifies the intuitive argument given in section [ secslowreact ] . it shows the effect of reaction rate , @xmath58 , on the mean concentration of received limiting molecules at the receiver , i.e. , @xmath285 where the expectation is made over 1000 trials . since we have used precoder , solving ( [ diff ] ) gives the difference of the concentrations at the receiver at the end of each time slot equal to @xmath286 . on the other hand , since the reaction is not complete , the concentration of limiting reactant at the receiver is not zero . it can be observed that as the reaction rate multiplied by the duration of reaction , i.e. , @xmath171 linearly increases , the number of received limiting molecules decreases exponentially fast . the restriction of positive signals makes handling of isi a challenge for molecular communication . in this paper , we proposed a new modulation scheme , named ts , to simulate negative signals . the modulation scheme places information on the difference of concentrations of @xmath4 and @xmath5 regardless of whether a chemical reaction is present or not . the difference of concentrations is a reaction - rate invariant " of chemical reactions such as @xmath287 or @xmath288 , meaning that the difference of concentrations follows a differential equations that is not affected by the reaction rate @xmath58 . the key property used here is that the stoichiometric coefficients of @xmath4 and @xmath5 are equal to each other in the above equations . on the one hand , this implies that the difference of concentrations follows a linear differential equation . on the other hand , invariance with respect to reaction rate implies that the modulation scheme is _ robust _ with respect to factors that influence the reaction rate such as existence of a catalyst . however , all of this comes at the cost of losing a degree of freedom by placing information on the difference of concentrations , and not utilizing the individual concentrations . the general idea of using the reaction rate invariants as signalling variables may be extended to other chemical reactions and employed in other communication settings . we showed that ts modulation allows for negative inputs , and hence we can employ classical ideas based on channel invertion to introduce a precoder at the transmitter side . we also discussed the effect of imperfections such as quantization and mismatch between diffusion coefficients of reactants on the performance of the proposed precoder . numerical results indicate that the ts scheme can strictly outperform previously proposed schemes , even with slow reaction rate and other mentioned imperfections . 35 m. pierobon and i. f. akyildiz , a physical end - to - end model for molecular communication in nanonetworks , " _ ieee journal on selected areas in communications _ , vol . 28 , pp . 602611 , 2010 . a. noel , k. c. cheung , and r. schober , improving receiver performance of diffusive molecular communication with enzymes , " _ ieee transactions on nanobioscience _ , vol . 13 , pp . 31 - 43 , 2014 . a. p. d. silva and n. d. mcclenaghan , proof - of - principle of molecular - scale arithmetic , " _ journal of the american chemical society _ 2000 , 122(16 ) , pp . 39653966 . s. wang , w. guo , and m. mcdonnell , transmit pulse shaping for molecular communications , " _ in proceedings of ieee conference on computer communications workshops _ , april 2014 , pp . 209 - 210 . n. farsad , and a. goldsmith , a novel molecular communication system using acids , bases and hydrogen ions , " _ _ arxiv:__1511.08957 , nov . b. tepekule , a. e. pusane , m. s. kuran , and t. tugcu , a novel pre - equalization method for molecular communication via diffusion in nanonetworks , " _ ieee communications letters _ , aug 2015 , pp . 1311 - 1314 . m. di francesco , , k. fellner , p. a. markowich , the entropy dissipation method for spatially inhomogeneous reaction diffusion - type systems , " _ in proceedings of the royal society of london a : mathematical , physical and engineering sciences _ , 464 ( 2100 ) : 32733300 , 2008 . m. bebendorf , a note on the poincar inequality for convex domains , " _ z. anal . anwendungen _ , 22 ( 4 ) : 751-756 , 2003 . l. desvillettes , k. fellner , b. q. tang , trend to equilibrium for reaction - diffusion systems arising from complex balanced chemical reaction networks , " _ arxiv_:1604.04536 , 2016 . l. desvillettes , k. fellner , exponential decay toward equilibrium via entropy methods for reaction - diffusion equations , " _ journal of mathematical analysis and applications _ , 319 ( 1 ) , 157176 , 2006 . c. v. pao , nonlinear parabolic and elliptic equations , 1992 , plenum press , new york . d. kilinc , and o. b. akan , receiver design for molecular communication , " _ ieee journal on selected areas in communications / supplement _ , part 2 , vol . 31 , no . 12 , dec . 2013 , m. pierobon and i. f. akyildiz , diffusion - based noise analysis for molecular communication in nanonetworks , " _ ieee transactions on signal processing _ , vol . 59 , pp . 25322547 , 2011 . h. mahdavifar and a. beirami , diffusion channel with poisson reception process : capacity results and applications , " _ in proc . of 2015 ieee international symposium on information theory ( isit 2015 ) _ , hong kong , china , june 2015 , pp . 1956 - 1960 . a. gohari , m. mirmohseni and m. nasiri - kenari , information theory of molecular communication : directions and challenges , " _ to appear in the ieee transactions on molecular , biological , and multi - scale communications , the shannon centennial special issue_. i. n. levine , _ physical chemistry _ , mcgraw - hill , 4th ed . , 1995 . cussler , e. l. , _ diffusion : mass transfer in fluid systems _ , new york : cambridge university press , 2007 . d. bothe , m. pierre , the instantaneous limit for reaction - diffusion systems with a fast irreversible reaction , " _ discrete and continuous dynamical systems series s _ , 8(1 ) , 4959 , 2011 . d. bothe , m. pierre , g. rolland , cross - diffusion limit for a reaction- diffusion system with fast reversible reaction , " _ communications in partial differential equations _ , 37:11 , 19401966 , 2012 . e. c. m. crooks , d. hilhorst , self - similar fast - reaction limits for reaction diffusion systems on unbounded domains , " _ journal of differential equations _ , 261 ( 3 ) , 22102250 , 2016 . e. h. degroot , h. sprengel , approximating the movement of a free chemical reaction interface , " _ international communications in heat and mass transfer _ , 17(3 ) , 377385 , 1990 . s. kakeya , on the limits of the roots of an algebraic equation with positive coefficients , " _ ohoku mathematical journal _ vol . 140 - 142 , 1912 . j. l. vazquez , the porous medium equation , mathematical theory , " clarendon - press , oxford university press , oxford ( 2007 ) . s. p. lloyd , least squares quantization in pcm , " unpublished bell lab . note , portions presented at the institute of mathematical statistics meet . , atlantic city , nj , sept . 1957 . also , _ ieee transactions on information theory _ ( special issue on quantization ) , vol . it-28 , pp . 129137 , mar . 1982 . r. mosayebi , h. arjmandi , a. gohari , m. nasiri - kenari , and u. mitra , receivers for diffusion - based molecular communication : exploiting memory and sampling rate , " _ ieee journal on selected areas in communications _ , vol . 23682380 , dec 2014 . r. leveque , finite difference methods for ordinary and partial differential equations : steady - state and time - dependent problems , " _ philadelphia , pa : siam _ , jul . 2007 . b. atakan and o. akan , on channel capacity and error compensation in molecular communication , " _ transaction on computational systems biology x _ , pp . 5980 , 2008 . a. tseng , k. chen , c. chen , and k. ma , electron beam lithography in nanoscale fabrication : recent development , " _ ieee transactions on electronics packaging manufacturing _ , vol . 2 , pp . 141149 , apr . 2003 . h. arjmandi , a. gohari , m. nasiri - kenari and farshid bateni , diffusion based nanonetworking : a new modulation technique and performance analysis , " _ ieee communications letters _ , 17(4 ) : 645 - 648 , 2013 . r. mosayebi , h. arjmandi , a. gohari , m. n. kenari , and u. mitra , on bounded memory decoder for molecular communications , _ in information theory and applications workshop ( ita ) _ , feb . m. u. mahfuz , d. makrakis , and h. t. mouftah , a comprehensive analysis of strength - based optimum signal detection in concentration - encoded molecular communication with spike transmission , " _ ieee transactions on nanobioscience _ , 1 , january 2015 l .- s . meng , p .- c . yeh , k .- c . chen , and i. f. akyildiz , on receiver design for diffusion - based molecular communication , " _ ieee transactions on signal processing _ , 22 , november 15 , 2014 . b. atakan , o. b. akan , on molecular multiple - access , broadcast , and relay channels in nanonetworks , " _ proceedings of the 3rd international conference on bio - inspired models of network , information and computing sytems _ , nov . t. nakano , j. moore , f. wei , v. vasilakos , and jianwei shuai , molecular communication and networking : opportunities and challenges , " _ ieee transactions on nanobioscience _ , vol . 2 , june 2012 . r. e. blahut , computation of channel capacity and ratedistortion functions , " _ ieee transactions on information theory _ 460473 , 1972 . s. m. moser , capacity results of an optical intensity channel with inputdependent gaussian noise , " _ ieee transactions on information theory _ , vol . 1 , pp . 207223 , 2012 . c. v. pao , on nonlinear reaction - diffusion systems , " _ journal of mathematical analysis and applications _ , 87(1 ) , 165198 , 1982 . consider the reaction diffusion equations given in and . when @xmath289 , _ i.e. , _ when the non - linear term does not exist , the system whose inputs are @xmath290 and @xmath291 and whose outputs are @xmath61 and @xmath62 is linear . as a result multiplying both @xmath290 and @xmath291 by a constant @xmath292 would affect @xmath61 and @xmath62 by the multiplicative constant @xmath292 . to understand the case of @xmath293 , it is helpful to consider the following change of variables : @xmath294 , @xmath295 , @xmath296 , @xmath297 , @xmath298 , @xmath299 , @xmath300 , and @xmath301 in equations and for some positive @xmath292 and @xmath302 . then , after the rescaling , equations and become @xmath303 note that this is the same reaction - diffusion equation with @xmath304 . let us begin with the case of @xmath305 , @xmath306 . we observe that multiplying both @xmath290 and @xmath291 by a constant @xmath292 would affect @xmath61 and @xmath62 by the same multiplicative constant @xmath292 _ only if @xmath58 is also multiplied by @xmath292 . _ when we employ @xmath307 times higher molecule production rates @xmath290 and @xmath291 , we should make the chemical reaction @xmath292 times faster if we want the concentrations @xmath61 and @xmath62 to be multiplied by @xmath292 . if we keep using the same reaction rate @xmath58 , molecules of type @xmath4 and @xmath5 would not cancel out as fast as @xmath308 and concentrations of molecules of type @xmath4 and @xmath5 will be multiplied by bigger multiplicative constants than @xmath292 . this is an effect of the non - linear term . finally , consider the case of @xmath309 , @xmath310 . observe the particular change of variable @xmath301 used . here , we see the intuitive relation between @xmath58 and time : if we multiply @xmath58 , @xmath53 and @xmath54 by two , we are making the reaction and diffusion twice as fast and we expect the same diffusion pattern in half the time . now , consider the use a time - slotted communication strategy with time steps of size @xmath95 . the above discussion shows that decreasing @xmath95 by a factor of two while at the same time increasing @xmath58 , @xmath53 and @xmath54 by a factor of two would keep the molecular concentrations ( and the error probabilities ) the same . consider the differential equations @xmath311 let @xmath312 and @xmath313 be two solutions to the above equations , where @xmath314 and @xmath315 are assumed to be hlder continuous functions . then by adding up equation for these two solutions , we obtain @xmath316 where in , we used the fact that @xmath317 and @xmath318 are non - negative . similarly , @xmath319 furthermore , @xmath320 have the zero initial and boundary conditions . now , consider the reaction - diffusion set of equations for molecule production rates @xmath321 and @xmath322 : @xmath323 these equations can be re - expressed in the standard form @xmath324 for @xmath325 @xmath326 observe that @xmath327 and @xmath328 and @xmath329 are quasi - monotone decreasing functions ( in the sense of @xcite ) . as a result , and imply that @xmath320 is an upper solution for and in the sense of @xcite . this completes the proof .

the wiretap channel model of wyner is one of the first communication models with both reliability and security constraints . explicit capacity - achieving schemes for various models of the wiretap channel have received considerable attention in recent literature . in this paper , we address the original version of this problem , showing that capacity of the general ( not necessarily degraded or symmetric ) wiretap channel under a `` strong secrecy constraint '' can be achieved using a transmission scheme based on polar codes . we also extend our construction to the case of broadcast channels with confidential messages defined by csiszr and krner , achieving the entire capacity region of this communication model . department of ece and institute for systems research , university of maryland , college park , md 20742 , email : [email protected] . research supported in part by nsf grant ccf1217245 ] department of ece and institute for systems research , university of maryland , college park , md 20742 , and iitp , russian academy of sciences , moscow , russia . email : [email protected] . research supported in part by nsf grants ccf1217245 , ccf1217894 , and ccf1422955 . + the results of this paper were presented in part at the ieee information theory workshop , jerusalem , israel , april 27-may 1 , 2015 . ] * _ _ * index terms : polar codes , chaining construction , strong secrecy , coordinate partition .

1410.3422

the wiretap channel model @xmath0 was introduced by wyner in 1975 @xcite . in this model , there are two receivers @xmath1 and a single transmitter @xmath2 . the transmitter aims at sending messages to receiver 1 through a communication channel @xmath3 . the information sent from @xmath2 to @xmath4 is also received by receiver 2 through another channel @xmath5 . the transmission problem in the system @xmath6 calls for designing a coding system that supports communication between @xmath2 and @xmath4 in a way that is both reliable and secure . the reliability requirement is the usual one for communication systems , namely , that the error probability of decoding the information by @xmath4 be made arbitrarily low by increasing the block length of the encoding . at the same time , the transmission needs to be made secure in the sense that the information extracted by receiver 2 about the message of @xmath2 approaches zero as a function of the block length . to describe the problem in formal terms , denote the input alphabet of the transmitter by @xmath7 , and the output alphabets of the channels @xmath3 and @xmath5 by @xmath8 and @xmath9 , respectively . the messages that the transmitter can convey to receiver 1 form a finite set denoted below by @xmath10 for transmission over the channel the message is encoded using a mapping @xmath11 where @xmath12 is an @xmath13-fold repetition of the input alphabet . we say that @xmath14 is a length-@xmath13 block encoder of the transmitter . capacity - attaining schemes for the wiretap channel rely on randomized encoding , i.e. , a mapping that sends @xmath15 to a probability distribution on @xmath16 in other words , the message @xmath17 is encoded as a sequence @xmath18 with probability @xmath19 and the encoder is defined as a matrix of conditional probabilities @xmath20 the decoder of receiver 1 is a mapping @xmath21 . we also denote by @xmath22 and @xmath23 the conditional distributions induced by the channels @xmath3 and @xmath5 , respectively , and define the induced distributions @xmath24 where , for instance , @xmath25 , where @xmath26 and @xmath27 refer to the @xmath28-th symbol of the vectors @xmath29 and @xmath30 , respectively . [ def:1.1 ] we say that the encoder - decoder pair @xmath31 _ gives rise to @xmath32-transmission _ over the wiretap channel @xmath0 if [ wiretap_definition ] @xmath33 where @xmath34 is the message random variable ( rv ) and @xmath35 is the rv that corresponds to the observations of receiver 2 . in definition [ wiretap_definition ] , eq . represents the reliability of communication condition while answers the security of transmission requirement . we note that in many works on transmission with a secrecy constraint the security condition was formulated in a more relaxed way , namely as the inequality @xmath36 this is particularly true about pre-1990s works in information theory , but also applies to some very recent works on the wiretap channel , e.g. , @xcite . however , as shown by maurer in @xcite , this constraint does not fulfill the intuitive security requirements in the system . more specifically , it is possible to construct examples in which inequality is satisfied and at the same time receiver 2 is capable of learning @xmath37 out of @xmath13 bits of the encoding @xmath38 in view of this , maurer suggested as a better alternative to condition . as a result , currently is called the `` weak security constraint '' as opposed to the stronger constraint . in this paper we design coding schemes that provide strong secrecy , so below we work only with condition . the secrecy capacity of the wiretap channel is defined as follows . [ def:1.2 ] the value @xmath39 is called an achievable rate for the wiretap channel @xmath0 if there exists a sequence of message sets @xmath40 and encoder - decoder pairs @xmath41 giving rise to @xmath42 transmission with @xmath43 and @xmath44 as @xmath45 . the secrecy capacity @xmath46 is the supremum of achievable rates for the wiretap channel . the following theorem provides an expression for @xmath46 . [ thm : dg ] ( @xcite ; see also @xcite ) the secrecy capacity of the wiretap channel @xmath0 equals @xmath47 , \label{eq : cs}\ ] ] where the maximum is computed over all rvs @xmath48 such that the markov condition @xmath49 holds true , and such that @xmath50 , @xmath51 . while most general constructive coding schemes for the wiretap channels rely on polar codes , there were some constructive solutions even before the publication of arikan s seminal work @xcite . at the same time , these schemes applied only to some special cases of the channels @xmath52 for instance the case when @xmath3 is noiseless and @xmath5 is a binary erasure channel was addressed in @xcite which show that in this case the capacity @xmath46 can be achieved using low - density parity - check codes . the results in @xcite are based on the weak security assumption while strong security is considered in @xcite . moreover , @xcite extends the construction to the cases when both @xmath3 and @xmath5 are erasure channels , and when @xmath3 is noiseless and @xmath5 is a binary symmetric channel . as usual with the application of low - density codes , the constructions are based on code ensembles and strictly speaking are not explicit . another special case of the wiretap channel relates to the combinatorial version of the erasure channel ( the so - called wiretap channel of type ii ) in which receiver 2 can choose to observe any @xmath53 symbols out of @xmath13 transmitted symbols . constructive capacity - achieving solutions for this case are based on mds codes @xcite or extractors @xcite . in @xcite , it is shown that @xmath46 is achievable with strong security using invertible extractors , if both @xmath3 and @xmath5 are binary symmetric channels . both encoding and decoding algorithms in @xcite have polynomial complexity . moreover , @xcite also claims that its proof method can be easily extended to other wiretap channels as long as both @xmath3 and @xmath5 are symmetric . after the introduction of polar codes by arikan , achieving @xmath46 via polar coding has been considered by different works , mostly under the degradedness assumption . recall that a channel @xmath54 is called _ degraded _ with respect to a channel @xmath55 if there exists a stochastic @xmath56 matrix @xmath57 such that for all @xmath58 @xmath59 the wiretap channel @xmath60 is called degraded if the channel to the eavesdropper is degraded with respect to the main channel . in this case theorem [ thm : dg ] affords a simpler formulation because there is no need in the auxiliary rv @xmath61 . namely , in the degraded case the secrecy capacity equals ( * ? ? ? * probl . 17.8 ) @xmath62 ( this specialization is true under slightly more general assumptions , but we will not need them below ) . communication over degraded wiretap channels using polar codes was considered in a number of papers , notably , @xcite . the main result of these works is that secrecy capacity can be attained under the weak security constraint . we note that the degraded case is easier to handle with polar codes because of the specific nature of the polar codes construction ( more on this below in sect . [ sect : closer ] ) . another step was made by @xcite which suggested a polar coding scheme that attains the rate @xmath46 of a symmetric degraded wiretap channel @xmath0 under the strong security requirement . more details about the results of @xcite are given in sect . [ sect : closer ] below . the problem of attaining secrecy capacity of the general wiretap channel under the strong secrecy condition and without the degradedness assumption was further studied in @xcite . a polar coding scheme suggested in this work attains a transmission rate of @xmath63.$ ] clearly @xmath64 for all @xmath2 since one can take @xmath65 in the markov chain @xmath66 which appears in theorem [ thm : dg ] . it is not immediately clear for which channels the result of @xcite actually attains the secrecy capacity of @xmath67 at the same time , the coding scheme employed in @xcite relies on two nested layers of the polarizing transform . the decoder for the second ( outer ) layer works with the probability distribution generated by the first decoder , which is not easily computable . thus , the low complexity decoding claim of the construction made in @xcite is not supported by the known decoding procedures for polar codes . for these reasons the construction in @xcite does not resolve the question of constructing an explicit capacity - achieving scheme for the nondegraded case of wiretap channels . in related works @xcite the problem of constructing capacity achieving schemes for wiretap channels was addressed for the case of quantum channels . the constructions suggested in these works attain symmetric secrecy capacity of quantum wiretap channels . these constructions require a shared secret key between the transmitter and receiver 1 . this requirement seems to be intrinsic to polar code constructions for this problem including our work . however the constructions in @xcite require a positive - rate shared key , which results in a communication scheme that transmits at rates separated from capacity . to summarize , to the best of our knowledge the question of constructing explicit capacity - achieving transmission schemes for the non - degraded wiretap channel with strong secrecy is an open problem . it is this problem that we aim to solve in this paper by removing the degradedness assumption . we also do not assume that either of the channels @xmath68 is symmetric . the main idea of our work is to exploit the markov chain conditions intrinsic to secure communication problems using polar codes . in section [ sect : wt ] we propose a polar coding scheme that attains the secrecy capacity under the strong security assumption . both the encoding and decoding complexity estimates of our construction are @xmath69 , where @xmath13 is the length of the encoding . in section [ csiszar - korner ] we generalize our construction to cover the case when a part of transmitter s message is public , i.e. , is designed to be conveyed both to receivers 1 and 2 . this model , called a broadcast channel with confidential messages , is in fact the principal model in the founding work of csiszr and krner @xcite on this topic . apart from the basic polar coding results @xcite , our solution of the described problems relies on the previous work on the wiretap channel @xcite , the polar coding scheme for the broadcast channel of @xcite , and the construction of polar codes for general memoryless channels in @xcite . a new idea introduced in our solution is related to a stochastic encoding scheme that emulates the random coding proof of the capacity theorem in @xcite , whereby polarization is used for the values of the auxiliary random variable @xmath61 in theorem [ thm : dg ] , followed by a stochastic encoding into a channel codeword . another insight , which is particularly useful for the broadcast channel result in sect . [ sect : bcc ] , is related to a partition of the coordinates of the transmitted block that enables simultaneous decoding of different parts of the transmitted message by both receivers , whereby the decoder of polar codes is used by the receivers according to their high- and low - entropy bits . it becomes possible to show that the receivers recover the bits designed to communicate with each of them with high probability , and that the secret part of the message is not accessible to the unintended recipient . we begin with recalling basic notation for polar codes and then continue with the scheme for capacity - achieving communication on discrete binary - input channels . let @xmath70 be a binary - input channel with the output alphabet @xmath71 input alphabet @xmath72 and the conditional probability distribution @xmath73 throughout the paper we denote the capacity and the symmetric capacity of @xmath70 by @xmath74 and @xmath75 , respectively . we say @xmath70 is symmetric if @xmath76 can be obtained from @xmath77 through a permutation @xmath78 such that @xmath79 note that if @xmath70 is symmetric then @xmath80 given a binary rv @xmath2 and a discrete rv @xmath4 supported on @xmath8 , define the bhattacharyya parameter @xmath81 as @xmath82 the value @xmath83 measures the amount of randomness in @xmath2 given @xmath4 in the sense that if it is close to zero , then @xmath2 is almost constant given @xmath4 , while if it is close to one , then @xmath2 is almost uniform on @xmath84 given @xmath4 . the bhattacharyya parameter @xmath85 of a binary - input channel @xmath70 is defined as @xmath86 clearly if @xmath87 then @xmath81 coincides with the value @xmath85 for the communication channel @xmath88 for @xmath89 and @xmath90 , define the polarizing matrix ( or the arikan transform matrix ) as @xmath91 , where @xmath921&1\end{array}\hspace*{-.05in}\big)$}}$ ] , @xmath93 is the kronecker product of matrices , and @xmath94 is a `` bit reversal '' permutation matrix @xcite . in @xcite , arikan showed that given a symmetric and binary input channel @xmath70 , an appropriate subset of the rows of @xmath95 can be used as a generator matrix of a linear code that achieves the capacity of @xmath70 as @xmath45 . the material in this section and section [ sect : asymmetric ] is well understood , but it merits some space in the present paper because it helps us to define the terminology that is useful for the main results below . given a symmetric binary - input channel @xmath70 , define the channel @xmath96 with input alphabet @xmath97 and output alphabet @xmath98 by the conditional distribution @xmath99 where @xmath100 is the conditional distribution that defines @xmath70 . define a combined channel @xmath101 by the conditional distribution @xmath102 in terms of @xmath101 , the channel seen by the @xmath28-th bit @xmath103 ( also known as the bit - channel of the @xmath28-th bit ) can be written as @xmath104 where @xmath105 . we see that @xmath106 is the conditional distribution of @xmath107 given @xmath108 provided that the channel inputs @xmath109 are uniformly distributed for all @xmath110 . the bit - channels thus defined are partitioned into good channels @xmath111 and bad channels @xmath112 based on the value of their bhattacharyya parameters . bearing in mind our notation choices later in the paper , we denote them by @xmath113 : z(w_i ) \leq \delta_n \}\\ \ch_{x|y}=\ch_{x|y}(n)&= \ { i\in [ n ] : z(w_i ) > 1-\delta_n \ } , \end{aligned}\ ] ] where @xmath114=\{1,2,\dots , n\}$ ] and @xmath115 as shown in @xcite , for any symmetric binary - input channel @xmath70 and any constant @xmath116 @xmath117 based on this equality , information can be transmitted over the good - bit channels while the remaining bits are fixed to some values known in advance to the receiver ( in polar coding literature they are called _ frozen bits _ ) . the transmission scheme can be described as follows : a message of @xmath118 bits is written in the bits @xmath119 the remaining @xmath120 bits of @xmath121 are set to 0 . this determines the sequence @xmath121 which is transformed into @xmath122 and the vector @xmath30 is sent over the channel . denote by @xmath29 the sequence received on the output . the decoder finds an estimate of @xmath121 by computing the values @xmath123 as follows : @xmath124 the results of @xcite imply the following upper bound on the error probability @xmath125 @xmath126 where @xmath127 is any number in the interval @xmath128 and @xmath129 this describes the basic construction of polar codes @xcite which attains symmetric capacity @xmath75 of the channel @xmath70 with asymptotically vanishing error rate . [ rem : on ] there is a subtle point about the limit relations in . even though asymptotically the bit channels are either good or bad , it is not true that @xmath130 because there is a nonempty subset of indices @xmath131 of cardinality @xmath132 that is neither good nor bad . this distinction has no import for the simple situation of transmitting over @xmath70 , but leads to complications in the multi - user systems considered below ; see , e.g. , in section [ sect : wt ] . let @xmath70 be a binary - input discrete memoryless channel @xmath133 and let @xmath134 be the capacity achieving distribution of @xmath70 . if @xmath134 is not uniform , then the basic scheme attains a transmission rate @xmath75 which is less than @xmath135 this scheme was extended by honda and yamamoto @xcite to cover the case of arbitrary distributions @xmath134 , attaining the rate @xmath74 for general binary - input channels . to explain the idea in @xcite , for a given block length @xmath13 define the sets @xmath136 : z(u_i|u^{i-1})\geq 1-\delta_n\ } \\ { \mathscr l}_x&=\{i\in [ n ] : z(u_i|u^{i-1})\leq \delta_n\ } \\ { \mathscr h}_{x|y}&=\{i\in [ n ] : z(u_i|u^{i-1},y^{n})\geq 1-\delta_n\ } \\ { \mathscr l}_{x|y}&=\{i\in [ n ] : z(u_i|u^{i-1},y^{n})\leq \delta_n\ } \end{aligned } \label{eq : l}\ ] ] where @xmath137 have the same meaning as above . it can be shown @xcite that @xmath138 we note that the set @xmath139 is the set of good bit channels defined in . unlike the case of uniform @xmath134 , it is not possible to use all of these channels to transmit information over @xmath140 this is because if @xmath141 , then @xmath108 can not be used to carry information conditioned on previous bits @xmath142 . hence @xcite argued that the set of information indices should be chosen as @xmath143 rather than @xmath139 . since @xmath144 and the number of indices that are neither in @xmath145 nor in @xmath139 is @xmath132 , we have @xmath146 i.e. , transmitting the information using the bits @xmath147 attains the capacity of the channel @xmath140 the code construction in @xcite makes use of the following partition of the coordinate set @xmath114 $ ] : @xmath148 { \mathscr i}={\mathscr h}_x \cap { \mathscr l}_{x|y } \end{array}\right\}\ ] ] where the superscript @xmath149 refers to the complement of the subset in @xmath114.$ ] in terms of this partition , the encoding is done as follows . the information bits are stored in @xmath150 as for the bits in the subset @xmath151 @xcite suggested to sample their values from the distribtion @xmath152 these values are shared with the receiver similarly to the `` frozen bits '' of the symmetric scheme of @xcite . once @xmath121 is determined , the transmitter finds @xmath153 and sends it over the channel . the receiver uses the following successive decoding function : for @xmath154 let @xmath155 note that this rule represents a map decoder for the @xmath28th subchannel , which for the symmetric case coincides with the ml decoder rule . the probability of decoding error can be bounded above similarly to : @xmath156 where the parameters have the same meaning as before . moreover , @xcite argues that there exists a set of deterministic maps @xmath157 such that holds true , stating the decoding rule in the form @xmath158 the mappings @xmath159 are shared between the transmitter and the receiver prior to communication . we shall use this form of the transmission scheme and the decoder throughout our paper . this completes the description of the capacity - achieving transmission scheme of @xcite . we will rely in part on these ideas in our construction of a coding scheme for the wiretap channel . to explain our proposal we will first discuss some of the schemes available in the literature . we begin with the transmission scheme of @xcite ( see also @xcite ) . as already remarked , these works are concerned with the special case when the channel @xmath5 is degraded with respect to @xmath3 and aim to attain the rate value with weak secrecy . let @xmath160 be a random uniform vector over @xmath97 . similarly to @xmath145 and @xmath139 given by , define the following subsets of indices : @xmath161 : z(u_i|u^{i-1},z^{n})\geq 1-\delta_n\}\\ { \mathscr l}_{x|z}&=\{i\in [ n ] : z(u_i|u^{i-1},z^{n})\leq \delta_n\ } \end{aligned}\ ] ] where @xmath162 , and @xmath35 is the output that receiver 2 observes when the transmitter sends @xmath160 . partition the set @xmath114 $ ] as follows : @xmath163 note that the degradedness assumption implies the inclusion @xmath164 . the coding scheme for the wiretap channel relies on the partition and is summarized in figure [ figure1 ] . the information is stored in the bits @xmath165 . the bits in the coordinates in @xmath166 are chosen randomly while the bits in @xmath167 form a subset of the frozen bits . attainability of the rate using this coding scheme is proved in the cited papers . an essential remark here is that the bits @xmath168 are randomly selected because fixing their values contradicts even the weak security constraint , let alone the stronger one . we note that generally @xmath169 and even though the number of coordinates in @xmath170 behaves as @xmath171 this constitutes an obstacle to achieving strong security . to bypass it , @xcite uses a different partition of the coordinates , namely @xmath172 apart from transmitting the information , the coding scheme aims to convey the bits in @xmath173 to receiver 1 using the good indices of receiver 1 , at the same time preserving the security requirement . this is accomplished using the `` chaining '' construction proposed in @xcite and shown in fig . [ figure2 ] . as the figure suggests , the bits in @xmath174 contained in block @xmath175 are transmitted over the channel as a part of the message of block @xmath176 for all @xmath177 this enables receiver 1 to recover these bits reliably as a part of the successive decoding procedure for block @xmath178 which is performed similarly to . at the same time , because of the inclusion @xmath179 receiver 2 does not have the resources for their reliable decoding , which provides the desired security . the analysis of the transmission is performed based on @xmath180 blocks of @xmath13 bits as opposed to a single block . the seed for the transmission is provided by choosing @xmath181 random bits which are shared with receiver 1 ( more on this below ) . in each of the blocks 1 to @xmath180 , the bits indexed by the set @xmath182 are used to send the message . here @xmath183 is an arbitrary subset of the set @xmath184 of size @xmath185 whose role is explained below . the bits in @xmath186 are selected randomly , and the bits @xmath187 are frozen , i.e. , assigned arbitrarily and shared in advance with receiver 1 ( they may be also known to receiver 2 without compromising secrecy ) . the assignment of bits in the set @xmath185 in block @xmath175 depends on the block index . in block 1 these bits are set equal to the message bits of the seed block . in block @xmath188 the bits indexed by the set @xmath185 are set to be equal to the bits in the set @xmath183 in block @xmath176 representing the chaining procedure . having formed the sequence @xmath189 in block @xmath190 , the encoder passes it through the polarizing transform and transmits the sequence @xmath153 over the channel . the only remaining problem is to convey to receiver 1 the bits of @xmath191 of the first block . this is done by performing the seed transmission of a block which encodes the @xmath192 bits using some error correcting code of length @xmath13 . as claimed in @xcite , it is possible to choose such a code to fulfill the reliability and security requirements because its rate can be made arbitrarily close to zero . the fact that the seed code needs to encode only a small number @xmath132 of message bits follows from the degradedness assumption , which is therefore essential in this construction . as shown in @xcite , this scheme satisfies both constraints and under the assumption that the channel to the eavesdropper is degraded with respect to the channel @xmath193 the rate of communication between the transmitter and receiver 1 can be made arbitrarily close to the value @xmath194 since the assumption that @xmath5 is degraded with respect to @xmath3 ensures that @xmath181 , i.e. , there is no asymptotic loss in rate by removing the bits @xmath195 from the message in order to support the strong security condition . in this section , we show that secrecy capacity for the wiretap channel given by theorem [ thm : dg ] is achievable using polar codes . for this purpose , we consider the rvs @xmath48 as described by theorem [ thm : dg ] , i.e. , we assume some fixed distributions @xmath196 and the conditional distributions @xmath50 , @xmath51 that satisfy the markov condition @xmath197 and maximize the expression in . define the rv @xmath198 where @xmath199 denotes @xmath13 independent realizations of @xmath61 . the transformation @xmath200 induces conditional distributions @xmath201 derived from the corresponding distributions of the rvs @xmath202 define the sets @xmath203 , @xmath204 , @xmath205 , @xmath206 as follows : @xmath207 : z(t_i|t^{i-1 } ) \geq 1-\delta_n\ } \\ \cl_v&= \ { i\in [ n ] : z(t_i|t^{i-1 } ) \leq \delta_n\}\\ { \mathscr h}_{v|y}&= \{i\in [ n ] : z(t_i|t^{i-1},y^{n } ) \geq 1-\delta_n\}\\ { \mathscr l}_{v|y}&= \{i\in [ n ] : z(t_i|t^{i-1},y^{n } ) \leq \delta_n\}\end{aligned}\ ] ] and define the sets @xmath208 , @xmath209 analogously . the cardinalities of these sets satisfy @xcite @xmath210 as @xmath211 define a partition of @xmath114 $ ] into the following sets which will be used to describe the coding scheme in this section in the sense different from sect . [ sect : closer ] . since both uses are localized to their respective sections , this should not cause confusion . ] @xmath212 the partition of @xmath114 $ ] that thus arises is illustrated in figure 3 . it will be seen that the subsets @xmath213 , @xmath214 , @xmath215 , @xmath216 in our coding scheme play the role similar to that of the analogously denoted subsets in . importantly , the cardinality of @xmath215 is not @xmath132 any more , which requires adjustments in the transmission scheme . moreover , there is an extra randomness needed to determine the sequence to be transmitted , as will be seen in the encoding algorithm below . partition of @xmath13 coordinates of the block for transmission over the wiretap channel @xmath0 ; see . the highlighted part of the top block represents high - entropy coordinates for the distribution @xmath217 . similarly , in the middle block we highlight the high - entropy coordinates of the distribution @xmath218 and in the bottom block the low - entropy coordinates for the distribution @xmath219 . [ fig : partition ] * _ _ * encoding : we build on the chaining idea of @xcite , connecting multiple blocks in a cluster whose performance in transmission will attain the desired goals . the cluster consists of a seed block and a number , @xmath180 , of other blocks . the seed block consists of @xmath220 random bits . even though the cardinality of the set @xmath215 constitutes a nonvanishing proportion of @xmath114 $ ] , the rate of the seed @xmath221 can be made arbitrarily small by choosing @xmath180 sufficiently large . ( for example , one can set @xmath222 for some @xmath223 , and let @xmath211 ) ] let us describe the encoding and decoding procedures . the transmission is accomplished using @xmath180 blocks of length @xmath13 each and the seed block . every block @xmath224 contains a group of almost deterministic bits , denoted by @xmath225 in figure [ figure3 ] . the values of these bits are assigned according to a family of deterministic rules @xmath226 described in . the bits @xmath227 in each of the @xmath180 blocks are determined similarly , based on @xmath228 . these bits are assigned the same values for each block , and are shared with receiver 1 . note that the rate @xmath229 can be made arbitrarily small similarly to @xmath230 . the remaining subsets of coordinates are filled as follows . for block 1 , the bits in the set @xmath231 are assigned the value of the bits of the seed block , while for blocks @xmath188 these bits are set to be equal to the bits in @xmath232 of block @xmath233 here , @xmath234 is a subset of @xmath235 having the same size as @xmath231 . the messages are stored in the bits indexed by @xmath236 the randomly chosen bits in @xmath234 are written in the coordinates that are good for receiver 1 and contained in the bad ( high - entropy ) set of receiver 2 . these bits are transmitted to receiver 1 in block @xmath175 and used for the decoding of the message contained in block @xmath237 for all @xmath238 finally , the bits in @xmath239 are assigned randomly and uniformly for each of the @xmath180 blocks . the diagram of the chaining construction for encoding is given in figure [ figure3 ] . once the blocks @xmath240 are formed , we find @xmath180 sequences @xmath241 by using the polarizing transform . finally , given @xmath242 the codeword to be sent over the wiretap channel will be chosen as @xmath243 with probability @xmath244 , where @xmath245 is the conditional distribution induced by the joint distribution of the rvs @xmath61 and @xmath2 . this logic is suggested by the proof of the capacity theorem , theorem [ thm : dg ] , which first considers `` transmitting '' the rv @xmath199 to the receivers , and then choosing @xmath160 so as to satisfy the markov chain condition in the statement . denote by @xmath246 the message sequence encoded in the seed block , and by @xmath247 the corresponding sequences in the other blocks ( see fig . [ figure3 ] ) . let @xmath248 be the sequences that receiver 1 observes on the output of the channel @xmath193 the decoding rule is as follows : @xmath249 where @xmath201 and @xmath250 are the conditional distributions induced by the joint distribution of the rvs @xmath251 and @xmath252 ( this rule is applied to each of the blocks @xmath190 , and @xmath175 is mostly omitted from the notation ) . let us show that the described scheme attains the secrecy capacity of @xmath67 namely , the following is true . [ prop : wiretap ] for any @xmath253 , @xmath254 and @xmath45 it is possible to choose @xmath180 so that the transmission scheme described above attains the transmission rate @xmath255 that is within @xmath256 of the secrecy capacity of @xmath0 and the information leaked to receiver 2 satisfies the strong secrecy condition . throughout the proof we assume that the rvs @xmath48 are as given in theorem [ thm : dg ] and denote by @xmath257 their joint distribution . the distribution @xmath258 refers to @xmath13 independent repetitions of the rvs . the rate of the proposed coding scheme is @xmath259 which approaches @xmath260 as @xmath211 according to theorem [ thm : dg ] , this is the target rate that we want to achieve for given @xmath61 and @xmath2 satisfying @xmath197 and @xmath50 , @xmath51 . now let us prove the reliability and security conditions . let us introduce the following rvs : let @xmath261 correspond to the sequence of message bits @xmath262 transmitted in blocks @xmath263 and let @xmath264 be a sequence of observations of receiver 2 as a result of the transmission of the @xmath180 blocks . further , let @xmath265 correspond to the bits contained in the subset @xmath266 the claim of low error probability for receiver 1 follows from the results of @xcite . since our communication scheme is more complicated compared to @xcite , we give some additional details . since we know the distribution @xmath267 we can compute the distributions @xmath268 now assume that the assignments of the bits indexed by @xmath167 and @xmath225 are done randomly by sampling from the distribution @xmath269 for each of the blocks @xmath270 let @xmath271 be the joint distribution of the correspoding sequences arising from this assignment . denote by @xmath272 the @xmath273 distance between the distributions . from the proof of lemma 1 in @xcite it follows that @xmath274 holds for all the @xmath180 blocks of transmission . moreover , since the message bits are entirely contained in the set of good bits for channel @xmath3 , the probability of error is bounded by @xmath275 for each individual block ( see @xcite , eqns ( 59)-60 ) ) . therefore , there exists a family of deterministic rules @xmath276 such that the overall error probability for the successive decoding procedure is at most @xmath277 . we conclude that the probability that receiver 1 decodes the information bits correctly approaches @xmath278 as @xmath13 tends to infinity . _ security : _ we will show that condition is fulfilled for the sequence of @xmath180 blocks of transmission . for that purpose , we will first prove the following lemma . let @xmath279 be a subset of coordinates , and let let @xmath280=\{t_i , i\in{\mathscr a}\}$ ] and @xmath281=\{t_i , i\in\ci\backslash{\mathscr a}\}.$ ] then @xmath282 ; t[{\mathscr a } ] , z^n)=o(n^3 2^{-n^{\beta } } ) . \label{lemma_ineq_wiretap}\ ] ] [ security_lemma_wiretap ] by definition in we have the inclusion @xmath283 . let us label the indices in @xmath235 as @xmath284 , and assume that @xmath285 . using the inequality @xmath286 @xcite , we obtain the estimate @xmath287 for all @xmath288 , where @xmath289 refers to the entropy under the distribution @xmath290 . our aim is to find an estimate on the mutual information in under the distribution @xmath291 . to find an upper bound for the entropy @xmath292 ( computed under @xmath291 ) , we use a standard estimate ( e.g. , ( * ? ? ? * theorem 17.3.3 ) ) , and write @xmath293 refers to the alphabet size for a single observation of the eavesdropper . then , we get from that @xmath294 observe that implies @xmath295 for all @xmath296 then we obtain @xmath297\,| t[{\mathscr a } ] , z^n)&= h ( t[\ci]\,| z^n)- h(t[{\mathscr a}]\,| z^n ) \nonumber\\ & \geq h ( t[\ci]| z^n)- |{\mathscr a}| \nonumber\\ & = \sum_{i=1}^{|\ci| } h(t_{a_i}|t_{a_1 } , t_{a_2 } , \dots , t_{a_{i-1 } } , z^n ) - |{\mathscr a}| \nonumber\\ & \geq |\ci | ( 1-o(n^2 2^{-n^{\beta } } ) ) - |{\mathscr a}| \label{lemma_ineq3_wiretap } \\ & \geq |\ci \backslash { \mathscr a}|- o(n^3 2^{-n^{\beta } } ) \nonumber \end{aligned}\ ] ] where is due to . this completes the proof of . let @xmath298 and @xmath299 for all @xmath300 , we have @xmath301 [ security_lemma_wiretap2 ] the proof is by induction on @xmath302 the base case @xmath303 follows from lemma [ security_lemma_wiretap ] . now assume that the claim of the lemma is true for @xmath304 and write @xmath305 using the chaining structure shown in figure [ figure3 ] we argue that the only part of the transmission that connects block @xmath306 to block @xmath307 is given by @xmath308 this implies that , conditional on @xmath309 , we have a markov chain @xmath310 so @xmath311 therefore , @xmath312 here is due to lemma [ security_lemma_wiretap ] and the induction hypothesis . this completes the proof . now we are ready to complete the proof of the strong security condition . for this purpose , consider the following sequence of inequalities : @xmath313 where and are implied by the chaining structure in figure [ figure3 ] , and is due to lemmas [ security_lemma_wiretap ] and [ security_lemma_wiretap2 ] . from , we have @xmath314 which implies @xmath315 thus , we observe that @xmath316 as required . we conclude that a secrecy rate of @xmath260 is achievable for any @xmath61 such that @xmath49 holds . therefore , the secrecy capacity @xmath46 given by is also achievable . in this section we observe that ideas of the previous section together with some earlier works enable us to extend our code construction to a more general communication model introduced in @xcite . consider a pair of discrete memoryless channels with one transmitter @xmath2 and two receivers @xmath317 as before , let @xmath318 and @xmath319 denote the channels and let @xmath320 and @xmath321 denote the input alphabet and the output alphabets . we assume that the system transmits three types of messages : 1 . a message @xmath322 from @xmath2 to @xmath323 for which there are no secrecy requirements ; 2 . a message @xmath324 from @xmath2 to @xmath4 which is secret from @xmath325 3 . a message @xmath326 from @xmath2 to @xmath4 and @xmath327 , called the `` common message '' . following @xcite , we call this communication scheme a _ broadcast channel with confidential messages _ ( bcc ) . as before , a block encoder for the bcc is a mapping @xmath328 a stochastic version of the encoder is a probability matrix @xmath329 with columns indexed by @xmath18 and rows indexed by the triples @xmath330 given such a triple , the stochastic encoder samples from the conditional probability distribution on @xmath16 in accordance with the problem statement , there are two decoders : the decoder of receiver 1 is defined by a mapping @xmath331 and the decoder of receiver 2 is a mapping @xmath332 . denote the rate of the common message @xmath53 by @xmath333 , and denote the rates of the secret and non - secret messages to @xmath4 by @xmath334 and @xmath335 respectively . the analogs of definitions [ def:1.1 ] , [ def:1.2 ] in this case look as follows . the encoder - decoder mappings @xmath336 give rise to _ @xmath32- transmission _ over the bcc if for every @xmath337 , @xmath338 , @xmath339 , decoder @xmath340 outputs the transmitted triple @xmath341 and decoder @xmath342 outputs the message @xmath53 with probability greater than @xmath343 , i.e. , @xmath344 @xmath345 is an achievable rate triple for the bcc if there exists a sequence of message sets @xmath346 , @xmath347 , @xmath348 and encoder - decoder triples @xmath349 giving rise to @xmath42 transmission with @xmath43 , such that @xmath350 where @xmath351 is the random variable that corresponds to the secret message . [ bcc ] note that our definition takes into account the formulation in @xcite and is slightly different from the one in @xcite ( where we write @xmath345 , @xcite has @xmath352 ) . r the following theorem gives the achievable rate region for the triple @xmath345 . [ thm : bcc]@xcite,@xcite the capacity region of the bcc consists of those triples of nonnegative numbers @xmath345 that satisfy , for some rvs @xmath353 with @xmath50 and @xmath51 , the inequalities @xmath354 \label{eq : r0}\\ r_s & \leq i(v;y|u)-i(v;z|u ) \label{eq : rs}\\ r_0+r_1+r_s & \leq i(v;y|u)+ \min [ i(u;y),i(u;z)].\label{eq : r } \end{aligned}\ ] ] moreover , it may be assumed that @xmath355 and the range sizes of @xmath356 and @xmath357 are at most @xmath358 and @xmath359 . [ main ] one can define the secrecy capacity @xmath46 in terms of the capacity region of the bcc , and then recover theorem [ thm : dg ] as a particular case of theorem [ main ] . in this section , we aim to show that the capacity region of the bcc can be achieved using polar codes . in the first two steps we design a scheme that achieves the rate pairs @xmath360 in - , and in the last step we show that for any such pair @xmath360 any rate value @xmath361-r_0-r_s\label{r_1}\ ] ] is also achievable . finally , the security condition in will be shown in proposition [ prop : sc ] below . the overall encoding scheme is stochastic and assumes some fixed joint distribution of the rvs @xmath362 such that the constraints of theorem [ thm : bcc ] are satisfied . since the results below are valid for any such distribution , this will enable us to claim achievability of the rate region in this theorem . the encoder is formed of two stages performed in succession . at the outcome of the first stage , which deals with the common message @xmath363 the encoder computes a sequence of @xmath180 blocks of @xmath13 bits denoted below by @xmath364 these blocks are used in the second stage to construct the data encoding that is going to be sent to both receivers . namely , it will be seen that the transformed blocks @xmath365 can at the same time encode the common message to both receivers and also encode side information for receiver 1 to ensure reliable transmission of the confidential message . the actual sequences to be transmitted are computed in the second stage based on the sequences @xmath366 this is done by first constructing sequences @xmath367 using the ideas developed in sect . [ sect : wt ] and by using a stochastic mapping of these sequences on the condewords @xmath368 upon transmitting , these codewords are received by receiver 1 as @xmath369 and by receiver 2 as @xmath370 we will argue that the receivers can independently perform decoding procedures that recover the three desired types of messages reliably ( and when appropriate , also securely ) . the proof of the fact that any @xmath333 satisfying is achievable follows from the polar coding scheme for the superposition region given in @xcite . given the rvs @xmath353 with @xmath50 and @xmath51 , let @xmath371 be @xmath13 independent repetitions of the rvs @xmath362 . set @xmath372 where @xmath95 is arikan s transform . as before , lowercase letters denote realizations of these rvs . define the sets @xmath373 , @xmath374 , @xmath375 as follows : @xmath376 : z(q_i|q^{i-1 } ) \geq 1-\delta_n\ } \\ { \mathscr l}_{u|y}&= \{i\in [ n ] : z(q_i|q^{i-1},y^{n } ) \leq \delta_n\ } \\ { \mathscr l}_{u|z}&= \{i\in [ n ] : z(q_i|q^{i-1},z^{n } ) \leq \delta_n\}.\end{aligned}\ ] ] the cardinalities of these sets , normalized by @xmath13 , approach respectively @xmath377 as @xmath211 now observe that for receiver 1 to recover @xmath378 correctly , the indices of the information bits should be a subset of @xmath379 . similarly , for receiver 2 to recover the sequence @xmath378 correctly , the information bits should be placed only in those positions of @xmath378 that are indexed by the set @xmath380 . therefore , choosing the indices of information bits as @xmath381 ensures that the message embedded into @xmath378 will be decoded correctly by both receivers . in this case , the rate of the common message is @xmath382 . given that @xmath383 we conclude the common message rate @xmath333 attains the value @xmath384 $ ] only if either @xmath385 or @xmath386 holds starting from some @xmath13 . however , generally this does not have to be the case . to overcome this problem , @xcite proposed the following coding scheme . define the sets @xmath387 without loss of generality , assume that @xmath388 , which implies that @xmath389 starting with some @xmath13 . to describe the encoding procedure , consider @xmath180 blocks of @xmath13 coordinates each . in block 1 , we use the positions indexed by @xmath390 to store message bits and assign the bits indexed by @xmath391 to some fixed values that are available to receiver 1 . in block @xmath392 we again use the positions indexed by @xmath390 to store message bits and copy the part of block @xmath393 indexed by the coordinates in @xmath390 into the positions indexed by a subset of coordinates @xmath394 in block @xmath175 ( thus @xmath395 ) . fill the remaining @xmath396 bits in each block @xmath397 with random and independent bits and communicate them to receiver 1 . these bits can be the same for each block as long as they are independent and uniform within the same block , so this part of the scheme has negligible impact on the overall rate . in the final block @xmath180 , we assign the bits indexed by @xmath390 to some fixed values that are available to receiver 2 and copy the bits in @xmath398 to the positions in @xmath399 the remaining @xmath396 coordinates in block @xmath180 are filled with random bit values . the bits in @xmath400 and @xmath401 are chosen based on deterministic rules @xmath159 similarly to sect . [ sect : wt ] . these bits are the same for each block and are shared with both receivers . the block diagram of the described coding scheme is shown by figure [ figure4 ] . ] the encoding stage described passes the sequences @xmath402 to the second stage which is responsible for actual communication . upon observing the channel outputs , both receivers perform decoding , which will be described below in sect . [ subsec : decoding ] . in this section we describe the construction of sequences @xmath30 that are sent by transmitter @xmath403 the construction relies on the sequences @xmath189 constructed by @xmath2 in the first stage . these sequences can be thought of as side information that enables receiver 1 to reconstruct the secret message . the transmission scheme we propose to achieve the rate @xmath334 that satisfies , is very similar to the scheme described for the wiretap channel problem in section [ sect : wt ] . our solution consists of choosing the indices of information bits and random bits appropriately and using a chaining scheme quite similar to the one shown in figure [ figure3 ] . let @xmath371 be as defined in section [ sect : common ] , and let @xmath404 . viewing @xmath356 as side information about @xmath61 , we define the sets @xmath405 : z(t_i|t^{i-1},u^{n } ) \geq 1-\delta_n\ } \\ { \mathscr l}_{v|u , y}&= \{i\in [ n ] : z(t_i|t^{i-1},u^{n},y^{n } ) \leq \delta_n\ } \\ { \mathscr h}_{v|u , z}&= \{i\in [ n ] : z(t_i|t^{i-1},u^{n},z^{n } ) \geq 1-\delta_n\ } \end{aligned}\ ] ] whose cardinalities , normalized by @xmath406 approach respectively the values @xmath407 as @xmath211 the intuition behind the construction presented below can be described as follows . first , note that the coordinates of @xmath408 indexed by @xmath409 can be decoded by receiver 1 , and so they can be used to send more data in addition to the common message . of these bits , the part indexed by @xmath410 can be used to transmit the confidential message . then , given that @xmath411 and @xmath412 we obtain @xmath413 this implies that the proposed scheme transmits the secret message at rates arbitrarily close to the rate given by ( provided that it also satisfies the strong security condition ) . building on this observation , we proceed to describe the coding scheme , adding some details that make the secrecy part work . define the sets , this should not cause confusion . ] @xmath414 note that the sets @xmath415 partition @xmath114 $ ] . this partition is basically the same as in ( see also figures 3 and [ fig : partition ] ) except for the fact that the high- and low - entropy subsets rely on entropy quantities that are additionally conditioned on @xmath356 . the transmission scheme that we propose is formed of multiple blocks joined in clusters of @xmath180 blocks . similarly to the wiretap coding scheme , there is a seed block shared between the transmitter and receiver 1 . the seed block consists of @xmath416 random bits . even if the set @xmath417 constitutes a nonvanishing proportion of @xmath114 $ ] , the rate of the seed @xmath418 can be made arbitrarily small by choosing @xmath180 sufficiently large . ( for example , one can set @xmath222 for some @xmath223 , and let @xmath13 tend to infinity . ) the encoding procedure is as follows . our aim is to construct @xmath180 blocks @xmath367 which will be used to form the transmitted sequences @xmath419 apart from the seed block , all the other blocks @xmath420 contain a group of almost deterministic bits , denoted by @xmath421 in . for block @xmath422 , the values of these bits are assigned according to a deterministic rule @xmath159 ( similarly to the earlier appearence of this mapping , see , e.g. , sect . [ sect : wt ] ) , namely @xmath423 for all @xmath424 . the set of frozen bits @xmath425 in each of the @xmath180 blocks is chosen similarly based on a family of deterministic rules @xmath426 these bits are going to be the same for the each block and shared with receiver 1 . note that the rate @xmath229 can be made arbitrarily small similarly to @xmath230 . the remaining subsets of coordinates in are filled as follows . for block 1 , the bits in the set @xmath417 are assigned the values of the bits of the seed block , while for block @xmath175 , @xmath188 these bits are set to be equal to the bits in @xmath427 of block @xmath233 here , @xmath234 is a subset of @xmath235 having the same size as @xmath231 . the messages are stored in the bits indexed by @xmath428 which are good for receiver 1 and contained in the bad ( high - entropy ) set of receiver 2 . the indices in the subset @xmath429 are still good for receiver 1 but bad for receiver 2 . nevertheless , they are filled with random bits that are used for decoding by receiver 1 in the same way as was done in sect . [ sect : wt ] . finally , the bits in @xmath430 are assigned randomly and uniformly for each of the @xmath180 blocks . we once again refer to fig . [ fig : partition ] which illustrates the described processing . once the blocks @xmath431 are formed , we compute @xmath180 sequences @xmath432 by using the polarizing transform . finally , the codewords to be sent by the transmitter are computed as follows . the codeword @xmath433 is sampled from @xmath434 according to the distribution @xmath244 , where @xmath245 is the conditional distribution induced by the joint distribution of the rvs @xmath61 and @xmath2 . assume that the transmitted sequence @xmath30 is received as @xmath29 by receiver 1 and as @xmath435 by receiver 2 . importantly , by our construction these sequences follow the conditional distributions @xmath22 and @xmath23 given by the channels @xmath3 and @xmath436 we describe the decoding procedures by receivers 1 and 2 . initially they perform similar operations aimed at recovering the common message . once this is accomplished , receiver 1 performs additional decoding to recover the secret message . we begin with the common - message part . in accordance with , receivers 1 and 2 decode the blocks @xmath437 relying on a iterative procedure . as the construction suggests , receiver 1 decodes in the forward direction , starting with block @xmath278 and ending with block @xmath180 , and receiver 2 decodes backwards , starting with block @xmath180 and ending with block @xmath278 . let @xmath438 denote the subblocks @xmath439 of the corresponding blocks ( see fig . [ figure4 ] ) . the processing by receiver 1 is as follows . for block @xmath278 , it computes @xmath440 for the remaining blocks @xmath188 receiver 1 computes the vector @xmath441 as follows : @xmath442 the processing by receiver 2 is quite similar except that it starts with block @xmath180 and advances `` backwards '' for @xmath443 for block @xmath180 the rule is as follows : @xmath444 for blocks @xmath445 , receiver 2 computes its estimates of the vector @xmath446 as follows : @xmath447 ( in - the notation is somewhat abbreviated to keep the formulas compact , e.g. , no reference is made to the index of the receiver , and the block index @xmath175 is sometimes omitted ) . the processing described above in - yields the sequences @xmath448 which are used by receiver 1 to recover the secret messages . denote by @xmath449 the message sequence encoded in the seed block , and let @xmath450 be the subblock of block @xmath175 indexed by the set @xmath451 for @xmath190 , the decoding rule is as follows : @xmath452 where @xmath453 is the conditional distribution induced by the joint distribution of the rvs @xmath454 , @xmath251 and @xmath252 ( the notation is again abbreviated similarly to ) . the rate of the common message achieved by the construction in sect . [ sect : common ] is equal to @xmath455.\ ] ] as @xmath13 increases , we obtain @xmath456 for sufficiently large @xmath180 this quantity is arbitrarily close to the common - message rate value given in . note that we have assumed that @xmath457 to handle the opposite case is suffices to interchange the roles of the pieces @xmath390 and @xmath391 in the common - message encoding and decoding procedures . as shown in @xcite , both receivers can decode the common message correctly with probability of error at most @xmath458 . this follows because in this stage , receivers 1 and 2 aim only at decoding the bits corresponding to the index sets @xmath459 and @xmath460 , respectively . that these bits can be recovered in a sucessive decoding procedure follows from the basic results on polar codes @xcite and @xcite . for the secret - message part of the communication the properties of the scheme are characterized by the following proposition . [ prop : sc ] for any @xmath253 , @xmath254 and @xmath45 it is possible to choose @xmath180 so that the transmission scheme described above attains a secrecy rate @xmath334 such that @xmath461 and the information leaked to receiver 2 satisfies the strong secrecy condition in definition [ bcc ] . assume that @xmath462 are as given by theorem [ main ] . the rate of proposed coding scheme is @xmath463 which converges to @xmath464 as @xmath13 goes to infinity . recalling theorem [ main ] , we note that this is the target value of the rate @xmath334 for given @xmath465 and @xmath2 satisfying @xmath466 and @xmath50 , @xmath51 . introduce the following rvs : let @xmath261 be a sequence of message bits @xmath467 sent in blocks @xmath263 and let @xmath264 be the rvs that represent the random observations of receiver 2 upon transmitting @xmath180 blocks @xmath468 further , let @xmath265 correspond to the bits contained in the subset @xmath469 for all @xmath470 _ reliability : _ the proof of reliable decoding by receiver 1 follows from the results of @xcite and is very similar to section [ sect : wt ] . let @xmath471 where @xmath472 is the joint distribution of the rvs @xmath362 appearing in theorem [ thm : bcc ] . let @xmath473 be the empirical distribution induced by the transmission scheme if the deterministic rules @xmath159 in both stages of the encoding are replaced by random assignments such that , for @xmath474 @xmath475 holds for the first and second stages , respectively . here @xmath476 is the rv defined in . then , from the proof of lemma 1 in @xcite , it follows that @xmath477 the second estimate for every @xmath478 hence , we conclude that @xmath479 holds for all the @xmath180 blocks of transmission . moreover , since the message bits are entirely contained in the set of @xmath480 , the successive decoding procedure has the probability of error bounded by latexmath:[\[\begin{aligned } for each individual block . thus , we observe that there exists a family of deterministic rules @xmath159 for each stage such that the overall error probability is at most @xmath482 we conclude that the probability that receiver 1 decodes the information bits correctly approaches @xmath278 as @xmath13 goes to infinity . _ security : _ we will show that condition is fulfilled for the sequence of @xmath180 blocks of transmission . note that receiver 2 observes not only a realization of @xmath483 , but also estimates the rvs @xmath484 through procedure - . for this reason the strong security condition to be proved takes the form @xmath485 the proof of is very similar to the proof of strong secrecy in section [ sect : wt ] . the counterparts of lemmas [ security_lemma_wiretap ] and [ security_lemma_wiretap2 ] for the bcc are provided below . the proofs are the same as the ones in section [ sect : wt ] and will be omitted . let @xmath280=\{t_i , i\in{\mathscr a}\}$ ] and @xmath281=\{t_i , i\in\ci\backslash{\mathscr a}\},$ ] where @xmath486 is any subset of @xmath235 . then @xmath282 ; t[{\mathscr a } ] , u^n , z^n)=o(n^3 2^{-n^{\beta } } ) \label{lemma_ineq_bcc}\ ] ] [ security_lemma_bcc ] let @xmath487 , @xmath488 and @xmath489 then , for all @xmath300 , we have @xmath490 [ security_lemma_bcc2 ] the rest of the proof also follows similarly to the proof in section [ sect : wt ] . repeating the inequalities which led to , we get @xmath491 which implies @xmath492 this completes the proof of . finally , let us show that the `` additional - message '' rate @xmath493 as given by is also achievable . we have seen that any rate pair @xmath360 satisfying - is achievable in the system . moreover , observe that receiver 1 decodes correctly messages at the rate of @xmath494 $ ] according to - , and additionally decodes messages at the rate of @xmath495 owing to the part of the encoding @xmath496 given by . since these two groups of information bits can be decoded simultaneously by receiver 1 , we conclude that it is possible to communicate to receiver 1 an additional message at rate @xmath493 . in this paper , we have considered the wiretap channel problem @xcite,@xcite . we proved that the secrecy capacity @xmath46 is achievable by polar codes under the strong security constraint , and without additional assumptions on the channel such as symmetry or degreadedness . we also showed that it is possible to build on this solution by adding a second layer of encoding which enables one to attain the capacity region of the broadcast channel with confidential messages introduced by csiszr and k " orner . e. hof and s. shamai , _ secrecy - achieving polar - coding _ , ieee information theory workshop , dublin , ireland , sept . 2010 , pp . j. honda and h. yamamoto , _ polar coding without alphabet extension for asymmetric models _ , ieee trans . theory * 59 * ( 2013 ) , no . 12 , 78297838 . o. o. koyluoglu and h. el gamal , _ polar coding for secure transmission and key agreement _ , ieee trans . forensics and security , * 7 * ( 2012 ) , no . 5 , pp . 14721483 . h. mahdavifar and a. vardy , _ achieving the secrecy capacity of wiretap channels using polar codes _ , ieee trans . theory , * 57 * ( 2011 ) , no . 10 , 64286443 . j. m. renes and m. m. wilde , _ polar codes for private and quantum communication over arbitrary channels _ , ieee trans . theory , * 60 * ( 2014 ) . 6 , 30903103 . e. aolu and a. vardy . _ a new polar coding scheme for strong security on wiretap channels . symposium on information theory , istanbul , turkey , july 2013 , pp.11171121 .

the problem of adaptive multivariate function estimation in the single - index regression model with random design and weak assumptions on the noise is investigated . a novel estimation procedure that adapts simultaneously to the unknown index vector and the smoothness of the link function by selecting from a family of specific kernel estimators is proposed . we establish a pointwise oracle inequality which , in its turn , is used to judge the quality of estimating the entire function ( `` global '' oracle inequality ) . both the results are applied to the problems of pointwise and global adaptive estimation over a collection of hlder and nikolskii functional classes , respectively .

1304.7668

this paper deals with multivariate functions estimation . for the proposed estimator we establish local as well as global oracle inequalities and show how to use them for deriving minimax adaptive results . _ model and setup . _ we observe @xmath0 following @xmath1 where @xmath2 , the noise @xmath3 are i.i.d . centered random variables satisfying a tail probability condition ( assumption [ assassumption - on - noise ] ) , and the design points @xmath4 are independent random vectors with common density @xmath5 with respect to the lebesgue measure . the sequences @xmath6 and @xmath7 are assumed to be independent . the density @xmath8 is known , however , in section [ secunknowndensity ] we discuss how to extend our results to the case of unknown design density . in addition , we assume that the function @xmath9 has a single - index structure , that is , there exist unknown @xmath10 and @xmath11 such that @xmath12 a minimal technical assumption about @xmath13 is that it belongs to some hlder ball , yet the knowledge of this ball will not be required for the proposed estimation procedure ; see the discussion after assumption [ asstechnical ] for more details . the paper aims at estimating the entire function @xmath14 on @xmath15 ^ 2 $ ] or its value @xmath16 , @xmath17 ^ 2 $ ] , from the data @xmath18 without any prior knowledge about the nuisance parameters @xmath19 and @xmath20 . the unit square is chosen for notation convenience ; and all the results remain true when @xmath15 ^ 2 $ ] is replaced by an arbitrary bounded interval of @xmath21 . throughout the paper we adopt the following notation . the joint distribution of the sequence @xmath22 will be denoted by @xmath23 , and those of @xmath24 by @xmath25 . in addition , @xmath26 and @xmath27 stand for the marginal distributions of @xmath28 and @xmath3 , respectively . to judge the quality of estimation , we use either the risk determined by the @xmath29 , @xmath30 , on @xmath31 ^ 2 $ ] with @xmath32 : @xmath33 a `` global '' risk ; or the `` pointwise '' risk defined as follows : @xmath34 ^ 2.\ ] ] here @xmath35 is an estimator , that is , an @xmath36-measurable function , and @xmath37 denotes the mathematical expectation with respect to @xmath38 . all the results established in the paper , except the lower bound given in theorem [ thpointwise - adaptation - lower ] , are obtained for @xmath39 . the principal difficulties with the case of arbitrary dimension are commented in remark [ remdifficulties ] . it is noteworthy that the single - index modeling , even if @xmath40 , is a direct generalization of the univariate regression model . therefore , our results , mainly presented in section [ subsecadaptive - estimation ] , generalize in several directions the existing ones obtained for the univariate random design regression ( see the discussion after theorem [ thglobal - adaptation ] ) . let us formulate the principal assumptions used in the sequel . they are imposed on the distributions of the design and noise variables as well as on the approximation property of the link function . [ assassumption - on - noise ] the random variable @xmath41 has a symmetric distribution with density @xmath42 with respect to the lebesgue measure . moreover , there exist @xmath43 , @xmath44 $ ] , and @xmath45 such that @xmath46 the assumption holds , for example , for the gaussian , laplace or , more generally , for the symmetrized weibull distribution . in the following , the functional class @xmath47 is considered as fixed . [ assassumption - on - design ] there exists @xmath48 such that @xmath49 ^ 2 } g(x ) \ge\underline{g}$ ] . the assumption holds obviously if the design points are uniformly distributed on any bounded borel set containing @xmath50 ^ 2 $ ] . the imposed condition is `` fitted '' to the estimation over @xmath51 ^ 2 $ ] that explains the set @xmath52 ^ 2 $ ] . when estimating over a rectangle @xmath53\times[c , e]\in\mathbb{r}^2 $ ] , the infimum should be taken over @xmath54\times[c-5/2,e+5/2 ] $ ] . if @xmath55 from assumption [ asstechnical ] below is known , the above condition can be relaxed to @xmath56\times[c-2,e+2 ] $ ] . we also remark that independently of the values @xmath57 assumption [ assassumption - on - design ] is fulfilled if @xmath58 and @xmath59 for any @xmath60 . [ asstechnical ] there exist @xmath61 and @xmath62 such that @xmath63 the latter assumption guarantees that the link function is smooth . however , it is important to emphasize that @xmath64 and @xmath65 are not supposed to be known a priori . in particular , they are not involved in our estimation procedure . on the other hand , both the parameters restrict the minimal sample size needed to justify the theoretical results involved . set for any @xmath66 @xmath67 in the sequel it will be assumed that @xmath68 , where @xmath69 to finish this section , we remark that all the presented results remain true if one assumes that @xmath70 , that is , is uniformly bounded , and @xmath65 is known . for clarity of presentation , it is assumed that the index vector , where @xmath71 stands for the unite sphere in @xmath72 . however , in section [ secextensions ] it is shown that our results can be extended to the case @xmath73 . the goal of our studies is at least threefold . we first seek an estimation procedure @xmath74 , @xmath75^{2 } $ ] , for @xmath76 which could be applicable to any function @xmath77 satisfying assumption ( [ single - index ] ) . moreover , we would like to bound the risk of this estimator uniformly over the set @xmath78 . more precisely , we want to establish for @xmath74 the so - called local oracle inequality at any point @xmath79^{2 } $ ] the risk of @xmath80 should be bounded as follows : @xmath81 here @xmath82 is completely determined by the function @xmath83 , vector @xmath84 and observations number @xmath85 , while @xmath86 is a constant independent of @xmath77 and @xmath87 . after being established , the local oracle inequality allows deriving minimax adaptive results for the function estimation at a given point . indeed , let @xmath88 , @xmath89 be a collection of functional classes such that @xmath90 . for any @xmath91 define @xmath92 where the infimum is taken over all possible estimators . the quantity @xmath93 is the minimax risk on @xmath94 . in the framework of minimax adaptive estimation , the task is to construct an estimator @xmath95 such that for any @xmath96 the estimator @xmath95 satisfying ( [ eqlocal - adaptation - intro ] ) is called optimally rate adaptive over the collection @xmath97 . subsequently , let ( [ eqlocal - oracle - intro ] ) be proved ; and let for any @xmath98 then one can assert that the estimator @xmath99 is adaptive over @xmath100 . thus , the first step is to prove ( [ eqlocal - oracle - intro ] ) . to the best of our knowledge , such results do not exist in the context of regression with random design not only under the single - index constraint , but also in univariate regression . next , ( [ eqlocal - oracle - intro ] ) is applied to minimax adaptive estimation over hlder classes , @xmath101 ; see section [ subsecadaptive - estimation ] for pertinent definitions . we will find the minimax rate over @xmath102 and prove that @xmath99 achieves it , that is , is optimally rate adaptive . this result is quite surprising because , if @xmath84 is fixed , say , @xmath103 , it is well known that an optimally adaptive estimator does not exist ; see @xcite for the gaussian white noise model , @xcite for density estimation , and @xcite for regression . local oracle inequality ( [ eqlocal - oracle - intro ] ) allows us to bound from above the `` global '' risk as well . indeed , for any @xmath104 , in view of jensen s inequality and fubini s theorem , @xmath105^r\le\mathbb{e}^{(n)}_{f } \| \widehat f - f \|^{r}_r= \|\mathcal r_{r,\cdot}^{(n ) } ( \widehat f , f ) \| _ r^r $ ] and , therefore , @xmath106 inequality ( [ eqglobal - oracle - intro ] ) is called the global oracle inequality , and in the considered framework it supplies new results . as local oracle inequality ( [ eqlocal - oracle - intro ] ) is a powerful tool for deriving minimax adaptive results in pointwise estimation , so inequality ( [ eqglobal - oracle - intro ] ) can be used for constructing adaptive estimators of @xmath14 . we will consider a collection of nikolskii classes @xmath107 ( see definition [ defnikolskii - class ] ) , where @xmath108 and @xmath109 . when considering these classes , we aim at estimating functions with inhomogeneous smoothness . this means that the underlying function can be very regular on some parts of its domain and rather irregular on the other sets . we will compute bounds for @xmath110 and show that , if @xmath111 , the rate of convergence is the minimax rate over @xmath112 . this means that our estimator @xmath99 is optimally rate adaptive over the collection @xmath113 whenever @xmath111 . in the case @xmath114 , we will show that the latter bound differs from the bound on the minimax risk by a logarithmic factor . following the contemporary language , we say that @xmath99 is `` nearly '' the construction of an optimally rate adaptive over the entire range of the nikolskii classes estimator under the single - index constraint ( [ single - index ] ) is an open question . all presented results are completely new . the adaptive estimation under the @xmath115 loss and single - index constraint , except the case @xmath116 in @xcite , was not studied . note , however , that the cited result was obtained under the gaussian errors model and over the hlder classes that do not admit the consideration of functions with inhomogeneous smoothness . it turns out that the adaptation to the unknown @xmath117 and @xmath118 can be viewed as selecting from a special family of kernel estimators in the spirit of that of @xcite , @xcite , @xcite . however , our selection rule is quite different from the aforementioned proposals , and it allows us to solve the problem of minimax adaptive estimation under the @xmath119 losses over a collection of nikolskii classes . it is worth mentioning that the single - index model is particularly popular in econometrics [ see , e.g. , @xcite , @xcite ] . the estimation , nevertheless , is usually performed under smoothness assumptions on the link function . one usually uses the @xmath120 losses , and the available methodology is based on these restrictions . to the best of our knowledge , the only exceptions are @xcite for the minimax estimation under the projection pursuit constraints , and @xcite for adaptation to unknown smoothness and structure . in section [ subsecoracle - approach ] we present our selection rule and establish for it local and global oracle inequalities . section [ subsecadaptive - estimation ] is devoted to the application of these results to minimax adaptive estimation . the proofs of the main results are given in section [ secproofs ] ; section [ secunknowndensity ] discusses an unknown design density , and the proofs of lemmas are moved to the supplementary material [ @xcite ] . in this section we motivate and explain our procedure and prove the local and global oracle inequalities . then we apply these results to adaptive estimation over a collection of hlder classes ( pointwise estimation ) and over a collection of nikolskii classes ( estimating the entire function with the accuracy of an estimator measured under the @xmath115 risk ) . let @xmath121 be a function ( kernel ) satisfying . with any such @xmath122 , any @xmath123 , @xmath124 $ ] and any @xmath125 , we associate the quantity @xmath126\,\mathrm{d}u\biggr{\vert}.\ ] ] note that the kernel smoother @xmath127/\delta ) f(u)\,\mathrm{d}u$ ] can be understood as an approximation of the function @xmath83 at the point @xmath128 . thus , @xmath129 is a monotonous approximation error provided by this kernel smoother . in particular , under assumption [ asstechnical ] , we have @xmath130 as @xmath131 . in what follows , @xmath132 , denotes the @xmath133 norm of @xmath122 and we will assume that the kernel @xmath122 satisfies the following condition . [ assassumption - on - kernel ] ( 1 ) @xmath134 $ ] , @xmath135 , @xmath122 is symmetric ; for any @xmath138 , denote the hardy littlewood maximal function of @xmath139 [ see , e.g. , @xcite ] by @xmath140 clearly , @xmath141 for any @xmath142 . now , let us define the oracle estimator . for any @xmath138 and @xmath143 defined in ( [ eqbandwidts ] ) , set @xmath144 | \sqrt{nh } \delta^{*}_{\mathcal{k},f}(h , y)\le \| \mathcal{k } \|_{\infty } \sqrt{\ln(n ) } \bigr\}.\ ] ] note that @xmath145 for any @xmath125 and any @xmath146 . hence , @xmath147 for any @xmath68 in view of ( [ eqrestriction - on - sample - size ] ) . next , assumption [ assassumption - on - kernel](2 ) implies that @xmath148 is continuous , hence , @xmath149.\end{aligned}\ ] ] here we have also used that @xmath150 in view of assumption [ assassumption - on - kernel](1 ) . the quantity similar to @xmath151 first appeared in @xcite for estimating univariate functions with inhomogeneous smoothness . some years later , this idea was further developed for multivariate function estimation ; see @xcite , @xcite and the more detailed discussion of the oracle approach therein . following their lead , we advance it for the estimation under the single - index constraint . the basic idea behind our selection rule is simple . for any @xmath152 $ ] , define the matrix @xmath153 and consider the family of kernel estimators with @xmath154 so that @xmath155 \biggr\}.\ ] ] we remark that assumptions [ assassumption - on - design ] and [ assassumption - on - kernel](1 ) assure well - definiteness of @xmath156 because @xmath157 @xmath158 ^ 2 $ ] and @xmath159 ^ 2 $ ] . the choice @xmath160 and @xmath161 leads to the _ oracle estimator _ @xmath162 . note that @xmath163 is not an estimator in the usual sense because it depends on the function @xmath77 to be estimated [ more precisely , on @xmath164 which determines @xmath77 ] . the meaning of @xmath166 is explained by the following result based on the straightforward application of rozenthal s inequality . [ proprisk - of - oracle - estimator ] for any @xmath167 , @xmath168 and @xmath68 , @xmath169^{-1/2}\qquad \forall t\in[-1/2,1/2]^{2 } , \ ] ] where @xmath170 is a numerical constant independent of @xmath87 . this result indicates that the `` oracle '' knows the exact value of @xmath84 and the optimal , up to @xmath171 , trade - off @xmath172 between the approximation error induced by @xmath173 and the stochastic error of the kernel estimator from @xmath174 with bandwidth @xmath172 . it explains why the `` oracle '' chooses the `` estimator '' @xmath175 . below we propose a `` real , '' based on the observation , estimator @xmath99 , which mimics the oracle for any @xmath176 , @xmath177 and @xmath68 , @xmath178^{-1/2}\qquad\forall t \in[-1/2,1/2]^{2},\ ] ] where @xmath179 is an absolute constant independent of @xmath87 and the underlying function @xmath77 . the latter result is a local oracle inequality . the construction of the estimator @xmath99 is based on the data - driven selection from the family @xmath174 . for any @xmath180 and any @xmath181 $ ] , define @xmath182 where @xmath183 a kernel estimator associated with the matrix @xmath184 is defined by @xmath185 the definition of @xmath186 is legitimate because @xmath187 @xmath188 ^ 2 $ ] and @xmath189 ^ 2 $ ] . for any @xmath190 , set @xmath191 and define @xmath192 where @xmath193 ^ 2 } |\widehat { f}(v ) | $ ] and @xmath194 is defined in ( [ eqbandwidts ] ) . put also @xmath195(\eta n)^{-1/2},\qquad \eta\in(0,1].\ ] ] the quantities @xmath196 , @xmath197 and @xmath198 are listed in section [ secconstants ] . set @xmath199 $ ] and let for any @xmath200 and @xmath201 , @xmath202_+ , \\ r^{(2)}_{t}(h)&=&\sup_{\eta\in\mathcal h_n\dvtx \eta\le h } \bigl[\sup _ { \theta\in\mathbb{s}^{1 } } \bigl|\widehat f_{(\theta , h)}(t)- \widehat f_{(\theta , \eta ) } ( t ) \bigr|-\operatorname{th}(\eta ) \bigr]_+.\end{aligned}\ ] ] subsequently , define @xmath203 as a solution of the following minimization problem : @xmath204\\[-8pt ] & & \qquad = \inf_{(\theta , h)\in\mathbb{s}^{1}\times\mathcal h_n } \bigl[r^{(1)}_{t}(\theta , h)+r^{(2)}_{t } ( h)+\operatorname{th}(h ) \bigr].\nonumber\end{aligned}\ ] ] then our final estimator is @xmath205 , where @xmath206 is obtained by minimizing ( [ eqselection - rule ] ) . [ remmeasurability ] we note that assumption [ assassumption - on - kernel](2 ) guarantees that all random fields involved in the description of selection rule ( [ eqselection - rule ] ) are continuous on @xmath207 . moreover , the set @xmath208 is finite . thus , @xmath203 is @xmath209-measurable and @xmath210 [ see @xcite ] . [ remdifficulties ] our selection rule ( [ eqselection - rule ] ) is defined in the case @xmath40 . the main difficulty in extending it to @xmath211 consists in the construction of the matrix @xmath212 for any vectors @xmath213 . indeed , analyzing the proof of theorem [ thlocal - oracle - inequality ] , we remark that the following properties should be fulfilled : @xmath214 where the class of matrices @xmath215 is defined in ( [ eqdefmatrixclass ] ) . if @xmath40 , these requirements hold . however , we were not able to construct a class of matrices obeying latter restrictions in the dimension strictly larger than 2 . note , nevertheless , that if such a class would be found , our results could be extended to @xmath211 without any additional consideration . we reinforce restriction ( [ eqrestriction - on - sample - size ] ) on the minimal sample size @xmath87 . let @xmath216 be defined as follows : @xmath217 where @xmath194 is defined in ( [ eqbandwidts ] ) and @xmath218 is given at the beginning of section [ secconstants ] . all our results below will be proved under the condition @xmath219 . first , we note that @xmath220 is well - defined since @xmath221 as @xmath222 . next , contrary to restriction ( [ eqrestriction - on - sample - size ] ) that relates the sample size @xmath87 to the quantities @xmath223 and @xmath55 from assumption [ asstechnical ] , restriction ( [ eqrestriction2-on - sample - size ] ) links the minimal value of @xmath87 with the quantity @xmath224 appearing in assumption [ assassumption - on - design ] . [ thlocal - oracle - inequality ] for any @xmath225 , @xmath226 and @xmath219 , @xmath227^{1/2 } + c_{2}n^{-1/2}\qquad\forall t\in[-1/2,1/2]^{2}.\ ] ] the constants @xmath228 and @xmath229 are independent of @xmath87 and @xmath77 and their explicit expressions can be extracted from the proof of the theorem . as already mentioned , the global oracle inequality is obtained by integrating the local oracle inequality . indeed , for any @xmath104 , using jensen s inequality and fubini s theorem , we have @xmath230 so @xmath231^{2 } } \biggl[\frac { \ln(n)}{nh^*_{\mathcal{k},f}(t^{\top}\theta^ * ) } \biggr]^{r/2}\,\mathrm{d}t \biggr\}^{1/r } + c_{2}n^{-1/2}.\ ] ] integration by substitution yields @xmath232^{2 } } \biggl[\frac { \ln(n)}{nh^*_{\mathcal{k},f}(t^{\top}\theta^ * ) } \biggr]^{r/2}\,\mathrm{d}t \le \int_{-1/2}^{1/2 } \biggl [ \frac { \ln(n)}{nh^*_{\mathcal{k},f}(z ) } \biggr]^{r/2}\,\mathrm{d}z,\ ] ] that leads to the following bound . [ thglobal - oracle - inequality ] for any @xmath176 , @xmath177 and @xmath219 , @xmath233 define @xmath234 , @xmath235 and let @xmath236 . obviously , for all @xmath237 we have@xmath238 that implies @xmath239 so the estimation of @xmath77 is equivalent to the estimation of @xmath240 . because @xmath241 , theorems [ thlocal - oracle - inequality ] and [ thglobal - oracle - inequality ] are applicable . to this end , it suffices to replace @xmath83 by @xmath242 in the definition of @xmath243 . in general , however , there is no universal way of expressing @xmath244 via @xmath245 , although in particular cases , mainly in adaptive estimation over classes of smooth functions , it is often possible . in this section we first apply the local oracle inequality given in theorem [ thlocal - oracle - inequality ] to the problem of pointwise adaptive estimation over a collection of h " older classes . next , we study adaptive estimation under the @xmath115 losses over a collection of nikolskii classes . the corresponding result is deduced from the global oracle inequality proved in theorem [ thglobal - oracle - inequality ] . assume throughout this section that the kernel @xmath122 obeys additionally assumption [ ass2assumption - on - kernel ] below ; we then introduce the following notation : for any @xmath246 , let @xmath247 be the maximal integer strictly less than @xmath248 . [ ass2assumption - on - kernel ] there exists @xmath249 such that @xmath250 we start with some definitions . [ defholder - class ] let @xmath251 and @xmath252 . a function @xmath253 belongs to the hlder class @xmath254 if @xmath255 is @xmath256-times continuously differentiable , @xmath257 for all @xmath258 , and @xmath259 the aim is to estimate the function @xmath260 at a given point @xmath261 ^ 2 $ ] under the assumption that @xmath262 , where @xmath263 the constant @xmath264 is from assumption [ ass2assumption - on - kernel ] , and @xmath2 is the dimension . we will see that @xmath264 can be an arbitrary number but it must be chosen a priori . [ thpointwise - adaptation ] let @xmath249 be fixed ; and let additionally assumptions [ assassumption - on - kernel ] and [ ass2assumption - on - kernel ] hold . then , for any @xmath265 , @xmath266 , @xmath168 and @xmath75^{2}$ ] , @xmath267 where @xmath268^{\beta/(2\beta+1)}$ ] and @xmath269 is independent of @xmath87 . the proof of the theorem is based on the evaluation of the uniform over @xmath270 lower bound for @xmath271 and on the application of theorem [ thlocal - oracle - inequality ] . we note that a similar upper bound for the minimax risk appeared in @xcite in the framework of gaussian white noise model , but the estimation procedure used there is different from our selection rule . the main question , however , is if @xmath272 coincides with the minimax rate for any given value of @xmath273 and @xmath274 ? to answer it , we need some additional assumptions on the densities of the noise variable @xmath41 and design variable @xmath275 . [ ass2on - noise ] there exist @xmath276 such that , for any @xmath277 $ ] , @xmath278 it is easy to see that the density of the normal law @xmath279 , @xmath280 , obeys the aforementioned assumption . in general , this assumption is fulfilled if the density @xmath42 is regular and decreases rapidly at infinity . more precisely , if the fisher information corresponding to the density @xmath42 is finite and the function @xmath281 ^ 2p^{-1}(y ) \,\mathrm{d}y$ ] is continuous at zero , assumption [ ass2on - noise ] is verified . [ ass2on - design ] there exist @xmath282 and @xmath283 such that , for all @xmath284 , @xmath285 . here @xmath286 is the euclidean vector norm on @xmath287 . we remark that the imposed assumption is very weak and holds for the majority of probability distributions used in statistical applications . [ thpointwise - adaptation - lower ] let assumptions [ ass2on - noise ] and [ ass2on - design ] be fulfilled . then , for any @xmath288^d$ ] , @xmath289 , @xmath168 , @xmath290 , and any @xmath66 large enough , @xmath291 where the infimum is over all possible estimators . here @xmath292 is a numerical constant independent of @xmath87 and @xmath274 , and @xmath293 is defined in theorem [ thpointwise - adaptation ] . to the best of our knowledge , this lower bound is new . it is worth mentioning that assumption [ ass2on - noise ] is close to being necessary . one can give examples where this condition does not hold and theorem [ thpointwise - adaptation - lower ] is not true anymore . theorems [ thpointwise - adaptation ] and [ thpointwise - adaptation - lower ] indicate that the estimator @xmath294 is minimax adaptive with respect to the collection @xmath295 . as already mentioned , this result is quite surprising . indeed , if , for example , @xmath296 , that is , is known , then @xmath297 , and the considered estimation problem reduces to estimation of @xmath83 at a point in the univariate regression model . as it is shown in @xcite , an adaptive estimator over @xmath298 does not exist and a price for adaption appears . the latter means that the asymptotic bound on the minimax risk provided by the adaptive estimator differs from the minimax rate of convergence by some factor . this factor for the majority of known results is @xmath171 . in addition , we would like to note that the assertion of theorem [ thpointwise - adaptation - lower ] is proved for arbitrary dimension . we begin by defining the relevant functional classes . [ defnikolskii - class ] let @xmath299 , @xmath300 and @xmath301 be fixed . a function @xmath253 belongs to the nikolskii class @xmath302 if @xmath255 is @xmath256-times continuously differentiable and @xmath303 it is also assumed that @xmath304 if @xmath305 . here , the target of estimation is the entire function @xmath77 under the assumption that @xmath306 , where @xmath307 let us briefly discuss the applicability of theorem [ thglobal - oracle - inequality ] requiring @xmath125 . to this end , we assume that @xmath308 . the latter assumption is standard for estimating functions with inhomogeneous smoothness [ see , e.g. , @xcite , @xcite , @xcite ] . if @xmath308 , the embedding @xmath309 with an absolute constant @xmath170 guarantees that @xmath125 with @xmath310 and @xmath311 . [ thglobal - adaptation ] let @xmath249 be fixed , and let assumptions [ assassumption - on - kernel ] and [ ass2assumption - on - kernel ] hold . then , for any @xmath266 , @xmath312 , @xmath313 and @xmath177 , @xmath314 where @xmath315 is independent of @xmath87 , and @xmath316 note that @xmath317 . indeed , the class @xmath318 can be viewed as a class of functions @xmath77 satisfying @xmath319 with @xmath296 . then , the problem of estimating such ( 2-variate ) functions reduces to the estimation of univariate regression functions . there are at least two observations arising in view of the latter remark . first , the upper bound of theorem [ thglobal - adaptation ] generalizes the results for the univariate regression [ @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ] in several directions . in particular , the majority of the papers treat the gaussian errors or the errors having exponential moment . an exception is @xcite , where some results are obtained under a very weak assumption on the noise ( weaker than our assumption [ assassumption - on - noise ] ) . nevertheless , these results are available only if @xmath320 . next , the rate of convergence for the latter problem , which can be found in @xcite , is also the lower bound for the minimax risk defined on @xmath321 . with the proviso that @xmath322 , the rate of convergence is given by @xmath323 the minimax rate of convergence in the case @xmath324 is not known , hence , the rate presented in the middle line above is only the lower asymptotic bound for the minimax risk . thus , the proposed estimator @xmath325 is adaptive whenever @xmath111 . in the case @xmath326 , we loose only a logarithmic factor with respect to the optimal rate and , as mentioned in the , the construction of an adaptive estimator over the collection @xmath327 in this case remains an open problem . in view of the latter remark , we conjecture that the presented lower bound is correct and , therefore , the upper bound result has to be improved . we now list the quantities that are involved in the description of the selection rule that led to the adaptive estimator @xmath328 . let @xmath329 . set @xmath330 \bigr)}+ 8r\ln(n)+394 , \\ c_2(n ) & = & 730\ln { \bigl(16n^2\tau\underline{g}^{-1/2 } [ 12q + \sqrt{2 } ] \bigr)}+ 8r\ln(n)+394 , \\ c_3(n)&=&365\ln { \bigl(5n^{2}q\underline{g}^{-1/2 } \bigr)}+8r\ln(n)+197 , \\ c_4(n)&=&365\ln { \bigl(5n^2\tauq \underline{g}^{-1/2 } \bigr)}+8r\ln(n)+197.\end{aligned}\ ] ] with @xmath331 given in ( [ eqbandwidts ] ) and @xmath332 , we define @xmath333 in spite of the cumbersome expressions , it is easy to see that @xmath334 to begin with we present upper bounds for the approximation errors of the estimators involved ( lemma [ lemboundsforbias ] ) and their stochastic errors ( lemma [ lemgauss - on - matrices ] ) . lemma [ leminegality - for - sup - norm ] allows us to proceed without knowledge of @xmath65 from assumption [ asstechnical ] . the proofs of the later two results are essentially based on proposition 1 of @xcite . the detailed proofs of these technical results are moved to the supplementary material [ @xcite ] . for any @xmath180 and @xmath335 $ ] , denote @xmath336 for ease of notation , we write @xmath337 . [ lemboundsforbias ] grant assumption [ assassumption - on - kernel ] . then , for any @xmath338 and any bandwidths @xmath339 $ ] satisfying @xmath340 , one has @xmath341 let @xmath342 with @xmath343 , a\ge1 $ ] , be a set of @xmath344 matrices satisfying @xmath345 here @xmath346 denotes the matrix @xmath347 norm . set , @xmath348 , @xmath349 and consider the following random fields defined on @xmath215 : @xmath350 \bigr\ } , \\ \xi_{n , t}(e ) & = & n^{-1/2}\sum_{i=1}^n j(x_i , e ) \varepsilon_i.\end{aligned}\ ] ] denote finally by @xmath351 the set of matrices @xmath215 with @xmath352 and @xmath353 . in what follows , we denote by @xmath354 ^ 2 } |f(x)| $ ] . [ lemgauss - on - matrices ] grant assumptions [ assassumption - on - noise][assassumption - on - kernel ] . then , for any @xmath355 and any @xmath177 , @xmath356\ge c_1(n)\|f\|_\infty+c_2(n ) \bigr\}\le(8 + \upsilon)n^{-4r}.\ ] ] the expressions for @xmath357 and @xmath358 are given in section [ secconstants ] . [ leminegality - for - sup - norm ] grant assumptions [ assassumption - on - noise][assassumption - on - kernel ] . then , for any @xmath359 , @xmath360 \bigr\}\le(8+\upsilon)n^{-4r}.\ ] ] the numbers @xmath361 are defined in ( [ eqrestriction - on - sample - size ] ) and @xmath362 is defined in section [ secconstants ] . in view of jensen s inequality , an upper bound for @xmath363 , @xmath364 , will suffice to complete the proof . let @xmath365 be a bandwidth such that @xmath366 . introduce the following random events : @xmath367 \bigr\}\ ] ] and let @xmath368 and @xmath369 denote the events complimentary to @xmath370 and @xmath371 , respectively . the proof is split into three steps . first , the following inclusion holds : @xmath373 indeed , the definition of the couple @xmath203 yields @xmath374 it remains to note that the mapping @xmath375 is decreasing so inclusion ( [ eq00proof - of - theorem - local ] ) follows . next , the triangle inequality yields @xmath376 we have in view of ( [ eq00proof - of - theorem - local ] ) and the definition of @xmath377 that @xmath378.\ ] ] the definition of @xmath379 implies that @xmath380 \nonumber \nonumber\\[-8pt]\\[-8pt ] & = & \mathrm{1}_{\mathcal a}\operatorname{th}\bigl(h^*\bigr).\nonumber\end{aligned}\ ] ] note that @xmath381 , for any @xmath382 and @xmath383 . hence , @xmath384 because @xmath122 is symmetric . the latter observation , inclusion ( [ eq00proof - of - theorem - local ] ) and the definition of @xmath385 yield @xmath386\\[-8pt ] & \le&\mathrm{1}_{\mathcal a } \bigl[r^{(1)}_{t } ( \hat{\theta } , \hat { h})+\operatorname{th}\bigl(h^*\bigr ) \bigr].\nonumber\end{aligned}\ ] ] from ( [ eq0proof - of - theorem - local ] ) , ( [ eq1proof - of - theorem - local ] ) , ( [ eq01proof - of - theorem - local ] ) and ( [ eq2proof - of - theorem - local ] ) , we obtain that @xmath387 + 3\operatorname{th}\bigl(h^*\bigr ) \\ & & { } + \bigl{\vert}\widehat f_{(\theta ^*,h^*)}(t)-f(t)\bigr{\vert}. \ ] ] in addition , the definition of @xmath388 guarantees that @xmath389 we then obtain @xmath390 note also that , for any @xmath391 , @xmath392(\eta n)^{-1/2 } \\ & \le & c_6\sqrt{(\eta n)^{-1}\ln(n)},\end{aligned}\ ] ] where @xmath393 and @xmath394 and @xmath395 are defined in ( [ eqasympt - bound - for - quantities ] ) . because @xmath396 , this bound and ( [ eq02proof - of - theorem - local ] ) yield @xmath397 note that @xmath398 , for any @xmath180 , @xmath181 $ ] . set @xmath399 the following `` approximation@xmath400stochastic part '' decomposition of @xmath401 will be useful in the sequel : @xmath402\\[-8pt ] & & { } + \sqrt{n^{-1}\det(e ) } \bigl[\eta_{n , t}(e)+\xi_{n , t}(e ) \bigr],\nonumber\end{aligned}\ ] ] where @xmath403 and @xmath404 are defined before the statement of lemma [ lemgauss - on - matrices ] . hence @xmath405 taking into account that @xmath406 in view of the definition of @xmath172 and using the third assertion of lemma [ lemboundsforbias ] , we obtain @xmath407.\end{aligned}\ ] ] applying the rosenthal inequality to @xmath408 which is a sum of centered independent random variables , from ( [ eq020proof - of - theorem - local ] ) we obtain @xmath409 where @xmath410 is independent of @xmath77 and @xmath87 . because @xmath142 and @xmath411 , we have @xmath412 hence , in view of the rosenthal inequality , we obtain @xmath413^{1/(2r ) } \le\tilde{c}_1 n,\ ] ] where @xmath414 is independent of @xmath77 and @xmath87 . the use of the cauchy schwarz inequality together with the statement of lemma [ leminegality - for - sup - norm ] leads to the following bound : @xmath415^{1/(2r ) } \nonumber\\[-8pt]\\[-8pt ] & \le & \tilde{c}_1(8+\upsilon)^{1/(2r)}n^{-1}.\nonumber\end{aligned}\ ] ] we note that @xmath417 first , let us bound from above @xmath418 . we have @xmath419 and , therefore , @xmath420 thus , denoting by @xmath421 , we obtain , for any @xmath422 , @xmath423 here we have also used that @xmath424 . note also that @xmath425 and , therefore , we obtain from ( [ eq10proof - of - theorem - local ] ) , for any @xmath426 satisfying @xmath427 , @xmath428 in view of lemma [ lemgauss - on - matrices ] . this bound and ( [ eq9proof - of - theorem - local ] ) yield @xmath429 now , let us bound from above @xmath430 . we have @xmath431 and , hence , @xmath432 similar to estimate ( [ eq10proof - of - theorem - local ] ) , with the use of ( [ eq03proof - of - theorem - local ] ) and the second assertion of lemma [ lemboundsforbias ] , we obtain , for any @xmath426 satisfying @xmath427 , that @xmath433 for any @xmath426 satisfying @xmath427 , bound ( [ eq010proof - of - theorem - local ] ) and lemma [ lemgauss - on - matrices ] yield @xmath434 together with ( [ eq12proof - of - theorem - local ] ) , the latter bound gives @xmath435 thus , we obtain from ( [ eq11proof - of - theorem - local ] ) and ( [ eq15proof - of - theorem - local ] ) that @xmath436 subsequently , this bound and ( [ eq08 - 1proof - of - theorem - local ] ) yield @xmath437^{1/(2r)}\le\tilde{c}_2n^{-1/2},\ ] ] where @xmath438 is independent of @xmath77 and @xmath87 . the assertion of the theorem follows from ( [ eq08proof - of - theorem - local ] ) , ( [ eq8proof - of - theorem - local ] ) and ( [ eq16proof - of - theorem - local ] ) . using the standard computation of the bias of kernel estimators , under assumptions [ assassumption - on - kernel ] and [ ass2assumption - on - kernel ] , we get , for any @xmath439 and any @xmath123 , @xmath440 since the right - hand side of the latter inequality is independent of @xmath128 , we have @xmath441 . this implies @xmath442 , for any @xmath123 , so the assertion of the theorem follows from theorem [ thlocal - oracle - inequality ] . we start this section with an auxiliary result used in the proof of the second assertion of the theorem . it was established in @xcite , corollary 2 of proposition 5 , and , for convenience , we formulate it as lemma [ lemklp - result ] below . the result cited below concerns a lower bound for estimators of an arbitrary mapping in the framework of an abstract statistical model . we do not present it in full generality and below a version reduced to the estimation at a given point is provided . let @xmath443 be a nonempty class of functions ; and let @xmath444 be an unknown function from model defined in ( [ model - regression])([single - index ] ) . the aim is to estimate the functional @xmath445 , @xmath17^d$ ] . introduce the following notation . for any given @xmath446 , set @xmath447.\ ] ] [ lemklp - result ] assume that , for any sufficiently large @xmath448 , there exist a positive integer @xmath449 , @xmath450 and functions @xmath451 such that @xmath452 then , for @xmath453 and any @xmath454^d$ ] , @xmath455 \lambda_n.\ ] ] the proof is based on the construction of @xmath456 satisfying conditions ( [ eqass1-klp - lemma])([eqass2-klp - lemma ] ) of lemma [ lemklp - result ] . first , we construct @xmath457 and verify ( [ eqass1-klp - lemma ] ) . let @xmath458 be a function such that @xmath459 , @xmath460 and @xmath461 . set @xmath462 , where @xmath463 will be chosen later , and define @xmath464 for @xmath465 , put @xmath466 assuming without loss of generality that @xmath467 is an integer . the value of @xmath468 will be determined later in order to satisfy ( [ eqass2-klp - lemma ] ) . let @xmath469 be defined as follows : @xmath470 finally , we set @xmath471 obviously , @xmath13 defined by ( [ linkfunctionforhypothesis ] ) belongs to @xmath472 , so all @xmath473 are in the class @xmath474 . moreover , for any @xmath475 , @xmath476 we see that ( [ eqass1-klp - lemma ] ) holds with @xmath477 . it is noteworthy that @xmath478^{2 } \\ & & \qquad= \frac{1}{n^2_n}\sum_{j=1}^{n_{n } } \mathbb { e}^{(n)}_{f_0 } \bigl[z^{2 } ( f_j , f_0 ) \bigr ] + \frac{1}{n^2_n}\mathop{\sum _ { j , k=1,}}_{j\neq k}^{n_{n } } \mathbb{e}^{(n)}_{f_0 } \bigl[z ( f_j , f_0 ) z ( f_k , f_0 ) \bigr].\end{aligned}\ ] ] it follows that @xmath479 & = & \biggl\{\int_{\mathbb{r}^{d+1 } } \frac{p^{2 } ( y - f_j(x ) ) } { p(y)}g(x)\,\mathrm{d}x\,\mathrm{d}y \biggr\}^n , \\ \mathbb{e}^{(n)}_{f_0 } \bigl[z ( f_j , f_0 ) z ( f_k , f_0 ) \bigr ] & = & \biggl\{\int_{\mathbb{r}^{d+1 } } \frac{p ( y - f_j(x ) ) p ( y - f_k(x ) ) } { p(y)}g(x)\,\mathrm{d}x\,\mathrm{d}y \biggr\}^n.\end{aligned}\ ] ] because @xmath480 , we have in view of assumptions [ ass2on - noise ] and [ ass2on - design ] , for all @xmath87 large enough , @xmath481g(x)\,\mathrm{d}x \,\mathrm{d}y\nonumber \\ & & \qquad \le 1+\mathfrak{q}\int_{\mathbb{r}^d}f^{2}_j(x)g(x ) \,\mathrm{d}x \label{eq01regression - lower } \\ & & \qquad \le 1+\mathfrak{q}\mathfrak{g}\int _ { \mathbb{r}^d}f^{2}_j(x ) \bigl(1+|x|_2^{\varpi } \bigr)^{-1}\,\mathrm{d}x,\nonumber \\ & & \int_{\mathbb{r}^{d+1 } } \biggl[\frac{p ( y - f_j(x ) ) p ( y - f_k(x ) ) } { p(y ) } \biggr]g(x ) \,\mathrm{d}x\,\mathrm{d}y\nonumber \\ & & \qquad \le 1 + \mathfrak{q}\int_{\mathbb{r}^d } \bigl|f_j(x)f_k(x ) \bigr|g(x)\,\mathrm{d}x\label{eq02regression - lower } \\ & & \qquad \le 1 + \mathfrak{q}\mathfrak{g}\int _ { \mathbb{r}^d } \bigl|f_j(x)f_k(x ) \bigr| \bigl(1+|x|_2^{\varpi } \bigr)^{-1}\,\mathrm{d}x.\nonumber\end{aligned}\ ] ] set @xmath482 and @xmath483 . denote for all @xmath484 by @xmath485 the orthogonal matrix @xmath486 , where @xmath487 , @xmath488 , are the canonical basis vectors in @xmath287 . integration by substitution with @xmath489 gives @xmath490 \bigl(1+| v|_2^{\varpi } \bigr)^{-1 } \,\mathrm{d}v \nonumber \\ & \le & c_{\varpi}l^2\|w\|^2_2 h^{2\beta+1}=\mathfrak{a}c_{\varpi } \|w\| ^2_2n^{-1 } \ln(n),\end{aligned}\ ] ] where we have denoted @xmath491 and @xmath492 . for @xmath85 sufficiently large , this bound , together with ( [ eq01regression - lower ] ) , leads to @xmath493 \le n^{\mathfrak{a } \mathfrak{q}\mathfrak{g}c_{\varpi } \|w\|^2_2}.\ ] ] for any @xmath494 , set @xmath495 . by changing of variables with @xmath496 , we have @xmath497w [ h^{-1}(v_2 -\vartheta_k^{\top}t ) ] | } { 1 + |\theta^{-1}_{j , k}v |_2^{\varpi}}\,\mathrm{d}v \\ & & \qquad \le \bigl|\det(\theta_{j , k})\bigr|^{-1 } c_{\varpi}l^2 h^{2\beta+2}\|w\|^2_1 , \ ] ] where @xmath498 and @xmath499 . note that @xmath500 for sufficiently large @xmath87 . we obtain @xmath501 hence , choosing @xmath502 , we obtain , for all @xmath87 large enough , that @xmath503 we have in view of ( [ eq02regression - lower ] ) and ( [ eq1proof - the - local - adaptive ] ) @xmath504 and , hence , ( [ eq2regression - lower ] ) and ( [ eq001proof - the - local - adaptive ] ) give @xmath505 choosing @xmath506 , we see that ( [ eqass2-klp - lemma ] ) holds with the constant @xmath507 . since @xmath508 appearing in ( [ eqass2-klp - lemma ] ) is chosen independently of @xmath274 , the assertion of the theorem follows from lemma [ lemklp - result ] . in the proof we exploit the ideas from @xcite . moreover , our considerations are , to a great degree , based on the technical result of lemma [ lemlp - norm - of - bais ] below . its proof is moved to the supplementary material [ @xcite ] . [ lemlp - norm - of - bais ] grant assumptions [ assassumption - on - kernel ] and [ ass2assumption - on - kernel ] . then , for any @xmath509 , @xmath510 and @xmath511 , we have @xmath512^{-1/\mathfrak{p}}\qquad \forall h>0.\ ] ] here @xmath513 is a dependent only of @xmath514 constant from the @xmath515-strong maximal inequality ; see @xcite for more details . proof of theorem [ thglobal - adaptation ] it is sufficient to prove the theorem in the case @xmath516 only . indeed , let us recall that the risk @xmath517 is described by the @xmath119 norm on @xmath518 $ ] , therefore , @xmath519 hence , the case @xmath520 can be reduced to the case @xmath521 . in view of theorem [ thglobal - oracle - inequality ] , in order to obtain the assertion of the theorem , it suffices to bound from above @xmath522 . set @xmath523\dvtx h^*_{\mathcal { k},f}(y)=1 \ } $ ] and @xmath524\dvtx\break h^*_{\mathcal{k},f}(y ) \in(2^{-k } , 2^{-k+1 } ] \cap [ h_{\min},1 ] \}$ ] , for @xmath525 in what follows , the integration over the empty set is supposed to be zero . we have @xmath526 for simplicity of notation , we denote by @xmath527 , @xmath528 , constants independent of @xmath87 , @xmath83 and @xmath274 . the definition of @xmath529 implies @xmath530^{r/2}.\ ] ] we have in view of ( [ eq1deforaclebandwidth ] ) , for any @xmath531 , @xmath532^{1/2}\qquad\forall y\in \gamma_k.\ ] ] let @xmath533 be a sequence whose choice will be done later . we obtain from ( [ eq3proof - theorem - global - adaptation ] ) @xmath534 to get the first inequality , we have used that @xmath535 is a monotonically increasing function . the computation of the quantity on the right - hand side of ( [ eq4proof - theorem - global - adaptation ] ) , including the choice of @xmath536 , will be done differently in dependence on @xmath537 and @xmath538 . _ case _ @xmath539 . put @xmath540^{1/(2\beta+1 ) } $ ] and choose @xmath541 if @xmath427 and @xmath542 if @xmath543 . by applying lemma [ lemlp - norm - of - bais ] with @xmath544 , @xmath545 and @xmath546 , we get @xmath547\\[-8pt ] & \le & \bar{c}_5 \biggl[l^{p } \bigl(n^{-1 } \ln(n ) \bigr)^{(r - p)/2}\sum _ { k\dvtx 2^{-k}\le h^*}2^{-k[\beta p-(r - p)/2 ] } + \biggl(\frac{\ln(n)}{nh^*}\biggr)^{r/2 } \biggr].\nonumber\end{aligned}\ ] ] because in the considered case @xmath548 , we obtain @xmath549.\ ] ] it remains to note that @xmath172 is chosen by balancing two terms on the right - hand side of the latter inequality . it yields @xmath550 the argument in the case @xmath551 is completed with the use of theorem [ thglobal - oracle - inequality ] , ( [ eq2proof - theorem - global - adaptation ] ) and ( [ eq5proof - theorem - global - adaptation ] ) . _ case _ @xmath324 . put @xmath552 and choose @xmath541 for all @xmath531 . repeating the computations that led to ( [ eq55proof - theorem - global - adaptation ] ) , we get @xmath553 here we have used that @xmath554 and that the summation in ( [ eq4proof - theorem - global - adaptation ] ) runs over @xmath426 such that @xmath555 , since otherwise @xmath556 . it remains to note that the equality @xmath324 is equivalent to @xmath557 and @xmath558 . the assertion of the theorem in the case @xmath324 follows now from theorem [ thglobal - oracle - inequality ] , ( [ eq2proof - theorem - global - adaptation ] ) and ( [ eq6proof - theorem - global - adaptation ] ) . _ case _ @xmath111 . set @xmath559 if @xmath427 and @xmath541 if @xmath543 , where @xmath172 will be chosen later . the following embedding holds [ see page 62 in @xcite ] : @xmath560 . thus , by applying lemma [ lemlp - norm - of - bais ] with @xmath561 , @xmath562 and @xmath563 , we obtain @xmath564\\[-8pt ] & = & \sum_{k\dvtx 2^{-k}\le h^*}\int \bigl ( \delta^*_{\mathcal{k},f } \bigl(2^{1-k},y \bigr ) \bigr)^{r}\,\mathrm{d}y \le \bar{c}_9 l^{r}\bigl(h^*\bigr)^{\beta r - r / p + 1}.\nonumber\end{aligned}\ ] ] by applying the same lemma with @xmath561 , @xmath545 and @xmath546 , we get @xmath565 } \\ & \le & \bar{c}_{11}l^p \bigl(n^{-1 } \ln(n ) \bigr)^{(r - p)/2}\bigl(h^*\bigr)^{\beta p-(r - p)/2}.\nonumber\end{aligned}\ ] ] here we have used that @xmath566 . in view of ( [ eq7proof - theorem - global - adaptation ] ) and ( [ eq8proof - theorem - global - adaptation ] ) , we choose @xmath172 from the equality @xmath567 , so that @xmath568 . finally , we obtain that @xmath569 the assertion of the theorem in the case @xmath111 follows now from theorem [ thglobal - oracle - inequality ] , ( [ eq2proof - theorem - global - adaptation ] ) and ( [ eq9proof - theorem - global - adaptation ] ) . in this section we briefly comment on the case when the design density @xmath570 is unknown . we provide changes to be done in the selection rule and in the presentation of the main result established in theorems [ thlocal - oracle - inequality ] and [ thglobal - oracle - inequality ] . we also explain basic ideas related to the proofs of the new results . in the context of the unknown design density , it is standard practice to use a plug - in estimator . this idea goes back to the nadaraya watson estimator and the problem considered in the paper is not an exception . suppose that an additional independent of an @xmath7 sample , say , @xmath571 , is available . alternatively , one can split the sample into two nonoverlapping parts . let @xmath572 stand for the probability law of @xmath573 . we reinforce assumption [ assassumption - on - design ] by the following condition : @xmath574 , where @xmath575 and @xmath576 denotes the supremum norm on @xmath21 . it is noteworthy that the constants @xmath577 and @xmath578 are both unknown . assume that based on @xmath579 we can construct an estimator of the design density @xmath570 , say , @xmath580 , having the following property . there exists a positive sequence @xmath581 as @xmath582 such that , for all sufficiently large @xmath85 , @xmath583 denote by @xmath584 ^ 2 $ ] the interval from assumption [ assassumption - on - design ] and introduce @xmath585 and @xmath586 , where @xmath587 tends to zero rather slowly . theoretically , @xmath587 can be chosen arbitrary , but a compromise allowing to keep our results under reasonable sample size is @xmath588 . in the definition of estimators @xmath589 and @xmath590 , the unknown now values @xmath591 , @xmath592 , should be replaced by their truncated estimators @xmath593 , @xmath594 . in the definition of all constants presented in section [ secconstants ] , the unknown now value @xmath595 has to be replaced by @xmath596 . let @xmath597 be obtained from @xmath598 by the replacement indicated in @xmath599 . then one should use @xmath600 in ( [ eqselection - rule ] ) . the right - hand sides of the local and global oracle inequalities established in theorems [ thlocal - oracle - inequality ] and [ thglobal - oracle - inequality ] will additionally contain a term @xmath601 , where @xmath508 is a numerical constant independent of @xmath77 , @xmath8 and the sample size @xmath87 . denote by @xmath602 the event @xmath603 . similar to the proof given in the step _ risk computation under @xmath369 _ of theorem [ thlocal - oracle - inequality ] , the computations under the event @xmath604 lead to the same reminder term in the oracle inequality . all computations under the event @xmath605 are done conditionally with respect to @xmath579 . it allows us to treat the estimator @xmath606 as nonrandom . analyzing the proof of `` probabilistic '' lemmas [ lemgauss - on - matrices ] and [ leminegality - for - sup - norm ] , we see that the results remain valid if we replace the function @xmath570 in the denominator of all expressions by an arbitrary function bounded from below and above on @xmath607 . thus , the use of @xmath608 in place of @xmath609 under @xmath610 is eligible . that leads to the similar assertion where @xmath611 is substituted with @xmath612 in all the constants involved . the latter quantity can be bounded under @xmath605 by @xmath613 , for @xmath85 large enough . however , such quantities can not be used in the definition of the threshold directly , because they incorporate the unknown @xmath614 . nevertheless , under the event @xmath605 , @xmath615 can be bounded by @xmath616 . moreover , we remark that all the quantities listed in section [ secconstants ] are increasing functions of @xmath595 so we can replace @xmath617 by the upper bound @xmath618 available under @xmath605 . it explains @xmath599 . the replacement of @xmath570 by @xmath619 leads to an additional `` approximation error '' bounded from above by @xmath620^{-1 } - 1 | . $ ] this quantity should be added to @xmath621 in order to preserve the proof of theorem [ thlocal - oracle - inequality ] . since @xmath622 depends on @xmath8 , one should instead use its upper bound @xmath623 which is available under @xmath605 . it gives @xmath624 for @xmath87 large enough . the use of @xmath625 in place of @xmath626 leads to an additional term @xmath627 in ( [ eq020proof - of - theorem - local ] ) . it explains @xmath628 and completes the sketch of the proof of theorem [ thlocal - oracle - inequality ] . since the global oracle inequality is obtained by the integration of the local one over bounded interval of @xmath21 , the assertion of theorem [ thglobal - oracle - inequality ] remains the same up to the term @xmath629 . if we suppose that @xmath630 , where @xmath631 is an isotropic hlder class of two - variate functions , then @xmath632^{\gamma/(2(\gamma+1))}$ ] , and this rate is attainable by a kernel estimator with properly chosen kernel and bandwidth this yields together with @xmath628 that if @xmath633 , the adaptive results established in theorems [ thpointwise - adaptation ] and [ thglobal - adaptation ] remain unchangeable . another possibility is to suppose that @xmath634 is a parametric family of densities . in this case , @xmath601 can be viewed as a reminder term . the authors wish to thank an associate editor and anonymous referees for their comments that substantially improved the paper .

as first paper of a series devoted to study the old stellar population in clusters and fields in the small magellanic cloud , we present deep observations of ngc121 in the f555w and f814w filters , obtained with the advanced camera for surveys on the _ hubble space telescope_. the resulting color - magnitude diagram reaches mag below the main - sequence turn - off ; deeper than any previous data . we derive the age of ngc121 using both absolute and relative age - dating methods . fitting isochrones in the acs photometric system to the observed ridge line of ngc121 , gives ages of gyr ( teramo ) , gyr ( padova ) and gyr ( dartmouth ) . the cluster ridge line is best approximated by the-enhanced dartmouth isochrones . placing our relative ages on an absolute age scale , we find ages of gyr ( from the magnitude difference between the main - sequence turn - off and the horizontal branch ) and gyr ( from the absolute magnitude of the horizontal branch ) , respectively . these five different age determinations are all lower by 2 3 gyr than the ages of the oldest galactic globular clusters of comparable metallicity . therefore we confirm the earlier finding that the oldest globular cluster in the small magellanic cloud , ngc121 , is a few gyr younger than its oldest counterparts in the milky way and in other nearby dwarf galaxies such as the large magellanic cloud , fornax , and sagittarius . if it were accreted into the galactic halo , ngc121 would resemble the `` young halo globulars '' , although it is not as young as the youngest globular clusters associated with the sagittarius dwarf . the young age of ngc121 could result from delayed cluster formation in the small magellanic cloud or result from the random survival of only one example of an initially small number star clusters .

0801.1111

characterizing old stellar populations provide important constraints on the early star formation histories of galaxies . only the satellite galaxies of the milky way ( mw ) are sufficiently close to resolve individual stars well below the oldest main - sequence turn - offs , which is a pre - condition for accurate photometric age dating of old stellar populations . all local group galaxies , for which adequate data exist , appear to contain stars older than 10 gyr @xcite . this result is based on main - sequence turn - off photometry of globular clusters and field populations in galactic satellites and a few more distant local group galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ) , as well as the detection of horizontal branch stars ( including rr lyrae variables ) in the local group and beyond . globular clusters are preferred as the basis for old stellar population age tracers since they are usually single - age , single - metallicity objects facilitating comparative studies . moreover , while globular cluster systems exhibit a range of ages ( e.g. , * ? ? ? * ) , the oldest ones may belong to the most ancient surviving stellar systems to have completed their formation in the youthful universe ( e.g. , * ? ? ? in those nearby galaxies where relative age dating based on main - sequence photometry was carried out in comparison to the oldest globular clusters in the milky way , no age difference within the measurement accuracy was found ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . the relative age dating of the oldest identifiable population ii objects thus indicates a common epoch of substantial early star formation in the milky way and its companions , although information about a putative , even older population iii remains to be uncovered in these objects . a galaxy that may _ not _ share this common epoch of early star formation at least not with respect to its globular clusters ( e.g. , * ? ? ? * ) is the small magellanic cloud ( smc ) . the smc is one of the closest and therefore best studied dwarf galaxies orbiting our galaxy . while the smc hosts a large number of intermediate - age and young star clusters , it only contains one old globular cluster , ngc121 , which is also the most massive star cluster . ngc121 is located @xmath7 2.4 ( @xmath73 kpc ) west of the smc bar at ( @xmath8 , @xmath9 ) = ( @xmath10 , @xmath11 ) . ngc121 is the only cluster in the smc that is sufficiently old to have developed an extended red horizontal branch @xcite and to contain rr lyrae stars . indeed , whether or not to call a star cluster a globular cluster is a matter of definition . in this case we refer to @xcite who consider lindsay1 as having a stumpy red clump and not a red horizontal branch . three rr lyrae stars were discovered in ngc121 by @xcite . @xcite found a fourth rr lyrae variable in the cluster and an additional 75 in a 1 @xmath12 1.3 square degree field centered on ngc121 . studies of various clusters in the lmc and in the mw showed that the presence of rr lyrae variables indicates that the parent population is as old as or older than @xmath7 10 gyr . an important question is whether ngc121 is as old as the typical old globular clusters in the large magellanic cloud ( lmc ) and in the mw . previous studies found ages ranging from 8 to 14 gyr for ngc121 @xcite using a variety of different techniques . studies based on the deepest available color - magnitude diagrams from hubble space telescope ( hst ) observations with the wide field and planetary camera 2 ( wfpc2 ) indicate an age of 10 to 10.6 gyr for ngc121 , suggesting that this globular cluster is several gyr younger than the oldest globulars in other nearby galaxies and in the mw @xcite . the capabilities of the advanced camera for surveys ( acs ) provide an improvement in both sensitivity ( depth ) as well as angular resolution , which is essential for a reliable photometric age determination in this dense star cluster . here we present deep photometry of ngc121 obtained with acs aboard the hst . we determine the age of ngc121 utilizing both absolute and relative methods ( e.g. , * ? ? ? the current study is the first in a series of papers based on hst studies of rich intermediate - age and old star clusters in the smc . in addition to ngc121 , six intermediate - age smc star clusters have been observed as part of our program : lindsay1 , kron3 , ngc339 , ngc416 , lindsay38 and ngc419 . we will derive fiducial ridgelines and fit isochrones to obtain accurate ages for each cluster using the same reduction techniques and isochrone models as described here ( see @xmath13 [ sec : obs]-[sec : age ] ) , and will present our results in future papers . in table [ tab : journalobs ] we list the cluster identification , date of observation , passband , exposure times and location of all clusters in our hst program ( go-10396 ; principal investigator : j. s. gallagher ) . in the next section we describe the data reduction procedure . in @xmath13 [ sec : cmd ] we present the color - magnitude diagram ( cmd ) of ngc121 and discuss its main features . in @xmath13 [ sec : age ] we describe our age derivation methods and present our results . the smc cluster ngc121 was observed with hst s acs on 2006 march 21 as part of our program focused on star clusters and field stellar populations in the smc . the program aims at exploring the star formation history and properties of the smc using both a number of carefully selected clusters and field regions . for ngc121 we obtained imaging in the f555w and f814w filters , which resemble the johnson v and i filters in their photometric properties @xcite . the images were obtained using the wide field channel ( wfc ) of acs and cover an area of @xmath14 with a pixel scale of @xmath7 0.05 arcsec . one set of exposures was taken at the nominal position of the cluster center . four long exposures were obtained in each filter for hot pixel removal and to fill the gap between the two halves of the 4096 @xmath12 4096 pixel detector . each pointing has an exposure time of 496 s in the f555w , and 474 s in the f814w filter . moreover , two short exposures were taken in each filter with an exposure time of 10 s in f555w and 20 s in f814w . ccccccc ngc121 & j96106030 & 2006@xmath1503@xmath1521 & f555w & 40.0 & @xmath16&@xmath17 + & j96106040 & & & 1984.0 & @xmath18 & @xmath19 + & j96106010 & & f814w & 20.0 & @xmath16&@xmath17 + & j96106020 & & & 1896.0 & @xmath18 & @xmath19 + lindsay1 & j96105030 & 2005@xmath1508@xmath1521 & f555w & 40.0 & @xmath20 & @xmath21 + & j96105040 & & & 1984.0 & @xmath22 & @xmath23 + & j96105010 & & f814w & 20.0 & @xmath20 & @xmath21 + & j96105020 & & & 1896.0 & @xmath22 & @xmath23 + kron3 & j96107030 & 2006@xmath1501@xmath1517 & f555w & 40.0 & @xmath24 & @xmath25 + & j96107040 & & & 1984.0 & @xmath26 & @xmath27 + & j96107010 & & f814w & 20.0 & @xmath24 & @xmath25 + & j96107020 & & & 1896.0 & @xmath26 & @xmath27 + ngc339 & j96104030 & 2005@xmath1511@xmath1528 & f555w & 40.0 & @xmath28 & @xmath29 + & j96104040 & & & 1984.0 & @xmath30 & @xmath31 + & j96104010 & & f814w & 20.0 & @xmath28 & @xmath29 + & j96104020 & & & 1896.0 & @xmath30 & @xmath31 + ngc416 & j96121030 & 2006@xmath1503@xmath1508 & f555w & 40.0 & @xmath32 & @xmath33 + & j96121040 & & & 1984.0 & @xmath34 & @xmath35 + & j96121010 & & f814w & 20.0 & @xmath32 & @xmath33 + & j96121020 & & & 1896.0 & @xmath34 & @xmath35 + lindsay38 & j96102030 & 2005@xmath1508@xmath1518 & f555w & 40.0 & @xmath36 & @xmath37 + & j96102040 & & & 1940.0 & @xmath38 & @xmath39 + & j96102010 & & f814w & 20.0 & @xmath36 & @xmath40 + & j96102020 & & & 1852.0 & @xmath38 & @xmath39 + ngc419 & j96103030 & 2006@xmath1501@xmath1505 & f555w & 40.0 & @xmath41 & @xmath42 + & j96103040 & & & 1984.0 & @xmath43 & @xmath44 + & j96103010 & & f814w & 20.0 & @xmath41 & @xmath42 + & j96103020 & & & 1896.0 & @xmath43 & @xmath44 + [ tab : journalobs ] the data set was processed adopting the standard space telescope science insitute acs calibration pipeline ( calacs ) to subtract the bias level and to apply the flat field correction . for each filter , the short and long exposures were co - added independently using the multidrizzle package @xcite . with this package the cosmic rays and hot pixels were removed and a correction for geometrical distortion was applied . the resulting ngc121 data consist of one 40 s and one 1940 s exposure in f555w and one 20 s as well as one 1896 s exposure in f814w . the two short exposures allowed us to measure brighter stars that are saturated in the long exposures . the photometric reductions were carried out using the daophot package in the iraf environment . we discarded saturated foreground stars and background galaxies using the _ source extractor _ package @xcite . due to the different crowding and signal - to - noise ratio properties of the long and the short exposure images , photometry involving point spread function ( psf ) fitting was only performed on the long exposures . for the short exposures we used aperture photometry , which turned out to yield smaller formal errors than psf photometry . we ran daophot on our data and set the detection threshold at 1 @xmath45 above the local background level in order to detect even the faintest sources . the list of stars detected in the f814w image was then used as coordinate input list to identify the stars in the f555w image and serve as our coordinate reference . 49493 sources were found to be common to both long exposure frames . for these sources , we performed aperture photometry using an aperture radius of 3 pixels . we then constructed a psf by combining 150 bright and isolated stars that were distributed fairly uniformlly across the image . finally , psf photometry was carried out . the photometric calibration was accomplished by converting the magnitudes of the individual stars to the standard acs magnitude system by using an aperture with a radius of @xmath46 ( or 10 pixels on the image ) , in combination with the aperture correction from the @xmath46 aperture radius to infinity and the synthetic zero points for the acs / wfc @xcite . the aperture correction was derived for each frame independently . the objects found in both images were cross - identified and merged with a software package written at the bologna observatory by p. montegriffo ( private communication ) . altogether we were able to cover a luminosity range of @xmath7 10 magnitudes after combining the resultant photometry of the short and long exposures . in figure [ fig : ngc121_sigma ] we show the photometric errors assigned by daophot . for stars measured on the short exposures , the formal photometric errors remain negligible over a wide range of magnitudes . in the long exposures , all the brighter stars with @xmath47 mag and @xmath48 mag are saturated . at @xmath49 19.6 mag , the short exposure ( _ blue dots _ ) and aperture photometry from the long exposure ( _ black dots _ ) samples were combined , and for stars fainter than @xmath49 22.2 mag , long exposure psf photometry ( _ red dots _ ) was used . we chose where to cut between the aperture and psf photometry catalogues based on the @xmath50 data and adopted the same value for @xmath51 so as to avoid a color slope associated with this division . for our study , we rejected all stars with a @xmath45 error larger than 0.2 mag and a daophot sharpness parameter @xmath52 in both filters . to obtain a superior cmd , we discarded all stars within a radius of @xmath53 from the cluster center , which excludes the very dense core of the cluster . with this selection , our final sample contains 17464 stars common in both filters . the resulting color - magnitude diagram ( cmd ) of ngc 121 and its surroundings is shown in figure [ fig : ngc121_letterfinal ] . our cmd for ngc121 reaches @xmath73.5 mag below the msto ( @xmath7 0.5 magnitudes deeper than the previous deepest available photometry ) , which allows us to carry out the most accurate age measurements obtained so far . the cmd shows a well - populated main sequence ( ms ) , subgiant branch ( sgb ) , red giant branch ( rgb ) , horizontal branch ( hb ) , and asymptotic giant branch ( agb ) . the gap on the rgb at @xmath54 20 mag is an artificial feature due to small number statistics resulting from our exclusion of crowded stars in the cluster center . ngc121 appears to be a single - age population object just as one would expect for a canonical star cluster . as expected , there is no obvious evidence for field star contamination by younger populations due to the location of ngc121 in a low - density area in the outer parts of the smc . within the field of view of the acs and at the high galactic latitude of the smc , galactic foreground contamination is very low ( e.g. , * ? ? ? another possible contamination source is the massive and extended galactic globular cluster 47tuc , which has a tidal radius of 42.86 arcmin @xcite and an angular distance from ngc121 of @xmath7 32 arcmin . we visually estimated the location of the center of ngc121 on the image and selected all stars within an annulus of @xmath53 and @xmath55 to create a bona fide sample . this cmd is displayed in figure [ fig : ngc121_lettercenter ] . there is no evidence for a binary sequence in ngc 121 , but we can not exclude their presence , due to the photometric error . the aforementioned traces of minor field contamination have mainly vanished . due to crowding , incompleteness becomes significant at the faint end of the ms : this affects particularly faint stars in the cluster center . hence in figure [ fig : ngc121_lettercenter ] the ms becomes less densely populated at fainter magnitudes . the red hb is well populated and extends into the rr lyrae instability strip ( clementini et al . 2007 , in prep ) . the presence of a red hb provides a circumstantial suggestion that ngc121 may be younger than old galactic and lmc globular clusters , since the hbs of the oldest globular clusters tend to extend farther into the blue ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? red hbs , however , can also be due to a second parameter other than age affecting the hb morphology ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? since a true hb is present , an age measurement for ngc121 can be made using the @xmath56 age measurement , which we will do in @xmath13 [ sec : relage ] . this method requires the determination of the apparent mean magnitude of the hb . our data yield @xmath57 = @xmath58 mag for this observable , which is in agreement with the magnitudes determined by shara et al . ( 1998 ) , alves & sarajedini ( 1999 ) and dolphin et al . ( 2001 ) . at @xmath50 = @xmath59 mag we find the ngc121 rgb bump ( @xmath60 ) which is 0.06 mag brighter than the magnitude found by @xcite . the difference in luminosity is due to the exclusion of the center stars . if we determine the @xmath60 on the entire sample , we obtain @xmath50 = @xmath61 mag , which is in excellent agreement with the magnitude found by @xcite . this feature is predicted by stellar evolution models , which also show that the luminosity of the rgb bump is dependent on the metallicity and age of the cluster . when the metallicity is known the difference between @xmath62 and @xmath63 can be used as an age indicator . above the ms turn - off , @xcite found 42 candidate blue straggler stars ( bss ) . evolved descendants of the bsss are important as possible sources of stars lying above the traditional hb ( e.g. , * ? ? ? . in our acs study we recovered the @xcite bss sample and also found more stars in the bss region of which some ( about 20 ) turned out to be pulsating variables ( dwarf cepheids ) . these stars will be discussed in clementini et al . ( 2007 , in preparation ) where we will present the results of an hst study of variable stars in ngc121 . age determinations of star clusters using isochrones depend crucially on the interstellar extinction , distance , and metallicity of the cluster , as well as on the chosen isochrone models . in fitting isochrones to the cmd of ngc121 we adopted the spectroscopic metallicity measurement of [ fe / h ] = @xmath64 from da costa @xmath65 hatzidimitriou ( 1998 , see also johnson et al . 2004 ) on the metallicity scale introduced by @xcite ( zw84 ) . the distance and the extinction were treated as free parameters . the smc distance modulus is @xmath66 = @xmath67 mag ( 60 kpc ) ( e.g. , * ? ? ? * ) , but due to the large depth extension of the smc along the line of sight we adjusted the distance modulus @xmath66 to produce the best isochrone fits to our cmd data . for easier comparison to the isochrones , we first derived a fiducial ridgeline ( table [ tab : ridgeline ] ) , which reproduces the mean location of the stellar distribution in the cmd ( exempting the hb ) . in order to determine the ridgeline , we separated the cluster center cmd into three sections : the ms , the sgb and the rgb . on the ms we determined the mode of the color distribution in magnitude bins of 0.3 mag width . for the sgb , we performed a linear least squares fit of a polynomial of 5th order to a hess diagram of this region in the cmd . finally , the rgb was fit by a third - order polynomial of the mean color , again in magnitude bins with a size of 0.3 mag each . the resulting ridgeline is shown in fig . [ fig : ngc121_lettercenter ] as a cyan line . we fitted our @xmath50 vs. @xmath50-@xmath51 cmd with three different isochrone models : padova isochrones ( girardi , `` private communication '' , girardi et al . 2000 ) , ] teramo isochrones @xcite , both with scaled solar isochrones ( [ @xmath4/fe ] = 0.0 ) , and dartmouth isochrones @xcite with both [ @xmath4/fe ] = 0.0 and + 0.2 . the padova isochrone grid has an age resolution of log(t)=0.05 , the teramo isochrone grid of 0.1 myr and the dartmouth isochrone grid of 0.5 gyr . our adopted spectroscopic metallicity of [ fe / h ] = @xmath68 corresponds most closely to z = 0.0004 in the padova models , to z = 0.0006 in the teramo models , and to [ fe / h ] = @xmath69 in the dartmouth models . all three sets of isochrone models are available in the standard acs color system . we fitted a large number of isochrones using different combinations of reddening , age , and distance . for each set of models , we selected the isochrone that best matched the observed data ( fig . [ fig : ngc121_letteriso ] , fig . [ fig : ngc121_dart ] ) . first we discuss figure [ fig : ngc121_letteriso ] . our best - fit age using padova isochrones is 11.2 gyr with @xmath66 = 19.05 mag and @xmath70 = 0.05 . the best fitting teramo isochrone yields an age of t = 11.8 gyr , @xmath66 = 19.02 mag , and @xmath70 = 0.04 . on the ms , both the teramo isochrone and the padova isochrones trace the ridge line almost perfectly . at the faint end of the ms , the padova isochrone continues further to the blue than the teramo isochrone and our derived ridge line ; however , this only becomes more apparent at magnitudes of @xmath50 @xmath7 25.5 mag and below . both isochones also provide an excellent approximation to the sgb and to the lower rgb up to about half a magnitude below the hb . at brighter magnitudes , the two isochrones deviate increasingly to the blue of the observed upper rgb . here the padova isochrone shows the strongest difference , deviating by approximately 0.38 mag in color from the observed tip of the rgb . the isochrone shows a magnitude for the base of the red hb that is about 0.5 mag fainter than the observed one . unlike teramo and dartmouth , the padova isochrone also models the agb and its tip , which is @xmath7 1 mag brighter than the tip of the rgb . the teramo isochrone is too blue by about 0.23 mag at the magnitude of the tip of the rgb and indicates a magnitude for the base of the red hb that is 0.2 mag too bright . if we had no prior knowledge of the metallicity of ngc121 and were to use the upper rgb as a metallicity indicator , a better fit would be obtained by choosing isochrones of a different metallicity or @xmath4 abundance . the problems of various isochrone models of given metallicities in reproducing the upper red giant branches of globular clusters with the same metallicities are a well - known problem ( e.g. , * ? ? ? * ; * ? ? ? our figure [ fig : ngc121_letteriso ] reflect the general failure of the chosen stellar evolutionary models to simultaneously reproduce the major features of cmds @xcite in spite of the excellent fit to the lower rgb , sgb , and ms . fortunately the latter are the most age - sensitive features of the cmd . the isochrone model provided by @xcite with [ @xmath4/fe ] = + 0.2 yield the best fit to the cmd ( fig . [ fig : ngc121_dart ] ) . the best - fit isochrone has the parameters _ t _ = 10.5 gyr , @xmath66 = 18.96 mag , and @xmath70 = 0.035 , using the @xmath4-enhanced isochrones of [ @xmath4/fe ] = + 0.2 . _ all _ the major features of the cmd are very well reproduced , including the upper rgb where the isochrone is offset slightly to the blue relative to the fiducial ridgeline . this offset is no more than 0.01 to 0.02 on average along the entire upper rgb ; i.e. , even the _ slope _ of the rgb is very well reproduced along its entire extent . unfortunately the stellar evolution models used here terminate at the he flash , and therefore do not fit the hb or the agb . is our use of @xmath4-enhanced models justified ? for ngc121 a value of @xmath71 } = + 0.24 $ ] has been measured , which is similar to the outer lmc cluster hodge 11 and to the old galactic outer halo clusters with @xmath71 } = + 0.3 $ ] @xcite . consequently , we assume that ngc121 also is enhanced in @xmath4-elements . we note that in this respect ngc121 differs from the general trend observed in red giant stars in the lmc and in dwarf spheroidal galaxies , where the [ @xmath4/fe ] ratios at a given [ fe / h ] tend to be lower by up to a few tenths of a dex than in the galactic halo ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? when we adopt the values for distance and reddening , but fit the cluster with an isochrone scaled solar , ngc121 gets a slightly older age of 11 gyr . the isochrone model with [ @xmath4/fe ] = 0.0 still provides a better fit than the teramo or padova models , but has an offset of @xmath7 0.05 on average along the upper rgb . past studies found that @xmath4-enhanced models imply a higher luminosity and temperature for the same mass than the solar scaled models and therefore an older age for the same magnitude ( e.g. , * ? ? ? * ) . the dartmouth models show exactly the opposite behavior . this is because in these models an increase in [ @xmath4/fe ] is accompanied by a corresponding increase of the total metallicity z , which makes the isochrones cooler at constant age and [ fe / h ] . finally , we note that the derived reddenings agree with the extinction @xmath72 = @xmath73 from the @xcite maps . the reddening law of @xcite is assumed . in the figures [ fig : ngc121_isocheck ] and [ fig : ngc121_isocheck_dart ] we show a range of isochrones for the three sets of stellar evolution models in order to illustrate the age uncertainty in a given model . the finally chosen , `` best '' isochrone is always displayed along with two younger and two older isochrones . for the cases of the teramo and padova models , the two isochrones that are one age step younger or older than the chosen , central isochrone provide an upper or lower envelope for the location of the high - density part of the sgb , the ms turn - off , and the base of the rgb . for the dartmouth isochrones the outermost isochrones provide this envelope . considering the high quality of the fit of the central isochrone in this cmd region in all models and the larger deviations of the adjacent isochrones , we estimate that the resultant age uncertainty is of the order of approximately @xmath74 gyr for the teramo and dartmouth isochrones and @xmath75 gyr for the padova isochrones . + cccc 1.7400 & 16.6859 & 1.0000 & 19.9538 + 1.7300 & 16.6916 & 0.9900 & 20.0855 + 1.7200 & 16.7011 & 0.9800 & 20.2218 + 1.7100 & 16.7107 & 0.9700 & 20.3630 + 1.7000 & 16.7224 & 0.9600 & 20.5090 + 1.6900 & 16.7339 & 0.9500 & 20.6601 + 1.6800 & 16.7483 & 0.9400 & 20.8162 + 1.6700 & 16.7605 & 0.9300 & 20.9774 + 1.6600 & 16.7834 & 0.9200 & 21.1440 + 1.6500 & 16.8001 & 0.9100 & 21.3159 + 1.6400 & 16.8177 & 0.9000 & 21.4932 + 1.6300 & 16.8381 & 0.8900 & 21.6762 + 1.6200 & 16.8554 & 0.8800 & 21.8899 + 1.6100 & 16.8706 & 0.8700 & 22.0599 + 1.6000 & 16.8867 & 0.8600 & 22.1921 + 1.5900 & 16.9098 & 0.8600 & 22.1933 + 1.5800 & 16.9289 & 0.8500 & 22.2305 + 1.5700 & 16.9540 & 0.8400 & 22.2520 + 1.5600 & 16.9712 & 0.8300 & 22.2660 + 1.5500 & 16.9874 & 0.8200 & 22.2770 + 1.5400 & 17.0008 & 0.8100 & 22.2890 + 1.5300 & 17.0153 & 0.8000 & 22.3050 + 1.5200 & 17.0309 & 0.7900 & 22.3180 + 1.5100 & 17.0477 & 0.7800 & 22.3280 + 1.5000 & 17.0658 & 0.7700 & 22.3380 + 1.4900 & 17.0851 & 0.7600 & 22.3480 + 1.4800 & 17.1057 & 0.7500 & 22.3570 + 1.4700 & 17.1276 & 0.7400 & 22.3670 + 1.4600 & 17.1508 & 0.7300 & 22.3770 + 1.4500 & 17.1753 & 0.7200 & 22.3880 + 1.4400 & 17.2013 & 0.7100 & 22.4010 + 1.4300 & 17.2286 & 0.7000 & 22.4120 + 1.4200 & 17.2575 & 0.6900 & 22.4260 + 1.4100 & 17.2877 & 0.6800 & 22.4520 + 1.4000 & 17.3195 & 0.6700 & 22.4710 + 1.3900 & 17.3528 & 0.6600 & 22.5020 + 1.3800 & 17.3877 & 0.6500 & 22.5190 + 1.3700 & 17.4241 & 0.6400 & 22.5410 + 1.3600 & 17.4621 & 0.6300 & 22.5720 + 1.3500 & 17.5018 & 0.6200 & 22.6240 + 1.3400 & 17.5432 & 0.6100 & 22.6890 + 1.3300 & 17.5962 & 0.6000 & 22.7610 + 1.3200 & 17.6378 & 0.5950 & 22.8500 + 1.3100 & 17.6904 & 0.5950 & 23.0310 + 1.3000 & 17.7246 & 0.6125 & 23.4000 + 1.2900 & 17.7605 & 0.6670 & 24.0000 + 1.2800 & 17.7983 & 0.7400 & 24.6000 + 1.2700 & 17.8379 & 0.8400 & 25.2000 + 1.2600 & 17.8794 & 0.9700 & 25.8000 + 1.2500 & 17.9231 & 1.1000 & 26.4000 + 1.2400 & 17.9690 & & + 1.2300 & 18.0171 & & + 1.2200 & 18.0676 & & + 1.2100 & 18.1205 & & + 1.2000 & 18.1761 & & + 1.1900 & 18.2343 & & + 1.1800 & 18.2952 & & + 1.1700 & 18.3590 & & + 1.1600 & 18.4257 & & + 1.1500 & 18.4955 & & + 1.1400 & 18.5685 & & + 1.1300 & 18.6447 & & + 1.1200 & 18.7242 & & + 1.1100 & 18.8071 & & + 1.1000 & 18.8936 & & + 1.0900 & 18.9837 & & + 1.0800 & 19.0775 & & + 1.0700 & 19.1752 & & + 1.0600 & 19.2767 & & + 1.0500 & 19.3823 & & + 1.0400 & 19.4920 & & + 1.0300 & 19.6059 & & + 1.0200 & 19.7241 & & + 1.0100 & 19.8367 & & + [ tab : ridgeline ] to check the reliability of the isochrone ages , we use a reddening - independent method to derive relative ages of ngc121 . this method is also independent of the photometric zeropoint of our data . this vertical method relies on the fact that the absolute magnitude of the msto depends on the age of the cluster ( e.g. , * ? ? ? * ) , while the absolute magnitude of the hb remains approximately age - independent for clusters older than @xmath76 10 gyr ( e.g. , * ? ? ? we measure the apparent magnitudes of the msto and hb , i.e. , @xmath77 and @xmath62 , in order to obtain the magnitude difference @xmath56 . with increasing age , a cluster has generally larger values of this parameter since the ms moves to fainter magnitudes . unfortunately , the determination of these two points comes with significant uncertainties . @xmath78 is hard to measure accurately both because of the width of the hb in luminosity and the msto s vertical extent in the turn - off region . first , we calculate the magnitude difference @xmath79 between the hb and the msto , as originally described by @xcite . because @xmath50 is proportional to v and we are only interested in the magnitude difference , which we measure at constant color , there is no need to transform the magnitudes from the acs system to _ v _ and _ i _ magnitudes @xcite . we follow the general definition of the msto as the bluest point along the ms ; in our case represented by the bluest point on the ridgeline . we find the msto at @xmath80 = @xmath81 mag , @xmath82 = @xmath83 mag . as described in section [ sec : cmd ] , the mean hb magnitude is @xmath84 = @xmath85 mag . this yields a @xmath86 mag . our result is only slightly lower than former values , e.g. , @xmath56 = 3.32 determined by @xcite , 3.33 by @xcite or 3.29 by @xcite . @xcite published a mean value of @xmath87 mag for old galactic halo globular clusters ( gc ) . the @xmath79 of ngc121 is 0.28 mag smaller than this value , which indicates that ngc121 is younger than most of the older _ galactic _ gcs . @xcite ( see also buonanno et al . 1989 ) found a relation between age and @xmath56 based on a study of 41 galactic globular clusters . adopting again a metallicity of [ fe / h ] = @xmath64 , we find with the formula @xmath88 } + 0.37 \delta v^{hb}_{to } - 0.24 $ ] an age of @xmath89 gyr for ngc121 . in order to compare our data directly with @xcite , we use the age calibration provided by @xcite as was done by these authors . for this purpose we must adopt the metallicity fe / h ] = @xmath91 of @xcite on the metallicity scale introduced by @xcite ( cg97 ) to be consistent with @xcite s calculation . note that elsewhere in the paper we are using metallicities on the zw84 scale which agrees reasonably well with the spectroscopically derived chemical abundances @xcite . using this method and the higher metallicity we determine an age of @xmath92 gyr for ngc121 , which is similar to the age that @xcite found . we have to emphasize that this age is younger than the absolute ages determined in section [ sec : isochrone ] due to the different metallicity scale . if we use our preferred zw84 metallicity scale with this method we obtain an older age of @xmath93 gyr . while the measurement of @xmath77 is affected by significant observational errors ( @xmath7 0.05 mag ) , the color of the msto is well - defined . @xcite found that the usage of a point on the sbg brighter than the msto and 0.05 mag redder ( @xmath94 ) provides more precise relative ages than @xmath77 . @xcite provide a conversion between @xmath95 and @xmath96 , @xmath97 data for a grid of five metallicities . we choose the conversion for [ fe / h ] @xmath98 because it is closest to the metallicity of ngc121 . in our data , we measured @xmath99 = @xmath100 at @xmath101 = @xmath102 mag . to convert @xmath99 to the absolute magnitude @xmath103 we use the distance modulus derived above . this yields @xmath104 mag . with the modified calibration by @xcite we obtain an age of @xmath105 gyr for ngc121 . we have summarized all our age results in table [ tab : ages ] . calculating @xmath106 = @xmath107 we find @xmath108 mag . there is a general trend of increasing rgb bump brightness with decreasing age @xcite , assuming that a different second parameter is not affecting the position of the rgb bump . @xcite presented a @xmath109 vs. [ fe / h ] diagram ( their fig.6 ) , where the brightness difference between the rgb bump and the hb is plotted against the cluster metallicity . if we assume that the hb magnitude does not critically depend on age , then ngc121 is slightly older than 10 gyr based on this relation . these comparative results are sensitive to the abundance of @xmath4-elements in ngc121 as compared to galactic globular clusters . many nearby galactic globular clusters are enhanced in @xmath4-elements relative to the solar value @xcite . as mentioned earlier , spectroscopic results for ngc121 indicate that this cluster is similarly enhanced in @xmath4 elements , which means that the relative ages should not be affected as long as we confine ourselves to the comparison of globular clusters with similar @xmath4-element ratios . the fact that ngc121 does not follow the trend of reduced [ @xmath4/fe ] ratios observed in other nearby dwarf galaxies facilitates both our relative age determinations . ccc @xmath5 & @xmath56 & @xcite ( zw ) + @xmath6 & @xmath95 & @xcite + @xmath93 & @xmath90 & @xcite + @xmath1 & isochrones & @xcite + @xmath110 & isochrones & girardi , `` private communication '' , + & & @xcite + @xmath3 & isochrones & @xcite + [ tab : ages ] in table [ tab : comp_ages_zw ] we compare the relative age of ngc121 with those for a sample of galactic globular clusters . while this comparison sample is located in the galactic halo , some objects may have formed outside the galaxy and might have been subsequently captured or accreted . we list the clusters by their identification in column ( 1 ) . the [ fe / h ] values are given in column ( 2 ) in the scale by @xcite . column ( 3 ) shows the @xmath111 and column ( 4 ) the ages obtained by using the @xcite calibration . finally , column ( 5 ) gives the relative age difference of these clusters to ngc121 @xmath112 . for @xmath111 we adopted the values from @xcite , unless differently stated ( see footnotes of table [ tab : comp_ages_zw ] ) . the clusters are listed in order of increasing @xmath111 and are divided into two groups . the first group shows nine pure gcs with similar metallicities as ngc121 . among these nine clusters is ngc2808 for which multiple mstos have been found @xcite . its @xmath111 value , derived prior to the study by @xcite , is comparatively small . even though those nine clusters have all similar metallicities , the spread in @xmath111 and therefore in age is quite large : ngc1262 shows the lowest @xmath111 = 3.24 and is similar in age to ngc121 , while ngc6656 has @xmath111 = 3.55 , which makes it @xmath73 gyr older than ngc121 . the second group includes a subset of galactic halo clusters that appear to be significantly younger than the average of the galactic globular cluster population ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? some members of this group are listed in the last part of table [ tab : comp_ages_zw ] . ngc362 and ngc288 are known to be a second parameter cluster pair of different ages , as reflected in their different @xmath111 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? as ngc362 has a similar @xmath113 and [ fe / h ] as ngc121 , it should therefore be of a similarly young age . other members of this group of young halo globulars are ic 4499 @xcite , ruprecht 106 ( assumed to be 35 gyr younger than the bulk of the galactic globulars with similar metallicities ; * ? ? ? * ; * ? ? ? * ) , arp 2 @xcite , terzan 7 @xcite , and pal 14 @xcite . ccccc ngc121 & -1.46 & 3.27 & 10.9 & 0 + ngc1261 & -1.32 & 3.24 & 10.4 & -0.5 + ngc5272 & -1.66 & 3.24 & 10.8 & 0.1 + ngc2808 & -1.36 & 3.25 & 10.6 & -0.3 + ngc3201 & -1.53 & 3.28 & 11.0 & 0.1 + ngc5904 & -1.38 & 3.34 & 11.4 & 0.5 + ngc6254 & -1.55 & 3.37 & 11.9 & 1.0 + ngc6218 & -1.40 & 3.48 & 12.9 & 2.0 + ngc6752 & -1.54 & 3.53 & 13.7 & 2.8 + ngc6656 & -1.41 & 3.55 & 13.7 & 2.8 + pal 12 & -0.94 & 3.17 & 9.45 & -1.3 + ig 4499 & -1.75 & 3.25 & 11 & 0.1 + ngc362 & -1.33 & 3.27 & 10.7 & -0.2 + pal 14 & -1.65 & 3.33 & 11.7 & 0.8 + ngc288 & -1.40 & 3.45 & 12.6 & 1.7 + [ tab : comp_ages_zw ] among the theories that try to explain the existence of these young objects is the model according to which they are intergalactic clusters captured by the mw @xcite , or clusters formed during interactions between the mw and the magellanic clouds assuming that they are on bound orbits @xcite . @xcite argued that the apparent young halo globular clusters formed in dwarf galaxies that later merged with the mw . hence the galactic globular clusters are assumed to be a mixture of objects that formed with the mw itself ( old halo group ) and others accreted from destroyed dwarf satellites ( young halo clusters ) ( see also mackey & gilmore 2004 ) . at least six globular clusters are believed to be associated with the sagittarius dwarf galaxy ( e.g. , * ? ? ? * and references therein ) , providing support for the accretion scenario . note , however , that sagittarius is contributing both old ( m54 , ter 8 , arp 2 ) and `` young '' ( ter 7 , pal 12 , whiting 1 ) globular clusters to the mw . similarly , the only other galactic dsph galaxy known to contain globular clusters , fornax , would contribute both kinds of globulars @xcite if it were to merge with our galaxy . this also holds for the lmc ( * ? ? ? * see also discussion in grebel , gallagher , & harbeck 2003 ) . if we take all the ages determined in this paper into account , we find that ngc121 is consistently 23 gyr younger than the oldest galactic globular clusters and lmc globular clusters . the age offset remains when comparing ngc121 to old galactic globular clusters in the same metallicity range ( see tab . [ tab : comp_ages_zw ] , upper panel ) . we also show that ngc121 is not as young as the youngest galactic and sagittarius globular clusters , some of which are @xmath7 2 gyr younger than ngc121 . we derived ages for the old smc globular cluster ngc121 based on our high dynamic range hst / acs photometry that extends at least three magnitudes below its msto . in order to obtain absolute ages , we applied three different isochrone models . these isochrone models yielded ages of @xmath110 gyr ( padova ) , @xmath114 gyr ( teramo ) , and @xmath3 gyr ( dartmouth ) . we find the @xmath4-enhanced dartmouth isochrones provide the closest approximation to the ms , sgb , and rgb , whereas the other models can not reproduce the slope of the upper rgb . high - resolution spectroscopy indicates that ngc121 is indeed @xmath4-enhanced @xcite , a property that it shares with many of the old outer galactic halo globulars . given the proximity of ngc121 to the smc on the sky and its distance , its physical association with the smc seems well - established . our determinations of relative ages for ngc121 are consistent with the results of our absolute age determination . relative age estimates , when converted to an absolute age scale , are @xmath5 gyr ( @xmath111 ) , @xmath93 ( @xmath90 ) and @xmath6 gyr ( @xmath115 ) . these numbers agree well with the absolute age derivations . our results confirm that ngc121 is 23 gyr younger than the oldest mw and lmc clusters ( as also found in earlier wfpc2 studies ) . ngc121 is similar in age to the youngest globular cluster in the fornax dsph @xcite , and to several of the young galactic halo clusters . on the other hand , ngc121 is not as young as some of the sgr dwarf galaxy s globular clusters or the youngest galactic globular clusters . it is intriguing that the smc in contrast to other galactic companion dwarf galaxies with globulars does not contain any old classical globular clusters . but given the existence of only one cluster and the question of star cluster survival , this could be a result of the one survivor from the smc s epoch of globular cluster formation randomly sampling an initial distribution of star cluster ages . on the other hand , in low - mass galaxies without bulges , spiral density waves , and shear it is much more difficult to destroy globular clusters through external effects . that this cluster is both younger than the galactic mean and enhanced in @xmath4-elements may have interesting implications for the early development of the smc . it also is intriguing that the only globular cluster in the smc is not very metal - poor . the smc must have experienced substantial enrichment prior to the formation of ngc121 . in the lmc , where two main epochs of the formation of populous compact star clusters have been found ( e.g. , * ? ? ? * ) , a few globular clusters are found that are old enough to exhibit blue hbs . interestingly , these globular clusters , which are similarly old as the oldest galactic globulars @xcite , have a similar metallicity to ngc 121 @xcite ( e.g. , ngc1898 , ngc2019 ) , indicating very early chemical enrichment . the mw also contains old classical globular clusters ( with blue hbs ) that have similarly high metallicities as the somewhat younger ngc 121 . evidently , the conditions for and the efficiency of star formation varied in these three galaxies at early epochs . after ngc 121 formed there was a hiatus in surviving stars clusters and thus possibly in cluster formation activity in the smc : the second oldest smc cluster is lindsay 1 with an age of @xmath7 8 gyr ( glatt et al . 2007 , in preparation ) . since then compact populous star clusters formed fairly continuously until the present day in the smc ( e.g. , da costa 2002 ) in contrast to both the lmc and the mw . in forthcoming papers on our acs photometry of smc clusters and field populations we will explore the evolutionary history of the smc in more detail . clearly , clues about the early star formation history of the smc will have to come from its old field populations . we thank the anonymous referee for extremely useful suggestions to improve our paper . we gratefully acknowledge support by the swiss national science foundation through grant number 200020 - 105260 and 200020 - 113697 . support for program go-10396 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 . we warmly thank paolo montegriffo to provide his software and leo girardi for the padova isochrones in the standard acs color system . gisella clementini and monica tosi have been partially supported by prin - miur-2004 and prin - inaf-2005 , and jay gallagher also obtained helpful additional support from the university of wisconsin graduate school . , d. r. , & sarajedini , a. 1999 , , 511 , 225 , m. , fusi pecci , f. , ferraro , f. r. , galleti , s. , catelan , m. , & landsman , w. b. 2001 , , 122 , 2569 , g. , mateo , m. , chiosi , c. , & bressan , a. 1992 , , 388 , 400 , e. , & arnouts , s. 1996 , , 117 , 393 , t. m. , et al . 2007 , , 658 , l95 , r. , corsi , c. e. & fusi pecci , f. 1989 , , 216 , 80 , r. , corsi , c. e. , fusi pecci , f. , richer , h. b. , & fahlman , g . g. 1993 , , 105 , 184 , r. , corsi , c. e. , pulone , l. , fusi pecci , f. , richer , h. b. , & fahlman , g . g. 1995b , , 109 , 663 , r. , corsi , c. e. , fusi pecci , f. , richer , h. b. , & fahlman , g . g. 1995a , , 109 , 650 , r. , corsi , c. , bellazzini , m. , ferraro , f. r. , & fusi pecci , f. 1997 , , 113 , 706 , r. , corsi , c. e. , zinn , r. , fusi pecci , f. , hardy , e. , & suntzeff , n. b. 1998 , , 501 , l33 , r. , corsi , c. e. , castellani , m. , marconi , g. , fusi pecci , f. , & zinn , r. 1999 , , 118 , 1671 , g. , zinn , r. , & moni bidin , c. 2007 , , 466 , 181 , e. , & gratton , r. g. 1997 , , 121 , 95 , m. , ferraro , f. r. , & rood , r. t. 2001a , , 560 , 970 , m. , bellazzini , m. , landsman , w. b. , ferraro , f. r. , fusi pecci , f. , & galleti , s. 2001b , , 122 , 3171 , m. 2005 , arxiv astrophysics e - prints , arxiv : astro - ph/0507464 , b. , demarque , p. , kernan , p. j. & krauss , l. m. 1996 , science , 271 , 957 , b. , demarque , p. , kernan , p. j. , krauss , l. m. , & sarajedini , a. 1996a , , 283 , 683 , b. , demarque , p. , & sarajedini , a. 1996b , , 459 , 558 , g. , held , e. v. , baldacci , l. , & rizzi , l. 2003 , , 588 , l85 , a. a. , et al . 2007 , , 659 , l17 , g. s. , armandroff , t. e. & norris , j. e. 1992 , , 104 , 154 , g. s. & hatzidimitriou , d. 1998 , , 115 , 1934 , f. , piotto , g. , cassisi , s. , busso , g. , recio - blanco , a. , salaris , m. , aparicio , a. , & rosenberg , a. 2005 , , 130 , 116 , a. e. , walker , a. r. , hodge , p. w. , mateo , m. , olszewski , e. w. , schommer , r. a. , & suntzeff , n. b. 2001 , , 562 , 303 , a. , chaboyer , b. , jevremovi , d. , baron , e. , ferguson , j. w. , sarajedini , a. , & anderson , j. 2007 , , 134 , 376 , i. , ferraro , f. r. , fusi pecci , f. , corsi , c. e. & buonanno , r. 1995 , , 275 , 1057 , j. p. 2002 , , 123 , 404 , f. , bellazzini , m. , cacciari , c. & ferraro f. r. 1995 , , 110 , 1664 , f. , bellazzini , m. , ferraro , f. r. , buonanno , r. , & corsi , c. e. 1996 , formation of the galactic halo ... inside and out , asp conf . morrison , h. l. & sarajedini , a. , 221 , c. , zoccali , m. , & aparicio , a. 2005 , , 43 , 387 , l. , & salaris , m. 2001 , , 323 , 109 , l. , bressan , a. , bertelli , g. , & chiosi , c. 2000 , , 141 , 371 , j. a. 1975 , , 87 , 641 , e. k. 1997 , reviews in modern astronomy , 10 , 27 grebel , e. k. 1999 , the stellar content of the local group , iau symp . 192 , eds . p. whitelock & r. cannon ( san francisco : asp ) , 17 , e. k. 2001 , astrophysics and space science supplement , 277 , 231 , e. k. , & gallagher , j. s. , iii 2004 , , 610 , l89 , d. , et al . 2001 , , 122 , 3092 , w. e. 1996 , , 112 , 1487 , e. v. , clementini , g. , rizzi , l. , momany , y. , saviane , i. , & di fabrizio , l. 2001 , , 562 , l39 , v. , franois , p. , spite , m. , primas , f. , & spite , f. 2000 , , 364 , l19 , i. j. , & faulkner , j. 1968 , , 153 , 101 , j. a. bolte , m. , stetson , p. b. , hesser , j. e. , & somerville , r. s. 1999 , , 527 , 199 , j. a. , bolte , m. , hesser , j. e. , ivans , i. i. , & stetson , p. b. 2004 , abundances in lmc and smc globular clusters , eds . a. mcwilliam & m. rauch ( pasadena : carnegie observatories ) , j. a. , ivans , i. i. , & stetson , p. b. 2006 , , 640 , 801 , a. , grebel , e. k. , gilmore , g. f. , wyse , r. f. g. , kleyna , j. t. , harbeck , d. , wilkinson , m. i. , & evans , n. w. 2007 , , submitted , a. m. , fruchter , a. s. , hook , r. n. , & hack , w. 2002 , in _ hubble _ after the installation of the acs and the nicmos cooling system , eds . s. arribas , a. koekemoer & b. whitmore ( baltimore : stsci ) , 337 , l. m. , & chaboyer , b. 2003 , science , 299 , 65 , y .- w . , demarque , p. , & zinn , r. 1994 , , 423 , 248 , a. d. & gilmore , g. f. 2004 , , 352 , 153 makarova , l. n. , et al . 2002 , , 396 , 473 , k. j. , sarajedini , a. , & frenck , r. s. 1998 , , 116 , 2395 , b. , diemand , j. , madau , p. , zemp , m. , & stadel , j. 2006 , , 368 , 563 , j. e. 1994 , , 422 , 158 , k. a. g. , hodge , p. w. , mateo , m. , olszewski , e. w. , schommer , r. a. , suntzeff , n. b. , & walker , a. r. 1998 , , 300 , 665 , e. w. , suntzeff , n. b. & mateo , m. 1996 , , 34 , 511 , a. , cassisi , s. , salaris , m. , & castelli , f. 2004 , , 612 , 168 , g. , bedin , l. r. , anderson , a. , king , i. r. , cassisi , s. , milone , a. p. , villanova , s. , pietrinferni , a. & renzini , a. 2007 , , 661 , l53 , b. j. , armandroff , t. e. , jacoby , g. h. , & da costa , g. s. 2005 , , 129 , 2232 , b. j. , venn , k. a. , & irwin , m. i. 2005 , , 130 , 2140 , k. u. , & bahcall , j. n. 1985 , , 59 , 63 , a. , saviane , i. , piotto , g. , & aparicio , a. 1999 , , 118 , 2306 , g. a. , hesser , j. e. , stetson , p. b. , mateo , m. , simard , l. , bolte , m. , friel , e. d. , & copin , y. 1997 , , 109 , 883 , m. , & girardi , l. 2002 , , 337 , 332 , m. , & weiss , a. 2002 , , 388 , 492 , a. 1997 , , 113 , 682 , a. , geisler , d. , harding , p. , & schommer , r. 1998 , , 508 , l37 , a. , et al . 2002 , , 567 , 915 , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 , m. m. , fall , s. m. , rich , r. m. , & zurek , d. 1998 , , 508 , 570 , m. d. , ct , p. , & sargent , w. l. w. 2001 , , 548 , 592 , m. , et al . 2005 , , 117 , 1049 , j. , carney , b. w. , gieren , w. p. , fouqu , p. , latham , d. w. , & fry , a. m. 2004 , , 415 , 531 , l. l. , da costa , g. s. & mould , j. r. 1985 , , 298 , 544 , a. d. 1958 , , 118 , 117 , a. 1998 , acta astronomica , 48 , 383 , d. a. 2000 , , 129 , 315 , d. a. , swenson , f. j. , rogers , f. j. , iglesias , c. a. , & alexander , d. r. 2000 , , 532 , 430 , a. r. 1989 , , 101 , 570 , a. r. 1991 , the magellanic clouds , iau symp . 148 , eds . r. haynes & d. milne ( los angeles : asp ) , 307 , a. r. 1992 , , 390 , l81 , r. , & west , m. j. 1984 , , 55 , 45 zinn , r. 1993 , the globular cluster - galaxy connection , asp conf . smith , g. h. & brodie , j. p. ( san francisco : asp ) , 302

x - ray binaries ( xrbs ) are probes of both star formation and stellar mass , but more importantly remain one of the only direct tracers of the compact object population . to investigate the xrb population in m31 , we utilized all 121 publicly available observations of m31 totalling over 1 ms from _ chandra s _ acis instrument . we studied 83 star clusters in the bulge using the year 1 star cluster catalogue from the panchromatic hubble andromeda treasury survey . we found 15 unique star clusters that matched to 17 x - ray point sources within 1 ( 3.8 pc ) . this population is composed predominantly of globular cluster low - mass xrbs , with one previously unidentified star cluster x - ray source . star clusters that were brighter and more compact preferentially hosted an x - ray source . specifically , logistic regression showed that the f475w magnitude was the most important predictor followed by the effective radius , while color ( f475wf814w ) was not statistically significant . we also completed a matching analysis of 1566 h ii regions and found 10 unique matches to 9 x - ray point sources within 3 ( 11 pc ) . the h ii regions hosting x - ray point sources were on average more compact than unmatched h ii regions , but logistic regression concluded that neither the radius nor h luminosity was a significant predictor . four matches have no previous classification and thus are high - mass xrb candidates . a stacking analysis of both star clusters and h ii regions resulted in non - detections , giving typical upper limits of erg s , which probes the quiescent xrb regime .

1406.4121

compact objects are the end - states of the evolution of massive stars and as such are signposts of the star formation history of a galaxy . x - ray binaries ( xrbs ) consist of a compact object , either a neutron star or black hole , which accretes matter from a companion star . active xrbs have luminosities of @xmath410@xmath5 erg s@xmath3 ( those @xmath6@xmath7 erg s@xmath3 are classified as ultra - luminous x - ray sources ) while those in quiescence are @xmath810@xmath9 erg s@xmath3 . xrbs are classified into two main categories based on the mass of the companion star : low - mass ( lmxb ) and high - mass ( hmxb ) @xcite . lmxbs accrete matter via roche lobe overflow while hmxbs transfer mass predominantly by bondi - hoyle ( wind ) accretion @xcite . lmxb formation is more efficient in globular clusters ( gcs ) than in the field of a galaxy due to the higher stellar densities in gcs @xcite . in the milky way , lmxb formation is approximately two orders of magnitude more efficient in gcs @xcite . conversely , hmxbs are associated with star - forming regions ( ob associations , h ii regions , and infrared - bright dusty regions ) as opposed to star clusters @xcite . due to their evolutionary timescales , lmxbs trace the stellar mass of a galaxy @xcite while hmxbs probe the star formation rate @xcite within the past @xmath4100 myr @xcite . the milky way s xrbs have been studied extensively in the galactic center , star clusters , and the field ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? faint x - ray sources ( quiescent xrbs , cataclysmic variables , millisecond pulsars , etc . ) in the milky way have been studied to 10@xmath10 erg s@xmath3 and lower @xcite as a result of _ chandra s _ subarcsecond resolution and optical counterpart identifications with the _ hubble space telescope _ ( _ hst _ ) . due to dust obscuration , we can only probe x - ray sources in the plane of the disk out to @xmath118 kpc from the solar system . this hinders our ability to obtain a complete sample of the galaxy s xrb population . fortunately , m31 provides a convenient nearby analogue that allows us to perform an analysis of the faint x - ray population in a large galaxy . m31 s xrb population has been studied numerous times over the past decades with various x - ray telescopes ( @xmath12 ( _ xmm - newton _ ) , @xmath13 ( _ rosat _ ) , @xmath14 ) . however , only the spatial resolution of _ chandra _ allows us to study low - luminosity x - ray sources by separating them from the diffuse emission in the disk and crowding in the bulge . the lmxbs in m31 have been extensively studied by various groups @xcite that report on x - ray luminosity functions , spatial distributions , variability , and spectral analysis . @xcite used _ xmm - newton _ to study 80% of the 416 confirmed gcs in m31 @xcite . they found 41 gcs associated with x - ray sources along with an additional 4 gcs identified with _ chandra _ and _ rosat _ , for a total of 11% of gcs with x - ray sources . s gcs with lmxbs were found to be brighter , redder ( metal - rich ) , and more compact than gcs without lmxbs , in agreement with other extragalactic xrb studies @xcite . @xcite predicted a large number of faint transient sources in the bulge of m31 , similar to the numerous faint x - ray sources ( accreting millisecond pulsars ) found in the galactic center @xcite . a recent study by @xcite with _ xmm - newton _ covered all of m31 for the first time and reported on the x - ray source population compared to all previous studies . the survey catalogued 1897 point sources ( 914 new ) and reached a limit of @xmath410@xmath15 erg s@xmath3 in the @xmath16 kev energy band . sources were classified / identified by x - ray hardness ratios , spatial extent and distribution , cross - correlations in other wavelengths , and variability ( using _ chandra _ and _ rosat _ observations ) . they found 10 field lmxbs and 26 field lmxb candidates by analyzing long - term x - ray source variability . in addition , another 36 lmxbs were associated with gcs and 17 lmxb candidates were associated with gc candidates . no hmxbs were identified - in fact none have been confirmed in m31 to date - although 2 new candidates were proposed in addition to the 18 candidates from @xcite . the comprehensive x - ray population analysis by @xcite has classified hundreds of sources and shed light on their long - term variability and spatial / flux distributions . however , they pointed out that @xmath465% of their point sources are classified as `` hard '' , meaning that they have no optical identification and their x - ray colors are ambiguous : they could be xrbs , supernova remnants , or background active galaxies . @xcite have attempted to address the issue of unclassified x - ray point sources by identifying them using x - ray data alone . they used structure functions , which estimate the mean intensity deviation of data over a time interval . by determining structure functions for each source they constrained its variability and were able to differentiate xrbs from active galactic nuclei . they found 220 x - ray sources above @xmath1710@xmath15 erg s@xmath3 with significantly more variability than expected in the structure function for active galactic nuclei . based on their analysis they classified these sources as xrbs , with an additional 30 xrb sources for a total of 250 probable xrbs ( 200 new ) out of their sample of 528 x - ray sources in the central 20 of m31 . low - luminosity xrbs are more variable than luminous xrbs and thus are well - suited to this classification technique given sufficient signal - to - noise data . one of the most powerful methods to confidently classify an x - ray source is identifying an optical counterpart . however , in crowded and extragalactic fields this generally requires the exquisite spatial resolution of _ hst_. the ability to observe a large fraction of identified x - ray sources is limited since _ _ has both high demand and a small field of view . the panchromatic @xmath18 andromeda treasury ( phat ) survey @xcite is a multicycle _ hst _ program to map one - third of m31 s disk and provide an unprecedented catalogue of star clusters and stars . while it does not cover all of m31 , it provides the best possible resource for determining optical counterparts to x - ray sources on such a large scale . m31 s bulge has been monitored extensively with _ chandra _ , whose 0.5 point spread function ( psf ) is superior to the 6 provided by _ xmm - newton_. to probe low - luminosity x - ray sources in m31 below the level achieved by @xcite , we will use the stacking method developed by @xcite and @xcite . by stacking star cluster positions obtained from the phat survey and h ii region positions ( see section [ sec : opt ] ) in a _ chandra _ x - ray image of m31 , we can determine , on average , if these objects host faint point sources . young star clusters and h ii regions will probe the hmxb population while gcs will probe the lmxbs . our study is not focused on the field population of xrbs . stacking analyses enable us to study the effect that metallicity has on faint x - ray sources by separating metal - rich and metal - poor classes of gcs . stacking is a unique way to investigate whether hmxbs in m31 exist but are below our current detection limits . by pushing extragalactic xrb luminosities into the quiescent regime ( @xmath1910@xmath9 erg s@xmath3 ) , we can reveal undetected populations that will help constrain binary evolution models ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we adopt a distance to m31 of 776 @xmath20 18 kpc as in ( * ? ? ? * hereafter j12 ) , which corresponds to a linear scale of 3.8 pc arcsecond@xmath3 . in section [ sec : obs ] , we describe the optical and x - ray observations and reduction procedure for the x - ray data . in section [ sec : match ] , we report results from matching star clusters and h ii regions to x - ray point sources . in section [ sec : stacking ] , we present results from an x - ray stacking analysis of star clusters and h ii regions . in sections [ sec : discuss ] and [ sec : summary ] we discuss our results and summarize our main conclusions . _ chandra _ has observed m31 numerous times , but the small field of view ( compared to _ xmm - newton _ ) of @xmath21 advanced ccd imaging spectrometer ( acis ) instrument ( @xmath22 for acis - s and @xmath23 for acis - i ) combined with the proximity of m31 require many pointings to cover an appreciable area of the galaxy . we use all 121 publicly available acis - s and acis - i observations in m31 for our matching and stacking analyses of star clusters and h ii regions , having a total exposure time over 1 ms . due to the degradation of the psf at large off - axis angles , we only use data from the s3 chip from the acis - s observations . the observations are listed in tables [ tab : acis - s ] and [ tab : acis - i ] and their fields of view shown in figure [ fig : acis - fov ] . cccccccc & 2000 - 06 - 01 & 10.688 & 41.27877 & 5129 & faint & 235678 & 87.5 + & 2000 - 07 - 02 & 10.68311 & 41.27876 & 5284 & faint & 235678 & 108.7 + & 2000 - 09 - 21 & 10.65816 & 40.87837 & 6002 & faint & 235678 & 162.9 + cccccccc & 1999 - 10 - 13 & 10.67806 & 41.26989 & 11890 & faint & 012367 & 193.0 + & 1999 - 12 - 11 & 10.68273 & 41.26395 & 4185 & faint & 01236 & 274.0 + & 1999 - 12 - 27 & 10.68409 & 41.2637 & 4189 & faint & 01236 & 285.4 + each observation was reprocessed with the _ chandra _ interactive analysis of observations ( _ ciao _ ) tools package version 4.5 @xcite and the _ chandra _ calibration database ( _ caldb _ ) version 4.5.5 @xcite . we started with the level-1 events file and produced a bad pixel file that identified observation - specific bad pixels , hot pixels , bright bias pixels , and afterglow events using @xmath24 . since we are studying faint sources , we ran @xmath25 to eliminate cosmic ray afterglows with only a few events . the level-1 events file was updated and filtered using the standard ( asca ) grades ( @xmath26 ) , status bits ( @xmath27 ) , good time intervals , charge transfer inefficiency ( cti ) , time - dependent gain , and pulse height with @xmath28 . we used the energy - dependent subpixel event repositioning algorithm to improve astrometry and , where necessary , set the vfaint option in @xmath28 . we then created merged event files for all 26 acis - s observations and all 95 acis - i observations with @xmath29 to obtain fully - reduced images in the soft , hard , and full - bands corresponding to energy ranges of @xmath30 kev , @xmath31 kev , and @xmath32 kev respectively . we produced exposure - corrected images ( in units of photons @xmath33 s@xmath3 pixel@xmath3 ) with the @xmath34 tool using a binsize of 1 to maintain the native resolution . weighted spectrum files ( corresponding to the soft , hard , and full energy bands ) were used to calculate instrument maps . the merged acis - s and acis - i images have dimensions ) large parts of the image are devoid of data . ] of 1.7@xmath35 @xmath36 1.7@xmath35 and 0.8@xmath35 @xmath36 0.85@xmath35 and correspond to exposures of @xmath4548 ks and @xmath4 510 ks respectively . the plate scale of the _ chandra _ images is 0.5 arcsecond pixel@xmath3 , corresponding to 1.9 pc pixel@xmath3 . _ chandra s _ absolute astrometry is @xmath40.6 at the 90% confidence level @xcite and has spatial resolution that ranges from 1 on - axis to 4 at 4 off - axis for 1.5 kev at 90% encircled energy fraction for acis . given that we have produced deep images of m31 s bulge , there may be new point sources that have not been previously identified . we used @xmath37 on the full , hard , and soft - band images with scales series from 1 to 8 we detected an additional 60 x - ray sources . however , this does not change the results of our matching analysis ( see section [ sec : res - matching ] ) . ] of 1 , 2 , 4 , 8 , and 16 as recommended by the chandra x - ray center to have a wide range of source sizes . this ensures that the variation of the psf at different off - axis angles is accounted for . exposure maps were also input to reduce false positives . the sigthresh parameter , which determines how many false detections are allowed , should not be larger than the inverse of the number of pixels in the image being analyzed . therefore it was set to @xmath38 for the acis - i merged image and @xmath39 for the acis - s merged image . these numbers are small since the acis - i merged image ( 0.68 deg@xmath40 ) covers an area larger than the bulge and the acis - s merged image ( 2.89 deg@xmath40 ) covers the majority of the galaxy even though only the nucleus and various pointings have data ( see figure [ fig : acis - fov ] and section [ sec : xraydata ] ) . in any case , setting the sigthresh parameter to the default value of @xmath41 did not change the number of x - ray sources matched to star clusters or h ii regions . we obtained source lists for the full , hard , and soft - band merged acis - i and acis - s images . from the six source lists we collated our results into a master source list with the @xmath42 tool from the tools for acis review and analysis ( tara ) package . we found a total of 1033 independent x - ray point sources across all 3 bands ( a source was not necessarily detected in more than one band ) for the acis - i merged observations and acis - s merged observations . we report the number of sources detected in each energy band for both acis - i and acis - s in table [ tab : xray - detect ] . while the @xmath42 tool filtered x - ray sources based on their positional uncertainties from @xmath37 ( e.g. retains only one source position if detected in all 3 energy bands ) , @xmath42 does not consolidate pairs of extremely close @xmath37 sources detected only in one band . we performed an internal match of our 1033 sources to remove sources within @xmath43 of another source . this found a further 112 matches that we removed to obtain 921 effective x - ray sources . c c c c c c acis - i & 435 & 344 & 388 & 1167 & + acis - s & 353 & 217 & 334 & 904 & + we input our list of sources to the acis extract package @xcite along with exposure maps ( calculated using weighted spectrum files ) so that each observation could be individually analyzed . acis extract then merged all observations ( acis - i and acis - s ) and produced the relevant source properties and statistics , such as x - ray luminosity . acis extract aligns each observation to a published astrometric catalogue and thus gives precision of @xmath40.1 for most point sources . we required acis extract s prob_no_source statistic for each x - ray source to be @xmath44 , except where by - eye inspection of the images showed source crowding that biased this statistic to lower values ( i.e. 5 sources included that were not @xmath44 ) . the majority of our sources ( @xmath475% ) are located in the bulge , where the _ chandra _ observations are deeper compared with the rest of the galaxy . we only processed ( through acis extract ) the 26 x - ray sources that matched to star clusters and h ii regions ( see section [ sec : res - matching ] for details ) . the luminosity range of these 26 x - ray sources was @xmath45 erg s@xmath3 to @xmath46 erg s@xmath3 . there is no meaningful , single value for the completeness limit since the exposure time varies significantly across the usable field of view . the deepest exposure of any 1 pixel reaches @xmath47 ks in both the merged acis - i and merged acis - s images ; this occurs in the nucleus and corresponds to a sensitivity limit of @xmath48 erg s@xmath3 . the complete x - ray point source catalogue will follow in a subsequent paper ( vulic et al . , in prep . ) . we use the star cluster catalogue compiled by from the phat survey . the survey operates using the advanced camera for surveys ( filters f475w and f814w ) as well as the wide field camera 3 ( filters f275w , f336w , f110w , and f160w ) on the _ hst_. the observational technique used by the survey is detailed in @xcite and comprises 23 areas known as `` bricks '' , with each one consisting of 18 different fields of view . the survey produces photometry with a signal - to - noise ratio of 4 at @xmath49 = 25.1 , @xmath50 = 24.9 , @xmath51 = 27.9 , @xmath52 = 27.1 , @xmath53 = 25.5 , and @xmath54 = 24.6 for single pointings in the uncrowded outer disk . the phat year 1 star cluster catalogue was compiled from bricks 1 , 9 , 15 , 21 , and parts of 17 and 23 . due to the limited _ chandra _ data we only utilize brick 1 and a small part of brick 9 ( proposal 12058 , pi : julianne dalcanton ) . even though figure [ fig : acis - fov ] shows numerous phat bricks coincident with _ chandra _ data , the only bricks with published star cluster catalogues that overlap are 1 and 9 . there are 83 star clusters in the field of view of the _ chandra _ observations that we use for our matching and stacking analyses ( indicated by crosses in figure [ fig : acis - fov ] ) . have corrected the absolute astrometry to agree with the 2mass reference system within @xmath460 mas . the psf ( full width at half maximum ) for f475w is @xmath55 @xcite . in addition , we use the new catalogue of h ii regions in m31 compiled by @xcite . m31 s entire disk is covered out to @xmath424 kpc from the center based on observations with the mayall 4 m telescope as part of the nearby galaxies survey @xcite . a total of 3691 h ii regions down to l@xmath56 erg s@xmath3 are identified in the 2.2 deg@xmath40 mosaic , with 1566 in the field of view of the _ chandra _ data ( indicated by circles in figure [ fig : acis - fov ] ) . the average psf ( full width at half maximum ) of the survey was 1 , which is generally better than _ chandra s _ range of @xmath57 but not as good as the @xmath55 for f475w from the phat survey . the astrometric calibration of the h ii region catalogue is good to @xmath58 , better than _ chandra s _ 0.6 but not as precise as the 60 mas for phat . cc prior to stacking star clusters and h ii regions we must exclude those with identified x - ray point sources . stacking probes faint x - ray point sources by co - adding star clusters ( or h ii regions ) that are undetected in x - rays . in this way one hopes to detect a signal in the source aperture ( centered on the position of the star cluster or h ii region ) above the background ( see section [ sec : stacking ] for more details ) . therefore , including a star cluster ( or h ii region ) with a detected x - ray point source in the complete stack of all undetected star clusters ( or h ii regions ) would defeat the purpose of detecting a faint x - ray source in the stack . figure [ fig : acis - fov ] shows where the star clusters and h ii regions we analyzed are located in m31 . the 83 star clusters in the field of view of _ chandra _ data are located mainly in the bulge ( phat brick 1 ) , while a handful are in the disk ( phat brick 9 ) ; the 1566 h ii regions are spread throughout the disk . here we summarize our matching results , beginning with star clusters . we used topcat @xcite to complete a matching analysis with the 83 star clusters and 921 x - ray point sources . a total of 15 star clusters were matched to 17 unique x - ray point sources within 1@xmath59 . the matching results are shown in table [ tab : scxrmatches ] . two star clusters ( phat ids 594 and 1374 ) each had two x - ray point source matches . figure [ fig : doubles ] shows these star clusters in the f475w band image from phat . the clusters are plausible double xrb hosts since the x - ray sources are separated by @xmath60.5 . even though we used a 1 matching radius , only 2 star clusters ( phat ids 1403 and 1415 ) had a separation @xmath60.5 from their matched x - ray point source . no additional clusters were matched when increasing the matching radius to @xmath610 . only one star cluster ( phat i d 1374 ) has not previously been associated with an x - ray source . in addition , one star cluster ( phat i d 1375 ) that was previously identified as a gc candidate x - ray source by @xcite was not associated with an x - ray source in our analysis . upon further inspection of our acis - i image it seemed that @xmath37 did not identify what was likely an x - ray source . to address the chance coincidence probability of our matches we created 25 source lists from our original x - ray catalogue that offset each source by @xmath20 5@xmath010@xmath59 @xcite . a false - match rate of @xmath42% was recovered ( @xmath601 false match ) . we also used the cumulative number counts of @xcite to estimate the expected background sources in the 83 star cluster regions of radius 1 each . in equation [ eq : bkgrnd ] , @xmath61 is the flux in erg s@xmath3 @xmath33 and @xmath62 is the number of background sources per deg@xmath40 above that flux , where we used the faintest x - ray point source from our matches ( @xmath63 erg s@xmath3 @xmath33 ) . this gave @xmath64 erg s@xmath3 @xmath33 ) = @xmath65 expected background sources , indicating that our matches are likely authentic . we also completed cone searches in the nasa / ipac extragalactic database ( ned ) within @xmath66 around each of the 17 x - ray point sources that were matched to star clusters and found no background galaxies . @xmath67 c c c c c c c c c c c c c c c c 594 & 10.856666 & 41.260284 & 30 & 10.856581 & 41.260256 & 0.00 & 1.00 & 4.3 & 0.4 & 33.32 & 1 & 2 & 0.25 & gc & ho + 594 & 10.856666 & 41.260284 & 31 & 10.856984 & 41.260252 & 0.00 & 1.00 & 4.3 & 0.4 & 33.37 & 1 & 2 & 0.87 & gc & ho + 1374 & 10.711676 & 41.285381 & 16 & 10.711668 & 41.285453 & 0.15 & 0.11 & 1.6 & 0.39 & 34.24 & 2 & 2 & 0.26 & & + 1374 & 10.711676 & 41.285381 & 17 & 10.711715 & 41.285303 & 0.25 & 0.70 & 1.6 & 0.39 & 27.88 & 2 & 2 & 0.30 & & + 612 & 10.749429 & 41.268238 & 21 & 10.749429 & 41.268268 & 0.01 & 0.00 & 3.5 & 0.9 & 37.61 & & & 0.11 & gc / bh ? & zh / ba + 616 & 10.766111 & 41.301309 & 24 & 10.766068 & 41.301343 & 0.01 & 0.00 & 2.0 & 0.39 & 37.42 & & & 0.17 & gc / bh ? & zh / ba + 642 & 10.794234 & 41.247575 & 27 & 10.794231 & 41.2476 & 0.02 & 0.00 & 3.1 & 0.39 & 37.82 & & & 0.09 & gc / bh ? & zh / ba + 663 & 10.762226 & 41.25627 & 23 & 10.7622 & 41.256281 & 0.01 & 0.00 & 3.2 & 0.4 & 37.37 & & & 0.08 & gc / bh ? & vo / ba + 682 & 10.814255 & 41.190284 & 28 & 10.814253 & 41.190228 & 0.13 & 0.00 & 7.0 & 0.9 & 35.98 & & & 0.20 & gc & vo + 1377 & 10.731713 & 41.309748 & 19 & 10.731691 & 41.309742 & 0.02 & 0.00 & 3.1 & 0.39 & 36.39 & & & 0.06 & fgstar / gc & vo / ko + 1381 & 10.692025 & 41.293409 & 13 & 10.692013 & 41.293426 & 0.03 & 0.00 & 1.5 & 0.39 & 35.91 & & & 0.07 & gc & ko + 1386 & 10.63027 & 41.327441 & 6 & 10.630254 & 41.327479 & 0.01 & 0.00 & 4.2 & 0.4 & 37.24 & & & 0.14 & gc & vo + 1395 & 10.672678 & 41.256623 & 10 & 10.672652 & 41.256609 & 0.01 & 0.00 & 1.0 & 0.39 & 36.51 & & & 0.09 & gc & vo + 1401 & 10.638559 & 41.295108 & 7 & 10.638544 & 41.295104 & 0.02 & 0.00 & 2.6 & 0.9 & 36.31 & & & 0.04 & gc & ho + 1403 & 10.623894 & 41.2993 & 4 & 10.624093 & 41.299152 & 0.39 & 0.00 & 3.5 & 0.9 & 34.79 & & & 0.76 & gc & zh + 1415 & 10.68207 & 41.211915 & 11 & 10.681932 & 41.211796 & 0.28 & 0.07 & 3.7 & 0.9 & 33.82 & & & 0.57 & gc & ho + 1444 & 10.58984 & 41.238797 & 3 & 10.589851 & 41.238801 & 0.20 & 0.00 & 4.2 & 0.4 & 35.85 & & & 0.03 & irsrc & ka + our results show that @xmath418% ( 15/83 ) of star clusters were matched to an x - ray point source within 1 . in a survey of eight fornax and virgo elliptical galaxies ( limiting luminosities typically few @xmath68 erg s@xmath3 ) @xcite found that 408 of 5904 ( @xmath47% ) gcs are associated with an x - ray source within 0.5 . similarly , a survey of 11 virgo cluster early - type galaxies ( limiting luminosities typically few @xmath68 erg s@xmath3 ) by @xcite found that 270 of 6488 ( @xmath44% ) gcs were associated with an x - ray source within 1 . our analysis was restricted to the bulge / central region of m31 and thus the concentration of x - ray point sources is larger than throughout the rest of the galaxy ( as @xcite found for the milky way ) . since lmxbs trace stellar mass @xcite , and stellar mass density is larger in the bulge , we would therefore expect to have a higher percentage of matches ( lmxbs ) compared to the value for the entire galaxy . additionally , our luminosity limit is a factor of @xmath69 fainter than previous surveys and therefore increases our percentage of matches . to more directly compare different galaxies , we compare the fraction ( and number ) of gcs with an lmxb versus those without above a given luminosity limit from our survey with those of @xcite and @xcite in table [ tab : xrb - compare ] . we compared our results to individual galaxies from these two surveys since most galaxies they studied had varying limiting @xmath70 ( i.e. their complete sample is not completeness - corrected ) . the fractions are comparable ( e.g. 6% for our work and 6.5% for @xcite ) at low-@xmath70 limits ( @xmath71 erg s@xmath3 ) . for @xmath72 erg s@xmath3 a fraction of 3.6% found in our study is much lower than the 8.3% of @xcite . in general , we and @xcite find a difference in the fraction that is lower by a factor of @xmath73 than @xcite for @xmath72 erg s@xmath3 , which could be attributed to small - number statistics . when analyzing the same galaxies as @xcite , @xcite found more gc - lmxbs as a result of deeper x - ray observations over an extended period of time , which also allowed for the identification of previously quiescent sources that went into outburst . of the 1566 h ii regions in the field of view , we found 10 that matched to 9 unique x - ray point sources within 3 ( @xmath411.2 pc , their average radius ) . the results are shown in table [ tab : htwoxrmatches ] . the false - match rate for h ii regions was @xmath435% , which is higher due to the increased matching radius and their larger physical size compared to star clusters . for each of the 9 x - ray point sources that were matched to h ii regions , we completed ned cone searches using a radius of 30 . only one source ( ae i d 38 , 2.3 from an h ii region ; see table [ tab : htwoxrmatches ] ) had any nearby entries in ned . these were a nova at a distance of 0.4 , an h ii region at a distance of 3.2 , and a background galaxy at a distance of 7 . we summarize our matching results and state completeness limits in table [ tab : match - results ] . c c c c c c c[!h ] this work & m31 & 6% ( 5 ) & 6% ( 5 ) & 6% ( 5 ) & 3.6% ( 3 ) & 2.4% ( 2 ) + + & ngc 4365 , ngc 4649 & & & & 8.3% ( 209 ) & + & ngc 4472 & & & & & 9.2% ( 70 ) + + & ngc 4374 , ngc 4697 & 6.5% ( 52 ) & & & & + & ngc 4365 , ngc 4406 & & 3.8% ( 49 ) & & & + & ngc 4486 , ngc 4552 & & & 3.8% ( 80 ) & & + & ngc 4526 & & & & & 2.9% ( 7 ) + c c c c c c c c c c c c c c c c c c 2165 & 10.8667 & 41.3092 & 10.99 & 36.87 & 33 & 10.866947 & 41.309129 & 0.14 & 0.00 & 7.8 & 0.5 & 35.94 & 1 & 2 & 0.7 & snr & wi / st + 2164 & 10.8667 & 41.3091 & 10.43 & 36.94 & 33 & 10.866947 & 41.309129 & 0.14 & 0.00 & 7.8 & 0.5 & 35.94 & 1 & 2 & 0.7 & snr & wi / st + 1686 & 10.6429 & 40.9527 & 13.33 & 35.31 & 8 & 10.643727 & 40.95233 & 0.46 & 0.00 & 4.2 & 0.9 & 35.87 & & & 2.6 & gc & pe / zh / st + 1798 & 10.7046 & 41.4017 & 7.83 & 36.72 & 14 & 10.704479 & 41.402351 & 0.28 & 0.00 & 7.9 & 0.9 & 35.53 & & & 2.4 & snr & st + 1805 & 10.7083 & 40.8914 & 8.62 & 34.44 & 15 & 10.707519 & 40.891337 & 0.59 & 0.00 & 3.7 & 0.9 & 34.86 & & & 2.1 & & + 1831 & 10.7233 & 41.4307 & 9.48 & 36.91 & 18 & 10.722953 & 41.430795 & 0.31 & 0.00 & 9.3 & 0.9 & 35.99 & & & 1.0 & snr & st / vo + 2080 & 10.8287 & 41.338 & 7.83 & 35.64 & 29 & 10.828635 & 41.338083 & 0.25 & 0.00 & 7.2 & 0.9 & 35.61 & & & 0.3 & sss ? & ho / st + 2241 & 10.885 & 41.3499 & 7.83 & 35.11 & 34 & 10.884631 & 41.349752 & 1.09 & 0.23 & 8.6 & 0.9 & 35.02 & & & 1.1 & & + 2471 & 10.9721 & 41.2017 & 10.43 & 36.03 & 36 & 10.972348 & 41.200971 & 0.62 & 0.00 & 6.9 & 0.9 & 35.44 & & & 2.7 & & + 3005 & 11.1367 & 41.4225 & 7.83 & 37.52 & 38 & 11.13587 & 41.422606 & 0.76 & 0.00 & 3.9 & 0.9 & 35.11 & & & 2.3 & & + c c c c phat star clusters & m@xmath74 & m@xmath75 & 15 & 17 + h ii regions & @xmath76 erg s@xmath3 & 10 & 9 + an x - ray color - color plot of the x - ray point sources matched to star clusters and h ii regions is shown in figure [ fig : xraycc ] , with the x - ray color classification scheme of @xcite overlaid . soft and hard colors are defined by the soft ( s , @xmath77 kev ) , medium ( m , @xmath78 kev ) , hard ( h , @xmath31 kev ) , and total ( t , @xmath32 kev ) band energies as @xmath79 and @xmath80 respectively . x - ray photometry was completed separately for each observation for each x - ray point source . the photometry results for each observation were then merged for each x - ray point source in each of the energy bands to determine colors . due to the variation in the spectral states of xrbs , the same source can end up in different areas of the diagram . also , only flux measurements account for the changing effective area of _ chandra _ , whereas counts do not . since the observations we used were spread over 13 years , using counts introduces another source of uncertainty . absorption also moves x - ray sources into different regions of the diagram and thus can result in misclassifications . most of the sources are clustered between the xrb and background source regions . however , given the low false match rate for star clusters and previous confirmations as gc sources it is likely that these are mostly xrbs . the x - ray source with ae i d # 3 has a luminosity typical of bright xrbs and was previously classified to match within 0.1 of an infrared source by @xcite . the remaining x - ray sources were already identified as gcs by various studies ( see table [ tab : scxrmatches ] for refs . ) with many that are likely black hole xrbs @xcite . a color - color diagram of the 83 star clusters in the _ chandra _ coverage region is shown in figure [ fig : sccc ] . a theoretical evolutionary track from the simple stellar population models of @xcite with @xmath81 is overlaid after applying a total reddening ( external and internal ) of @xmath82 @xcite . the 15 star clusters correspond to the 17 x - ray point sources they were matched to from figure [ fig : xraycc ] ( results in table [ tab : scxrmatches ] ) . in figure [ fig : sccc ] most star clusters are @xmath831 gyr old as expected for an old bulge population . to determine whether our matches correspond to brighter and redder star clusters , we created the color - magnitude diagram shown in figure [ fig : cmd ] . it is evident that the matches are all brighter and redder than star clusters without a detected x - ray point source . this is in agreement with the studies of gc - lmxbs in both elliptical and spiral galaxies mentioned earlier @xcite . the average f475w magnitude of matched star clusters was @xmath84 compared to @xmath85 for unmatched clusters , while the average effective radius of matched clusters was @xmath86 pc compared to @xmath87 pc for unmatched clusters so we computed the standard error of the mean . ] . this confirms the trend that more luminous , compact star clusters preferentially host an x - ray source . further evidence is shown in the left panel of figure [ fig : matches ] , where it is clear that our matches are all above the 80% completeness limit ( m@xmath88 ) mentioned in table [ tab : match - results ] . figure [ fig : histograms ] shows the x - ray luminosity of point sources that were matched to star clusters plotted against the f475w magnitude and effective radius @xmath89 . the histograms show that our sample ( dark portion ) has preferentially more bright clusters than the complete year 1 sample ( black outline - mostly disk clusters ) , while no significant variations in @xmath89 between the samples is seen . specifically , the f475w magnitudes of our sample are skewed towards the bright end compared to the complete sample . we performed a kolmogorov - smirnov ( ks ) test on the two distributions , namely our sample and the complete year 1 sample of star clusters , for both the f475w magnitude and @xmath89 . we found that the probability that the two distributions are drawn from the same sample is @xmath90 for the f475w magnitude and 0.09 for @xmath89 , confirming that the f475w magnitudes of the two distributions are different . to assess the statistical significance of our results we used the @xmath91 statistical software @xcite to complete a logistic regression on our data . this allows us to assess the impact that a parameter has on the probability of a star cluster hosting an x - ray source . the generalized linear model ( @xmath92 ) in @xmath91 is best - suited for this and has the form shown in equation [ eq : glm ] , where @xmath93 is the probability of success averaged over @xmath94 trials ( outcome of 1 or 0 , i.e. matched or not ) and the @xmath95 s are the regression coefficients of the variables @xmath96 that we model . @xmath97 for star clusters we modelled 3 variables , specifically the effective radius @xmath89 , f475w magnitude , and color ( f475w@xmath0f814w ) . we only included the 83 star clusters that were in the field of view of the _ chandra _ data , since only these star clusters could possibly have been matched to an identified x - ray point source . the results of the logistic regression are shown in table [ tab : logit ] . c c c c c @xmath89 & @xmath020.7241 & 9.0953 & @xmath02.279 & 0.02269 + f475w & @xmath02.9285 & 1.0582 & @xmath02.767 & 0.00565 + f475w @xmath0 f814w & 0.5696 & 2.2257 & 0.256 & 0.79802 + radius & -0.9283 & 2.7518 & -0.337 & 0.736 + h@xmath1 luminosity & 13.9974 & 36.3963 & 0.385 & 0.701 + since a variable in the model is generally significant for a @xmath98-value pr(@xmath99 ) @xmath19 0.05 ( i.e. null hypothesis is rejected ) , only @xmath89 and the f475w magnitude are statistically significant . from table [ tab : logit ] the coefficient gives the @xmath100 odds increase of a match for a unit increase in the respective variable . therefore , a unit increase in @xmath89 results in a decrease of @xmath101 in the @xmath100 odds . however , because each variable has a different range of values , a unit increase does not have a 1:1 correspondence across variables . therefore the smallest @xmath98-value indicates which variable is most important to the outcome . from this we see that the f475w magnitude is more significant when determining if a star cluster hosts a bright x - ray source , followed by the effective radius . however , star cluster color was not important , at odds with the surveys of gc - lmxbs in spiral and elliptical galaxies mentioned above . these surveys studied entire galaxy populations , meaning that even if gc catalogues were incomplete their numbers were not radically biased towards metal - poor or metal - rich . our result could be a consequence of our sample since it was restricted mostly to bulge clusters , which are metal - rich ( see figure [ fig : cmd ] ) . indeed , a ks test between the f475w@xmath0f814w color of star clusters in our sample and the complete year 1 phat clusters gives a probability of @xmath102 that the two distributions are drawn from the same sample . this confirms that bulge clusters are more metal - rich than those in the disk . complete surveys of m31 s gcs by @xcite and @xcite showed that metal - rich star clusters were more likely to host an x - ray source , indicating our result is biased by our sample . the 10 h ii regions matched to x - ray sources have average radii and h@xmath1 luminosities of @xmath103 pc and @xmath104 erg s@xmath3 compared to averages of @xmath105 pc and @xmath106 erg s@xmath3 for unmatched h ii regions luminosities so we computed the standard error of the mean . ] . while h ii regions that were matched to x - ray point sources seem to favour those that are more compact , we can not claim a trend exists with h@xmath1 luminosity due to the large errors associated with their average values . the right panel of figure [ fig : matches ] shows the distribution of h ii region luminosity and radius , with matches crowding the compact end . however , h ii regions have a much wider range of radii than globular clusters and generally are not well approximated by circular apertures @xcite . these differences complicate comparisons between the populations . the limiting @xmath107 for h ii regions from table [ tab : match - results ] applies across the entire @xmath89 range ; those with larger radii naturally have a larger luminosity and are not biased by incompleteness . as for star clusters we also completed logistic regression on the 1566 h ii regions that were in the field of view of the _ chandra _ data . we modelled the radius and h@xmath1 luminosity of h ii regions to assess their impact on the probability of hosting an x - ray source , with the results shown in table [ tab : logit ] . from the @xmath98-values of both variables we can see that neither is significant in determining the presence of an x - ray source as both are @xmath108 . in the right panel of figure [ fig : matches ] it would seem that at least the most compact h ii regions were more likely to host an x - ray source . however , based on the distribution of the population , specifically that most h ii regions have small radii , this impression is not supported by our statistical analysis . cc cc to study the faint population of x - ray sources we performed a stacking analysis of the 68 unmatched star clusters and 1556 unmatched h ii regions in m31 . using the techniques of @xcite and @xcite we stacked 31 by 31 ( 45 by 45 ) pixel regions centered on star cluster ( h ii region ) positions in x - ray images of m31 . star cluster ( h ii region ) positions closer than 16 ( 23 ) pixels to the edge of an x - ray image were excluded since each image in the stack has to be complete ( i.e. can not have null data in any pixels ) . this left a total of 54 star clusters and 1386 h ii regions in each sample . since there are only a handful of star clusters and they are all older than 1 gyr we do not subdivide the population by color . however , because brighter and smaller clusters preferentially host x - ray sources we do filter by @xmath89 and f475w magnitude ( see table [ tab : stackvals ] ) to take advantage of these trends . we followed a similar method for h ii regions . there are two peaks in the luminosity function for h ii regions in m31 , at 10@xmath15 erg s@xmath3 and @xmath109 erg s@xmath3 for b stars and o stars respectively @xcite . these values were used to separate the population by luminosity when stacking . we stacked star clusters and h ii regions in the full ( @xmath32 kev ) , hard ( @xmath31 kev ) , and soft ( @xmath30 kev ) energy bands . sample stacked images of star clusters and h ii regions are shown in figure [ fig : stack ] . cc the net counts in the source aperture @xmath110 , where @xmath111 and @xmath112 represent the counts and area respectively for the source ( s ) and background ( b ) regions . the source region for star clusters in a stacked image is 1 ( 2 pixels ) in radius , which corresponds to the radius used to match star clusters to x - ray point sources . keeping the source aperture radius consistent with the matching radius ensures that we do not include any x - ray sources outside of the original matching radius in our stacked images . the background region is defined to be outside the 90% encircled energy radius of the source aperture , which for star clusters that are 4 off - axis from the _ chandra _ aimpoint is 10 pixels4 of the aimpoint for acis observations ( centered on the supermassive black hole ) . ] . h ii regions are more extended and also randomly distributed throughout the acis field of view compared to star clusters . average off - axis angles for h ii regions are not much larger than 8 and so we use a 20 pixel diameter for the background region ( corresponding to 90% encircled energy radius for sources 8 off - axis ) . a negative source aperture luminosity results when the background - subtracted source aperture counts are @xmath190 . this means that the background has a larger average value per pixel than the source aperture . we use poisson statistics ( method of @xcite ) to calculate our uncertainties . we computed the 3@xmath113-clipped mean @xmath114 ( @xmath115 ) in the background , where pixels with counts @xmath116 + 3@xmath113 were excluded . upper and lower limits for both the source and background regions were calculated using equations ( 9 ) and ( 14 ) from @xcite , using @xmath117 ( number of gaussian @xmath113 ) and confidence level parameters @xmath118 and @xmath119 . uncertainties in the net counts c@xmath120 were calculated by normalizing background values to those of the source aperture . luminosities for stacked images were computed from fluxes for details on exposure - corrected images . ] ( photons @xmath33 s@xmath3 ) using a power - law spectrum with @xmath121=1.7 for each energy band and assuming a distance of 776 kpc . uncertainties for luminosities were determined by multiplying the ratio of luminosity to counts in the source aperture by the uncertainties in counts . we use a larger size for our stacked images of h ii regions ( 45 pixels ) than star clusters ( 31 pixels ) since h ii regions have larger average radii . a caveat to our method is that our merged x - ray images of m31 have an inconsistent psf at any point in the complete acis - i or acis - s image . this is a result of observations overlapping at different off - axis angles . this compounds the problem of stacking since we already introduce the same type of effect by stacking objects at various off - axis angles . all stacking completed on star clusters and h ii regions resulted in non - detections . the results are displayed in table [ tab : stackvals ] and shown graphically in figure [ fig : limits ] . correcting for foreground absorption ( assuming a galactic @xmath122 value of @xmath123 @xmath33 @xcite and a power - law spectrum of @xmath124 ) would increase all our luminosities by @xmath125 . since we analyzed the bulge of m31 , which is comprised of an old stellar population and thus globular clusters , it is not surprising that our x - ray point sources are predominantly gc - lmxbs ( table [ tab : scxrmatches ] ) . when comparing the averages of star cluster colour , magnitude , and effective radius our results showed that redder , brighter , and compact star clusters were more likely to host x - ray sources however , a logistic regression analysis showed that the f475w magnitude is the most significant predictor followed by the effective radius . the f475w@xmath0f814w color ( an indicator of metallicity in old stellar populations ) was not a significant predictor of whether a cluster hosted an x - ray source . a statistical analysis using logistic regression is a more robust method than comparing averages and as such accurately represents trends in the data . @xcite found different trends when using _ xmm - newton _ observations of m31 to identify 45 gc - lmxbs , namely that the stellar collision rate ( proportional to stellar density and core radius ) , luminosity , and metallicity are all significant . however , @xcite came to similar conclusions to ours when using _ hst _ data to derive structural parameters for 29 gc - lmxbs , and complementing the sample with published values for a total of 41 gc - lmxbs and 65 non - lmxb gcs . they found the probability of a gc hosting an x - ray source increased with increasing collision rate and proximity to the galaxy center . metallicity was not as important a predictor but an increasing cluster mass at fixed collision rate decreased the probability of hosting an x - ray source ( although the authors stress this may be a sample selection effect ) . the latter result does not agree with our positive trend between gc magnitude ( which scales with mass ) and a gc s probability of hosting an lmxb . this discrepancy could be due to our limited sample of bulge gcs that do not have as wide a range in metallicity as gcs at larger galactocentric distances . studies of the gc - lmxb connection in spiral galaxies are limited . @xcite studied xrbs in the antennae galaxies and found that massive young star clusters are more likely to host an x - ray source than less massive young clusters . when studying 32 gc - lmxbs in the spiral galaxy m104 with _ chandra _ , @xcite found that metal - rich gcs were more likely to host an x - ray source . @xcite found similar trends in the milky way . @xcite does mention that the observed trends in m31 are weaker than those found for elliptical galaxies , which further stresses the need to analyze the updated star cluster data from phat with _ chandra _ , along with surveys of other spiral galaxies . our stacking results gave upper limits of @xmath2 erg s@xmath3 across all bands for different stacks . when stacking star clusters that were brightest ( f475w magnitude ) , most compact ( smallest @xmath89 ) , and a combination of the two , we found that the detection significance increased across all bands , although not to a statistically significant level . due to the proximity of m31 the upper limits are approximately two orders of magnitude smaller than for star clusters in m51 @xcite . studies of globular clusters in the milky way have found a plethora of faint x - ray sources that include quiescent xrbs ( typically @xmath126 erg s@xmath3 ) @xcite . in gc 47 tucanae , five classified quiescent lmxbs had a mean @xmath127 kev ) @xmath2 erg s@xmath3 @xcite . therefore , assuming m31 s gcs are similar to those in the milky way , we should be probing this population with our stacking analysis . given that a single gc may only have a handful of faint lmxbs , even our stacked sample of 4 of the brightest and smallest gcs fails to provide a detection . at this point it seems that stacking star clusters to identify faint xrbs is proving difficult , at least in spiral galaxies / crowded fields . with the higher concentration of x - ray sources in galaxy bulges source crowding is a problem . along with diffuse emission in spiral disks , this complicates stacking analyses . specifically , the background level of each image in a stack is larger making it more difficult to detect faint sources in the aperture . ultimately a detailed , deep survey of a large sample of gcs in an early - type galaxy would be most promising for obtaining a stacking detection . h ii regions emit thermal x - rays at @xmath128 and @xmath129 kev due to shocks from ob star winds and supernovae / supernova remnants @xcite . the lifetimes of h ii regions ( a few myr ) are long enough that hmxbs can form , providing another possible source of x - ray emission . therefore we would expect x - ray emission in h ii regions to come from various sources , with bright hmxbs and supernova remnants ( if present ) dominating . @xcite found an hmxb in the giant h ii region ic131 in m33 , while @xcite found 2 hmxb candidates in h ii regions in m83 . @xcite studied ultraluminous x - ray sources in nearby galaxies and concluded that 18 of 27 were located inside ob associations or star forming regions . however , most hmxbs are not found in h ii regions or ob associations . in the milky way , @xcite found clustering between 79 hmxbs and 458 ob associations ( 7@xmath113 for distances @xmath130 kpc ) . the average offset of 0.4 @xmath20 0.2 kpc between hmxbs and ob associations was consistent with compact object natal kicks of 100 @xmath20 50 km s@xmath3 . the average kinematical age ( time from supernova to hmxb phase ) was @xmath44 myr . a follow up to this work by @xcite found the upper limit on kinematical age to be @xmath43 myr . therefore unless natal kicks are small hmxbs could be located @xmath11 6.5 ( 400 pc ) from h ii regions in m31 . based on these results one would expect most x - ray emission from h ii regions to be thermal rather than from xrbs . from table [ tab : htwoxrmatches ] most of the previous classifications for h ii regions hosting x - ray point sources in m31 have been supernova remnants as opposed to hmxbs . however , these are only supernova remnant candidates that have been found using optical line ratios ( h@xmath1/s ii ) . in addition , our x - ray color - color plot ( figure [ fig : xraycc ] ) shows that none of the x - ray point sources appear in the supernova remnant regime . when comparing the average radii of matched and unmatched h ii regions those that were more compact were the most likely hosts of x - ray sources . nevertheless , a logistic regression analysis showed that both the luminosity and radius of an h ii region were not significant predictors of it hosting an x - ray source . the four h ii region matches that are previously unidentified can be classified as hmxb candidates and require further analysis to determine their nature . as with star clusters our stacking results of h ii regions gave upper limits of @xmath2 erg s@xmath3 across all bands for most stacks . we saw the most improvement in detection significance ( soft band ) when filtering by the smallest radii ( @xmath131 pc ) and @xmath132 erg s@xmath3 , although not to a statistically significant level . while most of the _ chandra _ observations have been restricted to the central region of m31 , an extension of the data to the disk is necessary to probe both young star clusters and the numerous h ii regions . the resolution of x - ray data that does cover most of m31 ( i.e. _ xmm - newton _ and _ rosat _ ) is not sufficient to distinguish individual x - ray point sources in crowded regions or reliably associate them with clusters . more specifically , with the exquisite resolution of _ hst _ observations such as the phat survey only _ chandra _ will allow for robust identification of x - ray counterparts . figure [ fig : acis - fov ] is truly striking . chandra _ acis - i data does cover a large portion of the first 8 bricks of the phat survey , the remaining bricks and almost the entire northeast quadrant of m31 are unobserved . in m31 , _ chandra _ has been used mostly for a long - term monitoring program of the supermassive black hole in the nucleus . other observations throughout the galaxy were used to study supersoft x - ray sources . due to the degradation of the psf at large off - axis angles , data outside of the acis - s3 chip has poor resolution and therefore is not useful even for stacking . further motivation lies in the absence of a confirmed hmxb in m31 , which can be addressed by using both the h ii region and star cluster catalogues available to identify candidates . investigating the spiral arm regions that the phat survey has observed with ( future ) complementary _ chandra _ data would be invaluable to furthering our understanding of the xrb population in m31 . we used all 121 publicly available _ chandra _ acis observations of m31 to complete a study of star clusters and h ii regions . specifically , we analyzed 83 star clusters in the bulge of m31 from the phat survey and found 15 unique matches ( 18% ) to 17 x - ray point sources within 1 ( 3.8 pc ) . when correcting our percentage of matches to completeness limits of past surveys we found similar results . the most compact , brightest , and reddest clusters preferentially hosted x - ray sources , on average . logistic regression showed that the star cluster f475w magnitude was the most important predictor of finding a star cluster x - ray source . this follows from the fact that the most luminous and therefore most massive star clusters are more likely to host xrbs . a less significant predictor was @xmath89 , which associated smaller star clusters ( probes compactness to some extent ) with x - ray sources . the link between these two properties and xrbs stems from the high densities and larger number of stars that create dynamical conditions most favorable for the formation of xrbs . the f475w@xmath0f814w color , an indicator of metallicity in old stellar populations , was not a significant predictor for star clusters hosting x - ray point sources . the majority of clusters in our sample are metal - rich and thus we do not have an evenly distributed sample of blue and red clusters . a more complete survey with a larger spread in metallicity is required to confirm this result . a stacking analysis resulted in non - detections across all energy bands and average upper limits of @xmath2 erg s@xmath3 . this is consistent with quiescent xrb luminosities in the milky way . from 1566 h ii regions in the field of view of the _ chandra _ data only 10 matched to 9 unique x - ray point sources within 3 ( 11 pc ) . on average , the matches corresponded to the most compact h ii regions . however , a logistic regression analysis showed that neither the radius nor h@xmath1 luminosity was a significant predictor of an h ii region hosting an x - ray source . from previous optical emission - line classifications most of these sources were supernova remnant candidates . four sources that have no previous optical identifications are hmxb candidates . a stacking analysis resulted in upper limits of @xmath2 erg s@xmath3 for all energy bands . to advance our understanding of the xrb population in m31 , specifically the disk , and probe the faintest extragalactic xrbs , a complete _ chandra _ survey of the phat region is necessary . we thank the referee for detailed comments that improved the manuscript . we also thank ben williams for helpful comments and cliff johnson for the phat footprint . support for this work was provided by discovery grants from the natural sciences and engineering research council of canada and by ontario early researcher awards . we have also used the canadian advanced network for astronomical research ( canfar ; @xcite ) . 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 acknowledge the following archives : the hubble legacy archive ( hla.stsci.edu ) , chandra data archive ( cda.harvard.edu/chaser ) , and 2mass ( ipac.caltech.edu/2mass ) . + _ facilities : _ hst ( acs , wfc3 ) , cxo ( acis ) , s. , armstrong , p. , ball , n. , et al . 2011 , in astronomical society of the pacific conference series , vol . 442 , astronomical data analysis software and systems xx , ed . i. n. evans , a. accomazzi , d. j. mink , & a. h. rots , 61 , d. e. , evans , i. n. , glotfelty , k. , et al . 2006 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 6270 , the chandra x - ray observatory calibration database ( caldb ) : building , planning , and improving none & 1386 & 1720 & 1.889 & 2.18 & 0.018 & 2.389 & 1.122 & 0.012 & 1.421 & 0.892 & 0.011 + r @xmath1917 pc & 711 & 865 & 1.116 & 2.257 & 0.023 & 0.117 & 1.123 & 0.015 & 13.02 & 0.854 & 0.013 + l@xmath133 erg s@xmath3 & 476 & 617 & 3.186 & 2.032 & 0.041 & 6.678 & 0.983 & 0.025 & 3.435 & 0.674 & 0.021 + r @xmath634 pc & 83 & 106 & 113.128 & 1.794 & 0.08 & 35.079 & 0.175 & 0.011 & 379.804 & 0.064 & 0.005 + l@xmath134 erg s@xmath3 & 1231 & 1515 & 1.3 & 2.16 & 0.02 & 2.087 & 1.104 & 0.013 & 1.121 & 0.872 & 0.012 + l@xmath133 erg s@xmath3 & r @xmath1917 pc & 132 & 172 & 105.83 & 1.442 & 0.061 & 63.978 & 0.131 & 0.007 & 28.521 & 0.04 & 0.003 + none & 54 & 86 & 24.058 & 4.641 & 0.054 & 0.632 & 1.573 & 0.03 & 0.187 & 3.054 & 0.045 + [email protected] pc & 33 & 51 & 2.302 & 4.42 & 0.069 & 0.74 & 1.473 & 0.038 & 0.553 & 2.871 & 0.057 + f475w @xmath620.5 & 21 & 28 & 1.678 & 3.073 & 0.09 & 1.157 & 1.283 & 0.05 & 1.158 & 1.651 & 0.073 + [email protected] pc & 21 & 35 & 0.711 & 4.952 & 0.089 & 2.156 & 1.693 & 0.05 & 0.07 & 3.276 & 0.074 + f475w @xmath1919 & 15 & 25 & 1.044 & 6.528 & 0.113 & 3.587 & 1.685 & 0.055 & 0.818 & 4.723 & 0.097 + [email protected] pc & f475w @xmath1919 & 4 & 7 & 3.315 & 7.475 & 0.222 & 2.827 & 1.565 & 0.091 & 2.413 & 5.602 & 0.197 +

_ strictly chordality- graphs ( ) _ are graphs which are either cycle - free or every induced cycle is of length exactly . strictly chordality-3 and strictly chordality-4 graphs are well known chordal and chordal bipartite graphs , respectively . for , the study has been recently initiated in and various structural and algorithmic results are reported . in this paper , we show that maximum independent set ( mis ) , minimum vertex cover , minimum dominating set , feedback vertex set ( fvs ) , odd cycle transversal ( oct ) , even cycle transversal ( ect ) and steiner tree problem are polynomial time solvable on graphs , . we next consider-free graphs and show that fvs , oct , ect , steiner tree problem are polynomial time solvable on subclasses of-free graphs . + * keywords : * strictly chordality graphs ,-free graphs , feedback vertex set , odd ( even ) cycle transversal , steiner tree .

1610.00853

strictly chordality @xmath0 graphs ( @xmath1 graphs ) are graphs which are either cycle - free or every induced cycle is of length @xmath0 . this graph class was introduced very recently by dhanalakshmi et al . in @xcite by generalizing chordal and chordal bipartite graphs in a larger dimension . @xmath5 and @xmath6 graphs are well known chordal graphs and chordal bipartite graphs , which are well studied as it helps to identify the gap between np - complete input instances and polynomial - time solvable input instances on many problems . problems such as clique , independent set , coloring have polynomial - time algorithms restricted to @xmath7 graphs . on the similar line , authors of @xcite have explored @xmath8 in detail from both structural and algorithmic front . in @xcite , polynomial - time algorithms for problems such as testing , hamiltonian cycle , coloring , tree - width , and minimum fill - in have been presented . in this paper , we revisit @xmath1 graphs and study classical problems such as mis , dominating set , fvs , oct , ect and steiner tree . in recent times , these problems are extensively studied in the context of parameterized complexity @xcite . also , cycle hitting problems such as fvs , oct , ect have polynomial - time algorithms restricted to chordal and chordal bipartite graphs @xcite . further , independent set and vertex cover also have polynomial - time algorithms in chordal @xcite and chordal bipartite graphs . steiner tree , a generalization of classical minimum spanning tree problem and dominating set are known to be np - complete in chordal and chordal bipartite graphs . it is important to highlight that chordal ( chordal bipartite ) graphs have a special ordering , on vertices namely _ perfect vertex elimination ordering _ ( _ perfect edge elimination ordering _ ) and this ordering is greatly used in solving all of the above combinatorial problems . for @xmath9 graphs , a _ vertex cycle ordering ( vco ) _ is proposed in @xcite . it would be an interesting attempt to see whether vco helps in solving the above mentioned combinatorial problems restricted to @xmath1 graphs . this is the first focus of this paper . the second focus of this paper is to study subclasses of @xmath4-free graphs from minimal vertex separator ( mvs ) perspective and analyze the complexity of cycle hitting problems in @xmath4-free graphs . @xmath4-free graphs have received good attention in the literature as it is a subclass of @xmath10-free graphs and a superclass of split graphs . interestingly , steiner tree @xcite and dominating set @xcite is np - complete on @xmath4-free graphs and other classical problems are polynomial - time solvable @xcite . in this paper , we investigate the complexity of cycle hitting problems and steiner tree on subclasses of @xmath4-free graphs and present polynomial - time algorithms for all of them . * organization of the paper : * in section [ sec::defn ] , we introduce basic terminologies and theorems used in this paper . the algorithmic results on @xmath1 graphs ; maximum independent set , odd ( even ) cycle transversal , feedback vertex set , dominating set and steiner tree are presented in section [ sec::algs ] . in section [ 2k2 ] , we present the structural and algorithmic results on the subclasses of @xmath4-free graphs . we follow the notation as in @xcite . let @xmath11 be a simple , connected and undirected graph with the non - empty vertex set @xmath12 and the edge set @xmath13= \{\{@xmath14 } @xmath15 @xmath16 and @xmath17 is adjacent to @xmath18 in @xmath11 and @xmath19}. the @xmath20 of a vertex @xmath18 of @xmath11 , @xmath21(@xmath18 ) , is the set of vertices adjacent to @xmath18 in @xmath11 . the degree of the vertex @xmath18 is @xmath22 . let @xmath23 , we define @xmath24 as @xmath25 . a _ cycle _ @xmath26 on @xmath27-vertices is denoted as @xmath28 , where @xmath29 and @xmath30 . the graph @xmath11 is said to be @xmath31 if every pair of vertices in @xmath11 has a path and if the graph is not connected it can be divided into disjoint connected @xmath32 @xmath33 , @xmath34 , where @xmath35 denotes the set of vertices in the component @xmath36 . the graph @xmath11 is said to be _ k - connected _ ( or _ k - vertex connected _ ) if there does not exist a set of @xmath37 vertices whose removal disconnects the graph . the graph @xmath38 is called a @xmath39 of @xmath11 if @xmath40 @xmath41 @xmath12 and @xmath42 . the subgraph @xmath38 of a graph @xmath11 is said to be @xmath43 @xmath39 , if for every pair of vertices @xmath17 and @xmath18 of @xmath38 , \{@xmath14 } @xmath44 @xmath45 if and only if \{@xmath14 } @xmath44 @xmath13 and it is denoted by @xmath46 $ ] . an @xmath43 @xmath47 is a cycle that is an induced subgraph of @xmath11 . the graph @xmath11 is said to be _ cycle free _ if there is no induced cycle in @xmath11 . @xcite [ thm::sckcons ] a graph @xmath11 is a @xmath1 graph if and only if it can be constructed iteratively by any one of the following operations . * @xmath48 is an @xmath1 graph . * @xmath49 is an @xmath1 graph . * if @xmath11 is an @xmath1 graph , then the graph @xmath50 , where , @xmath51 , @xmath52 such that @xmath53 and @xmath17 is any vertex in @xmath12 , is also an @xmath1 graph . * if @xmath11 is an @xmath1 graph , then the graph @xmath50 , where , @xmath54 , @xmath55 such that @xmath56 and @xmath17 is any vertex in @xmath12 , is also an @xmath1 graph . * if @xmath11 is an @xmath1 graph , then the graph @xmath50 , where , @xmath57 , @xmath58 such that @xmath59 and @xmath60 is any edge in @xmath13 , is also an @xmath1 graph . * if @xmath11 is an @xmath1 graph and @xmath61 , then the graph @xmath50 , where , @xmath62 , @xmath63 such that @xmath64 and @xmath65 is any path of length @xmath66 contained in no induced cycle in @xmath11 or in any one induced cycle @xmath67 of length @xmath0 in @xmath11 such that there does not exist an induced cycle @xmath68 in @xmath11 with @xmath69 , @xmath70 for some @xmath71 and for at least one @xmath72 , @xmath73 . throughout this subsection , the graph @xmath11 refers to an @xmath1 graph , @xmath74 . it is clear from the above bi - implication that we can get a vertex cycle ordering ( vco ) for any @xmath11 in at most @xmath27 iterations , where @xmath27 is the number of vertices in @xmath11 @xcite . let @xmath75 , @xmath76 , be the ordering of @xmath11 . if @xmath77 , then either @xmath11 is a trivial graph or a cycle of length @xmath0 . if @xmath78 , then the label(@xmath79 ) , @xmath80 , denotes the + ( a ) pendant vertex if it satisfies the condition ( iii ) of _ theorem [ thm::sckcons ] _ , + ( b ) 0-pendant cycle if it satisfies the condition ( iv ) of _ theorem [ thm::sckcons ] _ and if @xmath17 is not a part of any cycle in @xmath11 , + ( c ) 1-pendant cycle if it satisfies the condition ( iv ) of _ theorem [ thm::sckcons ] _ and if @xmath17 is part of at least one cycle , + ( d ) 2-pendant cycle if it satisfies the condition ( v ) of _ theorem [ thm::sckcons ] _ and + ( e ) @xmath81-pendant cycle if it satisfies the condition ( vi ) of _ theorem [ thm::sckcons ] _ w.r.t the induced graph on @xmath82 . note that , in a @xmath81-pendant cycle @xmath83 , @xmath83 can have either @xmath84 or @xmath85 as a cut vertex but not both . a graph @xmath11 is said to be a cage graph of size @xmath27 denoted as @xmath86 if there exist @xmath87 such that @xmath88 for all @xmath89 and @xmath90 is a path of length @xmath91 . graph . one of the vertex cycle ordering for this graph is ( @xmath92 , @xmath93 , @xmath94 , @xmath95 , @xmath96 , @xmath97 @xmath98 , @xmath99 , @xmath100 ) , where the vertices @xmath101 and @xmath102 are said to be pendant , @xmath103 is a 0-pendant cycle , @xmath104 is a 1-pendant cycle , @xmath105 as 2-pendant cycle and @xmath106 is a @xmath107-pendant cycle . the graph induced on the vertex set @xmath108 is the @xmath109.,width=170,height=132 ] @xcite a connected graph is @xmath4 free if and only if it forbids @xmath110 , @xmath111 and @xmath112 as an induced subgraphs . let @xmath11 be a graph and @xmath113 . a vertex @xmath114 is said to be a _ universal vertex _ if @xmath115 . an edge @xmath60 is said to be a _ universal edge _ if @xmath116 , either @xmath117 or @xmath118 . @xcite [ 2k2pre ] let @xmath11 be a connected graph and @xmath83 be any minimal vertex separator of @xmath11 . let @xmath119 , @xmath120 be the connected components in @xmath121 . @xmath11 is @xmath4 free if and only if it satisfies the following conditions : * @xmath121 contains at most one non - trivial component . further , if @xmath121 has a non - trivial component , say @xmath122 , then the graph induced on @xmath123 does not contain @xmath124 , @xmath125 , @xmath126 as an induced subgraphs . * every trivial component of @xmath121 is universal to @xmath83 . * every edge in the non - trivial component of @xmath121 is universal to @xmath83 . * the graph induced on @xmath127 is either connected or has at most one non - trivial component . further , if the graph induced on @xmath127 has a non - trivial component , say @xmath128 , then the graph induced on @xmath129 does not contain @xmath124 , @xmath125 , @xmath126 as an induced subgraphs . * if @xmath83 and @xmath121 has a non - trivial component , say @xmath128 and @xmath122 , respectively , then every edge in @xmath128 is universal to @xmath130 , where @xmath131 . let @xmath11 be a strictly chordality @xmath0 graph , @xmath3 , and let @xmath132 be the vco of @xmath11 , @xmath133 . each algorithm makes use of a vco and picks the _ desired _ vertices . at every stage of the algorithm , _ pruning _ of undesired vertices is also done . our algorithms are based on dynamic programming paradigm . for each @xmath79 , @xmath134 , we define _ label(x@xmath135 ) _ that denotes the associated vertices in @xmath79 . for figure [ fig::vco ] , @xmath136 , where @xmath137 , @xmath138 , @xmath139 . + * problem 1 * _ maximum independent set ( mis)_. given an @xmath1 graph @xmath11 , @xmath3 , an independent set @xmath140 such that @xmath141 . the objective is to find an independent set in @xmath11 of maximum cardinality . we now present an algorithm to find a mis . * let @xmath75 , @xmath133 , be the vco of @xmath11 * find an mis @xmath142 for @xmath143 . add @xmath142 to @xmath83 . * remove @xmath144 from @xmath11 and let the resulting graph be @xmath50 . * update @xmath145 and repeat _ steps 2 _ and @xmath146 . let @xmath147 denote the independent set of @xmath11 with maximum size . then , @xmath148 where , @xmath149 . + * computing @xmath150 : * [ x1pendvertex ] @xmath151 if @xmath152 is a pendant vertex . on the contrary , assume that @xmath17 is not a part of any maximum independent set of @xmath11 . since @xmath17 is a pendant vertex , @xmath153 is a singleton set , say @xmath154 . if @xmath18 is also a pendant vertex , then there is nothing to prove . assume that @xmath18 is not a pendant vertex . it is clear from the definition of @xmath147 that either @xmath155 or @xmath156 . by our assumption , thus , @xmath156 . by choosing @xmath18 , we are forced not to add the vertices in @xmath158 , whose cardinality is strictly greater than zero . this will contradict the maximality of @xmath147 unless @xmath11 is either @xmath159 or @xmath160 where @xmath161 i s the first vertex of degree at least three in @xmath11 . @xmath162 [ x10/1pendcycle ] let @xmath163 be the 0-pendant cycle ( or 1-pendant cycle ) where @xmath164 , @xmath165 and @xmath166 . then @xmath167 if @xmath0 is odd and @xmath168 if @xmath0 is even . it is clear that , the maximum size of an independent set of a cycle @xmath49 is @xmath169 . the cardinality of the given set @xmath150 is @xmath169 . thus , @xmath150 is the maximum independent set of @xmath143 . it remains to show that the set @xmath150 does not affect the maximality of @xmath147 . i.e. , to prove that the maximality of @xmath147 is affected if we choose @xmath170 when @xmath0 is odd and @xmath171 when @xmath0 is even . it is enough to prove that @xmath84 is not part of @xmath147 . since @xmath164 , any mis @xmath172 containing @xmath84 has the property that @xmath173 . thus , if we choose @xmath84 for @xmath150 , then the cardinality of the resultant independent set for @xmath11 is either @xmath174 or less than @xmath175 . @xmath162 [ x12pendcycle ] let @xmath163 be the 2-pendant cycle where @xmath165 and @xmath166 , @xmath164 and @xmath176 . then @xmath177 if @xmath0 is odd and @xmath178 if @xmath0 is even , where @xmath179 , @xmath180 , @xmath181 and @xmath182 , @xmath183 . we prove this lemma by splitting @xmath0 into odd and even . * case 1 : * when @xmath0 is odd . the size of the set @xmath150 is @xmath169 , which is the maximum size of an independent set in an odd cycle of length @xmath0 . an argument similar to _ lemma [ x10/1pendcycle ] _ proves that the set @xmath150 does not affect the maximality of @xmath147 . * case 2 : * when @xmath0 is even . the size of both the sets @xmath184 and @xmath185 are @xmath186 , which is the maximum size of an independent set in an even cycle of length @xmath0 . in order to get the maximum independent set for @xmath11 , the maximum is taken over @xmath187 , @xmath188 , and the conclusion follows . we consider @xmath189 , not to contradict the maximality of @xmath147 due to the presence of both @xmath84 and @xmath190 . @xmath162 [ x1k/2 + 1pendcycle ] let @xmath163 be the @xmath81-pendant cycle where @xmath191 , @xmath166 , @xmath164 and @xmath192 . then @xmath193 if @xmath194 and @xmath195 if @xmath196 , where @xmath179 , @xmath180 and @xmath197 , @xmath198 . the @xmath81-pendant cycle forms a @xmath199 , @xmath200 . it is clear from the definition of @xmath81-pendant cycle that either @xmath84 is a cut vertex or @xmath85 is a cut vertex but not both and the degree of each vertices in the set @xmath201 is two . we prove this lemma by partitioning the @xmath0 into the following two cases : * case 1 : * @xmath194 . the size of the set @xmath193 is @xmath186 , which is maximum . moreover , the set does not include @xmath84 and @xmath85 and this concludes the proof of this case . * case 2 : * @xmath196 . the size of both @xmath184 and @xmath185 are @xmath186 , which is maximum , where @xmath184 is the set containing @xmath84 and @xmath185 is the set containing @xmath85 . by the definition of @xmath81-pendant cycle , it is enough to take the maximum of @xmath202 and @xmath203 , to get @xmath147 . @xmath162 let @xmath11 be an @xmath1 graph . a maximum independent set can be found in polynomial time . further , a minimum vertex cover can be computed in polynomial time . the claim follows from _ lemmas [ x1pendvertex]-[x1k/2 + 1pendcycle ] _ and the fact that vco can be computed in polynomial time @xcite . a minimum vertex cover for @xmath11 can be obtained by taking the complement of a maximum independent set of @xmath11 , which can be obtained in polynomial time . thus , the theorem . @xmath162 * problem 2 * _ minimum dominating set_. given an @xmath1 graph @xmath11 , @xmath3 , the objective is to find a vertex subset @xmath83 of @xmath11 with minimum cardinality such that for every @xmath204 , either @xmath205 or @xmath206 for some @xmath207 . the algorithm for a minimum dominating set : start by finding the vco for a given @xmath1 graph @xmath11 , say @xmath75 , @xmath133 . now , find the minimum dominating set for the first element in the ordering . this immediately suggests us to remove the chosen vertices along with its neighbors from @xmath11 and we recursively compute the dominating set . @xmath208 where @xmath209 denotes a dominating set of @xmath11 with minimum size and @xmath210 + * computing @xmath211 : * [ dx1pendvertex ] @xmath212 if @xmath152 is a pendant vertex and @xmath213 . the pendant vertex @xmath17 can be dominated either by choosing its neighbor @xmath18 or by choosing the vertex @xmath17 itself . by choosing @xmath18 , we can dominate more vertices in @xmath11 , which helps us to minimize the size of the dominating set for @xmath11 . @xmath162 [ dx10/1pendcycle ] let @xmath163 be the 0-pendant cycle ( or 1-pendant cycle ) where @xmath164 , @xmath165 and @xmath166 . then @xmath214 where @xmath215 . it is clear that , the minimum size of a dominating set of a cycle @xmath49 is @xmath216 . the cardinality of the given set @xmath211 is @xmath216 . thus , @xmath211 is the minimum dominating set of @xmath217 and the set does not affect the minimality of @xmath209 as @xmath211 contains @xmath84 . this completes the proof of the lemma . @xmath162 [ dx12pendcycle ] let @xmath163 be the 2-pendant cycle where @xmath165 and @xmath166 , @xmath164 and @xmath176 . then @xmath218 , where @xmath219 , @xmath220 , @xmath221 , @xmath222 , @xmath223 . the size of both the sets @xmath224 and @xmath225 are @xmath226 , which is the minimum dominating set in a cycle of length @xmath0 . in order to get the minimum dominating set for @xmath11 , the minimum is taken over @xmath227 , @xmath198 , and the conclusion follows . @xmath162 [ dx1k/2 + 1pendcycle ] let @xmath163 be the @xmath81-pendant cycle where @xmath191 , @xmath166,@xmath164 and @xmath192 . then @xmath228 where @xmath229 is the minimum dominating set for @xmath217 including @xmath84 , @xmath230 is the minimum dominating set for @xmath217 including @xmath85 , @xmath221 , @xmath222 . the argument similar to _ lemma [ dx12pendcycle ] _ establishes the claim . @xmath162 thus , we get a polynomial - time algorithm to find a minimum dominating set using _ lemmas [ dx1pendvertex]-[dx1k/2 + 1pendcycle]_. + * problem 3 * _ odd cycle transversal_. given an @xmath1 graph @xmath11 , @xmath3 , the objective is to find a vertex subset @xmath83 of @xmath11 with minimum cardinality such that @xmath121 is a bipartite graph ( every induced cycle is even ) . since the @xmath1 graphs does not contain a odd cycle when @xmath0 is even , the set @xmath83 is empty in this case . hence , our problem is to find the set @xmath83 for @xmath231 graph , @xmath232 . let @xmath233 , @xmath133 , be the vco of @xmath11 . thus , the recursive solution is : + @xmath234 + where , @xmath235 is the required set @xmath83 . + * problem 4 * _ even cycle transversal_. given an @xmath1 graph @xmath11 , @xmath3 , the objective is to find a vertex subset @xmath83 of @xmath11 with minimum cardinality such that @xmath121 is a graph where every induced cycle is of odd length . since the @xmath1 graphs does not contain an even cycle when @xmath0 is odd , the set @xmath83 is empty in this case . let @xmath233 , @xmath133 , be the vco of @xmath11 . thus , the recursive solution is : + @xmath236 + where , @xmath237 is the required set @xmath83 . @xmath235 and @xmath237 yield an optimum oct and ect , respectively . arguments similar to _ lemmas [ dx1pendvertex]-[dx1k/2 + 1pendcycle ] _ establishes this claim and thus , @xmath235 and @xmath237 can be computed in polynomial time . @xmath162 * problem 5 * _ feedback vertex set_. given an @xmath1 graph @xmath11 , @xmath3 , the objective is to find a vertex subset @xmath83 of @xmath11 with minimum cardinality such that @xmath121 is a forest . it is easy to see that fvs is precisely oct when @xmath0 is odd , and ect when @xmath0 is even . thus , fvs can be computed in polynomial time . + * problem 6 * _ steiner tree_. given an @xmath1 graph @xmath11 , @xmath3 , and a terminal set @xmath238 , steiner tree asks for a tree @xmath239 spanning the terminal set . the objective is to minimize the number of additional vertices ( @xmath240 , also known as steiner vertices ) . let @xmath67 be the @xmath241-pendant cycle in @xmath11 such that there exist a cycle @xmath68 in @xmath11 , where either @xmath242 or @xmath243 or @xmath244 . let @xmath245 . the _ removal of a s - pendant cycle _ @xmath67 from @xmath11 yields the induced subgraph @xmath246 . note that for each @xmath67 , there is a corresponding @xmath247 and @xmath248 corresponds to the graph @xmath249 . we now present an algorithm to find a minimum steiner set . * remove all the pendant vertices and pendant cycles which do not contain any terminal vertex and update @xmath11 . return @xmath11 if @xmath11 is acyclic . * let @xmath75 , @xmath133 , be the vco of @xmath11 * find a steiner set @xmath142 for @xmath143 . add @xmath142 to @xmath83 . a desired vertex @xmath250 for the @xmath143 is added to @xmath247 . * remove @xmath143 from @xmath11 and let the resulting graph be @xmath50 . * update @xmath145 and repeat _ steps 1 - 4_. let @xmath251 denote the vertex set of steiner tree @xmath239 which spans @xmath238 with a minimum number of steiner vertices . @xmath252 * computing @xmath253 : * [ sx1pendvertex ] if @xmath152 is a pendant vertex , then @xmath254 and @xmath255 , where @xmath18 is the vertex of some @xmath49 in @xmath11 and the first vertex of @xmath256 in a path from @xmath17 in @xmath11 . we add the vertex @xmath18 to the terminal set because the required tree @xmath239 should be connected . now , the only possible steiner tree @xmath239 containing pendant vertex @xmath17 and @xmath18 is @xmath257 . @xmath162 [ sx10/1pendcycle ] let @xmath163 be the 0-pendant cycle ( or 1-pendant cycle ) where @xmath164 , @xmath165 and @xmath166 . let @xmath258 be the set of terminal vertices in @xmath143 . then @xmath259 and @xmath260 where @xmath261 is the induced path obtained by removing the internal vertices of @xmath262 , @xmath263 from @xmath143 , @xmath264 and @xmath265 is obtained by removing the internal vertices of @xmath266 and @xmath267 from @xmath143 , respectively . we add the vertex @xmath84 to the terminal set because the required tree @xmath239 should be connected . the minimum of all possibilities over the @xmath143 is considered to get a minimum steiner tree . @xmath162 [ sx12pendcycle ] let @xmath163 be the 2-pendant cycle where @xmath165 and @xmath166 , @xmath164 and @xmath176 . let @xmath268 be the set of terminal vertices in @xmath143 . then @xmath250 is either @xmath84 or @xmath190 and @xmath269 where @xmath261 is the induced path obtained by removing the internal vertices of @xmath262 , @xmath263 from @xmath143 , @xmath264 and @xmath265 is obtained by removing the internal vertices of @xmath270 and @xmath271 from @xmath143 , respectively . we add either @xmath84 or @xmath190 to the terminal set to get the connected graph @xmath239 . we list all the possibilities by adding @xmath84 to the terminal set and by adding @xmath190 to the terminal set , separately . finally , we choose the minimum of all in order to get a minimum steiner tree . @xmath162 [ sx1k/2 + 1pendcycle ] let @xmath163 be the @xmath81-pendant cycle where @xmath191 , @xmath166,@xmath164 and @xmath192 . let @xmath268 be the set of terminal vertices in @xmath143 . then @xmath250 is either @xmath84 or @xmath85 and @xmath272 where @xmath261 is the induced path obtained by removing the internal vertices of @xmath262 , @xmath263 from @xmath143 , @xmath264 and @xmath265 is obtained by removing the internal vertices of @xmath266 and @xmath273 from @xmath143 , respectively . the argument similar to _ lemma [ sx12pendcycle ] _ establishes the claim . @xmath162 thus , we get a polynomial - time algorithm to find a minimum steiner set using _ lemmas [ sx1pendvertex]-[sx1k/2 + 1pendcycle]_. steiner tree can be obtained by finding a minimum spanning tree of the induced subgraph on @xmath251 . + it is known from @xcite that steiner tree and dominating set are np - complete on @xmath4-free graphs . in this section , we study subclasses of @xmath4-free graphs where these two problems are polynomial - time solvable . further , on such subclasses , we show that fvs and oct are also polynomial - time solvable . to the best of our knowledge , this line of study has not been explored in the literature on these problems . @xmath274-free graphs form a proper subclass of @xmath4-free graphs , where every induced cycle is of length 5 . we observed the following structural properties and conclude that it is a trivial graph class . [ c3c4 ] if @xmath11 is a connected @xmath274-free graph , then for any minimal vertex separator @xmath83 of @xmath11 satisfies the following properties : * @xmath83 is an independent set . * if @xmath275 , then @xmath121 have exactly one trivial component . * if @xmath121 has a non - trivial component , say @xmath122 , then for every edge @xmath276 , @xmath277 and @xmath278 . i.e. , for every vertex @xmath279 , @xmath280 is an independent set . * every vertex in a non - trivial component is adjacent to exactly one vertex in @xmath83 . * on the contrary , assume that @xmath83 has at least one edge , say @xmath281 . let @xmath36 be a trivial component in @xmath121 and let @xmath282 . since , @xmath11 is a @xmath4-free graph , @xmath283 ( by _ theorem [ 2k2pre].(ii ) _ ) . thus , @xmath284 forms an induced @xmath285 , which is a contradiction to the definition of @xmath11 . hence , @xmath83 is an independent set . * on the contrary , assume that @xmath121 has at least two trivial components , say @xmath36 and @xmath286 . let @xmath287 and @xmath288 . let @xmath289 be any two vertices in @xmath83 . by @xmath290 , @xmath291 and by _ theorem [ 2k2pre].(ii ) _ , @xmath292 , @xmath293 , @xmath294 , @xmath295 . thus , @xmath296 forms an induced @xmath297 , which is a contradiction to the definition of @xmath11 . hence , @xmath121 have exactly one trivial component if @xmath298 . * by _ theorem [ 2k2pre].(iii ) _ , every edge @xmath276 is universal to @xmath83 , thus , @xmath278 . moreover , if @xmath299 , then every vertex in @xmath300 forms an induced @xmath285 together with @xmath17 and @xmath18 . hence , @xmath277 . * on the contrary , assume that exist a vertex @xmath18 in a non - trivial component such that @xmath301 @xmath302 . by @xmath303 , there exist a trivial component in @xmath304 , say @xmath305 . let @xmath306 . therefore , @xmath307 forms an induced @xmath297 , which is a contradiction to the definition of @xmath11 . @xmath162 if @xmath11 is a connected @xmath274-free graph , then @xmath11 is either a tree or @xmath308 . from _ theorem [ c3c4 ] _ , we can observe that the only possible structure of a non - trivial component after the removal of any minimal vertex separator from @xmath11 is @xmath309 and @xmath310 . further , if @xmath311 , then the graph is @xmath312-free . if @xmath313 and if @xmath121 has a non - trivial component , then the graph is an induced @xmath308 . @xmath162 thus , fvs , oct , steiner tree problem and a dominating set can be solved in @xmath314 time when the input is restricted to @xmath274-free graphs . @xmath315-free graphs are @xmath4-free graphs which are either acyclic or every induced cycle is of length 4 . further , these graphs are @xmath4-free chordal bipartite graphs . we shall study this graph class from mvs perspective . [ c3c5 ] if @xmath11 is a connected @xmath315-free graph , then for any minimal vertex separator @xmath83 of @xmath11 satisfies the following properties : * @xmath83 is an independent set . * if @xmath121 has a non - trivial component , say @xmath122 , then for every vertex @xmath279 , @xmath280 is an independent set . * for every edge @xmath316 in a non - trivial component @xmath122 of @xmath121 , @xmath17 is universal to @xmath83 and @xmath317 . * let @xmath239 be the set of all vertices in the trivial components of @xmath121 . then the graph induced on the vertex set @xmath318 is a complete bipartite graph . * let @xmath319 and @xmath320 be the set of all vertices in a non - trivial component which are universal and non - universal to @xmath83 , respectively . then , there exists a vertex @xmath321 such that @xmath17 is universal to @xmath320 . * the argument is similar to the proof in _ theorem [ c3c4].(i)_. * the argument is similar to the proof in _ theorem [ c3c4].(iii)_. * on the contrary , assume that there exists an edge @xmath276 such that @xmath322 , @xmath323 and @xmath324 . since , @xmath11 is @xmath4-free graph , @xmath278 and there exists a trivial component in @xmath121 , say @xmath305 . let @xmath306 . by our assumption , @xmath17 is adjacent to some vertex in @xmath83 , say @xmath161 and @xmath18 is adjacent to some vertex in @xmath83 , say @xmath325 , such that @xmath326 . thus , @xmath327 forms an induced @xmath308 , which is a contradiction to the definition of @xmath11 . * this is true by the fact that @xmath83 is independent and every trivial component is universal to @xmath83 . * by ( iii ) , @xmath122 is a bipartite graph where @xmath319 and @xmath320 are the independent sets . let us prove the statement by mathematical induction on the cardinality of @xmath319 . + _ base case : _ since @xmath122 is connected , the statement is true for @xmath328 . + _ hypothesis : _ assume that the statement is true for @xmath329 . + _ induction step : _ let @xmath330 . + for some @xmath321 , the graph @xmath331 has a vertex @xmath332 universal to @xmath320 , by the hypothesis . if @xmath333 , then there is nothing to prove . assume that @xmath334 . for arbitrary @xmath335 . if @xmath336 , then @xmath18 is the required vertex which is universal to @xmath320 . if @xmath337 , then @xmath17 is the required vertex which is universal to @xmath320 . @xmath162 although , it is known that the problem of finding a minimum feedback vertex set in chordal bipartite graphs , a super class of @xmath4-free chordal bipartite graphs , is polynomial time solvable @xcite , using the above observation we provide a different approach for this problem in @xmath315-free graph . moreover , our approach takes linear time in terms of the input size . also , it is easy to see that fvs is precisely ect . [ fvsc3c5 ] let @xmath11 be a connected @xmath315-free graph and @xmath83 be any minimal vertex separator of @xmath11 , then the cardinality of any minimum feedback vertex set @xmath338 is * @xmath339 , if @xmath121 has only trivial components , and @xmath239 is the set of all trivial components in @xmath121 . * @xmath340 , if @xmath121 has a non - trivial component @xmath122 , which is cycle - free , and @xmath319 is the set of all vertices in @xmath122 which are universal to @xmath83 . * @xmath341 , if @xmath121 has a non - trivial component @xmath122 and @xmath122 has at least one cycle . * if @xmath11 is a cycle - free graph , then either @xmath311 or @xmath342 . thus , @xmath343 , which is minimum . without loss of generality , assume that @xmath11 has at least one cycle and @xmath304 has only trivial components , say @xmath344 , @xmath345 . by our assumption , @xmath346 and by _ theorem [ c3c5 ] _ , @xmath83 is an independent set . let @xmath347 . clearly , @xmath348 results in a forest , where @xmath338 consists of @xmath349 - 1 vertices from @xmath83 and @xmath350 - 1 vertices from @xmath239 . now , our claim is to prove the set @xmath338 is minimum . * * @xmath351 + on the contrary , assume that @xmath338 is not minimum , then the removal of @xmath142 vertices from @xmath11 results in a forest , where @xmath352 . i.e. , @xmath83 has at least two vertices in @xmath348 , say @xmath353 . clearly , @xmath354 forms an induced @xmath297 , which is a contradiction to the definition of @xmath338 . * * @xmath355 + on the contrary , assume that @xmath338 is not minimum , then the removal of @xmath356 vertices from @xmath11 results in a forest , where @xmath357 . i.e. , @xmath239 has at least two vertices in @xmath348 , say @xmath358 . let @xmath161 and @xmath325 be any two vertices in @xmath83 . clearly , @xmath354 forms an induced @xmath297 , which is a contradiction to the definition of @xmath338 . + hence , @xmath338 is a minimum fvs if @xmath121 has only trivial components . * all possible structures of @xmath122 is given in _ figure [ fig : g123]_. from the structures of @xmath122 , it is clear that @xmath338 is a minimum fvs . it follows from _ theorem [ c3c5 ] _ that no more structures of @xmath122 are possible . + when @xmath122 is cycle - free ] * we prove this case separately for @xmath311 and @xmath298 . * * @xmath311 and let @xmath359 . + it is clear that , every cycle of @xmath11 lies in @xmath122 . thus , @xmath360 and the removal of @xmath361 vertices from @xmath319 results in a forest . now , our claim is to prove that @xmath338 is minimum . on the contrary , assume that removing at most @xmath362 - 2 vertices from @xmath319 results in a forest . i.e. , @xmath363 has at least two vertices in @xmath319 , say @xmath364 . since , @xmath11 is @xmath4-free @xmath365 . note that , @xmath366 because every edge in @xmath122 is between an universal vertex and a non - universal vertex in @xmath122 , by _ theorem [ c3c5].(iii)_. similarly , @xmath367 . thus , the only possibility is @xmath368 . therefore , @xmath369 forms an induced @xmath297 , which is a contradiction to @xmath338 . * * @xmath275 . @xmath83 has at least two vertices , say @xmath353 . our claim is to prove that @xmath83 is minimum . * * * @xmath370 + on the contrary , assume that for some @xmath371 there exists a set @xmath372 such that @xmath373 and @xmath374 is a forest . let @xmath375 . then @xmath376 forms an induced @xmath297 , which is a contradiction . let @xmath377 and @xmath378 . then @xmath379 forms an induced @xmath297 , which is a contradiction . * * * @xmath380 + on the contrary , assume that for some @xmath332 there exists a set @xmath381 such that @xmath373 and @xmath374 is a forest . let @xmath382 and @xmath383 . then @xmath384 forms an induced @xmath297 , which is a contradiction . let @xmath385 . then for any @xmath371 , @xmath376 forms an induced @xmath297 , which is a contradiction . from all the above cases , it is proved that @xmath338 is a minimum fvs . hence , the theorem . @xmath162 [ stc3c5 ] let @xmath11 be a connected @xmath315-free graph , @xmath386 be the terminal set of @xmath11 and @xmath83 be any mvs of @xmath11 . let @xmath239 be the set of all trivial components in @xmath121 , @xmath319 and @xmath320 be the set of universal and non - universal vertices in a non - trivial component of @xmath121 , respectively . if @xmath247 is connected , then the steiner tree @xmath251 is the graph induced on the vertex set @xmath247 . if @xmath247 is not connected , then the steiner tree @xmath251 is the graph induced on the vertex set * @xmath387 , for some @xmath279 , if @xmath388 is connected or when @xmath247 is the subset of @xmath239 or @xmath319 or ( @xmath389 ) . * @xmath390 , for some @xmath371 , when @xmath247 is the subset of @xmath83 * @xmath391 , where @xmath332 is universal to @xmath320 , when @xmath247 is the subset of @xmath320 or ( @xmath392 ) or ( @xmath393 ) or ( @xmath394 ) or ( @xmath395 ) or ( @xmath396 ) . * @xmath397 , for some @xmath279 and a vertex @xmath332 universal to @xmath320 , if @xmath388 is connected or @xmath398 . trivially follows from _ theorem [ c3c5]_. @xmath162 [ dsc3c5 ] let @xmath11 be a connected @xmath315-free graph and @xmath83 be any minimal vertex separator of @xmath11 . let @xmath239 be the set of all trivial components in @xmath121 , @xmath319 and @xmath320 be the set of universal and non - universal vertices in a non - trivial component of @xmath121 , respectively . if @xmath121 has only trivial components , then the dominating set is @xmath399 , for some @xmath279 and @xmath371 when @xmath400 , and the dominating set is @xmath83 when @xmath401 . if @xmath121 has a non - trivial component , then the dominating set is @xmath402 , for some @xmath279 and @xmath321 is universal to @xmath320 . trivially follows from _ theorem [ c3c5]_. @xmath162 _ theorem [ fvsc3c5 ] _ , _ theorem [ stc3c5 ] _ and _ theorem [ dsc3c5 ] _ naturally yields an algorithm to find a minimum fvs , steiner tree and dominating set , respectively , in @xmath403 time , which is linear in the input size . @xmath404-free graphs are @xmath4-free graphs where every induced cycle is of length 3 . this graphs can also be called as @xmath4-free chordal graphs . note that @xmath4-free chordal graphs are known as split graphs . we know that the structural of any minimal @xmath405-vertex separator in chordal graphs is a clique . it is important to highlight that , the feedback vertex set problem is solvable in polynomial time , for chordal graphs @xcite , a superclass of split graphs . [ fvsc4c5 ] let @xmath11 be a connected @xmath404-free graph and @xmath83 be any mvs of @xmath11 , then a minimum fvs @xmath406 is * @xmath407 , if @xmath11 is a complete graph , for some @xmath408 . * @xmath409 , for some @xmath410 , if @xmath121 has only trivial components . * @xmath121 has a non - trivial component @xmath122 and @xmath122 is a tree . if there exist a vertex @xmath205 such that @xmath411 , then @xmath412 . if for every vertex @xmath205 , @xmath413 , then @xmath414 . * @xmath415 , if @xmath121 has a non - trivial component @xmath122 and @xmath122 has at least one cycle . * the proof is obvious from the definition of complete graphs . * since , @xmath83 is a clique , we have to remove at least @xmath416 vertices from @xmath83 . assume that the remaining edge in @xmath83 is @xmath316 , after the removal of @xmath417 vertices . we know that @xmath121 has at least two components and given that every component in @xmath121 is a trivial component . thus , we have to remove any one vertex from @xmath316 such that all cycles formed between trivial components and an edge @xmath316 are removed . * by ( ii ) , it is clear that we have to remove at least @xmath418 vertices from @xmath83 . if there exists a vertex , @xmath18 , in @xmath83 whose neighborhood in a non - trivial component is a singleton set , then the removal of @xmath419 from @xmath11 creates a forest and thus , @xmath420 . if every vertex in @xmath83 has more than one vertex in @xmath122 as its neighbor , then @xmath17 forms at least one cycle along with @xmath122 , thus , @xmath421 . * we enumerate all possible feedback vertex set in @xmath422 , whose removal from @xmath11 results in a forest , and we choose the minimum among them . @xmath162 _ theorem [ fvsc4c5 ] _ naturally yields an algorithm to find a minimum fvs in @xmath423 time . it is important to highlight that , the feedback vertex set problem is solvable in polynomial time , @xmath424 , for chordal graphs @xcite , a superclass of split graphs . @xmath425-free graphs are @xmath4-free graphs where every induced cycle is of length @xmath107 or 5 . a structural observation is given below : let @xmath11 be a connected graph and @xmath83 be a minimal vertex separator for @xmath11 . let @xmath426 be the connected components of @xmath121 . for some @xmath427 , @xmath428 denotes the shortest path between @xmath17 and @xmath18 in a graph @xmath11 such that all internal vertices belongs to @xmath35 . [ c3 ] if @xmath11 is a connected @xmath425-free graph , then for any minimal vertex separator @xmath83 of @xmath11 satisfies the following properties : * @xmath83 is an independent set . * if @xmath121 has a non - trivial component @xmath122 , then for every vertex @xmath279 , @xmath280 is an independent set . moreover , @xmath429 , for all @xmath430 . * if @xmath346 and @xmath121 has a non - trivial component @xmath122 , then @xmath130 is @xmath431-free , where @xmath432 . moreover , @xmath38 is independent and there exist a unique vertex @xmath433 such that @xmath17 is universal to @xmath38 . * if @xmath121 has a non - trivial component , say @xmath122 , then @xmath122 is @xmath308-free . further , the graph induced on @xmath434 is @xmath308-free . * the argument is similar to the proof in _ theorem [ c3c4].(i)_. * the argument is similar to the proof in _ theorem [ c3c4].(iii)_. let @xmath17 and @xmath18 be any two vertices in @xmath280 . our claim is to prove that @xmath429 . on the contrary , assume that @xmath435 ( since , @xmath11 is @xmath4-free , @xmath436 ) , say @xmath437 . we know that in a @xmath4-free graph , every edge in a non - trivial component is universal to @xmath83 . thus , either @xmath438 or @xmath439 . if @xmath438 , then @xmath440 forms an @xmath285 or if @xmath439 , then @xmath441 forms an @xmath285 , which is a contradiction to the definition of @xmath11 . * on the contrary , assume that @xmath130 has an induced @xmath431 , say @xmath442 . choose any two vertices @xmath161 and @xmath325 from @xmath83 . either @xmath443 , where @xmath444 forms an induced @xmath10 ( @xmath445 is induced by @xmath303 ) or @xmath446 , where @xmath447 forms an induced @xmath10 ( @xmath448 is induced by @xmath303 ) , which is a contradiction to the definition of @xmath11 . @xmath38 is independent because of the fact every edge in @xmath122 is universal to @xmath83 . the existence of universal vertex to @xmath38 in @xmath130 is true by the fact @xmath11 is @xmath4-free and it is unique by ( ii ) . * on the contrary , assume that @xmath122 has an induced @xmath449 . choose a vertex @xmath207 . since , every edge in @xmath122 is universal to @xmath83 , any one of the following is true : * * @xmath450 , then @xmath451 forms a @xmath285 . * * @xmath452 , then the edge @xmath453 is not universal to @xmath83 . both contradicts the definition of @xmath11 . since @xmath83 is independent , the graph induced on @xmath454 is also @xmath308-free . @xmath162 _ theorem [ c3 ] _ naturally yields an algorithm to find the fvs , which is described as follows . finding a fvs in a @xmath455-free graph is same as finding a fvs in @xmath456 , say @xmath457 , and in @xmath458 , which is a recursive call and the recursion bottoms out when it returns a bipartite graph , @xmath459-free graph . this can be done in polynomial time . [ stc3 ] let @xmath11 be a connected @xmath425-free graph , @xmath386 be the terminal set of @xmath11 and @xmath83 be any minimal vertex separator of @xmath11 . let @xmath239 be the set of all trivial components in @xmath121 . if @xmath247 is connected , then the steiner tree @xmath251 is the graph induced on the vertex set @xmath247 . if @xmath247 is not connected , then the steiner tree @xmath251 is the graph induced on the vertex set * @xmath387 , for some @xmath279 , if @xmath460 . * @xmath390 , for some @xmath371 , if @xmath461 . * @xmath462 , r\cup \{x_i\})\}$ ] , if @xmath463 . * @xmath464 , r)$ ] , if @xmath247 is the subset of @xmath122 or @xmath465 or @xmath466 . trivially follows from _ theorem [ c3]_. @xmath162 [ dsc3 ] let @xmath11 be a connected @xmath425-free graph and @xmath83 be any minimal vertex separator of @xmath11 . let @xmath239 be the set of all trivial components in @xmath121 . if @xmath121 has only trivial components , then the dominating set is @xmath399 , for some @xmath279 and @xmath371 when @xmath400 , and the dominating set is @xmath83 when @xmath401 . if @xmath121 has a non - trivial component , then the dominating set is @xmath467 , where @xmath371 and @xmath468 is the dominating set of the graph induced on @xmath469 , which is @xmath4-free chordal bipartite graph . trivially follows from _ theorem [ c3]_. @xmath162 it is easy to see that the _ theorem [ stc3 ] _ and _ theorem [ dsc3 ] _ yields a linear time algorithm to find a steiner tree and dominating set , respectively . @xmath470-free graphs are @xmath4-free graphs where every induced cycle is of length 3 or 5 . the structural observations for this graph class are as follows : [ c4 ] if @xmath11 is a connected @xmath470-free graph , then for any minimal vertex separator @xmath83 of @xmath11 satisfies the following properties : * @xmath83 is connected except if @xmath11 is an induced @xmath308 or @xmath471 . * @xmath83 is connected and has a non - trivial component , @xmath122 , in @xmath121 . if a vertex @xmath472 is adjacent to a vertex @xmath473 , then @xmath474 . * if @xmath83 is not a clique , then @xmath121 has exactly one trivial component . moreover , every vertex in a non - trivial component of @xmath121 is not universal to any non - adjacent pair of vertices in @xmath83 . * if @xmath83 is not a clique , then the only possibility of a non - trivial component of @xmath121 is @xmath309 . * the size of the maximum independent set of the graph induced on @xmath83 is at most two . * @xmath83 contains neither @xmath431 nor @xmath475 , @xmath476 . * on the contrary , assume that @xmath477 $ ] has at least two components . choose two vertices @xmath161 and @xmath325 from different components of @xmath477 $ ] . if @xmath121 has only trivial components , then @xmath161 , @xmath325 and any two trivial components from @xmath121 forms @xmath297 , which is a contradiction . if @xmath121 has a non - trivial component , @xmath122 , then choose an edge @xmath276 . if @xmath313 , then either @xmath17 is universal to @xmath83 or @xmath18 is universal to @xmath83 . w.l.o.g , assume that @xmath17 is universal to @xmath83 . thus , @xmath478 and a trivial component in @xmath121 forms a @xmath297 , which is a contradiction to the definition of @xmath11 . if @xmath479 , then either @xmath480 or @xmath481 . w.l.o.g , assume that @xmath480 . let @xmath482 . thus , @xmath478 and a trivial component in @xmath121 forms a @xmath297 , which is a contradiction to the definition of @xmath11 . * on the contrary , assume that @xmath483 for some @xmath484 . since , @xmath11 is @xmath4-free and @xmath122 is a non - trivial component , there exists a vertex @xmath485 such that @xmath486 . thus , @xmath487 forms an induced @xmath297 , which is a contradiction . * on the contrary , assume that @xmath121 has more than one trivial component . let @xmath488 and @xmath154 be any two trivial components in @xmath121 . since @xmath83 is not a clique , @xmath83 contains a @xmath489 . since , @xmath11 is a @xmath4-free graph , @xmath490 . thus , @xmath491 forms an induced @xmath297 , which is a contradiction to the definition of @xmath11 . moreover , if there exists a vertex , @xmath17 , in a non - trivial component of @xmath121 is universal to some non - adjacent pair @xmath492 in @xmath83 and if @xmath154 is a trivial component of @xmath121 , then @xmath491 forms an induced @xmath297 , which is a contradiction . * since @xmath83 is not a clique , @xmath83 contains a @xmath493 , say @xmath494 . on the contrary , assume that the non - trivial component of @xmath121 , @xmath122 , contains either @xmath495 or @xmath496 . consider an edge @xmath316 , since every edge in @xmath122 is universal to @xmath83 , either @xmath497 or @xmath498 . w.l.o.g , assume that , @xmath497 . now , consider the edge @xmath499 , since , @xmath500 either @xmath501 or @xmath502 . by ( iii ) , @xmath483 . thus , the only possibility is @xmath502 . if @xmath503 is a path , then @xmath504 forms an induced @xmath297 , which is a contradiction . if @xmath503 is @xmath505 , then consider the edge @xmath506 , either @xmath507 or @xmath508 . by ( iii ) , both @xmath509 . thus , the edge @xmath506 is not universal to @xmath83 , which is a contradiction . * on the contrary , assume that there exists at least three mutually independent vertices , say @xmath510 in @xmath83 . it is clear that @xmath83 is not a clique , therefore by ( iii ) and ( iv ) , there exists a trivial component and a non - trivial component , i.e. , a @xmath511 , in @xmath121 . by ( iii ) , @xmath512 can be adjacent to at most one vertex in @xmath513 . w.l.o.g , assume that @xmath514 . this implies , neither @xmath17 is adjacent to the vertex @xmath515 nor @xmath18 is adjacent to the vertex @xmath515 , which is a contradiction to _ theorem 1.(iii)_. * on the contrary , @xmath83 contains either @xmath431 or @xmath475 , @xmath476 . if @xmath83 contains a @xmath516 : by ( iii ) , @xmath121 has exactly one trivial component and a non - trivial component @xmath309 , say @xmath517 ( by ( iv ) ) . by ( iii ) , either @xmath518 or @xmath519 . w.l.o.g , assume that @xmath518 . since , @xmath11 is @xmath4-free , every edge in @xmath122 is universal to @xmath83 . therefore , @xmath520 . by ( iii ) , @xmath521 . hence , @xmath522 forms an induced @xmath297 , which is a contradiction . proof for @xmath83 does not contains @xmath523 directly follows from ( v ) . @xmath162 by the _ theorem [ c4 ] _ , it is clear that @xmath83 is @xmath308-free . hence , the feedback vertex set in a @xmath524-free graph can be determined as follows : [ fvsc4 ] let @xmath11 be a connected @xmath470-free graph and @xmath83 be any minimal vertex separator of @xmath11 , then the cardinality of a minimum fvs , @xmath338 , is equal to * @xmath525 , if @xmath11 is a complete graph , for some @xmath526 . * @xmath527 , if @xmath83 is a independent set , for some @xmath528 . * @xmath529 , if @xmath83 is neither clique nor a independent set , for some @xmath528 . * @xmath530 , if @xmath83 is a clique , where @xmath122 is a non - trivial component in @xmath121 . * the proof is obvious from the definition of complete graphs . * from _ theorem [ c4].(i ) _ , it is clear that @xmath83 is independent only when @xmath531 . also , @xmath313 and @xmath532 says that @xmath533 , for some @xmath534 i.e. , @xmath535 . thus , the removal of a vertex from @xmath11 makes @xmath11 a tree and it is minimum . * we know that , @xmath83 is a split graph . since @xmath83 is not complete , @xmath121 has a non - trivial component , @xmath536 and a trivial component , @xmath537 . thus , finding @xmath406 is equivalent to finding @xmath538 and @xmath539 . the possible structures of @xmath540 are ( a ) @xmath541 ( b ) @xmath542 ( c ) @xmath543 ( d ) @xmath309 ( e ) @xmath493 ( by _ theorem [ c4 ] _ ) . in all the cases , we are forced to pick exactly @xmath544 , for some @xmath545 , vertices . thus , @xmath546 , for some @xmath528 . * since , @xmath83 is a clique and @xmath121 has at least one trivial component , @xmath406 contains at least @xmath547 , for some @xmath528 , vertices . if @xmath121 has a non - trivial component , @xmath122 , then we are forced to find @xmath548 in order to compute @xmath406 . thus , @xmath549 . it is minimum because we are varying @xmath550 for all vertices in @xmath83 and picking up the minimum . @xmath162 * remark : * minimum dominating set and steiner tree problem are np - complete restricted to @xmath4-free graphs @xcite . in this paper , we have identified two non - trivial subclasses of @xmath4-free graphs where these problems are polynomial time solvable . in this section , we consider the complexity of connected dominating set and connected fvs using the results presented in _ sections [ sec::algs]-[2k2]_. it is interesting to observe that every minimum connected dominating set contains a minimum dominating set as a vertex subset . it is natural to ask , can we use a minimum dominating set as a terminal set and call steiner tree algorithm as a black box to get a minimum connected dominating set . surprisingly , this observation holds good for @xmath1 graphs and subclasses of @xmath4-free graphs . a similar observation is true for connected vertex cover and connected fvs . further , maximum leaf spanning tree problem is also polynomial time solvable restricted to @xmath1 and subclasses of @xmath4-free graphs . due to page constraint , the proof details are missing in this paper . pranabendu misra , venkatesh raman , m. s. ramanujan and saket saurabh : parameterized algorithms for even cycle transversal , graph - theoretic concepts in computer science : 38th international workshop , springer berlin heidelberg publishers , 172 - 183 ( 2012 ) . f@xmath551nic@xmath551 gavril : algorithms for minimum coloring , maximum clique , minimum covering by cliques , and maximum independent set of a chordal graph , siam journal of computing , vol . 1 , 180 - 187 ( 1972 ) . f.r.k . chung , a. gy@xmath552rf@xmath552s , w.t . trotter and z. tuza : the maximum number of edges in @xmath4-free graphs of bounded degree . discrete math . , 81(2 ) , 129 - 135 ( 1990 ) . z. bl@xmath552zsik , m. hujter , a. pluh@xmath552r and z. tuza : graphs with no induced @xmath297 and @xmath4 , discrete math , 115 , 51 - 55 , ( 1993 ) . d. meister : two characterisations of minimal triangulations of @xmath4-free graphs , discrete mathematics , 306 , 3327 - 3333 , ( 2006 ) .

we present results of multi - epoch vlbi observations with vera ( vlbi exploration of radio astrometry ) of the 22 ghz ho masers associated with a young stellar object ( yso ) iras 22198 + 6336 in a dark cloud l1204 g . based on the phase - referencing vlbi astrometry , we derive an annual parallax of iras 22198 + 6336 to be 1.309.047 mas , corresponding to the distance of 764 pc from the sun . although the most principal error source of our astrometry is attributed to the internal structure of the maser spots , we successfully reduce the errors in the derived annual parallax by employing the position measurements for all of the 26 detected maser spots . based on this result , we reanalyze the spectral energy distribution ( sed ) of iras 22198 + 6336 and find that the bolometric luminosity and total mass of iras 22198 + 6336 are 450 and 7 , respectively . these values are consistent with an intermediate - mass yso deeply embedded in the dense dust core , which has been proposed to be an intermediate - mass counterpart of a low - mass class 0 source . in addition , we obtain absolute proper motions of the ho masers for the most blue - shifted components . we propose that the collimated jets aligned along the east - west direction are the most plausible explanation for the origin of the detected maser features .

0808.0546

one of the crucial issues in astronomy and astrophysics is to understand formation processes of stars , planetary systems , and their clusters . they are the most basic constituents in the galaxy and hence , play essential roles in its formation and evolution . in order to understand physics and dynamics in the star - formation processes , it is necessary to obtain accurate physical properties of target sources such as size , mass , and luminosity . because all of these values strongly depend on the adopted distance , the distance is the most fundamental parameter for quantitative discussion . in spite of its importance , however , it has long been difficult to measure accurate distances to star - forming regions without assumptions . the most reliable and direct way to determine the distance is an annual trigonometric parallax method , based on precise measurements of the position and motion of the object . in 1990 s , the hipparcos satellite measured annual parallaxes for more than 100 000 stars with a typical precision of 1 mas level @xcite . nevertheless , accurate distance measurements with hipparcos for star - forming regions were limited to nearby sources within a few hundred pc from the sun ( e.g. lombardi et al . 2008 ) . although distances to several ob clusters associated with nearby star - forming regions up to @xmath4600 pc have been reported @xcite , hipparcos could not directly determine the distances to newly born stars deeply embedded in dusty molecular cloud cores , because hipparcos was capable of observing only in the optical wavelengths . in the last few years , phase - referencing vlbi astrometry has been developed drastically , yielding the annual parallaxes of nearby star - forming regions . for instance , radio continuum sources in the taurus @xcite , ophiuchus @xcite , and orion @xcite regions were observed with the nrao very long baseline array ( vlba ) to derive their annual parallaxes with typical uncertainty of better than a few percent . in addition , we have started the initial scientific project `` measurements of annual parallaxes of nearby molecular clouds '' with vera ( vlbi exploration of radio astrometry ) . vera is a japanese vlbi network dedicated to astrometric observations aimed at revealing 3-dimensional structure of the galaxy @xcite . in our project for nearby molecular clouds , we specially focus on the accurate distance measurements of star - forming regions within 1 kpc from the sun @xcite . our results will contribute to refine all of the observational studies on nearby star - forming regions by providing their most accurate and directly measured distances . we have already conducted astrometric observations of the h@xmath0o maser sources in the orion @xcite , ophiuchus @xcite , and perseus @xcite regions . all of the results provide direct measurements of their distances with much higher accuracy than the previous photometric methods and the parallax measurements with hipparcos . as a part of this project , we present results of astrometric observations of the h@xmath0o maser source iras 22198 + 6336 in a nearby dark cloud core l1204 g . l1204 is located in the cepheus - cassiopeia molecular cloud complex @xcite and is associated with a well known high - mass star - forming region s140 . it is also identified to be a molecular cloud wb176 @xcite . according to the molecular line observations of l1204 , it consists of several high density molecular cores labeled by a - g @xcite . a deeply embedded yso iras 22198 + 6336 is associated with one of the dense cores , l1204 g , and it is identified to be a powering source of molecular outflow @xcite . the centimeter and millimeter counterpart of iras 22198 + 6336 is detected with the vla @xcite . the h@xmath0o masers were detected @xcite and their distributions were investigated by the vla @xcite and vlba @xcite . however , they did not measure the proper motions of the h@xmath0o masers because observations by @xcite could not achieve sufficient high resolution and @xcite observed only at 1 epoch . therefore , the kinematics of the circumstellar materials around iras 22198 + 6336 remains still unclear . the distance to l1204 is estimated by a photometric method to be 910 pc @xcite assuming that a b0v star hd211880 is the ionization source of the h region s140 , which is also associated with l1204 . on the other hand , @xcite identified hd211880 ( hip110125 ) as a member of the cepheus ob2 cluster . using the hipparcos data , @xcite determined the distance to the cepheus ob2 cluster to be closer value of 615@xmath135 pc from the sun . the kinematic distance of 1.3 kpc is sometimes adopted in several literatures @xcite according to the systemic velocity of @xmath5 km s@xmath6 @xcite . depending on the adopted distance , the luminosity of iras 22198 + 6336 could differ by a factor of 4 , and hence , its physical properties sometimes contain large uncertainties . based on our accurate astrometric observations with vera , we will provide the parallactic distance to iras 22198 + 6336 in l1204 g , together with the kinematics of the circumstellar gas traced by the h@xmath0o masers . we will also discuss the basic properties of the powering source of the h@xmath0o masers by adopting our new distance to l1204 g . the vera observations of the h@xmath0o masers ( @xmath7-@xmath8 , 22235.080 mhz ) associated with iras 22198 + 6336 have been carried out since nov . 2006 at intervals of about 1 month and are still ongoing . we present the results from a total of 9 observing sessions from nov . 2006 to dec . 2007 ( 2006/312 , 2006/361 , 2007/047 , 2007/096 , 2007/130 , 2007/215 , 2007/282 , 2007/322 , and 2007/353 ; hereafter an observing session is denoted by year / day of the year ) . all 4 vera stations participated in all sessions , providing a maximum baseline length of 2270 km . observations were made in the dual beam mode ; the h@xmath0o masers associated with iras 22198 + 6336 and a reference source j2223 + 6249 ( @xmath922h23m18.096661s , @xmath10d4933.80393 " ; @xcite ) were observed simultaneously @xcite . the reference source j2223 + 6249 is separated south from iras 22198 + 6336 by 1.1 degrees . j2223 + 6249 was detected with the peak flux density of about 200 mjy in all the epochs as shown in figure [ fig - j2223 ] . left - handed circular polarization was received and sampled with 2-bit quantization , and filtered using the vera digital filter unit @xcite . the data were recorded onto magnetic tapes at a rate of 1024 mbps , providing a total bandwidth of 256 mhz in which one if channel and the rest of 15 if channels with 16 mhz bandwidth each were assigned to iras 22198 + 6336 and j2223 + 6249 , respectively . a bright continuum source , 3c454.3 , was observed every 80 minutes for bandpass and delay calibration . system temperatures including atmospheric attenuation were measured with the chopper - wheel method @xcite to be 100 - 500 k , depending on weather conditions and elevation angle of the observed sources . the aperture efficiencies of the antennas ranged from 45 to 52% depending on the stations . a variation of the aperture efficiency of each antenna as a function of elevation angle was confirmed to be less than 10% even at the lowest elevation in the observations ( @xmath420 degrees ) . correlation processing was carried out on the mitaka fx correlator @xcite located at the naoj mitaka campus . for h@xmath0o maser lines , the spectral resolution was set to be 15.625 khz , corresponding to the velocity resolution of 0.21 km s@xmath6 . data reduction was performed using the nrao astronomical image processing system ( aips ) . amplitude and bandpass calibrations were done for the h@xmath0o maser source ( iras 22198 + 6336 ) and reference source ( j2223 + 6249 ) independently . for phase - referencing , fringe fitting was done with the aips task fring on the phase reference source ( j2223 + 6249 ) , and the solutions were applied to the target source ( iras 22198 + 6336 ) . because the reference source j2223 + 6249 showed the double - peaked structure in the synthesized images as shown in figure [ fig - j2223 ] , we carried out self - calibration to correct for the structure effect on the phase calibration . in addition , we applied the results of the dual - beam phase calibration , in which difference between instrumentally added phases in the two beams were measured by injecting artificial noise sources into both beams at each station during the observations @xcite . we also corrected for the approximate delay model adopted in the correlation processing . the difference in the optical path length error between our newly applied model and the calc9 model developed by the nasa / gsfc vlbi group are confirmed to be less than @xmath42 mm . in this correction , drifts of the visibility phase caused by the earth s troposphere were estimated based on the gps measurements @xcite . as discussed in @xcite , the tropospheric zenith delay consists of `` dry '' and `` wet '' parts , and only latter one is variable and hence significantly cause the phase errors in the calibrations . the typical tropospheric `` wet '' zenith delay ( path length ) was 5 - 10 cm while it was sometimes as large as 40 cm , in particular at the ogasawara and ishigaki stations in summer . on the other hand , the `` dry '' component of the tropospheric zenith delay was stable about 2.3 m. ( 80mm,80mm)fig1.eps synthesis imaging and deconvolution ( clean ) were performed using the aips task imagr with the uniform weighting to achieve the highest spatial resolution , allowing us to resolve complex spatial structures of the masers . the resultant synthesized beam size ( fwmh ) was typically 1.2 [email protected] mas with a position angle of -40 degrees . the peak positions and peak flux densities of masers were derived by fitting elliptical gaussian brightness distributions to each image using the aips task sad . the formal uncertainties in the maser positions given by sad were mostly better than 0.05 mas . ( 80mm,80mm)fig2.eps ( 140mm,140mm)fig3.eps figures [ fig - total ] and [ fig - spectra ] show an example of a total - power spectrum and scalar - averaged cross - power spectra of the h@xmath0o masers associated with iras 22198 + 6336 , respectively . the h@xmath0o maser lines were detected within the lsr ( local standard of rest ) velocity range from @xmath12 km s@xmath6 to @xmath13 km s@xmath6 . the intense spectra at the lsr velocity of about @xmath14 km s@xmath6 were detected in all the observed epochs , whereas other velocity components were detected at rarer occurrences . in this paper , we refer to the brightest components at @xmath14 km s@xmath6 as the main components . our result is consistent with those of long - term monitoring observations made by @xcite and @xcite that the most prominent maser feature in iras 22198 + 6336 was at the lsr velocity of about @xmath15 km s@xmath6 , while those around the systemic velocity of @xmath5 km s@xmath6 @xcite rarely appeared . according to their long - term monitoring observations , the radial velocities of the h@xmath0o masers in iras 22198 + 6336 changed with time over the 11-year monitoring period at a rate of 0.2 km s@xmath6 yr@xmath6 . the lsr velocity of the @xmath15 km s@xmath6 component has been gradually red - shifted to be about @xmath16 km s@xmath6 @xcite . this trend predicts @xmath17 km s@xmath6 of the velocity of the main component at around 2007 , as observed in our observations . however , we can not find any signature of the velocity drift within our observing period probably due to our relatively shorter time span . since the spectrum taken at the first epoch , 2006/312 , showed most of the velocity components , we first searched for the h@xmath0o masers at the first epoch by producing the synthesized images for all the individual velocity channels with the field of view of @xmath44@xmath114 . as a result , we found roughly 4 clusters of maser spots , or features , spreading over the 300 mas@xmath11300 mas region . hereafter we define a spot " as emission occurring in a single velocity channel and a feature " as a group of spots which are spatially coincident within the beam size and are detected for at least two consecutive velocity channels . based on this coarse map , we again carried out synthesis imaging for all the epochs with the field of view of 25.6 [email protected] mas centered at the positions for each of the identified features . the results of the mapping are shown in figure [ fig - allspot ] , and examples of phase - referenced channel maps are shown in figure [ fig - chmap ] . we labeled the groups of features a - m in the order of their lsr velocities . some of the weak components were not detected in the phase - referenced maps but only detected in the self - calibrated images , in which we carried out phase - calibration using the brightest maser spot rather than the reference source j2223 + 6249 , due to the lower dynamic range of the phase - referenced images . for such spots , we derived their absolute positions by referencing to the absolute position of the brightest spot in the main component that is significantly detected in the phase - referenced images at all the epochs . the maser distribution shows roughly a circular structure with the radius of about 150 mas as seen in figure [ fig - allspot](a ) . overall spatial and radial velocity distribution of h@xmath0o maser features are in good agreement with those by the vla @xcite and the vlba @xcite , but we resolve the previously detected components into several spatially distinct maser spots due to our higher spatial resolution . the main components which are most blue - shifted and brightest , j - m , coincide with both the position and velocity of the c2 component labeled by @xcite , while the most red - shifted components , a - b , coincide in the position and velocity with the c4 component of @xcite . the northern components near the systemic velocity , f - i , are also coincident with both the position and velocity of the c1 component of @xcite . on the other hand , another group of components c - e is spatially coincident with c3 of @xcite , whereas their lsr velocities , from -7 to -10 km s@xmath6 , are significantly red - shifted by more than 5 km s@xmath6 . here we emphasize that the accuracy of our absolute positions with an uncertainty of only 1 mas , which is limited mainly by the position uncertainty of the reference source j2223 + 6249 , 0.90 mas and 1.31 mas in right ascension and declination , respectively @xcite , are about two orders of magnitude higher than those by the previous vla observations @xcite and by the fringe - rate maps obtained with the vlba @xcite . the properties of the features a - i are summarized in table [ tab - features ] . these parameters are derived from the phase - referenced images except for the two features , d and e , as noted in table [ tab - features ] . they are compact and almost no significant spatial structures , and hence , we only summarize the parameters for the peak spot in each feature . all of the features listed in table [ tab - features ] were detected each only at a single one different epoch probably due to their high variability . on the other hand , the brightest main features , j - m , show complex spatial structures . we summarize the parameters for the selected 26 spots in table [ tab - spots ] , which are detected in at least three epochs , and hence , are usable to obtain reliable proper motions as discussed later . the parameters for the spots l1 - 3 are derived from the self - calibrated images while others are from the phase - referenced images . the main features are resolved into four subgroups ; j @xmath18 , k @xmath19 , l @xmath20 , and m @xmath21 in right ascension and declination , respectively , and all of them consist of multiple spots . in particular , the component m shows the elongated structure along the north - south direction from 0 mas to @xmath22 mas in declination , as shown in figures [ fig - central ] and [ fig - chmap ] . the relatively large scale structures found in these features have been inferred from the fact that the h@xmath0o masers are highly resolved - out with the longer baselines in the vlba @xcite . we also found that the flux densities of the cross - power spectra of the masers are typically only about 30% of those of the total - power spectra as shown in figures [ fig - total ] and [ fig - spectra ] , which are suggestive of the missing flux . although the peak velocities of the main features , @xmath14 km s@xmath6 were stable , their internal structures were found to be highly variable . the features k and l appeared later than 2007/215 , while m was detected only at the earlier epochs than 2007/215 . in contrast , the feature j was highly variable but detected throughout whole of the observing period . ( 160mm,160mm)fig4.eps ( 80mm,80mm)fig5.eps as depicted in figure [ fig - allspot](a ) , we found a radial velocity gradient along the east - west direction when we consider the clusters of features a - b and j - m ; blue - shifted ( j - m ) and red - shifted(a - b ) components tend to be distributed at the western and eastern sides of the map , respectively . the similar radial velocity gradient has been reported by @xcite and @xcite which is attributed to the bipolar outflow . on the other hand , the larger - scale ( @xmath231 ) velocity gradient along the northwest - southeast direction has been found in the spectral line observations of nh@xmath24 @xcite and c@xmath25o @xcite . although they attributed these velocity gradients to the molecular outflow , their velocity structures showed an opposite trend to our h@xmath0o maser distribution ; the blue - shifted and red - shifted components traced by the nh@xmath24 and c@xmath25o lines tend to be located at southeast and northwest , respectively . it would be suggested that the large - scale velocity gradient does not directly related to that traced by the h@xmath0o masers . detailed discussion about the spatial and velocity structures of the h@xmath0o masers will be presented later . lccrrr & @xmath26 & detected & @xmath27 & @xmath28 & @xmath29 + no & [ km s@xmath6 ] & epoch & [ mas ] & [ mas ] & [ jy beam@xmath6 ] + a & -3.15 & 2006/312 & 293.43 & 60.22 & 1.9 + b & -3.57 & 2006/312 & 302.30 & 47.19 & 0.74 + c & -8.20 & 2006/312 & 171.42 & -128.00 & 3.9 + d@xmath31 & -8.41 & 2007/353 & 173.85 & -141.05 & 2.2 + e@xmath31 & -8.83 & 2007/353 & 174.73 & -136.19 & 2.7 + f & -9.68 & 2006/312 & 36.88 & 145.81 & 11.5 + g & -13.05 & 2007/215 & -24.59 & 207.43 & 1.9 + h & -14.31 & 2006/312 & -5.22 & 199.21 & 15.7 + i & -14.31 & 2007/353 & -39.40 & 218.13 & 7.2 + + + + + + lccrrrrr & @xmath26 & detected & @xmath27 & @xmath28 & @xmath29 & @xmath32 & @xmath33 + no & [ km s@xmath6 ] & epochs & [ mas ] & [ mas ] & [ jy beam@xmath6 ] & [ mas yr@xmath6 ] & [ mas yr@xmath6 ] + j1 & -16.42 & 1 * 3**67 * 9 & 4.15 & -23.17 & 3.7 & -2.33(25 ) & -2.32(32 ) + j2 & -16.63 & 1 * 3**67 * 9 & 4.14 & -23.31 & 6.4 & -2.48(25 ) & -2.20(32 ) + j3 & -16.84 & 1 * 3**67 * 9 & 4.03 & -23.33 & 3.0 & -2.27(25 ) & -2.10(32 ) + k1 & -17.05 & * * * * * * 789 & 14.21 & 20.60 & 2.7 & -2.02(168 ) & 3.00(213 ) + k2 & -17.26 & * * * * * 6789 & 15.53 & 20.48 & 4.7 & -0.99(83 ) & 3.18(105 ) + k3 & -17.47 & * * * * * 6789 & 15.51 & 20.44 & 8.2 & -0.58(83 ) & 3.71(105 ) + k4 & -17.68 & * * * * * 6789 & 15.48 & 20.46 & 8.5 & -0.55(83 ) & 3.71(105 ) + k5 & -17.89 & * * * * * 6789 & 15.51 & 20.46 & 6.8 & -0.40(83 ) & 3.79(105 ) + k6 & -18.10 & * * * * * 67 * 9 & 15.62 & 20.36 & 2.7 & -1.29(88 ) & 3.80(110 ) + l1@xmath31 & -17.68 & * * * * * 6789 & -0.99 & -11.88 & 12.8 & -1.45(83 ) & -2.63(105 ) + l2@xmath31 & -17.89 & * * * * * 6789 & -0.95 & -11.94 & 10.1 & -2.53(83 ) & -1.85(105 ) + l3@xmath31 & -18.10 & * * * * * 6789 & -1.02 & -11.77 & 2.5 & -2.00(83 ) & -2.44(105 ) + m1 & -17.05 & 123456 * * * & -0.90 & -3.40 & 19.6 & -3.44(41 ) & -2.00(50 ) + m2 & -17.26 & 123456 * * * & -0.88 & -3.37 & 60.2 & -3.45(41 ) & -1.23(50 ) + m3 & -17.47 & 123456 * * * & -0.87 & -3.53 & 48.8 & -3.43(41 ) & -1.35(50 ) + m4 & -17.68 & * 2 * 456 * * * & -1.52 & -5.48 & 10.4 & -2.98(55 ) & -0.75(69 ) + m5 & -17.89 & * 2 * 456 * * * & -1.51 & -5.47 & 8.9 & -2.96(55 ) & -0.26(69 ) + m6 & -18.10 & * * * 456 * * * & -0.12 & -5.75 & 0.6 & -3.73(98 ) & 0.70(125 ) + m7 & -17.47 & 1**456 * * * & -0.68 & -1.82 & 42.6 & -3.49(45 ) & 0.26(55 ) + m8 & -17.68 & 1**456 * * * & -0.72 & -2.28 & 20.4 & -3.44(45 ) & 1.11(55 ) + m9 & -17.89 & * * * 456 * * * & -0.08 & -2.55 & 1.2 & -3.28(98 ) & 0.83(125 ) + m10 & -18.10 & * 2 * 456 * * * & -2.48 & -2.67 & 3.3 & -0.90(55 ) & 0.30(69 ) + m11 & -17.05 & 1 * 3**6 * * * & -0.52 & -0.35 & 2.7 & -3.89(45 ) & -0.01(56 ) + m12 & -17.68 & 1234 * * * * * & -0.50 & -0.39 & 175.7 & -3.39(78 ) & 0.80(96 ) + m13 & -17.89 & 12345 * * * * & -0.47 & -0.22 & 146.1 & -3.62(60 ) & -0.24(72 ) + m14 & -18.10 & 1234 * * * * * & -0.45 & -0.21 & 59.5 & -3.24(78 ) & 0.99(96 ) + + + + + + + + + ( 160mm,160mm)fig6.eps we conducted monitoring observations of the h@xmath0o masers for about one year , and the absolute motions of the maser spots were successfully obtained by referencing to the extragalactic position reference source j2223 + 6249 . the movement of each maser spot can be expressed by the sum of a linear motion and the annual parallax . it can be assumed that all the maser features or spots in l1204 g have the same annual parallax within the astrometric accuracy of vera . therefore , we can simultaneously determine the proper motions ( @xmath34 , @xmath35 ) and the initial positions ( @xmath36 , @xmath37 ) in right ascension and declination for the @xmath38-th feature or spot , along with a common annual parallax of l1204 g , @xmath39 , by the least - squares analysis of all the detected features or spots . because only the features in the main components , j - m , were detected at more than two epochs , we excluded the features a - i in the following astrometric analysis . to obtain the accurate annual parallax , we attempted to analyze the data by adopting four different datasets ; ( 1 ) both the right ascension and declination data for all the spots , ( 2 ) only the right ascension data for all the spot , ( 3 ) same as ( 1 ) but for all the features , and ( 4 ) same as ( 2 ) but for all the features . as already defined , the feature is the sum of the spatially coincident maser spots integrated over at least two consecutive velocity channels . thus , we determined the peak position of each feature by the gaussian fitting of the velocity - integrated phase - referenced intensity map . on the other hand , the peak position of each spot is derived from the individual channel map obtained from the phase - referenced analysis by assuming that the velocity of the maser spot is constant , i.e. the maser spots are regarded to be identical if their velocities are the same . it would not always be valid if the masers show significant velocity drift @xcite . however , we applied this assumption to the identification of the detected spots between different epochs because we could not find significant velocity drift in our observations as mentioned above . comparison of former and latter datasets shows the effect of the structures in the maser features on our astrometric accuracy . as mentioned above , we found significant spatial structures in the features j - m . in particular , the elongated structure in the feature m can be clearly seen in figure [ fig - central ] , in which the peak positions of the maser spots do not always coincide with those of the features . these complex spatial and velocity structures within the feature m would degrade the accuracy of astrometry as reported in the previous results of vera for nearby sources @xcite . instead , we resolved the internal structure of the feature m by producing the channel maps of all the velocity components . they could better trace the motion of the masing gas in the multi - epoch astrometric analysis . according to their positions and velocity channels , we identified in total 5 features ( 1 in feature j - l for each and 2 in feature m ) and 26 spots in the main components j - m as listed in table [ tab - spots ] , which were detected in at least three observing epochs . some of the spots , for instance , the spots j1 - 3 , k6 , m4 - 5 , m7 - 8 , and m10 - 11 , were not always detected probably due to the variability and the insufficient sensitivity . however , our identification of the maser spots is proved to be appropriate because their motions are well fitted to the linear proper motions and annual parallax which is common for other spots , as shown in figure [ fig - parallax ] . thus , all of the proper motions identified in our analysis really trace the bulk motion of the masing gas . in addition , we evaluate the effect of the atmospheric zenith delay residual by comparing the datasets 1 and 3 with those of datasets 2 and 4 . according to the previous results with vera @xcite , the accuracy of the derived annual parallax is significantly improved when only the data for right ascension are used , as this data is less affected by atmospheric modeling errors . therefore , we ran the fitting by making use of right ascension data alone and both right ascension and declination data for comparison . all the results for different four datasets can be compared in table [ tab - results ] . the results of the fitting are also displayed in figure [ fig - parallax ] for the dataset 1 . the standard deviations of the post - fit residuals are @xmath40.2 mas and @xmath40.3 mas in right ascension and declination , respectively . if we employ the formal uncertainties in the gaussian fitting of the maser spots , @xmath40.05 , as the positional errors in our observations , a reduced-@xmath40 is as large as 20 . this result would suggest that we underestimate the positional errors and that these post - fit residuals represent the actual positional accuracy of the astrometric observations with vera . thus , we took into account @xmath41 and @xmath42 listed in table [ tab - results ] in the least - squares analysis to give a weight , which is inversely proportional to the square of the error , for the right ascension and declination data , respectively . as can be seen in table [ tab - results ] , all of the 4 datasets yield the consistent parallax values within the mutual errors . although none of the spots were persistent during whole of the observing period due to their variability , we could determine the annual parallax with an uncertainty of smaller than 10% by successfully identifying multiple spots during almost 1 year . the highest accuracy is achieved by adopting both right ascension and declination data for all of the 26 detected spots , dataset 1 . the resultant value of the annual parallax for the dataset 1 is [email protected] mas , corresponding the distance to iras 22198 + 6336 of 764@xmath127 pc . the associated uncertainties of only @xmath44% for the dataset 1 is smaller than the others , and this dataset provides the most reliable value . detailed discussions about the possible astrometric error sources will be presented in the section [ sec - error ] . ( 140mm,140mm)fig7.eps lcccc parameter & dataset 1 & dataset 2 & dataset 3 & dataset 4 + dataset & ra and dec & ra only & ra and dec & ra only + & 26 spots & 26 spots & 5 features & 5 features + total number of data & 218 & 109 & 52 & 26 + numbers of fitted parameters & 105 & 53 & 21 & 11 + degrees of freedom & 113 & 56 & 31 & 15 + @xmath39 ( mas ) & 1.309(47 ) & 1.289(66 ) & 1.243(80 ) & 1.288(107 ) + @xmath43 ( pc ) & 764(27 ) & 776(40 ) & 805(52 ) & 777(65 ) + @xmath41 ( mas ) & 0.23 & 0.23 & 0.22 & 0.23 + @xmath42 ( mas ) & 0.29 & & 0.32 & + + + + + along with the annual parallax , proper motions of the maser spots are derived , as summarized in table [ tab - spots ] and illustrated in figure [ fig - allspot](b ) . the proper motion of 1 mas yr@xmath6 corresponds to 3.6 km s@xmath6 at the distance of 764 pc . although the proper motions for some of the features contain large uncertainties , we can see a systematic trend in the direction and magnitude of the proper motion vector in figure [ fig - allspot](b ) . the kinematics traced by the proper motions as well as the radial velocities of the masers will be discussed in the section [ sec - model ] . as discussed in the recent works on astrometry with vlba and vera @xcite , it is difficult to estimate individual error sources induced in the vlbi astrometry quantitatively . in this paper , we regard the post - fit residuals of the least - squares analysis listed in table [ tab - results ] as the typical positional errors , which results in a reduced-@xmath40 of unity . because they are significantly larger than the formal uncertainties in the gaussian fitting of the maser spots , @xmath40.05 mas , some systematic errors would have significant effects on the astrometry . here we discuss about the possible error sources in the present results . one of the serious error sources is the atmospheric zenith delay residuals mainly due to the tropospheric water vapor . this effect is caused by the difference in the optical path lengths through the atmosphere between the target and reference sources because their elevation angles are usually different . @xcite presented results of detailed simulations to calculate the positional errors caused by the atmospheric zenith delay residuals in astrometry with vera . they demonstrated that the positional error due to the zenith delay residual of 2 cm , which is a typical value of vera observations , is better than 0.030 mas and 0.018 mas for right ascension and declination , respectively , in the case of the declination of 60 degrees , separation angle ( sa ) between the two sources of 1 degree , and the position angle ( pa ) on the sky from the target source to the reference source of 180 degrees , which are valid for the iras 22198 + 6336 and j2223 + 6249 pair aligned along the north - south direction . it turns out that the simulated positional error for this pair can not fully explain the observed positional uncertainties , unless we introduce an unrealistically large scatter of the zenith delay residual of @xmath430 cm for each of the observation . therefore , it is unlikely that the atmospheric zenith delay residuals is the most dominant source of our astrometric errors . alternatively , we propose that the principal error source in the present results would be the temporal variation of the internal structure of the masers , as proposed by some previous results from the vera astrometry @xcite . this effect would be more serious for nearby sources than the distant ones . in fact , we can see spatial structure of the features in figure [ fig - central ] , and some of the spots in figure [ fig - parallax ] show significant deviations from the best fit model . even in the case of the fitting of the spots rather than the features , the internal structures in the individual spots might be still significant . although this effect would limit the positional accuracy of each observation , the accuracy of the annual parallax measurements can be improved with increasing number of spots used in the analysis if the variation of the structure behaves in the random way . indeed , the post - fit residuals are reduced almost inversely proportional to the square root of the degrees of freedom in the fitting as listed in table [ tab - results ] . on the other hand , we confirmed that the uncertainties in the station positions , path length errors due to ionosphere , and the delay model employed in our data analysis , which is consistent with the calc9 model , would have negligible effects on our astrometry with vera @xcite . the uncertainties in the absolute position of the reference source j2223 + 6249 , 0.90 mas and 1.31 mas in right ascension and declination , respectively @xcite , do not affect the derived annual parallax and proper motion because they would add only a constant offset to the position of the maser features . the spatial structure and its temporal variation of the reference source j2223 + 6249 possibly contribute to the positional errors of the maser features . however the structure effect of the reference source would be much less significant than those of the masers because we performed self - calibration of j2223 + 6249 to solve its structure . according to the dynamic range of the self - calibrated image of j2223 + 6249 ( @xmath420 ; figure [ fig - j2223 ] ) , the positional uncertainty caused by the structure of the reference source would be less than 0.025 mas ( hpbw/2snr where hpbw and snr represent the synthesized beam width and signal - to - noise ratio , respectively ) . in summary , we conclude that the astrometric accuracy in our observations is mainly limited by the structure of the maser features , and that the atmospheric zenith delay residuals also contribute to some extent to the error sources . as suggested by @xcite , tropospheric zenith delay residual also plays as a random error because they are correlated only over a few days . thus , one can reduce the post - fit residuals by employing longer observing epochs as well as using larger number of masers . thus , the longer monitoring observatios will be helpful to achieve higher precision and accuracy of the parallax measurements , which is still underway . we derived the distance to l1204 g to be 764@xmath127 pc from the sun . this value is smaller than the earlier photometric result , 910 pc @xcite , and the kinematic distance , 1.3 kpc @xcite . while they did not report the errors in the derived distances , photometric and kinematic distances would generally contain larger uncertainties by a factor of a few tens of percent than the parallactic distances ( e.g. hirota et al . 2008 , sato et al . 2008 , and references therein ) . the closer distance , 615@xmath135 pc , is reported by @xcite based on the hipparcos measurements of the annual parallaxes of the cepheus ob2 cluster members , in which one of the members hd211880 is the ionization source of s140 that is possibly associated with l1204 . this value is offset from our measured distance of iras 22198 + 6336 by 149@xmath144 pc and is slightly larger than the apparent diameter of the cepheus ob2 cluster of about 130 pc ( @xmath44 degrees in the plane of the sky ; de zeeuw et al . if l1204 g is physically associated with the cepheus ob2 cluster , our result suggests that l1204 g would be located at the farthest side of the cepheus ob2 cluster . our parallactic distance to iras 22198 + 6336 is found to be very close to the well studied h@xmath0o maser sources in the cepheus - cassiopeia region such as ic1396n @xcite and cepheus a @xcite , although they are not based on the parallax measurements . together with parallactic distances of the maser sources in this region including ngc281 reported by @xcite and further measurements such as for ic1396n and cepheus a , we will be able to reveal the 3-dimensional structure of this molecular cloud complex and refine the statistical studies of star - formation in this region @xcite . although the adopted fir luminosity of iras 22198 + 6336 is different from literature to literature , it has been included in the sample of massive ysos judging from the relatively large far - infrared or bolometric luminosity @xcite . for instance , @xcite adopted @xmath45 to be 250@xmath2 while @xcite adopted @xmath46 to be 580@xmath2 by assuming the same distance @xmath4900 pc . in contrast , @xcite assumed the kinematic distance of 1.3 kpc and derived the larger fir luminosity of 1240@xmath2 . very recently , @xcite reported results of a survey of millimeter and centimeter continuum emission from high - mass to intermediate - mass ysos including iras 22198 + 6336 , of which they adopted the luminosity of 1300@xmath2 . based on the radio continuum observations with the vla of iras 22198 + 6336 along with the observed flux densities reported by previous literatures , @xcite presented basic physical properties of the radio counterpart of the iras source , vla 2 , such as mass and luminosity . however , all of these results employed the larger distances than our present result . here we refine the physical properties of the possible powering source of the h@xmath0o masers , iras 22198 + 6336 vla 2 , based on our newly derived distance and previously reported flux values by multiplying a scaling factor proportional to an appropriate power of the distance @xmath43 . for instance , a factor of ( 764 pc/1300 pc)@xmath47=0.35 is adopted for correcting the luminosity reported by @xcite . first of all , we stress that the bolometric luminosity is reduced to be 450@xmath2 , which is only 35% of previously adopted value , 1300@xmath2 . such a low luminosity would argue against a classification of iras 22198 + 6336 as a massive yso . the derived luminosity set limits on the stellar mass of iras 22198 + 6336 . if it is a zero - age main sequence ( zams ) star , the spectral type of late - b ( between b6 and b7 ) with the stellar mass of about 5@xmath3 @xcite well accounts for the derived luminosity . on the other hand , if it is in a pure accretion phase , the stellar mass of 7@xmath3 is derived from the accretion luminosity with the typical mass accretion rate of 10@xmath48 yr@xmath6 and the stellar radius of 5@xmath49 @xcite . the newly obtained total ( gas+dust ) mass of iras 22198 + 6336 , ( 7 - 10)@xmath3 , which is only 35% of that derived from the sed fitting @xcite , is in good agreement with the above values . thus , our result implies that iras 22198 + 6336 is not a massive but an intermediate - mass yso with the mass of no more than 7@xmath3 . the lack of the iras 12 @xmath50 m emission as well as no near - infrared counter part in the 2mass images support the hypothesis that iras 22198 + 6336 is deeply embedded intermediate - mass yso , which is analogous to low - mass class 0 source @xcite . according to @xcite , iras 22198 + 6336 is associated with centimeter radio continuum emission , which could arise either in an ultra - compact h region or a shock - ionized region in the outflow . they concluded that the contribution of shock - ionized gas around iras 22198 + 6336 should be only a small fraction of the observed centimeter emission because of an extremely high ratio of the centimeter continuum luminosity @xmath51 to the outflow momentum rate , @xmath52 , with an efficiency factor of 90% . however , if we assume the distance of 764 pc , the ratio is reduced by a factor of 0.59 because the luminosity is proportional to @xmath53 while the outflow momentum rate , which is usually derived from the linear size , radial velocity , and inclination angle of the outflow , is proportional to @xmath43 . even if this is the case , the efficiency of 50% is still higher than the typical low - mass ysos , @xmath410% @xcite , suggesting that it would be significantly higher in the very early - stage of intermediate - mass ysos . the mass and luminosity of iras 22198 + 6336 refined by our newly derived distance is consistent with that of the intermediate - mass deeply embedded yso with the spectral type of late - b . nondetection of the 6.7 ghz ch@xmath24oh masers , which is thought to be a signpost of massive star - formation site , toward iras 22198 + 6336 @xcite also supports our conclusion that iras 22198 + 6336 is not a massive yso . although the oh masers , which is another signature of massive star - formation associated with an ultra - compact h region , are detected toward iras 22198 + 6336 @xcite , relatively lower resolution of the oh maser observations , @xmath43 , might give argument against their association with the same powering source of the h@xmath0o masers . further observational studies with the higher - resolution and higher - sensitivity centimeter , millimeter , and submillimeter interferometers would shed light on the properties of the central star in l1204 g more clearly . such detailed observational studies on iras 22198 + 6336 would be quite important because it is one of the rare sources in the very early stage of intermediate - mass yso associated with the h@xmath0o masers ( e.g. patel et al . 2000 ) and hence , good target sources for the complete understanding of the very early stage of intermediate - mass star - formation processes . together with the annual parallax , we successfully measured the proper motions of the maser spots around iras 22198 + 6336 . these results enable us to discuss in detail about the kinematics of the circumstellar materials . prior to discussions , we should note that the proper motions obtained with vera do not represent the `` intrinsic '' proper motions of the masers because they are measured with respect to the sun and hence , include the contribution of the solar motion and the galactic rotation . assuming the solar motion relative to the lsr based on the hipparcos satellite data , ( @xmath54)=(10.00 , 5.25 , 7.17 ) km s@xmath6 @xcite , we estimated the contribution of the solar motion to the observed absolute proper motion to be 3.612 mas yr@xmath6 and @xmath55 mas yr@xmath6 in right ascension and declination , respectively . in addition , contributions of the galactic rotation is calculated to be ( @xmath56 mas yr@xmath6 , @xmath57 mas yr@xmath6 ) on the assumption of the flat rotation with @xmath58 of 8.0 kpc @xcite and @xmath59 of 236 km s@xmath6 @xcite for l1204 g at the distance of 764 pc . subtracting these values from the observed proper motions listed in table [ tab - results ] and shown in figure [ fig - allspot](b ) , we obtained the intrinsic proper motions of the h@xmath0o masers corrected for the solar motion and galactic rotation . the origin of these proper motions is attributed to either due to the systemic motion of the host cloud l1204 g itself , the internal motions of the masers with respect to the powering source iras 22198 + 6336 , such as jets and rotating disk associated with it , or both . regarding to the systemic motion , observed proper motions might trace the peculiar motion of l1204 g with respect to the galactic rotation . the discrepancy between the kinematic distance of l1204 g , which is obtained by assuming that l1204 g exactly follows the galactic rotation curve , and our parallactic distance would be suggestive of the effect of the peculiar motion of l1204 g rather than the galactic rotation . indeed , l1204 g ( @xmath60 degrees ) is situated at 75 pc above the galactic plane with a half - thickness of 87 pc @xcite , which seems too large to apply the flat rotation model . however , the systemic motion of l1204 g is quite uncertain because of a lack of the proper motion measurements of iras 22198 + 6336 , as suggested in the previous result with vera of ngc 1333 @xcite . therefore , here we only consider several possibilities for the origin of internal proper motions of the masers and the kinematics in the circumstellar material . as shown in figure [ fig - allspot](a ) , the proper motions of the features j - m provide a hint of systematic motion directed outward from the center of the circular structure with the approximate diameter of @xmath4300 mas or 200 au . it is naturally interpreted that the most blue - shifted and red - shifted components trace the interacting region with the ambient gas and the collimated jets aligned in the east - west direction , and that the powering source is located at the center of the maser distribution ( see figure [ fig - allspot](a ) ) . this is roughly consistent with the position and elongation of the vla source observed at the 7 mm band @xcite , although their spatial resolution is much larger than our vlbi observations . the possible trend of the northward proper motion vectors in figure [ fig - allspot](b ) could be due to the unknown systemic motion of the powering source . the close - up view of the main features shows an arc - like structure indicative of a bow shock , as seen in figure [ fig - allspot](b ) . such bow shock structures at the tips of the collimated jets have been found in low - mass class 0 protostars iras 05413 - 0104 @xcite and s106 fir @xcite . in this case , the origin of the other maser spots close to the systemic velocity , which seem to be distributed perpendicular to the possible jets , would be accounted for by the rotating disk . the radial velocity difference along the north - south direction , @xmath48 km s@xmath6 and their separation , @xmath61 au ( @xmath4300 mas ) , implies the keplerian rotation with the centrally enclosed mass of 1.6@xmath3/(cos @xmath38)@xmath47 where @xmath38 is the inclination angle between the rotation axis and the line - of - sight . the inclination angle of the disk @xmath38 can be estimated from that of the east - west jets traced by the most blue - shifted and red - shifted masers as follows . the mean proper motions of the most blue - shifted masers are derived to be @xmath62 mas yr@xmath6 and @xmath63 mas yr@xmath6 in right ascension and declination , respectively . if we adopt the proper motions only in the right ascension direction as the outward velocity of the jets , the mean transverse velocity of 5.5 km s@xmath6 toward west and the mean radial velocity of -6.6 km s@xmath6 with respect to the systemic velocity for the maser spots j - m give the inclination angle @xmath38 of 50 degrees . similarly , if we adopt the proper motions of both the right ascension and declination direction , the mean transverse velocity of 12.6 km s@xmath6 yields the inclination angle @xmath38 of 66 degrees . the enclosed mass is estimated to be 3.9@xmath3 and 9.7@xmath3 for former and latter case , respectively , and these values are almost consistent with the result of the sed fitting @xmath64 . on the other hand , the radio continuum emission in the 1.3 cm and 7 mm bands showed elongated structure along the northwest - southeast direction @xcite . according to their sed analysis , the 1.3 cm emission is dominated by the free - free emission from the shock - ionized jets while the 44% of the 7 mm flux comes from the thermal dust emission in the circumstellar disk , and hence , the 7 mm emission would be due to a superposition of the ionized jet and the dust disk @xcite . the direction of the radio continuum emissions are consistent with that of the large - scale velocity gradient @xcite , and hence , @xcite attributed them to the free - free emission from the ionized radio jets . however , the large - scale velocity gradient showed opposite trend to our h@xmath0o maser distribution as mentioned in section [ sec - distribution ] . the maser distributions might trace an interface between the ambient material and the outflow lobe which is parallel to the radio continuum source ( e.g. moscadelli et a. 2006 ) , whereas the asymmetric distribution of the masers with respect to the outflow axis can not be accounted for by the simple model only . it is worth considering another possibility of the rotating disk around the yso with a rotation axis along the north - south direction . if we assume the radial velocity gradient along the east - west direction , @xmath416 km s@xmath6 per @xmath61 au ( @xmath4300 mas ) , as a cause of rotation , the enclosed mass is estimated to be at least 6.4@xmath3 , which is comparable to the total mass or stellar mass of iras 22198 + 6336 as discussed in the previous section . in this case , the northern and southern components c - i would trace the jets from the central yso . their radial velocity close to the systemic velocity would indicate that the jets are aligned almost within the plane of the sky and hence , the disk is almost edge - on view . the directions of the radio continuum emissions in the 1.3 cm and 7 mm bands @xcite are in good agreement with that of the h@xmath0o maser features c - i . nevertheless , this model can not explain the observed velocity drift of 0.2 km s@xmath6 yr@xmath6 for the @xmath15 km s@xmath6 feature @xcite because the velocity drift at the edge of the disk ( at the radius of 100 au ) is estimated to be only 0.01 km s@xmath6 yr@xmath6 in the case of the yso with the mass of 5@xmath3 . thus , the rotating disk with the axis of north - south direction is inadequate to explain the kinematics in the circumstellar materials around iras 22198 + 6336 probed by the blue - shifted maser features j - m . alternatively , distributions and proper motions of the features j - m could also be interpreted as a part of the smaller - scale expanding shell as shown in figure [ fig - allspot](b ) , with the radius of [email protected] mas ( [email protected] au ) and the expanding velocity of [email protected] km s@xmath6 in the plane of the sky . in this case , the dynamical time of the expanding shell is calculated to be only 8 years . basic characteristics of this expanding shell is quite similar to the case of cepheus a @xcite and w75n @xcite . meanwhile , the nature of the possible powering source of the shell is not consistent with the position and elongation of the radio continuum emission @xcite . in addition , it is unlikely that such a shell is traced only by the most blue - shifted components . thus , we rule out the possibility of the small - scale expanding shell as a kinematical model of the blue - shifted maser features j - m . @xcite proposed three possible mechanisms to account for the velocity drift of the maser emission ; a rotating circumstelar disk , acceleration / deceleration of the jets , or a precession of the jets . according to our discussion , most plausible explanation for the proper motions , radial velocity distribution , and velocity drifts of the maser features is the collimated bipolar jets along the east - west direction . thus , the origin of the velocity shift would be the deceleration ( because the radial velocity of the blue - shifted features became closer to the systemic velocity ) or precession of the jets . to distinguish between possible kinematic models , proper motion measurements of red - shifted components located at the eastern end of the maser feature distributions would be crucial . in addition , further high resolution observations of centimeter to submillimeter continuum emission as well as near- and mid - infrared observations of the central yso and collimated jets , if they are associated , would help to reveal the overall circumstellar structure of this source with accuracies of much better than 0.1 . in this paper , we presented the results of multi - epoch vlbi astrometry with vera of the 22 ghz h@xmath0o masers associated with iras 22198 + 6336 in l1204 g . using the absolute positions of total 26 maser spots that were detected in at least three observing sessions , we derived the annual parallax of iras 22198 + 6336 to be [email protected] mas , corresponding to the distance of 764@xmath127 pc from the sun . this is the most accurate distance to l1204 g with the uncertainty of only 4% . because the newly derived distance is closer than those well adopted in the previous literatures ( 910 pc or 1.3 kpc ) , the bolometric luminosity of iras 22198 + 6336 should be significantly reduced . according to the sed of iras 22198 + 6336 , the bolometric luminosity and total mass of iras 22198 + 6336 were refined to be 450@xmath2 and 7@xmath3 , respectively . these values are not consistent with a massive yso but are in agreement with an intermediate - mass yso . we confirmed that iras 22198 + 6336 would be a deeply embedded intermediate - mass yso analogous to a low - mass class 0 source @xcite . together with the annual parallax , we measured the absolute proper motions of the h@xmath0o masers . we discussed several kinematical models of the circumstellar gas such as a small scale expanding shell , collimated jets , or rotating disk . we proposed that the distributions , proper motions , and the previously reported velocity drifts @xcite of the maser features associated with iras 22198 + 6336 are consistent with the collimated bipolar jets and possibly rotating disk perpendicular to them . iras 22198 + 6336 is one of the rare sources in the very early stage of intermediate - mass ysos . therefore , further high - resolution and high - sensitivity observations of the continuum emission from centimeter to submillimeter wavelength , in particular with the future interferometers such as evla and alma , and the proper motion measurements of the h@xmath0o masers and continuum sources will be essential to reveal the kinematics and more detailed properties of iras 22198 + 6336 , which will contribute to the understanding of very early stage of intermediate - mass star - formation processes . we thank the anonymous referee for valuable comments and suggestions . we also thank guillem anglada and carlos carrasco - gonzalez for providing us with information on the vla observations of iras 22198 + 6336 . we are grateful to the staff of all the vera stations for their assistance in observations . th is financially supported by grant - in - aids for scientific research from the ministry of education , culture , sports , science and technology ( 13640242 , 16540224 , and 20740112 ) . anglada , g. 1995 , rev . mexicana astron . conf . , 1 , 67 brand , j. , cesaroni , r. , comoretto , g. , felli , m. , palagi , f. , palla , f. , & valdettaro , r. 2003 , a&a , 407 , 573 chikada , y. , et al . 1991 , in frontiers of vlbi , ed . h. hirabayashi , m. inoue , & h. kobayashi ( tokyo : universal academy press ) , 79 claussen , m. j. , marvel , k. b. , wootten , a. , & wilking , b. a. 1998 , apj , 507 , l79 crampton , d. & fisher , w. a. 1974 , pub . , 14 , 283 dame , t. m. , et al . 1987 , apj , 322 , 706 dehnen , w. & binney , j. j. 1998 , mnras , 298 , 387 de zeeuw , p. t. , hoogerwerf , r. , de bruijne , j. h. j. , brown , a. g. a. , blaauw , a. 1999 , aj , 117 , 354 edris , k. a. , fuller , g. a. , & cohen , r. j. 2007 , a&a , 465 , 865 fukui , y. 1989 , in low - mass star formation and pre - main - sequence objects , ed . b. reipurth ( garching : eso ) , 95 furuya , r. s. , kitamura , y. , wootten , h. a. , claussen , m. j. , saito , m. , marvel , k. b. , & kawabe , r. 2000 , apj , 542 , l135 hachisuka , k. et al . 2006 , apj , 645 , 337 hirota , t. et al . 2007 , pasj , 59 , 897 hirota , t. et al . 2008 , pasj , 60 , 37 honma , m. et al . 2003 , pasj , 55 , l57 honma , m. et al . 2007 , pasj , 59 , 889 honma , m. et al . 2008a , submitted to pasj honma , m. , tamura , y. , & reid , m. j. 2008b , pasj , 60 , in press iguchi , s. , kurayama , t. , kawaguchi , n. , kawakami , k. 2005 , pasj , 57 , 259 imai , h. et al . 2007 , pasj , 59 , 1107 jenness , t. , scott , p. f. , & padman , r. 1995 , mnras , 276 , 1024 kawaguchi , n. , sasao , t. , manabe , s. 2000 , in proc . spie , 4015 , radio telescopes , ed . h. r. butcher ( washington : spie ) , 544 loinard , l. , mioduszewski , a. j. , rodrguez , l. f. , gonzlez , r. a. , rodrguez , m. i. , torres , r. m. 2005 , apj , 619 , l179 loinard , l. torres , r. m. , mioduszewski , a. j. , rodrguez , l. f. , gonzlez - lpezlira , r. a. , lachaume , r. , vzquez , v. , & gonzlez , e. 2007 , apj , 671 , 546 loinard , l. , torres , r. m. , ; mioduszewski , a. j. , & rodrguez , l. f. 2008 , apj , 675 , l29 lombardi , m. , lada , c. j. , & alves , j. 2008 , a&a , 480 , 785 menten , k. m. , reid , m. j. , forbrich , j. , & brunthaler , a. 2007 , a&a , 474 , 515 migenes , v. et al . 1999 , apjs , 123 , 487 molinari , s. , brand , j. , cesaroni , r. , & palla , f. 1996 , a&a , 308 , 573 moscadelli , l. , testi , l. , furuya , r. s. , goddi , c. , claussen , m. , kitamura , y. , & wootten , a. 2006 , a&a , 446 , 985 palla , f. , brand , j. , cesaroni , r. , comoretto , g. , & felli , m. 1991 , a&a , 246 , 249 patel , n. a. , greenhill , l. j. , herrnstein , j. , zhang , q. , moran , j. m. , ho , p. t. p. , & goldsmith , p. f. 2000 , apj , 538 , 268 perryman , m. a. c. , et al . 1997 , a&a , 323 , l49 petrov , l. , kovalev , y. y. , fomalont , e. , & gordon , d. 2005 , aj , 129 , 1163 reid , m. j. 1993 , ara&a , 31 , 345 reid , m. j. & brunthaler , a. 2004 , apj , 616 , 872 snchez - monge , . , palau , a. , estalella , r. , beltrn , m. , & girart , j. m. 2008 , a&a , 485 , 497 sandstrom , k. m. , peek , j. e. g. , bower , g. c. , bolatto , a. d. , & plambeck , r. l. 2007 , apj , 667 , 1161 sato , m. et al . 2007 , pasj , 59 , 743 sato , m. et al . 2008 , pasj , 60 , in press stahler , s. w. & palla , f. 2004 , the formation of stars ( weinheim : wiley - vch ) szymczak , m. , hrynek , g. , & kus , a. j. 2000 , a&as , 143 , 269 tafalla , m. , bachiller , r. , & martn - pintado , j. 1993 , apj , 403 , 175 tofani , g . , felli , m . , taylor , g. b. , & hunter , t. r. 1995 , a&as , 112 , 299 torrelles , j. m. et al . 2003 , apj , 598 , l115 torrelles , j. m. et al . 2001 , nature , 411 , 277 torres , r. m. , loinard , l. , mioduszewski , a. j. , & rodrguez , l. f. 2007 , apj , 671 , 1813 ulich , b. l. , haas , r. w. 1976 , apjs , 30 , 247 valdettaro , r. , palla , f. , brand , j. , cesaroni , r. , comoretto , g. , felli , m. , & palagi , f. 2002 , a&a , 383 , 244 wilking , b. a. , blackwell , j. h. , mundy , l. g. , & howe , j. e. 1989 , apj , 345 , 257 wouterloot , j. g. a. & brand , j. 1989 , a&as , 80 , 149 yonekura , y. , dobashi , k. , mizuno , a. , ogawa , h. , & fukui , y. 1997 , apjs , 110 , 21

new visible polarization data combined with existing ir and fir polarization data are used to study how the magnetic field threading the filamentary molecular cloud gf 9 connects to larger structures in its general environment . we find that when both visible and nir polarization data are plotted as a function of extinction , there is no evidence for a plateau or a saturation effect in the polarization at 1.3 as seen in dark clouds in taurus . this lack of saturation effect suggests that even in the denser parts of gf 9 we are still probing the magnetic field . the visible polarization is smooth and has a well - defined orientation . the ir data are also well defined but with a different direction , and the fir data in the core region are well defined and with yet another direction , but are randomly distributed in the filament region . on the scale of a few times the mean radial dimension of the molecular cloud , it is as if the magnetic field were ` blind ' to the spatial distribution of the filaments while on smaller scales within the cloud , in the core region near the iras point source psc 20503 + 6006 , polarimetry shows a rotation of the magnetic field lines in these denser phases . hence , in spite of the fact that the spatial resolution is not the same in the visible / nir and in the fir data , all the data put together indicate that the field direction changes with the spatial scale . finally , the chandrasekhar and fermi method is used to evaluate the magnetic field strength , indicating that the core region is approximately magnetically critical . a global interpretation of the results is that in the core region an original poloidal field could have been twisted by a rotating elongated ( coreenvelope ) structure . there is no evidence for turbulence and ambipolar diffusion does not seem to be effective at the present time .

astro-ph0608188

with the measurement of visible linear polarization of field stars , followed by developments of near and far infrared ( respectively nir and fir ) and submillimetric polarimetry , it is now possible to trace magnetic field structures from the scale of the milky way to that of star forming regions and cores . in the diffuse interstellar medium ( ism ) , the general assumption that elongated dust grains usually have their smallest axis aligned in a direction comparable to that of the local magnetic field has lead to a toroidal representation of the galactic magnetic field . a similar topology of the magnetic field can also sometimes be observed in other galaxies ( e.g. @xcite ) . on the smaller scales of molecular clouds , a similar relation exists between grain alignment and magnetic field directions ( see @xcite for an exciting review on the subject ) and many topologies of the magnetic field are suspected to exist . among others , helicoidal , toroidal , poloidal and hourglass configurations have been proposed to interpret observed polarization patterns , taking care of the physical conditions in these environments ( e.g. , orion a , m17 ) . however , one of the most important limitations of linear polarimetry is that we only get a two - dimensional projected average representation of the magnetic field topology because of the integration on the plane of the sky of the contribution to the polarized emission due to each aligned grain , or its dichroic absorption for visible / nir observations . this leads to a degeneracy problem since many three dimensional ( 3d ) topologies of the field can produce identical polarization patterns . a method to get rid of this degeneracy problem is the use of models . on the scale of relatively nearby filamentary molecular clouds and cores , this method has been used and allows in some cases to select one field topology among many possible ones ( e.g. , @xcite , @xcite ) . in addition to models , other diagnostics of magnetic fields can be used to lift the degeneracy due to the plane - of - sky integration . when column densities are high enough , observations of the zeeman effect can yield the line - of - sight component of the magnetic field strength ( e.g. @xcite ) . finally , by determining the ion - to - neutral molecular line - width ratio in non turbulent regions with a linear flow , it is possible to determine the angle between the field direction and the line of sight ( @xcite , @xcite , @xcite ) . by combining all three methods together , we get the 3d field topology ( e.g. , @xcite for m17 , @xcite for orion a ) . the observation of zeeman splitting of spectral lines requires bright dense cores , but even in other regions , it is possible to get the 3d geometry of the field even if the field strength is unknown . these methods can inform us about how magnetic fields can contribute to cloud dynamics , but our ultimate goal is to understand the competition between gravitational , thermal , magnetic , and turbulent forces . more specifically , magnetic fields and turbulence are both competing to slow down the star formation process , otherwise stars would form much faster than observations show . combining polarization modeling techniques ( e.g. , @xcite ) and predictions for magnetized simulations of turbulence ( e.g. , @xcite , @xcite ) , one can hope to be able to determine the relative contributions of magnetic fields and turbulence . to reach this goal , and to understand how magnetic fields thread the dark filaments , and to put observational constraints on modeling we propose to make comparisons of magnetic fields as seen in and around individual objects . this approach , namely a multi - scale analysis of the magnetic field , has begun to be done via comparisons of visible , nir , fir and/or submm polarization data in the omc1 ( @xcite ) , omc3 ( @xcite ) and gf 9 regions ( @xcite ) . in these works , aspects of the galactic magnetic field extending on tens of degrees are compared with aspects of the clouds probed in the densest regions of the clouds with resolutions @xmath2 , but there is generally a lack of information between these two scales . here we present observations at an intermediate scale , as much as the presence of field stars allows it , to have a clearer view of the topology of the field at the interface ism / filamentary molecular clouds . from a general point of view , this approach should provide new insights on the way magnetic fields and clouds are dynamically coupled and shaped . in this paper we address this problem for the gf 9 filamentary molecular cloud region . a forthcoming paper will address the more complex situation for the orion a region . a description of data available about gf 9 is given in section [ gf9region ] . observations and observational results are presented in section [ obsres ] . an analysis and a discussion follow in section [ discuss ] , and our conclusions are given in section [ conclu ] . the gf 9 dark filamentary structures have been catalogued first as b150 by @xcite , l1082 by @xcite and gf 9 by @xcite . a first estimate of the distance to these condensations as @xmath3 pc was determined by @xcite ( see @xcite ) but an estimate of @xmath4 pc to the extreme class 0 object gf 9 - 2 based on star counts was reported more recently by @xcite . the first estimate is reported in catalogs or works on relatively large core samples by @xcite , @xcite , @xcite , @xcite , @xcite , while the second one is reported in works by @xcite , @xcite and @xcite . assuming @xmath4 pc , gf 9 - 2 ( see figure [ skymap ] ) has a luminosity @xmath5 , an envelope mass @xmath6 , a ratio @xmath7 and a bolometric temperature @xmath8k . no outflow manifestation , nor structures are detected but infall motions are observed ( @xcite , @xcite , @xcite and @xcite ) . gf 9 - 2 must be particularly young since it has a h@xmath9o maser but no detected outflow ( @xcite , @xcite ) . two @xmath10 regions , namely gf 9-core and gf 9-fila , have been observed by @xcite , @xcite . the two areas are shown in figure [ skymap ] by blue boxes identified by c and f. the gf 9-core region is associated with the iras point source class 0 protostar psc 20503 + 6006 , but also with gf 9 - 2 and an object named [ lm99]351 . jhk photometry of these two regions probe similar masses but the filament region is consistent with a uniform - density cylindrical cloud , while the core region contains a centrally condensed core with the highest extinction region lying to the north of the iras point source . co , @xmath11co and cs surveys of these two regions give consistent views with these two pictures , but also reveal infall / outflow motions in order to explain cs line broadening in the core region where n@xmath12 @xmath13 ( assuming @xmath14 pc ) . in the filament region , n@xmath15 @xmath13 ( see table 2 in @xcite ) . co and nh@xmath16 observations of three regions with an infrared association are presented by @xcite . these sources are iras 20520 + 6003 , 20526 + 5958 and 20503 + 6006 , and they are associated respectively with the ldn 1082a , b and c cores . an analysis of these data by @xcite shows that motions consistent with uniform rotation in dense cores are present in ldn 1082a and ldn 1082b , while not in ldn 1082c . however recent works by @xcite show a velocity gradient along the major axis in the core in gf 9 - 2 which could be due to a rotation of the core . @xmath11co observations by @xcite have also been done at 3 positions in gf 9 . referring to @xcite , objects named [ lm99]349 and [ lm99]350 ( see figure [ skymap ] ) have no associated embedded young stellar objects ( eyso ) while [ lm99]351 ( @xmath17 west of gf 9 - 2 ) and [ lm99]352 have associated eyso . first @xmath18 @xmath19 m fir polarimetry observations from space with iso in gf 9 were done in the core and filament regions by @xcite . additional measurements at @xmath20 @xmath19 m on 9 stars shining through the same regions of gf 9 are also reported by @xcite . finally , 850 @xmath19 m polarimetry observations were done at the jcmt by jane s. greaves in 1998 at one position in the filament . initially , the relatively small heliocentric distance of gf 9 , its filamentary shape , and potentially interesting fir measurements in this region were the major factors which motivated us to conduct visible polarimetry around these filamentary dark structures . the observations were carried out on the 1.6 m telescope at the observatoire du mont - mgantic ( omm ) , qubec , canada , between 2000 september and 2003 july using a 8.2 arcsecond aperture hole and a broad band red filter ( rg645 : 7660 central wavelength , 2410 fwhm ) . polarization data were taken with beauty and the beast , a two - channel photoelectric polarimeter , which uses a wollaston prism , a pockels cell , and an additional quarter wave plate . the data were calibrated for polarization instrumental efficiency , instrumental polarization ( due to the telescope mirrors ) and zero point of position angles using a prism , non polarized and polarized standard stars , respectively . on average , the instrumental polarization was @xmath21 and was subtracted . the observational errors were calculated from photon statistics and also include uncertainties introduced by the previously mentioned calibrations . the final uncertainty on individual measurements of the polarization @xmath22 is usually around @xmath23 . for more details on the instrument and the observational method , see @xcite . the jcmt data archive has been used and a treatment of the data taken in 1998 was done with commonly used scuba data reduction routines . data were taken in relatively strong water vapour absorption ( @xmath24 ) conditions , and we finally got very low signal to noise polarization ratios and no significant results out of noise were found . most of the stars were chosen with the gsc interface at the canadian astronomy data center ( cadc ) . the others were selected directly on digitized sky survey ( dss ) red plates . all stars observed are compiled in table [ pvisdata ] . column 1 gives the gsc number . column 2 shows a gf 9 star designation number used in this paper . columns 3 and 4 give equatorial coordinates at epoch 2000 . degrees of polarization and equatorial position angles with their uncertainties are given in columns 5 and 6 respectively . column 7 gives the visible extinction coefficient @xmath25 from the @xcite atlas . finally , 1 is tabulated in column 8 when @xmath26 ( meaning @xmath27 ) , otherwise 0 is tabulated in this column . we note that when a star has no gsc number , not much information is generally available about it . figure [ polmap ] presents a polarization map of stars for which the polarization degree meets the condition @xmath26 . this map is to be compared to figure [ skymap ] . the histogram of the degree of polarization is shown in figure [ phisto ] . if we only consider data for which @xmath28 ( data shown with full lines ) , the mean and the standard deviation of the distribution are @xmath29 and @xmath30 , respectively . the histogram of the position angles is shown in figure [ anghisto ] . if we only consider data for which @xmath27 ( data shown with full lines ) , the mean position angle and the standard deviation of this well peaked distribution are @xmath31 and @xmath32 , respectively . this means a standard deviation of 0.445 rad , comparable to those found for clouds with embedded clusters ( see myers @xmath33 goodman 1991b ) . stars s35 , s70 , s75 , s86 and s92 are identified on both maps in figures [ skymap ] and 2 . stars s35 and s86 are suspected to have variable linear polarization . these stars are located in the continuity of two filaments but in gaps where extinction appears to be relatively low . the data relative to these stars are compiled in table [ vardata ] with the respective julian date ( j.d . ) of each measurement appearing in the last column . the position angle of stars s70 , s75 and s92 differ significantly from the mean direction of the other stars ( @xmath31 with @xmath34 ) . star s70 is a visual binary whose components fit both in the aperture hole used for the observations ; its position angle is perpendicular to this mean position angle . thus , we may be in the presence of a physical binary system with circumstellar material , or of a single system presenting intrinsic polarization and a background star . the difference of the position angles of stars s75 and s92 with the average value may be related to their position relative to the clouds as seen on figure [ skymap ] . ( @xmath35 ) is located north - east of iras 20520 - 6003 at the edge of two filamentary structures mainly parallel . comparatively , star s92 is also located in a dense region ( @xmath36 ) in a small filamentary structure oriented mostly north - south . given their large extinction , these two sources could either be deeply embedded in their clouds or be young stellar objects surrounded with disks in which multiple scattering can produce a relatively high degree of polarization ( e.g. , @xcite ) . without multiwavelength data it is difficult to distinguish between these possibilities . three stars in our sample , gf 9 s1 , gf 9 s2 and gf 9 s15 have measured spectral types and visual magnitudes . however we could not get reliable distances from these values . the @xmath37 values obtained are negative or yield very large distances . similarly using @xmath37 values from @xcite , we could not get reliable distances . in order to study the variations of the degree of polarization with visible extinction , we used the @xcite catalog to determine @xmath25 , in the direction of each star for which polarization measurements are available . this catalog is the first version of the atlas and catalog of dark clouds derived by using the optical database digitized sky survey i ( dss ) , and applying a traditional star - count technique to 1043 plates contained in the dss . using the visible extinction coefficient @xmath25 listed in table [ pvisdata ] , values of @xmath39 against @xmath40 , where @xmath41 , are shown in figure [ pav ] . for all stars , except gf 9 s35 and gf 9 s86 ( suspected variables ) , data directly observed at this wavelength are shown with crosses . assuming nir data from @xcite and visible data probe the same types of grains , the serkowski law , @xmath42,\ ] ] where @xmath43 , is used to derive the relation @xmath44 . this relation is used to convert the h - band data to equivalent @xmath45 data , assuming @xmath46 . these equivalent data are listed in table [ irtovis ] and are shown with diamonds in figure [ pav ] . as was previously mentionned by @xcite , no saturation of @xmath22 can be seen at about @xmath47 ( or @xmath48 ) , and @xmath49 , as observed in cold dark clouds in taurus ( see @xcite ) . data in the visible confirm this analysis but over the whole filament and the region surrounding it , and not just in the core region . thus , while not universal , it is possible in some regions to probe magnetic fields inside dark clouds via polarimetry of background stars as suggested here by data taken in the visible and in the nir . as can be seen in figure [ polmap ] and in the histogram of position angles in figure [ anghisto ] , the polarization pattern in the vicinity of gf 9 is relatively well oriented on the plane of the sky with a mean direction of @xmath31 and a standard deviation @xmath50 . since the filament has an incurved concentric shape reminiscent of supernova remnants ( @xcite ) , one can ask if a possible object at the origin of this shape could be located somewhere in the region delimited by the black circle traced in figure [ skymap ] . a possible candidate seems to be the post agb star glmp 1012 ( see figure [ skymap ] ) but given its distance estimate @xmath51 3 kpc ( @xcite ) compared with those of gf 9 , even by considering @xmath52 km s@xmath53 winds , it is difficult to defend the view that this source can be responsible for the shape of the filaments . however since the problem is tridimensional , a source at the origin of such a shape could also be located outside of the domain defined by this circle . maybe such a source exists but has not been detected , or the apparent concentric shape of the dense ism is only a projection effect and other mechanisms , such as turbulent fluctuations in the ism or density waves , are at the origin of its formation . a quick look in the catalogs reveals no potential candidate in a circle a few degrees in diameter around the position r.a.(2000 ) = 20@xmath54 and dec.(2000 ) @xmath55 . if now one compares the direction of the inferred magnetic field with the orientation of the filamentary structures on the plane of the sky , it seems that the magnetic field has approximately the same orientation over an arc circle which covers about 135@xmath56 , from a position angle ( p.a . ) = 270@xmath56 to 45@xmath56 as measured from the center of the circle in figure [ skymap ] . when looking at both maps from east to west in figure [ skymap ] and 2 , a slow rotation of the direction of the field is present , the average directions near the core and near the filament regions differ by about 10@xmath56 ( see table [ means ] ) , significantly under the variation in p.a . of the filament over the whole region . moreover from east to west , there is a great dispersion in the orientation of the filamentary structures , thus it is hard with this two dimensional representation to see a direct impact of the field on the apparent shape of the filaments . this is in agreement with the fact that in other regions there is no prefered projected angle on the plane of the sky between the directions of the ambient magnetic field and the orientation of the filamentary structures ( e.g. @xcite , @xcite ) . see also @xcite for a discussion on this subject . comparisons of the orientation of the magnetic field according to the visible , ir and fir data around and in the core and the filament regions can be done by looking at the maps in figures [ mapcore ] and [ mapfila ] respectively . in the core region , when going from visible to ir , to fir data , one can see an anticlockwise rotation of the p.a . by about 63@xmath56 in total . for each of these two regions , the means and dispersions of the degree of polarization and of the p.a . of the visible , ir and fir data are compiled in table [ means ] . the p.a . for the fir data corresponds to the observed polarization , rotated by 90@xmath56 to give the direction of the magnetic field . in figures [ mapcore ] and [ mapfila ] , only fir data for which the p.a . is approximately the same when using the klc instrumental polarization and the c - off instrumental polarization ( clemens 1999 ) are shown . the degree of polarization of the fir data is not considered in the present analysis . table [ means ] also includes similar information for the whole gf 9 region and for the local galactic scale , to be discussed below . _ comparison of visible and ir data _ first , visible and ir data do not probe the same regions on the plane of the sky . visible data probe the more diffuse parts around prestellar cores while the ir data probe the denser phases where protostellar cores are present . secondly we note that in the core region the ir vectors have approximately the same orientation which implies that the ir polarization is not intrinsic to each star but produced by dichroic absorption by aligned grains . this statistical argument , however , does not apply in the filament region where only one ir measurement with sufficient signal - to - noise ratio is available . thirdly , in both cases the resolution used with each technique is comparable . fourthly , the simple fact that the ` ir ' magnetic field has not the same orientation than the ` visible ' one implies that if the grains observed with these two techniques have identical properties , and if their dichroic absorption properties are the same at both ir and visible wavelengths , then the net polarization produced by aligned grains located in the densest parts of the cloud dominates the net polarization produced by aligned grains located in the foreground diffuse ism . all these facts , in addition to the dependence of @xmath22 with @xmath25 mentionned in section [ pversusav ] , suggest that there is a real rotation of the magnetic field lines in the core region when moving across the plane of the sky . _ comparison of ir and fir data _ firstly , ir and fir data probe the same regions on the plane of the sky in both the core and the filament regions . secondly we note that in the core region the fir vectors are very well aligned but not with the same p.a . than the ir vectors . ir data result from a magnetic field oriented at p.a . @xmath57 while fir data suggest a magnetic field oriented at p.a . thirdly , the resolution with these two techniques is not the same . the fir data resolution is @xmath59 , many times the ` pencil ' beam resolution of the ir and visible data . finally , fir observations probe the radiation emitted by grains mainly located in the densest parts of the cloud while ir observations probe grains located between the stars and the observer . when looking at figure [ mapcore ] , we see that ir and fir data cover a common area @xmath60 in size . thus if these two observational techniques probe the same column densities but at two different resolutions , one should be able to explain the rotation of the magnetic field lines as seen at these two wavelengths . a first hypothesis is that a given magnetic field morphology leads to several orientations according to the resolution at which it is observed ( see @xcite , but also @xcite ) . however , here we see that ir data are spatially distributed through fir data , thus by using a simple principle of juxtaposition it seems impossible to reproduce the mean orientation of fir vectors seen through @xmath59 beams by combining several ir vectors with very small cross sections . on the other hand , since the mean position angle of the stars observed in the visible is not the same than the one for stars observed in the ir , these ir - observed stars should not be foreground to the clouds , but rather embedded in or background to the clouds . in what follows we review possible explanations for the shift in position angles . \(1 ) the stars observed in the ir could be located inside the gf 9 core . if this was the case , the impact of their presence on their immediate environment should be observed . iras observations of these stars show that their luminosities satisfy @xmath61 ( see section [ gf9region ] ) , which would argue against them being class 0 sources , i.e. , very young . therefore , we reject the hypothesis that these ir stars are all embedded in the gf 9 core . \(2 ) the stars observed in the ir could be background to the gf 9 core region . this is suggested by their position in the @xmath62 diagram shown in figure [ pav ] . two possibilities can be invoked to explain the difference in position angles : + ( a ) a first and may be naive approach is assume that grains along the line of sight to the gf 9 core are all at the same temperature , and that both the ir and fir observational techniques probe all grains with the same efficiency . if this is right , the mean position angles observed at both wavelengths should be similar , except for the known 90@xmath63 difference between them . in this scenario , the difference in position angles is explained by a second cloud or filamentary structure located behind the gf 9 core , and in which dust grains would be aligned differently . + ( b ) a second and probably more realistic scenario is to emphasize differences related to the intrinsic nature of both observational techniques . based on point ( 1 ) mentioned above , the assumption of a dust temperature gradient for an externally heated cloud seems reasonable . in this case , the fir measurements should be more sensitive to `` cold '' dust grains aligned in the densest regions of the core than to `` hot '' dust grains aligned in the envelope and in the diffuse ism . on the other hand , ir observations should probe all the grains along the line of sight , independently of their temperature . thus grains aligned differently in the envelope than in the denser core should both contribute to the ir polarization position angle , explaining the offset in position angles . while scenario ( 2b ) does not reject the possibilty of a second cloud or filament evoked in ( 2a ) , it has the benefit to propose a simple and consistent explanation . thus we prefer scenario ( 2b ) , in which the intermediate position angles of ir vectors shown in figure [ mapcore ] may result from a vectorial addition of both fir and visible vectors covering this region and its vicinity . _ comparisons of the magnetic fields orientations with the cores elongations _ we compare in table [ offsets ] the orientations of five protostellar cores spread into the filaments with the magnetic field orientations . orientations of the cores l1082a , b and c were determined with the @xmath64 lines of nh@xmath16 while orientations of the lm cores are based on the visible dss data for l1082c and visible extinction for lm351 . ] at first sight we see that , except for l1082b , these dark condensations have a tendency to be elongated along the filaments as was noticed by @xcite ( see figures [ skymap ] and [ polmap ] ) . names , positions and p.a . of these cores are shown in table [ offsets ] in columns 1 through 4 respectively . for objects lm351 and l1082c , located in the core region , and for objects lm349 and lm350 , located in the filament region , the offset , @xmath65 , on the plane of the sky between the p.a . of each protostellar core and the mean magnetic field orientation probed with visible data is shown in column 5 . the same is done with the ir and fir measurements in columns 6 and 7 respectively . these offsets were not estimated for l1082b since the magnetic field direction is undersampled around this object . on larger scales , the magnetic field is mostly along the minor axis of the cores , as can be seen in figure [ polmap ] . however , on smaller scales , if we accept that ir measurements can probe the magnetic field in dense regions as it is suggested by the variations of @xmath22 with @xmath66 shown in figure [ pav ] , we see that a rotation of the field is apparent in the core region . on this smaller spatial scale the field is well aligned in the core region in a direction mainly parallel to the cores major axes . turbulence models such as those developed by @xcite show that matter may flow onto cores along filaments with the field being stretched along the filaments . in such a case , the field would tend to be along the major axis of the elongated low - density structures connected with higher density cores . however , such models show that the polarization pattern probing dust in the low density structures would be chaotic rather than smooth and well aligned . moreover , @xcite showed recently that a velocity gradient of 2.3 km s@xmath53 pc@xmath53 is present along the major axis of the core gf9 - 2 but none along its minor axis . therefore , if we assume that sources named lm351 , l1082c and gf 9 - 2 are associated to a common spatial volume traced with different tracers , and if we assume that the velocity gradient along the major axis of gf 9 - 2 is due to rotation of the core , this picture suggests that an original poloidal field has been twisted on small scales by the rotation of the core into a toroidal field . this would be consistent with a magnetic support model . on the other hand , this effect does not seem to be present in the filament region where the magnetic field probed with ir data is mainly parallel to the small axes of the cores detected in this area , in approximately the same direction than the magnetic field seen at greater scales . but on the other hand if fir measurements are reliable , the chaotic fir polarization pattern suggests that turbulence plays an active role in the filament region . finally comparison of both regions suggests a magnetic field interacting differently , and maybe less with its surrounding medium in the filament than it does in the core region , in agreement with the evolutionary stage of these two regions . additional ir data would be necessary to confirm this point , and the quality of fir measurements should of course be confirmed in both regions . the mean orientation of the ambient magnetic field can also be compared to the local galactic magnetic field orientation . figure [ mapgal ] shows a polarization map in galactic coordinates based on measurements compiled by @xcite with data for which @xmath67 . the region presents a complicated picture because we are looking down a spiral arm and possibly because of the influence of the cygnus star - forming complex . in their works @xcite and @xcite have used a large set of polarization data and studied the inferred magnetic field geometry in this part of the galaxy . at first sight , the simple picture of a toroidal magnetic field implies that the magnetic field lines should point in the direction toward @xmath68 . in their statistical study @xcite show that this trend is effectively real and are also able to detect the curvature of the local spiral arm . as mentionned by @xcite , since gf 9 is at a relatively close distance it is probably threaded by the magnetic field in the local spiral arm . jones compares the orientation of the mean ir vector in the core region with the visible vectors on a greater scale , and shows that this vector points in the direction of the ` vanishing point ' approximately located at @xmath68 . we can see this vector denoted by ` ir ' in figure [ mapgal ] . we also show the mean visible vector ` v ' in this map . it does not point in the same direction than the ` ir ' vector but it is approximately parallel to its nearest neighbours . all these facts suggest that it is preferable to use data probing the diffuse parts of the ism in the environments of molecular clouds to compare ambient to the cloud and galactic magnetic field s orientations . in order to estimate if the core region observed in cs by @xcite is magnetically supercritical or magnetically subcritical , we use the @xcite method to estimate the magnetic field strength . the magnetic field strength in the plane - of - sky is estimated with the expression : @xmath69 where @xmath70 is the polarization p.a . dispersion in degrees , @xmath71 is the molecular hydrogen density in molecules @xmath13 , @xmath72 is the fwhm line width in km s@xmath53 , and @xmath73 is the velocity dispersion ( see @xcite ) . we use only the ir data for the core region since we have no way of estimating the turbulent velocity dispersion for the larger area corresponding to the visible data . the value @xmath74 is taken from table [ means ] using the jones ir data . this value is provided by a relatively small set of six measurements for which the mean of the uncertainties is @xmath75 . if instead of a direct average , we use a weighted average for estimating the dispersion in p.a . , we get @xmath76 . finally , working with the p.a . directly instead of the vectors , the dispersions are @xmath77 and @xmath78 , weighted and unweighted respectively . all these values are close to each other and reflect the fact that there is little dispersion in p.a . in the cloud . therefore , we believe that despite the small number of measurements , we nevertheless have a representative value for the dispersion . the various parameters that we used for the cs core region and our results are shown in table [ strengths ] . once @xmath79 is derived , we estimate @xmath80 , the observed mass to flux ratio @xmath81 normalized to its critical value . this observed normalized mass to flux ratio is subject to geometrical bias , as explained by @xcite . in the case of estimating the total field from the plane - of - sky component of the field , two of the three components are known and the correction factor for @xmath80 is then @xmath82 , which corresponds to @xmath83 given in table [ strengths ] . this correction is statistical in nature since we do not know the appropriate angle for any individual object . in fact , there may not be a line of sight component to the field . using @xmath84 for the dispersion in p.a . is the quadratic sum of the dispersion @xmath85 and the mean uncertainty @xmath75 ] to take into account the uncertainties in the measurements would decrease the magnitude of @xmath79 and increase the magnitude of @xmath80 and @xmath87 by about 50% . for comparison , the dispersion in p.a . for the visible data around the core region is @xmath88 ( table [ means ] ) . with this value ( even though we are outside the area for which the cs line - widths have been measured ) , the values of @xmath80 and @xmath83 are larger by about 110% . these values are probably too large because the visible data cover a much larger area than just the core region . also , a bias exists in the chandrasekhar - fermi method that tends to cause estimated field strentghs to be too large and thus to underestimate values of @xmath80 ( e.g. @xcite ) . what is the effect of the distance @xmath89 on these results ? the column density @xmath90 is used to get the gas mass , so @xmath91 , where @xmath92 is the radius of the cloud , assumed to be spherical . the density @xmath93 . since @xmath94 , @xmath95 , @xmath96 and @xmath97 . if the distance is 200 pc instead of 440 pc as assumed for the values given in table [ strengths ] , @xmath98 would be closer to unity by about @xmath99 . is there additional possible evidence for a subcritical core ? @xcite identified two other tests to distinguish between ambipolar diffusion and turbulence as supporting mechanisms for star formation . if ambipolar diffusion is acting , one would expect an hourglass configuration for the magnetic field . but the distribution of the ir polarization vectors ( see fig [ mapcore ] ) suggests a smooth field , or even a slight curvature in the opposite direction than expected for an hourglass shape . a third test identified by @xcite , if ambipolar diffusion is acting one expects @xmath100 with @xmath101 approaching 1/2 for larger densitites ( @xmath102 @xmath13 ) . the values derived here are more or less compatible with this relation ( see figure 1 in @xcite ) . in addition , the comparison of the magnetic field orientations with the elongations of the cores seen in table [ offsets ] is not in general compatible with ambipolar diffusion , since in that case the cores would be perpendicular to the direction of the magnetic field . taking all the above arguments into account , our data suggest that the gf 9 core region is approximately magnetically critical . ambipolar diffusion does not appear to be supported . also , turbulence would produce a randomness in the polarization vectors which is not observed . the picture presented in section [ tosmall ] of an original poloidal field which is transformed into a toroidal field by rotation appears most plausible . visible linear polarization data taken at the mont - mgantic observatory on 78 stars in the vicinity of the dark filamentary cloud gf 9 were presented and compared with available nir and fir polarization data covering part of the same region . our main conclusions are : there is no evidence for saturation of polarization as a function of extinction for both the visible and nir data , suggesting that even in the denser regions the same grain properties and grain alignment mechanisms apply , and that the magnetic field is still being probed adequately there . all these facts put together suggest that _ in gf 9 the magnetic field changes direction with the spatial scale . _ the field differs by @xmath103 in the gf 9 core compared to the surrounding region . ambipolar diffusion does not appear to be supported given the current evidence . the effects expected from turbulence are definitely missing . a global interpretation of the results is that an original poloidal field could have been twisted into a toroidal configuration by a rotating elongated structure during its collapse . future work should check if the trend in polarization position angles to rotate counterclockwise with increasing wavelength still continues in the submm . also , as the lack of saturation of polarization as a function of extinction suggests , if dust grains are magnetically aligned there should be no polarization hole in the submm in the core and the filament regions . as an additional benefit , submm polarization would afford a higher spatial resolution than the fir data discussed here . we thank the conseil de recherche en sciences naturelles et en gnie du canada for supporting this research . the authors thank j. s. greaves for providing information about observations of gf 9 with scubapol at the jcmt , as well as b. malenfant and g. turcotte for their helpful and friendly support during observations at mont - mgantic observatory . we also thank an anonymous referee for his constructive comments and helpful suggestions . this work made extensive use of the simbad database at the canadian astronomy data center , which is operated by the dominion astrophysical observatory for the national research council s herzberg institute of astrophysics . andr , ph . ; ward - thompson , d. and barsony , m. 2000 , protostars and planets iv , p59 arce , h. g. ; goodman , a. a. ; bastien , p. ; manset , n. and sumner , m. 1998 , apj 499 , l93 barnard , e. e. 1927 , atlas of selected regions in the milky way , carnegie institute of washington bastien , p. and mnard , f. 1990 , apj 364 , 232 benson , p. j. @xmath33 myers , p. c. 1989 , apjss , 71 , 89 chandrasekhar , s. and fermi , e. 1953 , apj 118 , 113 ciardi , d. r. ; woodward , c. e. ; clemens , d. p. ; harker , d. e. and rudy , r. j. 2000 , aj 120 , 393 ciardi , d. r. ; woodward , c. e. ; clemens , d. p. ; harker , d. e. and rudy , r. j. 1998 , aj 116 , 349 clemens , d. ; kraemer , k. and ciardi , d. 1999 , workshop on iso plarisaton observations , eds . r.j . laureijs @xmath33 r. siebenmorgen . esa - sp 435 , p.7 crutcher , r. m. 2004 , astrophysics and space science , v. 292 , issue 1 , p. 225 crutcher , r. m. 1999 , apj 520 , 706 crutcher , r. m. ; nutter , d. j. ; ward - thompson , d. and kirk , j. m. 2004 , apj 600 , 279 dobashi , k. ; uehara , h. ; kandori , r. ; sakurai , t. ; kaiden , m. ; umemoto , t. and sato , fumio 2005 , pasj 57 , sp1 , 1 dobashi , k. ; bernard , j. p. ; yonekura , y. and fukui , y. 1994 , apjss 95 , 419 fiege , d. j. and pudritz , r. e. 2000 , apj 544 , 830 froebrich , d. 2005 , apjs 156 , issue 2 , pp . 169 - 177 furuya , r. s. ; kitamura , y ; wootten , a. ; claussen , m. j. and kawabe , r. 2003 , apjss , 144 , 71 furuya , r. s. ; kitamura , y ; shinnaga , h. 2005 , poster presented at protostars and planets v , no 8087 goodman , a. a. ; benson , p. j. ; fuller , g. a. and myers , p. c. 1993 , apj 406 , 528 goodman , a. a. ; jones , t. j. ; lada , e. a. and myers , p. c. 1995 , apj 448 , 748 gsten , r. 1994 , in _ the cold universe _ , eds . th . montmerle et al . ( editions frontires : gif - sur - yvette ) , p.169 heiles , c. ; goodman , a. a. ; mckee , c. f. and zweibel , e. g. 2000 , protostars and planets iv , p279 heiles , c. 2000 , aj 199 , 923 heiles , c. 1996 , apj 462 , 316 heitsch , f. ; zweibel , e. g. ; mac low , m - m . ; li , p. ; norman , m. l. , 2001 , apj 561 , 800 heyer , m. h. ; vrba , f. j. ; snell , r. l. ; schloerb , f. p. , strom , s. e. , goldsmith , p. f. & strom , k. m. 1987 , apj 321 , 855 hilton , j. @xmath33 lahulla , j.f . 1995 , aass 113 , 325 houde , m. ; bastien , p. ; dotson , j. l. ; dowell , c. d. ; hildebrand , r. h. ; peng , r. ; phillips , t. g. ; vaillancourt , j. e. and yoshida , h. 2002 , apj 569 , 803 houde , m. ; bastien , p. ; peng , r. ; phillips , t. g. and yoshida , h. 2000a , apj 536 , 857 houde , m. ; peng , r. ; phillips , t. g. ; bastien , p. and yoshida , h. 2000b , apj 537 , 245 houde , m. ; phillips , thomas g. ; bastien , p. ; peng , r. and yoshida , h. 2001 , apj 547 , 311 houde , m. ; dowell , c. d. ; hildebrand , r. h. ; dotson , j. l. ; vaillancourt , j. e. ; phillips , t. g. ; peng , r. and bastien , p. 2004 , apj 604 , 717 jones , t. j. 2003 , aj , 125 , 3208 lazarian , a. 2003 , jqsrt 79 - 80 , 881 - 902 lee , c. w. @xmath33 myers , p. c. 1999 , apjss 123 , 233 lynds , b. d. 1962 , apjs 7 , 1 manset , n. and bastien , p. 1995 , pasp 107 , 483 matthews , b. c. ; wilson , d. w. and fiege , j. d. 2001 , apj 562 , 400 myers , p. c. , fuller , g. a. , goodman , a. a. , @xmath33 benson , p. j. 1991a , apj , 376 , 561 myers , p. c. and goodman , a. a. 1991b , apj 373 , 509 padoan , p. ; goodman , a. a. ; draine , b. t. ; juvela , m. ; nordlund , . and rgnvaldsson , . e. 2001 , apj 559 , 1005 preite - martinez , a. 1988 , a@xmath33a 76 , 317 scarrott , s. m. ; rolph , c. d. ; @xmath33 semple , d. p. 1990 , in galactic and interstellar magnetic fields , iau symposium 140 , edited by r. beck and p. p. kronberg ( kluwer , dordrecht ) , p. 245 serkowski , k. , mathewson , d. l. , @xmath33 ford , v. l. 1975 , apj vol . 196 , p. 261 - 290 . schleuning , d. a. 1998 , apj 493 , 811 schneider , s. @xmath33 elmegreen , b. g. 1979 , apjss 41 , 87 valle , j. p. ; greaves , j. s. and fiege , j. d.2003 , apj 588 , 910 viotti , n. r. 1969 , mem . soc . astron . italiana , 40 , 75 walawender , j. m. , zweibel , e. g. @xmath33 heiles , c. 2001 , baas , 198 , 41.05 wiesemeyer , h. 1998 , asp conf . series , vol . , 189 wiesemeyer , h. 1997 . the spectral signature of accretion in low mass protostars . phd dissertation , university of bonn lrrrrrrrrcc no&91&20:47:23.5&+60:12:59.5&[email protected]&[email protected]&1.61&1 + 0424600447 & 40 & 20:47:25.8 & + 60:17:41.2&[email protected]&[email protected]&1.34&1 + no&120&20:47:34.0&+59:51:54.4&[email protected]&[email protected]&2.09&1 + 0424601275&46&20:47:35.4&+60:03:35.1&[email protected]&[email protected]&1.89&0 + no&92&20:47:47.6&+60:01:02.2&[email protected]&[email protected]&2.14&1 + 0424601151&57&20:47:48.0&+60:06:49.4&[email protected]&[email protected]&1.78&1 + 0424600411&31&20:47:50.2&+60:14:40.6&[email protected]&[email protected]&1.39&1 + 0396300254&48&20:48:01.0&+59:57:01.1&[email protected]&[email protected]&1.89&1 + no&98&20:48:01.1&+60:00:47.1&[email protected]&[email protected]&2.13&0 + 0424600967 & 47&20:48:08.4&+60:00:24.6 & [email protected]&[email protected]&1.66&1 + 0424600359 & 39&20:48:11.5 & + 60:18:52.2&[email protected]&[email protected]&1.14&1 + 0424601339 & 36&20:48:17.9 & + 60:06:26.9 & [email protected]&[email protected]&2.22&1 + 0424600585 & 18&20:48:20.5 & + 60:01:15.2 & [email protected] & [email protected]&1.66&1 + 0424600629 & 17&20:48:25.5 & + 60:10:11.9 & [email protected] & [email protected]&1.97&0 + 0424600469 & 11 & 20:48:27.7 & + 60:32:54.5 & [email protected] & [email protected]&0.64&1 + 0424601143&13 & 20:48:30.0 & + 60:24:59.4 & [email protected] & [email protected]&0.98&1 + 0424601171 & 16&20:48:32.5 & + 60:15:08.4 & [email protected] & [email protected]&1.43&1 + 0424600521 & 10 & 20:48:40.8 & + 60:32:15.1 & [email protected] & [email protected]&0.66&1 + 0424600727 & 35&20:48:44.2 & + 60:09:52.9 & see table [ vardata]&&2.17 & + 0424601053 & 34&20:48:46.5 & + 60:09:42.4 & [email protected]&[email protected]&2.17&1 + 0424600659&54&20:48:47.1&+60:12:49.4&[email protected]&[email protected]&1.73&1 + 0424600961&37&20:48:47.3&+60:01:45.4 & [email protected]&[email protected]&1.24&1 + 0424601093&53&20:48:51.3&+60:13:46.1&[email protected]&[email protected]&1.69&1 + 0424600621 & 15 & 20:48:51.4&+60:15:27.6 & [email protected] & [email protected]&1.38&1 + 0424600109 & 14 & 20:49:03.7 & + 60:18:57.7 & [email protected] & [email protected]&1.27&1 + 0424601169&52&20:49:07.1&+60:05:05.0&[email protected]&[email protected]&1.17&1 + 0424600721&55&20:49:09.6&+60:18:43.4&[email protected]&[email protected]&1.56&1 + 0424600857 & 9 & 20:49:13.6 & + 60:30:03.6 & @xmath1050.11 & [email protected]&0.75&1 + no&90&20:49:14.1&+60:12:44.5&[email protected]&[email protected]&2.19&1 + 0424600123 & 8&20:49:14.4 & + 60:29:21.0 & 2.70@xmath104 0.12 & [email protected]&0.80&1 + 0424601087 & 38&20:49:15.7&+60:16:26.1 & [email protected]&[email protected]&1.97&1 + 0424600963&51&20:49:18.6&+60:05:50.6&[email protected]&[email protected]&1.17&1 + 0424600281&56&20:49:18.7&+60:20:20.9&[email protected]&[email protected]&1.42&1 + 0396300376 & 19&20:49:33.3 & + 59:56:54.5 & [email protected] & [email protected]&0.82&1 + 0424600981 & 7 & 20:49:33.7 & + 60:27:11.5 & [email protected] & [email protected]&0.85&1 + 0396300020 & 20&20:49:41.5 & + 59:56:15.7 & [email protected] & [email protected]&0.82&1 + 0424601133 & 21&20:49:44.1 & + 60:03:12.3 & [email protected] & [email protected]&0.84&1 + 0424601015 & 6 & 20:50:04.3 & + 60:24:04.9&[email protected] & [email protected]&1.35&1 + 0424601131 & 42&20:50:25.7&+60:24:15.8 & [email protected]&[email protected]&1.56&1 + 0424601183&50&20:50:25.8&+60:10:52.6&[email protected]&[email protected]&1.04&1 + 0424601341 & 33&20:50:31.9&+60:13:00.4 & [email protected]&[email protected]&1.49&1 + 0424601127 & 44&20:50:36.8&+60:26:15.6 & [email protected]&[email protected]&1.06&1 + 0424601041 & 32&20:50:37.2 & + 60:15:38.6 & [email protected]&[email protected]&1.84&0 + no&86&20:50:45.3&+60:20:01.4&see table [ vardata]&&2.36 & + 0424600477 & 41&20:50:46.3&+60:23:26.3&[email protected]&[email protected]&1.39&1 + 0396300338 & 22&20:50:48.3 & + 59:58:22.0 & [email protected] & [email protected]&0.77&1 + 0396300620 & 23&20:50:56.2 & + 59:59:30.6 & [email protected] & [email protected]&0.73&1 + 0424600655&29&20:51:07.8&+60:11:24.2&[email protected]&[email protected]&1.34&1 + 0424601145&27&20:51:11.2&+60:06:26.2&[email protected]&[email protected]&1.06&1 + 0424601213&28&20:51:12.6&+60:08:53.9&[email protected]&[email protected]&1.16&1 + 0424601111 & 4&20:51:12.8 & + 60:25:15.5 & [email protected] & [email protected]&1.33&1 + 0424600103 & 5 & 20:51:16.2 & + 60:33:19.9 & [email protected]&[email protected]&1.14&1 + no&70&20:51:32.3&+60:41:41.3&[email protected]&[email protected]&1.02&1 + 0424701143 & 43&20:51:33.6&+60:25:05.5&[email protected]&[email protected]&1.44&1 + 0424701033&45&20:51:34.7&+60:14:03.5 & [email protected]&[email protected]&2.07&1 + 0424701049&30&20:51:43.7&+60:11:50.0&[email protected]&[email protected]&1.59&0 + 0396300164 & 24&20:51:49.2 & + 59:54:15.5 & [email protected] & [email protected]&1.01&0 + 0424701155&26&20:51:51.7&+60:05:31.0&[email protected]&[email protected]&1.13&1 + 0424701211 & 3 & 20:51:53.4 & + 60:25:18.5&[email protected] & [email protected]&1.36&1 + no&85&20:51:55.4&+60:19:00.0&[email protected]&[email protected]&2.29&0 + 0424701127 & 2 & 20:51:57.1 & + 60:22:44.6&[email protected] & [email protected]&1.87&0 + no&71&20:52:01.8&+60:38:13.2&[email protected]&[email protected]&1.13&1 + 0424701215 & 1 & 20:52:02.7&+60:22:27.9&[email protected] & [email protected]&1.87&0 + 0424701115&25&20:52:04.5&+60:03:55.6&[email protected]&[email protected]&1.11&1 + 0424701085&72&20:52:23.2&+60:32:48.4&[email protected]&[email protected]&1.36&1 + no&75&20:53:25.7&+60:16:32.6&[email protected]&[email protected]&3.29&1 + no&68&20:53:42.0&+60:24:07.8&[email protected]&[email protected]&2.39&1 + no&66&20:53:52.0&+60:26:37.7&[email protected]&[email protected]&2.52&1 + no&59&20:53:52.8&+60:35:44.6&[email protected]&[email protected]&1.45&1 + no&74&20:53:55.6&60:15:51.6&[email protected]&[email protected]&3.20&0 + no&69&20:53:55.7&+60:22:55.9&[email protected]&[email protected]&2.72&1 + no&69bis&20:53:59.0&+60:22:50.0&[email protected]&[email protected]&2.72&1 + no&61&20:54:01.3&+60:39:55.3&[email protected]&[email protected]&1.15&1 + no&77&20:54:27.8&+60:11:16.5&[email protected]&[email protected]&1.93&1 + no&65&20:54:28.8&+60:26:53.5&[email protected]&[email protected]&2.32&1 + 0424700062&62&20:54:31.3&+60:32:40.7&[email protected]&[email protected]&1.90&1 + no&64&20:54:53.0&+60:29:35.7&[email protected]&[email protected]&2.28&0 + 0424700686&73&20:56:15.6&+60:17:41.9&[email protected]&[email protected]&1.94&1 + lrrrr 35&[email protected]&[email protected]&2,452,054.64583 + 35&[email protected]&@xmath1068.7&2,452,141.72222 + 86&[email protected]&[email protected]&2,452,830.75347 + 86&[email protected]&[email protected]&2,452,835.78125 + lrrrr c1&[email protected]&@xmath107&2.34 + c2&[email protected]&@xmath108&2.34 + c3&[email protected]&@xmath109&2.60 + c4&[email protected]&@xmath110&2.34 + c5&[email protected]&@xmath111&2.60 + c6&[email protected]&@xmath112&2.34 + f2&[email protected]&@xmath113&2.19 + lrrrr @xmath114 ( @xmath115)&@xmath116 & @xmath117 & @xmath118 & @xmath119 + @xmath120 ( @xmath121 ) & @xmath122&@xmath123&@xmath124&@xmath125 + @xmath126 ( @xmath115)&@xmath127&@xmath128 & & + @xmath129 ( @xmath121 ) & @xmath130&@xmath131 & & + @xmath132 ( @xmath115 ) & ... & ... & & + @xmath133 ( @xmath121 ) & @xmath134&random & & + lrrrrrrrr lm349&20 49 6.9&60 11 48&47&@xmath135&@xmath136&random&f + lm350&20 49 46.5&60 15 40&49&@xmath137&@xmath138&random&f + lm351&20 51 28.0&60 18 33&43&@xmath139&@xmath140&@xmath141&c + lm352&20 53 49.9&60 11 27 & ... & ... & ... & ... & ... + l1082c&20 51 27.4&60 19 00&7&@xmath142&@xmath143&@xmath144&c + l1082a&20 53 29.6&60 14 41 & ... & ... & ... & ... & ... + l1082b&20 53 50.3&60 09 47&48 & ? & & ... + ll @xmath145 ( h@xmath9 @xmath13 ) & 15000 @xmath104 3000 + @xmath90 ( h@xmath9 @xmath146 ) & 9.9 @xmath147 + @xmath148 & 0.16 + @xmath149 @xmath150 & @xmath151 + @xmath152 & @xmath153 + @xmath154 ( @xmath155 ) & 0.88@xmath104 0.20 + @xmath70 @xmath156 & @xmath157 + @xmath79 ( @xmath158 ) & 170@xmath10456 + @xmath80 & @xmath159 0.44 + @xmath83 & 0.35 +

morphological and spectroscopic studies of high redshift clusters indicate that a significant fraction of present - day early - type galaxies was transformed from star forming galaxies at . on the other hand , the slow luminosity evolution of early - type galaxies and the low scatter in their color - magnitude relation indicate a high formation redshift of their stars . in this paper we construct models which reconcile these apparently contradictory lines of evidence , and we quantify the effects of morphological evolution on the observed photometric properties of early - type galaxies in distant clusters . we show that in the case of strong morphological evolution the apparent luminosity and color evolution of early - type galaxies are similar to that of a single age stellar population formed at , irrespective of the true star formation history of the galaxies . furthermore , the scatter in age , and hence the scatter in color and luminosity , is approximately constant with redshift . these results are consequences of the `` progenitor bias '' : the progenitors of the youngest low redshift early - type galaxies drop out of the sample at high redshift . we construct models which reproduce the observed evolution of the number fraction of early - type galaxies in rich clusters and their color and luminosity evolution simultaneously . our modelling indicates that% of early - type galaxies were transformed from other galaxy types at , and their progenitor galaxies may have had roughly constant star formation rates prior to morphological transformation . the effect of the progenitor bias on the evolution of the mean ratio and color can be estimated . the progenitor bias is a linear function of the scatter in the color - magnitude relation produced by age variations , and is maximal if the observed scatter is entirely due to age differences . after correcting the observed evolution of the mean ratio for the maximum progenitor bias we find that the mean luminosity weighted formation redshift of stars in early - type galaxies for and , and for and . our analysis places the star formation epoch of early - type galaxies later than previous studies which ignored the effects of progenitor bias . the results are consistent with the idea that ( some ) ly - break galaxies are star forming building blocks of massive early - type galaxies in clusters . 50min 054ms1054 0t_0 keywords galaxies : evolution , galaxies : elliptical and lenticular , cd , galaxies : structure of , galaxies : clusters

astro-ph0101468

early - type galaxies ( elliptical and s0 galaxies ) constitute @xmath12% of the galaxy population in the central regions of nearby rich clusters ( dressler 1980 ) , and studies of nearby and distant clusters have provided strong constraints on their evolution and formation . the cluster observations have been used to constrain the stellar ages of early - type galaxies . on one hand , there is very good evidence that early - type galaxies form a very homogeneous , slowly evolving population : the scatter in their colors is small , both at low redshift ( e.g. , bower , lucey , & ellis 1992 ) , and at high redshift ( ellis et al . 1997 ; stanford , eisenhardt , & dickinson 1998 ) , the scatter in their mass - to - light ratios is equally small ( e.g. , lucey et al . 1991 , pahre , djorgovski , & de carvalho 1998 , kelson et al . 2000 ) , and the evolution of their mass - to - light ratios is slow ( van dokkum & franx 1996 , kelson et al . 1997 , bender et al . 1998 , van dokkum et al . these results indicate a very high redshift of formation for the stars in early - type galaxies . as an example , van dokkum et al ( 1998a ) found from the evolution of the fundamental plane relation to @xmath13 that the stars in massive early - type galaxies must have formed at @xmath14 for @xmath15 , @xmath7 , and a salpeter ( 1955 ) imf . furthermore the low scatter in colors and @xmath3 ratios implies that the scatter in ages at any redshift is very small ( e.g. , ellis et al . 1997 , stanford et al . 1998 ) . on the other hand , evidence is accumulating that many early - type galaxies in clusters were relatively recently transformed from star forming galaxies . dressler et al . ( 1997 ) report a high fraction of spiral galaxies in clusters at @xmath16 . these galaxies are nearly absent in nearby rich clusters ( dressler 1980 ) , and hence must have transformed into early - type galaxies between @xmath17 and @xmath18 . other studies ( e.g. , couch et al . 1998 , van dokkum et al . 2000 ) have confirmed this trend , and extended it to @xmath19 . dressler et al.(1997 ) found that the increased fraction of spiral galaxies at high redshift was accompanied by a low fraction of s0 galaxies , and concluded that the @xmath20 spiral galaxies transformed into s0 galaxies . other studies have found evidence for merging and interactions in high redshift clusters ( lavery & henry 1988 ; lavery , pierce , & mcclure 1992 ; dressler et al . 1994 ; couch et al . 1998 ; van dokkum et al . these transformations may provide a way to form young elliptical galaxies in the clusters at late times . the morphological evidence for recent and rapid evolution in the cluster galaxy population is supported by photometric and spectroscopic studies , which have found a significant increase of the number fractions of star forming and post star burst galaxies from @xmath18 to @xmath21 ( e.g. , butcher & oemler 1978 , 1984 , dressler & gunn 1983 , couch & sharples 1987 , postman , lubin , & oke 1998 , poggianti et al . 1999 , van dokkum et al 2000 ) . hence the evidence appears secure that many early - type galaxies in clusters were transformed from star forming galaxies at fairly recent times . the purpose of this paper is to produce models which can bring these apparently different lines of evidence regarding the ages into agreement . specifically , we analyse the effects of late morphological transformations as found by dressler et al . , and others , on the observed evolution of colors and luminosities with redshift . we show that the effect of `` progenitor bias '' ( franx & van dokkum 1996 ; van dokkum et al . 2000 ) can be very significant : if early - type galaxies were transformed from other galaxy types at late times , the youngest progenitors of present - day early - type galaxies drop out of the sample of early - type galaxies at high redshift . hence the samples of galaxies at low and high redshift are not the same , and in particular , the high redshift samples might be biased towards the oldest progenitors of present - day early - type galaxies . the purpose of this paper is to quantify the effects of progenitor bias , and to estimate the corrections which are required for models which ignore morphological evolution . to that end , we construct relatively simple models for the history of star formation and morphological type of cluster early - type galaxies . we show that the observed slow luminosity evolution and low scatter in the color - magnitude relation and the fundamental plane are natural consequences of recent and ongoing assembly of early - type galaxies . we correct the observed evolution for the progenitor bias , and derive the corrected mean age of the stars in early - type galaxies . our models are extensions of the earlier work by van dokkum et al ( 1998b ) , bower et al . ( 1998 ) , and shioya & bekki ( 1998 ) . these authors considered complex star formation histories of cluster galaxies , but did not investigate the effects of morphological transformations on the observed properties of early - type galaxies at high redshift . the progenitor bias is implicitly included in the very detailed semi - analytical models by , for example , kauffman & charlot ( 1998 ) and baugh , cole , & frenk ( 1996 ) . in these complex models it is difficult to isolate and quantify specific effects , such as the progenitor bias . the main contribution of our work is that we use simple analytic models with only three free parameters , which allows us to provide explicit estimates for the effects of morphological evolution on the observed colors and luminosities of early - type galaxies . we construct simple models for the evolution of early - type galaxies , which parameterize both their star formation history and morphological evolution . our models allow the full history to be quantified by three parameters . they are not based on a full physical description of the formation and evolution of galaxies as derived by , e.g. , kauffmann , white , & guiderdoni ( 1993 ) and baugh et al . ( 1996 ) . we assume that early - type galaxies have similar histories of star formation and morphology . the history of each individual galaxy can be described by three parameters : the time when star formation starts , the time when star formation stops , and the variation of star formation between the start and end . the star formation is allowed to increase or decrease with time . it is assumed that the galaxy is classified as an early - type galaxy some time after the cessation of starformation . we assume that the galaxies do not stop their star formation at the same time , but rather have a distribution of cessation times . this description is fairly generic for many detailed theories of galaxy formation . we do not define the processes driving the morphological transformation and the cessation of star formation . various processes have been invoked , including gas stripping of spiral galaxies ( e.g. , gunn & gott 1972 , tytler & vidal 1978 , abadi , moore , & bower 1999 , kodama & smail 2000 ) , tidal interactions ( e.g. , fried 1988 , moore et al . 1996 ) , and mergers ( e.g. , lavery et al . 1992 , van dokkum et al.1999 ) . it does not matter in our analysis what process causes the cessation of star formation , and we allow for bursts of star formation to accompany this event . the transition phase between star forming galaxy and early - type galaxy is assumed to last for a significant time . immediately after the cessation of star formation galaxies are likely to show tidal features ( in case of an interaction or merger ) , or weak spiral arms . the morphologies of `` e+a '' or post star burst galaxies seem to be consistent with this scenario . they generally are classified as early - type spirals or tidally distorted systems ( poggianti et al.1999 ; fabricant et al . 2000 ; van dokkum et al . it is not well known how long it takes before these structures fade or disperse ; we assume that early - type galaxies are first classified as such @xmath22gyr after star formation in their progenitors ceased . in the current section we do not specify whether the models apply to elliptical galaxies , s0 galaxies , or both . observationally , the early - type galaxy fraction is more robust than the relative numbers of elliptical galaxies and s0 galaxies ( e.g. , fabricant , franx , & van dokkum 2000 , van dokkum et al . if different histories apply to the two classes of galaxies , two separate models will be needed for each sub - class . we explore such models in section 5 . the generic effects of late morphological transitions are illustrated in fig . [ modeltracks.plot ] . in panel ( a ) , the evolution of the mass - to - light ratio is shown for a population of galaxies with a small range in age , without any late morphological transitions . as can be seen , galaxies with young stellar populations evolve more rapidly than galaxies with old stellar populations , and hence the spread in @xmath3 ratios increases with redshift ( panel b ) . the mean evolution is faster than that of a single age population formed at @xmath23 , which is indicated by the dashed line in ( b ) . this type of modelling has been used extensively to constrain the ages of early - type galaxies in clusters : if there are no morphological transitions , the evolution of the mean color or luminosity constrains the mean age of early - type galaxies , and the scatter constrains the spread in ages ( e.g. , bower , lucey , & ellis 1992 , ellis et al . 1997 , van dokkum et al . 1998(a , b ) , bower et al . 1998 , stanford et al . 1998 , ferreras & silk 2000 ) . in panel ( c ) , a model is shown in which galaxies are transformed into early - type galaxies at a constant rate . the broken lines indicate galaxies before they are recognized as early - type galaxies . it can easily be seen that the sample of early - type galaxies at high redshift is a small subsample of all galaxies which are classified as early - type galaxies at @xmath18 . the two samples are therefore not directly comparable , and erroneous results are derived if morphological evolution is ignored . as a result , the evolution in the mean @xmath3 ratio of the early - type galaxies is very slow ( panel d ) . the slope of the @xmath3 @xmath24 relation is comparable to the slope for a single stellar population which formed at @xmath25 ( indicated by the dashed line ) , even though the mean formation redshift of the stars in all early - type galaxies at @xmath18 is low at @xmath26 . even more remarkable is the fact that the scatter in @xmath3 ratios is virtually constant . these effects are caused by the fact that the youngest galaxies continuously drop out of the sample going to higher redshifts . in the subsections below we specify a broader range of models , and explore the consequences . the full models are quantified by three parameters : @xmath27 , the time when star formation starts , @xmath28 , the time scale which characterizes the distribution of times when star formation stops , and @xmath29 , which describes the star formation rate between the start and end of star formation . the last two parameters are defined in the following way . the parameter @xmath28 determines the probability distribution of @xmath30 , the time when star formation stops for an individual galaxy : @xmath31 we show later that this expression can provide a satisfactory fit to the data . if @xmath32 , with @xmath33 the present age of the universe , star formation in the progenitors terminated at very high redshift . at the other extreme , @xmath34 corresponds to a constant transformation rate . for each individual galaxy , the morphological transformation to early - type galaxy occurs at @xmath35 , i.e. , @xmath36 gyr after truncation of star formation . the parameter @xmath29 characterizes the variation of the star formation rate . the star formation history of the galaxies can include bursts , and increasing or decreasing continuous formation rates as a function of time . after star formation ceases , the luminosity and color evolution of such a complex population is well approximated by a single age population of stars with the same luminosity weighted mean age ( e.g. , van dokkum et al . 1998b ) . hence for each galaxy we approximate the evolution of the complex population by that of a single age population formed at @xmath37 , with @xmath38 for @xmath39 , the stellar population is dominated by a burst at the end of the star formation history , and for @xmath40 the population is dominated by a burst at the start of the star formation history . if the star formation rate is approximately constant from @xmath41 to @xmath42 , then @xmath43 . we tested the accuracy of our approach by calculating the luminosity and color evolution of galaxies with complex star formation histories ( e.g. , an exponentially declining star formation rate followed by a star burst ) , and comparing the results to predictions from single burst models with the same values of @xmath44 . after transformation to early - type galaxy ( i.e. , @xmath45gyr after star formation has ceased ) our approximation of luminosity and color evolution is accurate to a few percent . three possible star formation histories of early - type galaxies are shown in fig . [ fstar.plot ] . the model with @xmath46 has a star burst at the end of its star formation history , and may be appropriate for spiral galaxies falling into clusters , or mergers . although these histories are quite complex , @xmath47gyr after star formation has ceased their evolution is similar to that of single age stellar populations formed at @xmath48 . our parameterization allows straightforward computation of luminosities and colors of galaxies , because the luminosity evolution of a single age stellar population can be approximated by a power law ( e.g. , tinsley 1980 , worthey 1994 ) . therefore , to good approximation , @xmath49 the coefficient @xmath50 depends on the passband , the imf , and the metallicity . in this paper we will limit the discussion to rest frame @xmath51 band luminosities , and rest frame @xmath52 colors . our results can easily be expressed in other ( rest frame ) bands . the worthey ( 1994 ) models give @xmath53 and @xmath54 for solar metallicity and a salpeter ( 1955 ) imf . the color evolution can be approximated by @xmath55 expressed in magnitudes , color and luminosity evolution are related through @xmath56 in these models . the evolution of the mean color and scatter in the color will therefore be similar to the scaled evolution of the mean @xmath3 ratio and its scatter . in the following , we use monte - carlo simulations to calculate model predictions for given values of @xmath28 , @xmath29 , and @xmath27 . this approach has the advantage that the mean and scatter in colors and luminosities at a given time can be computed using the same methods as for the observations . in the simulations , all galaxies start out as star forming objects at @xmath41 . each galaxy is assigned a value for @xmath30 , the time when star formation ceases . the distribution of @xmath30 is given by eq . [ first.eq ] , with boundary condition @xmath57 , with @xmath33 the present age of the universe . at each timestep @xmath58 , objects which satisfy the condition @xmath59 are selected as early - type galaxies . using eq.[levo.eq ] and [ colevo.eq ] these galaxies are assigned a luminosity and a color . the central value and the spread of the color and luminosity distribution at time @xmath58 are calculated with the biweight statistic ( beers , flynn , & gebhardt 1990 ) . this statistic was also used by stanford et al . ( 1998 ) and van dokkum et al.(1998b , 2000 ) to calculate the scatter in the high redshift color - magnitude relation . we note that the evolution of the mean @xmath3 ratio and colors of early - type galaxies can also be evaluated analytically . in appendix a we give approximate expressions which can be used for most applications . we calculate the predicted evolution of the @xmath3 ratio and its scatter for several models . to demonstrate the effects of progenitor bias we consider two classes of models , one with mild morphological evolution ( @xmath60 ) and one with strong morphological evolution ( @xmath61 ) . figure [ modeldata.plot ] shows the results . in models with @xmath60 morphological evolution is not very important to @xmath21 , and @xmath62% of present - day early - type galaxies were already in place at that redshift . in models with @xmath61 , on the other hand , only @xmath63% of early - type galaxies were in place at @xmath21 . in all panels of fig . [ modeldata.plot ] broken lines show the class of models with mild morphological evolution , and solid lines show models with strong morphological evolution . panels ( a ) , ( b ) , and ( c ) show the effect of changing the star formation history , while keeping the time of onset of star formation fixed at @xmath64 . the results for the evolution of the mean @xmath3 ratio are shown in panel ( a ) . quite remarkably , nearly all model predictions lie very close to the line for old populations which formed at the start of the universe ( @xmath65 ) . we find little dependence on the tranformation time scale @xmath28 or on @xmath44 , which characterizes the star formation history . this result illustrates the dramatic effects of the progenitor bias , as the mean stellar age of the galaxies at @xmath18 can be as low as half the age of the universe ( for @xmath66 ) , and yet the apparent evolution follows the relation for very old populations formed at infinite redshift . the reason for this behaviour is that the formation models are almost scale free . in models with a constant transformation rate , at any time the youngest half of the galaxies have been added to the sample in the last half of the age of the universe at that time . as a result , the mean stellar age of the galaxies is an almost constant fraction of the age of the universe , independent of the star formation histories of individual galaxies , and the mean mass - to - light ratios and colors will evolve as if the stars formed at very high redshift . in the appendix we demonstrate this effect analytically . the absolute colors and mass - to - light ratios at @xmath18 will be different for these models , but these are notoriously difficult to interpret uniquely ( e.g. , worthey 1994 , faber et al . 1999 ) . fortunately , the scatter in @xmath3 ratios and colors is much more sensitive to the transformation timescale and star formation histories . we show the scatter in @xmath3 ratios in panel ( b ) . for a given morphological transformation rate the variation in the average scatter is large , because it is a strong function of the star formation history of galaxies . as we will see later , this implies that observations of the scatter put important constraints on the models . for galaxies with later star formation , @xmath44 increases , and the scatter increases almost linearly with @xmath44 . for the limiting case of a constant transformation rate ( @xmath34 ) , we find that the scatter is well approximated by @xmath67 ( see appendix ) . the scatter is not strongly dependent on redshift because of the scale - free nature of the models . in such scale - free models , the relative age differences between early - type galaxies at any epoch is fairly constant . the scatter is proportional to the age differences between galaxies relative to the mean age ( e.g. , van dokkum et al . as galaxies are continuously added to the sample , the range in ages increases at the same rate as the mean age ; hence the spread in ages relative to the mean age remains constant . in the appendix this effect is demonstrated analytically . only in models with mild morphological evolution to @xmath21 ( indicated by the broken lines in fig . [ modeldata.plot ] ) we find an increase of the scatter with redshift . in these models the scatter is very low at @xmath18 , and at @xmath68 reaches values similar to those in models with strong morphological evolution . the progenitor bias of the models is shown in panel ( c ) . we define the progenitor bias at given redshift as the difference between the @xmath3 ratio of early - type galaxies and the @xmath3 ratio of all progenitors of present - day early - type galaxies . this is the `` error '' in the observed @xmath3 ratio that is caused by the late addition of early - type galaxies to the sample . as the figure shows , the bias increases with increasing tranformation time scale @xmath28 , and increasing @xmath44 . both parameters also cause the scatter to increase . for models with strong morphological evolution the progenitor bias can be approximated by @xmath69 this approximation is accurate to @xmath70% for @xmath71 . this result suggests that the progenitor bias can be estimated on the basis of the observed transformation rate , and the observed scatter . the star formation history is the individual galaxies , as parametrized by @xmath44 , is not needed to estimate the effect . this is a very useful result , as it is difficult to constrain the value of @xmath44 directly from the observed colors and luminosities . panels ( d ) , ( e ) , and ( f ) show the effect of changing the time of onset of star formation @xmath27 , while keeping the star formation history constant at @xmath72 ( i.e. , approximately constant star formation from @xmath27 to @xmath30 ) . the evolution of the mean @xmath3 ratio for the various models is shown in panel ( f ) . the evolution is very sensitive to the time of onset of star formation : there is an almost linear relation between the time when star formation commences and the rate of @xmath3 evolution . the reason for this behaviour is that the mean age of the stellar population in all galaxies is lower for higher values of @xmath27 . as will be shown in sect . [ mc.sec ] the observed evolution of the mean @xmath3 ratio places strong constraints on the time when star formation commenced in early - type galaxies . the scatter and the progenitor bias are shown in panels ( e ) and ( f ) . they both depend on the value of @xmath27 , such that the scatter is higher and the progenitor bias stronger for later onset of star formation . as a result of this dual dependence the relation between the progenitor bias and the observed scatter ( eq . [ progz.eq ] ) is not very sensitive to the value of @xmath27 , once again indicating that the observed rate of morphological evolution and the observed scatter suffice to estimate the progenitor bias . the models described in the previous section can be applied to observations of early - type galaxies in clusters at @xmath73 . parameters to fit are the evolution of _ ( i ) _ the early - type galaxy fraction , _ ( ii ) _ the mean @xmath4 ratio , as determined from the fundamental plane , _ ( iii ) _ the scatter in @xmath4 , and _ ( iv ) _ the scatter in the color - magnitude relation . free parameters are the morphological transformation rate ( parameterized by @xmath28 ) , and the star formation histories of galaxies prior to the transformations ( parameterized by @xmath29 and @xmath27 ) . the evolution of the early - type galaxy fraction in clusters is shown in fig . [ bestfit.plot](a ) ( taken from van dokkum et al . 2000 ) . data points are from dressler ( 1980 ) , andreon , davoust , & heim ( 1997 ) , dressler et al . ( 1997 ) , lubin et al . ( 1998 ) , fabricant et al . ( 2000 ) , and van dokkum et al . the early - type fraction decreases by a factor @xmath74 from @xmath18 to @xmath21 . the scatter around the downward trend is significant , and it will be interesting in the future to explore systematic differences in the evolution depending on cluster type . in this section we do not distinguish elliptical galaxies and s0 galaxies within the class of early - type galaxies . in sect . [ es0.sec ] we explore models in which both classes are modeled separately . the evolution of the early - type galaxy fraction is possibly influenced by the inclusion of low mass galaxies undergoing star bursts . since the samples are magnitude selected , the presence of such galaxies at high redshift would decrease the fraction of early - type galaxies , and would be unrelated to the evolution of massive galaxies . for the cluster 054 at @xmath13 we have a large data set of confirmed members with accurate colors ( van dokkum et al . 2000 ) , and we tested the importance of this effect by determining the early - type galaxy fraction among red galaxies alone . if we limit the analysis to red galaxies with @xmath75 the early - type fraction changes from 45% to 52% . this result indicates that the brightening of low mass galaxies has a minor effect on the fraction of early - type galaxies . other effects may have the opposite effect . as an example , biases introduced by the selection of the clusters themselves probably cause us to _ under_estimate the evolution of the early - type galaxy fraction ( see , e.g. , kauffmann 1995 ) . the evolution of the rest frame @xmath4 ratio with redshift is shown in fig . [ bestfit.plot](b ) and is taken from van dokkum et al.(1998a ) data are from jrgensen et al . ( 1996 ) , van dokkum & franx ( 1996 ) , kelson et al . ( 1997 ) , and van dokkum et al.(1998a ) . the @xmath3 ratio evolution is derived from the evolution of the zeropoint of the fundamental plane relation ( see van dokkum & franx 1996 ) . the evolution is well determined , because the fundamental plane has very small scatter . the scatter in @xmath76 is shown in fig . [ bestfit.plot](c ) . the data point at @xmath77 is from jrgensen et al . data points at higher redshift are from kelson et al . ( 2000 ) ( @xmath78 ) , kelson et al . ( 1997 ) ( @xmath79 ) , and van dokkum et al . ( 1998a ) ( @xmath13 ) . the highest redshift points have considerable uncertainty , because they are derived from small samples . the scatter in the @xmath52 color - magnitude relation is taken from van dokkum et al . ( 2000 ) and shown in fig . [ bestfit.plot](d ) . data are from stanford et al . ( 1998 ) , bower , lucey , & ellis ( 1992 ) , ellis et al . ( 1997 ) , van dokkum et al . ( 1998b ) , and van dokkum et al . the observations were brought to a common ( rest frame ) band by using @xmath80 and @xmath81 , as derived from the worthey ( 1994 ) models . the scatter is roughly constant with redshift , at @xmath82 magnitudes . the observed evolution of the early - type galaxy fraction directly constrains the transformation time scale @xmath28 . this is demonstrated in fig . [ modeldata.plot](a ) , which shows the redshift dependence of the transformed galaxy fraction for different values of the transformation time scale . we varied the transformation time scale between @xmath83 and @xmath34 . if @xmath84 all morphological transformations occur at very high redshift and if @xmath34 the transformation rate is constant . the observed evolution of the early - type galaxy fraction , shown in fig . [ bestfit.plot](a ) , gives a best fitting value of @xmath85 . values between @xmath86 and @xmath87 are consistent with the data at the 90% confidence level . the clusters in fig . [ bestfit.plot](a ) form a very heterogeneous sample , and a potential worry is that the evolution is severely affected by selection biases . as a test of the robustness of the transformation rate we fitted all clusters with high x - ray luminosities separately . solid symbols in fig . [ bestfit.plot](a ) are clusters with @xmath88ergss@xmath89 ( see van dokkum et al . this sample gives a very similar best fit value of @xmath90 . we note , however , that the morphological transformation rate is one of the main uncertainties in our study . we will return to this issue in sect . [ es0.sec ] . we next explore models with the best fit transformation time @xmath61 , and we vary the time of onset of star formation @xmath27 and the parameter describing the star formation history @xmath29 . as demonstrated in the previous section , the rate of evolution of the mean @xmath3 ratio has an approximately linear dependence on the value of @xmath27 . the observed evolution , shown in fig . [ bestfit.plot](b ) , is very slow . the dashed line indicates a simple single burst model with @xmath91 . this model provides a good fit to the data , as was previously shown by van dokkum et al . ( 1998a ) . we find a best fitting value of @xmath92 . values between @xmath64 and @xmath93 are consistent with the data at the @xmath94% confidence level . the implication is that star formation commenced at high redshift ( @xmath95 ) in early - type galaxies . the line in fig.[bestfit.plot](b ) shows the best fitting model with @xmath96 . as was demonstrated in sect . [ model.sec ] the star formation history between @xmath27 and @xmath30 is constrained by the observed scatter in @xmath3 ratio and color . this strong dependence of the scatter on the star formation history was discussed earlier by van dokkum et al . ( 1998b ) and bower et al . ( 1998 ) in the context of `` traditional '' models which do not include morphological evolution . if all scatter in the color - magnitude relation and the fundamental plane is produced by age variations , the best fitting model has @xmath97 ( i.e. , an approximately constant star formation rate ) , with values to @xmath98 consistent with the data at the 90% confidence level . the lines in fig . [ bestfit.plot](c ) and ( d ) show the model with @xmath99 . for values of @xmath29 lower than 0.5 the scatter is underpredicted , and additional scatter due to , e.g. , merging , metallicity variations , or dust is allowed . as an example , the broken lines show a model with @xmath61 , @xmath64 , and @xmath100 . the preferred values of @xmath101 correspond to gradual star formation histories similar to those depicted in the upper two panels of fig . [ fstar.plot ] . our best fitting model has @xmath102 , @xmath103 , and @xmath104 . the progenitor bias , expressed as the difference between the observed @xmath3 ratio of early - type galaxies and the @xmath3 ratio of all progenitors of early - type galaxies in nearby clusters , is shown in fig.[bestfit.plot](e ) . the progenitor bias is maximal if @xmath99 , in which case the scatter in the cm relation and the fp is entirely produced by age variations . in this extreme model the progenitor bias is approximately @xmath105 in @xmath106 , or @xmath107 in @xmath52 color . we stress that these values are upper limits , because the scatter can be caused by other effects than age . as shown in the previous section , the progenitor bias scales linearly with the scatter introduced by age variations . for example , if @xmath100 the scatter due to age is only half the observed scatter . in this case the progenitor bias is only @xmath108 in @xmath106 ( indicated by the broken line in fig . [ bestfit.plot]e ) . it is interesting to compare these results to the simple scaling with the observed scatter that was derived in sect . [ progz.sec ] . using the observed scatter of @xmath109 in @xmath106 eq . [ progz.sec ] gives a progenitor bias of @xmath110 , identical to the value we derived from the full fitting of the models to the observations . this remarkable agreement demonstrates quantitatively that the observed transformation rate and the observed scatter suffice to predict the progenitor bias . in particular , knowledge of the detailed star formation histories of the galaxies ( @xmath27 and @xmath29 in our models ) is not required . we used @xmath6 and @xmath7 to calculate the model predictions and the data points shown in fig . [ bestfit.plot ] . the models depend on the cosmological model because of the conversion from time to redshift . the observed evolution of the mean @xmath3 ratio also depends on the cosmology , because effective radii of galaxies are measured in arcseconds . as demonstrated in bender et al . ( 1998 ) and van dokkum et al . ( 1998a ) the dependencies can be quite strong , and the observed evolution of @xmath111 provides an upper limit on @xmath112 ( assuming the slope of the imf is not significantly steeper than the salpeter ( 1955 ) value ) . note that the models and data shown in fig.[bestfit.plot ] are not dependent on the hubble constant , because the luminosity evolution of stellar populations can be approximated by a power law and all data points are relative to nearby clusters . our results are very similar for a universe with a cosmological constant ( @xmath6 and @xmath9 ) . the predicted evolution of the early - type galaxy fraction remains virtually unaffected , and the best fitting value for the transformation time scale remains @xmath85 . the observed evolution of the mean @xmath4 ratio is more rapid in this cosmology , and is no longer well fitted by models with extremely high formation redshift of the stellar population ( see also van dokkum et al . models with @xmath96 predict too slow evolution , and good fits are obtained for @xmath113 . because the stellar populations of early - type galaxies are younger for @xmath114 the predicted scatter in @xmath3 ratios and colors is higher for a given value of @xmath29 . as a result , the observed scatter is best fitted for @xmath115 in this cosmology , as compared to @xmath116 for a universe without cosmological constant . the evolution of the @xmath3 ratio can be corrected for progenitor bias . our best fitting model indicates that the observed evolution of the @xmath3 ratio underestimates the true evolution by @xmath117 in @xmath118 , and @xmath119 in @xmath120 . van dokkum et al . ( 1998b ) found @xmath121 for @xmath6 and @xmath7 , or @xmath122 for @xmath6 and @xmath9 . the evolution after applying the maximum correction for progenitor bias is @xmath123 , or @xmath124 for positive cosmological constant . we conclude that the progenitor bias has a small ( @xmath125% ) , but non - negligible effect on the observed evolution of the mean @xmath3 ratio from @xmath18 to @xmath21 . the corrections are upper limits , because it is assumed that the scatter in luminosities and colors is entirely due to age variations among the galaxies . the observed evolution of the mean color and @xmath3 ratio of early - type galaxies indicates a very high redshift of formation for their stars ( e.g. , van dokkum & franx 1996 , schade , barrientos , & lopez - cruz 1997 , stanford et al.1998 , van dokkum et al . 1998a , kodama et al . because the true evolution of early - type galaxies is underestimated as a result of the progenitor bias , the ages of their stars are _ over_estimated . we constrain the formation epoch of the stars in early - type galaxies from the corrected evolution of the mean @xmath3 ratio . first we assume that the scatter in the color - magnitude relation and the fundamental plane is entirely caused by age variations . using @xmath126 in eq.[levo.eq ] we find that the mean luminosity weighted formation time of the stars in early - type galaxies @xmath127 , or @xmath5 when expressed in redshift . for @xmath6 and @xmath9 @xmath128 , and we find @xmath8 . as a check on our procedure , we also computed the mean formation epoch directly from the monte - carlo simulation which gave the best fit to the observations . the age derived from the simulation is identical to the age derived from the corrected evolution of the @xmath3 ratio . if the scatter in the color - magnitude relation and the fundamental plane is not entirely caused by age variations , the mean age of the stars is higher . both the mean age and the scatter have an approximately linear dependence on @xmath29 ( eq . [ tform.eq ] and [ scatevo.eq ] ) . therefore , for a given morphological transformation rate the mean age of the stars is to good approximation a linear function of the scatter at @xmath18 . from the monte - carlo simulations with @xmath61 we find @xmath129 for @xmath6 and @xmath9 . if the contribution of age variations to the scatter is negligible the correction for progenitor bias is negligible as well , and we find @xmath130 for @xmath6 and @xmath7 , and @xmath131 for @xmath6 and @xmath9 ( cf . van dokkum et al . unfortunately it is observationally difficult to disentangle the contributions of age and metallicity to the observed scatter ( e.g. , trager et al . in the previous sections we have treated early - type galaxies as one class . however , dressler et al . ( 1997 ) separated the evolution of elliptical and s0 galaxies , and found that the ratio of the fractions of s0 and elliptical galaxies evolves rapidly from @xmath18 to @xmath132 . figure [ dressler.plot ] shows the evolution of the number fractions of s0 galaxies and elliptical galaxies separately , for the dressler et al . ( 1997 ) clusters ( solid symbols ) . the dotted lines are linear fits to the data . as can be seen , the s0 fraction increases by a factor of @xmath133 5 towards lower redshift , whereas the elliptical fraction drops by @xmath134% from redshift @xmath135 to @xmath18 . dressler et al . ( 1997 ) interpret the rapid evolution of the s0 fraction as evidence for transformation of spiral galaxies to s0 galaxies since @xmath135 . as pointed out by fabricant et al . ( 2000 ) the decrease in the elliptical galaxy fraction over time requires substantial accretion onto clusters since @xmath135 . we model this evolution in the following way : we assume that the fraction of ellipticals is constant with time , and that s0 galaxies are formed from the spiral galaxies that are present in the clusters at @xmath135 . we ignore the fact that infalling spiral galaxies are also transformed into s0 galaxies . the evolution of the s0 galaxies is approximated reasonably well by a model with a constant transformation rate ( fig . [ dressler.plot ] ) . the overprediction of the fraction of s0 galaxies at high redshift implies that we underestimate the differences between the elliptical and s0 galaxies . in our first models , we assume that the star formation rate for the spiral galaxies is constant in time ( @xmath136 ) , and that star formation starts at @xmath96 for all galaxies . figure [ es0.plot ] shows the evolution of the mean @xmath3 ratio in this model , for elliptical galaxies ( dotted line ) , s0 galaxies ( dashed line ) , and elliptical and s0 galaxies combined ( solid line ) . the model reproduces the observed evolution of the @xmath3 ratio of the combined sample of elliptical galaxies and s0 galaxies . however , elliptical galaxies in this model have systematically higher @xmath3 ratios and redder colors than s0 galaxies at all redshifts . the difference is @xmath137 in @xmath106 , or @xmath138 in @xmath52 color . this difference corresponds to a difference of @xmath139% in luminosity weighted age ( e.g. , worthey et al . this large difference arises because all elliptical galaxies formed their stars at extremely high redshift in this model , whereas the s0 galaxies are continuously transformed from star forming galaxies . this result is in conflict with studies of luminous early - type cluster galaxies at low and intermediate redshift , which indicate very little difference in luminosity weighted age between luminous elliptical and s0 galaxies in the cores of rich clusters ( e.g. , jrgensen , franx , & kjrgaard 1996 , ellis et al . 1997 , van dokkum et al . 1998a , jones , smail , & couch 2000 , kelson et al . other studies have reported s0 galaxies with younger stellar populations , but these tend to have low luminosities ( e.g. , caldwell et al . 1993 , van dokkum et al.1998b , kuntschner & davies 1998 ) . there are several ways in which this conflict can be solved . first , we can adapt the starformation histories . the model assumes that star formation in elliptical and s0 galaxies commenced at @xmath25 , and that their star formation rate was approximately constant from @xmath27 to @xmath30 . if star formation in elliptical galaxies commenced at a later time the systematic difference between elliptical and s0 galaxies is reduced . if @xmath140 for elliptical galaxies ( i.e. , star formation commenced at @xmath141 ) the systematic difference in @xmath3 ratio between elliptical and s0 galaxies at @xmath18 is very small . however , it increases to @xmath142 in @xmath76 at @xmath19 . this corresponds to a systematic difference in @xmath52 color of @xmath143 , which is inconsistent with the low scatter in the cm relation of early type galaxies at @xmath13 ( van dokkum et al . 2000 ) . an alternative explanation for the small observed difference between luminous elliptical and s0 galaxies is that most of their stars were formed at very high redshift . in our models , this corresponds to low values of @xmath29 , the parameter describing the star formation history . the systematic difference in @xmath3 ratio scales linearly with @xmath29 , and for values of @xmath144 the systematic age difference between elliptical and s0 galaxies is @xmath145% at all redshifts . this explanation can be tested with the help of direct measurements of the star formation rates of the spiral galaxies in intermediate redshift clusters . a third possibility is that the effects of metallicity cancel the effects of age ( e.g. , worthey , trager , & faber 1996 , shioya & bekki 1998 , trager et al . the spiral galaxies would be enriched in metals compared to the elliptical galaxies of the same luminosity , and hence their colors would be redder , and their @xmath3 ratios higher than that of elliptical galaxies with the same age . this solution requires exact fine tuning between age and metallicity to keep the scatter small . it can be tested by direct observations of absorption line strengths of elliptical and s0 galaxies in nearby and distant clusters . the first studies do not appear to indicate significant differences between massive elliptical and s0 galaxies ( e.g. , kuntschner & davies 1998 , jones et al . 2000 , kelson et al . 2001 ) . finally , the separation of ellipticals and s0 s is very difficult , and it is not clear whether a clear distinction can be made between elliptical galaxies and s0 galaxies with @xmath146 luminosities ( e.g. , jrgensen & franx 1994 , rix & white 1990 ) . the latter study demonstrated that many elliptical galaxies contain significant disks , and would be classified as s0 galaxies when seen edge - on . jorgensen and franx argued that the coma elliptical and s0 galaxies can be approximated well by a population of galaxies which has a homogeneous distribution in the ratio of bulge light to total light . in this case , no clearcut distinction would exist between s0 galaxies and elliptical galaxies , and the evolutionary histories would not be so different between the galaxies . this might also explain the fact that the classifications between different authors agree well when the total fraction of elliptical and s0 galaxies is considered , but not when the classes are separated ( e.g. , andreon 1998 , fabricant et al . quantitative classifications would be able to resolve this issue , especially if detailed attention is given to the classification uncertainties due to projection effects . in this paper we have quantified the effects of morphological evolution on the observed properties of early - type galaxies in clusters at @xmath73 . a manifestation of morphological evolution is the `` progenitor bias '' : at high redshift , the progenitors of the youngest present - day early - type galaxies drop out of the sample . in general , the observed evolution of the mean luminosity and color of early - type galaxies is slower than the true evolution of all progenitors of early - type galaxies , and the scatter in luminosity and color is smaller . if morphological evolution is strong the observed mean evolution is similar to that of a single age stellar population formed at @xmath25 , and the scatter is approximately constant with redshift . these results can be derived analytically ( cf . appendix [ equations.sec ] ) . in traditional models it is difficult to explain the low number fractions of early - type galaxies in distant clusters and the homogeneity and slow evolution of their stellar populations simultaneously . we have shown that the observed constant scatter in the cm relation and the fp and the slow evolution of the mean @xmath3 ratio are natural consequences of morphological evolution . observations of early - type galaxies in clusters at @xmath73 are well fitted by our models , and we can constrain the morphological transformation rate and the star formation histories of galaxies prior to transformation from the fits . the observed evolution of the early - type galaxy fraction constrains the transformation rate , the evolution of the mean @xmath3 ratio constrains the time of onset of star formation , and the scatter in the color - magnitude relation and the fundamental plane constrain the star formation histories of galaxies prior to transformation . the best fits are obtained if approximately 50% of present - day early - type galaxies were transformed from other morphological types at @xmath0 , star formation commenced at early times ( @xmath95 ) , and star formation was approximately constant prior to transformation . such gradual star formation histories are consistent with the low numbers of star burst galaxies seen in intermediate redshift clusters ( e.g. , abraham et al . 1996 , balogh et al . 1998 , ellingson et al . 2000 ) , although some star bursts could be concealed by dust ( poggianti et al . 1999 ) . the observed color and luminosity evolution of early - type galaxies can be corrected for the effects of progenitor bias . we have shown that for a given morphological transformation rate the progenitor bias is proportional to the scatter in @xmath3 ratio or color . the correction is maximal if the scatter in the cm relation and the fp at @xmath18 is entirely due to age variations between galaxies . this maximum correction to the evolution of rest frame @xmath118 is approximately @xmath147 , and the correction to the evolution of rest frame @xmath148 is approximately @xmath149 . these corrections can easily be computed for other rest frame bands using stellar population synthesis models . as an example , the worthey ( 1994 ) and bruzual & charlot ( 2000 ) models give @xmath150 for solar metallicity and salpeter ( 1955 ) imf . therefore , the correction to @xmath151 is @xmath152 . the corrections are upper limits , because variations in dust content and metallicity probably contribute to the scatter . on the other hand , as pointed out by , e.g. , trager et al . ( 2000 ) variations in metallicity and age may conspire to produce a low scatter . if this is confirmed the corrections may be larger than the values given here . because the mean luminosity and color evolution are underestimated , the mean age of the stars in early - type galaxies that is derived from them is overestimated in traditional models . we find that the mean luminosity weighted formation redshift of early - type galaxies can be as low as @xmath153 for @xmath6 and @xmath7 , or @xmath8 for @xmath6 and @xmath9 , if the scatter in the cm relation and the fp is entirely caused by age variations . this result is important , because previous studies which ignored the effects of progenitor bias found much earlier formation epochs of the stars in early - type galaxies ( e.g. , ellis et al . 1997 , stanford et al . 1998 , van dokkum et al . the redshift range 2 3 is of interest because it is observationally accessible . in particular , our results are consistent with the idea that ( some ) ly - break galaxies are the star forming building blocks of present - day early - type galaxies in clusters ( baugh et al . 1998 ) . in our main analysis we treated the evolution of early - type galaxies as one class , but we also explored a model which separates the two classes . in our model , only the s0 galaxies evolve morphologically from spiral galaxies , as advocated by dressler et al . the model predicts that colors and @xmath3 ratios of luminous s0 galaxies in nearby clusters are systematically offset from those of elliptical galaxies , in the sense that elliptical galaxies are redder and have higher @xmath3 ratios . this result is in conflict with studies of bright early - type galaxies in the cores of rich clusters , which generally find very little difference between elliptical and s0 galaxies ( e.g. , bower et al . 1992 , ellis et al . there are several ways to resolve this conflict : star formation may have started later in the elliptical galaxies , or most of the star formation in elliptical and s0 galaxies must have occurred at very early times , long before morphological transformation to early - type galaxy . alternatively , there might be a conspiracy between age and metallicity ( e.g. , trager et al . 2000 ) such that s0 galaxies have systematically higher metallicities than elliptical galaxies of the same luminosity . this can be tested by absorption line measurements ; current measurements do not indicate such an effect for cluster galaxies . finally , the separation between elliptical and s0 galaxies may not be very sharp . rix & white ( 1990 ) showed that many s0 galaxies will be classified as elliptical galaxies when not viewed edge - on . jrgensen and franx ( 1994 ) showed that the elliptical and s0 galaxies in coma can be modeled by a population of galaxies with a homogeneous distribution of bulge - to - total light ratio . hence , the galaxies which are classified as elliptical galaxies may also have evolved since @xmath21 . it is interesting to compare our results to predictions of semi - analytical galaxy formation models in cdm cosmologies ( white & frenk 1991 ; cole 1991 ) . because the morphologies of galaxies change quite rapidly in these models ( e.g. , baugh , cole , & frenk 1996 , kauffmann 1996 ) progenitor bias is an important ingredient . in particular , the models provide reasonable fits to the observed scatter in the cm relation ( kauffmann & charlot 1998 ; cole et al . 2000 ) and the evolution of the mean @xmath3 ratio of massive galaxies ( diaferio et al . 2000 ) , even though galaxies in these simulations have quite complex ( star ) formation histories . however , in these complex models it is difficult to isolate and study specific effects such as the progenitor bias . the main contribution of our work is that we use very simple models with only three free parameters . we show that the effects of progenitor bias can be constrained directly from the observations , and it is straightforward to reproduce our results . the main uncertainty in the present paper is the rate of morphological evolution . more high resolution , large field studies of distant clusters are needed to constrain the evolution of the early - type galaxy fraction better . detailed simulations are needed to determine the effects of infall on the observed properties and morphological mix of cluster galaxies ( e.g. , ellingson et al . furthermore , studies of the evolution of the mass function of rich clusters to @xmath21 are necessary to determine whether high redshift clusters are representative progenitors of present - day rich clusters . as pointed out by , e.g. , kauffmann ( 1995 ) significant evolution in the mass function of rich clusters may result in a `` progenitor bias '' at the cluster level . we thank the referee , richard ellis , for constructive comments which improved the presentation and clarity of the paper significantly . p. g. v. d. acknowledges support by nasa through hubble fellowship grant hf-01126.01 - 99a 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 . combining eq . [ first.eq ] [ levo.eq ] gives the following approximate expression for the observed evolution of the mean @xmath4 ratio of early - type galaxies : @xmath154}{\int_0^{t-0.1\t0 } \exp \left ( - \frac{t'}{\taustop } \right ) \left [ t-(\fstar t'+\tstart)\right]^{-\kappa_b } dt'}.\ ] ] similarly , the color evolution can be approximated by @xmath155}{\int_0^{t-0.1\t0 } \exp \left ( - \frac{t'}{\taustop } \right ) \left [ t-(\fstar t'+\tstart)\right]^{\kappa_b-\kappa_u } dt'}.\ ] ] for given @xmath29 the integrals can be evaluated analytically . if @xmath34 and @xmath96 ( i.e. , in the case of a constant transformation rate ) evaluation of the integral in eq . [ allml.eq ] gives @xmath156^{1-\kappa_b}}.\ ] ] for @xmath157 this reduces to @xmath158 this approximation is accurate to better than @xmath159% for @xmath160 and @xmath161 . as a consequence , @xmath162 similarly , the color evolution reduces to @xmath163 these expressions are identical to the evolution of a single age stellar population that was formed at @xmath25 . for @xmath96 the scatter in @xmath164 at time @xmath58 can be approximated by @xmath165 for a constant morphological transformation rate the distribution of @xmath30 is a simple top hat , bounded by @xmath166 and @xmath167 . the @xmath168 spread of a top hat distribution is @xmath169 , and the mean is @xmath170 . as a result , @xmath171 for @xmath172 ( i.e. , at low redshift ) and small @xmath29 , this reduces to @xmath173 in the rest frame @xmath51 band . similarly , the scatter in the color can be approximated by @xmath174 the scatter is to good approximation independent of redshift , and a strong function of @xmath29 , the parameter describing the star formation history of individual galaxies . abadi , m. g. , moore , b. , & bower , r. 1999 , , 308 , 947 abraham , r. g. , smecker - hane , t. a. , hutchings , j. b. , carlberg , r. g. , yee , h. k. c. , ellingson , e. , morris , s. , oke , j. b. , rigler , m. 1996 , , 471 , 694 andreon , s. , davoust , e. , & heim , t. 1997 , a&a , 323 , 337 andreon , s. 1998 , , 501 , 533 balogh , m. l. , schade , d. , morris , s. l. , yee , h. k. c. , carlberg , r. g. , ellingson , e. 1998 , , 504 , l75 baugh , c. m. , cole , s. , & frenk , c. s. 1996 , , 283 , 1361 baugh , c. m. , cole , s. , frenk , c. s. , & lacey , c. g. 1998 , , 498 , 504 beers , t. c. , flynn , k. , & gebhardt , k. 1990 , , 100 , 32 bender , r. , saglia , r. p. , ziegler , b. , belloni , p. , greggio , l. , hopp , u. , & bruzual , g. 1998 , , 493 , 529 bower , r. g. , lucey , j. r. , & ellis , r. s. 1992 , , 254 , 601 bower , r. g. , kodama , t. , & terlevich , a. 1998 , , 299 , 1193 bruzual , g. , & charlot , s. 2000 , in preparation butcher , h. , & oemler , a. 1978 , , 219 , 18 butcher , h. , & oemler , a. 1984 , , 285 , 426 caldwell , n. , rose , j. a. , sharples , r. m. , ellis , r. s. , bower , r. g. 1993 , , 106 , 473 cole , s. 1991 , , 367 , 45 cole , s. , lacey , c. g. , baugh , c. m. , frenk , c. s. 2000 , , 319 , 168 couch , w. j. , & sharples , r. m. 1987 , , 229 , 423 couch , w. j. , barger , a. j. , smail , i. , ellis , r. s. , sharples , r. m. 1998 , , 497 , 188 diaferio , a. , kauffmann , g. , balogh , m. l. , white , s. d. m. , schade , d. , & ellingson , e. 2000 , preprint ( astro - ph/0005485 ) dressler , a. 1980 , , 236 , 351 dressler , a. , & gunn , j. e. 1983 , , 270 , 7 dressler , a. , oemler , a. , jr . , sparks , w. b. , & lucas , r. a. 1994 , , 435 , l23 dressler , a. , oemler , a. , jr . , couch , w. j. , smail , i. , ellis , r. s. , barger , a. , butcher , h. , poggianti , b. m. , & sharples , r. m. 1997 , , 490 , 577 ellingson , e. , lin , h. , yee , h. k. c. , & carlberg , r. g. 2000 , , in press ( astro - ph/0010141 ) ellis , r. s. , smail , i. , dressler , a. , couch , w. j. , oemler , a. , jr . , butcher , h. , & sharples , r. m. 1997 , , 483 , 582 faber , s. m. , trager , s. c. , gonzalez , j. j. , worthey , g. 1999 , ap & ss , 267 , 273 fabricant , d. , franx , m. , & van dokkum , p. g. 2000 , , 539 , 577 ferreras , i. , & silk , j. 2000 , , 532 , 193 franx , m. , & van dokkum , p. g. 1996 , in new light on galaxy evolution ( iau 171 ) , r. bender & r. l. davies , eds . , kluwer , p. 233 fried , j. w. 1988 , a & a , 189 , 42 gunn , j. e. , & gott , j. r. 1972 , apj , 176 , 1 jones , l , smail , i. , & couch , w. j. 2000 , , 528 , 118 jrgensen , i. , & franx , m. 1994 , , 433 , 553 jrgensen , i. , franx , m. , & kjaergaard , p. 1996 , , 280 , 167 kauffmann , g. 1995 , , 274 , 153 kauffmann , g. 1996 , , 281 , 487 kauffmann , g. , & charlot , s. 1998 , , 294 , 705 kauffmann , g. , white , s. d. m. , & guiderdoni , b. 1993 , , 264 , 201 kelson , d. d. , van dokkum , p. g. , franx , m. , illingworth , g. d. , & fabricant , d. 1997 , , 478 , l13 kelson , d. d. , illingworth , g. d. , van dokkum , p. g. , & franx , m. 2000 , , 531 , 184 kelson , d. d. , illingworth , g. d. , franx , m. , & van dokkum , p. g. 2001 , , submitted kodama , t. , arimoto , n. , barger , a. j. , & aragon - salamanca , a. 1998 , a&a , 334 , 99 kodama , t. , & smail , i. 2000 , preprint ( astro - ph/0008139 ) kuntschner , h. , & davies , r. l. 1998 , , 295 , l29 lavery , r. j. , & henry , j. p. 1988 , , 330 , 596 lavery , r. j. , pierce , m. j. , & mcclure , r. d. 1992 , , 104 , 2067 lubin , l. m. , postman , m. , oke , j. b. , ratnatunga , k. u. , gunn , j. e. , hoessel , j. g. , & schneider , d. p. 1998 , , 116 , 584 lucey , j. r. , guzman , r. , carter , d. , & terlevich , r. j. 1991 , , 253 , 584 moore , b. , katz , n. , lake , g. , dressler , a. , & oemler , a. , jr . 1996 , nature , 379 , 613 pahre , m. a. , djorgovski , s. g. , & de carvalho , r. 1998 , , 116 , 1591 poggianti , b. , smail , i. , dressler , a. , couch , w. , barger , a. , butcher , h. , ellis , r. , & oemler , a. , jr . 1999 , , 518 , 576 postman , m. , lubin , l. m. , & oke , j. b. 1998 , , 116 , 560 rix , h .- w . , & white , s. d. m. 1990 , , 362 , 52 salpeter , e. 1955 , , 121 , 161 schade , d. , barrientos , l. f. , & lopez - cruz , o. 1997 , , 477 , l17 shioya , y. , & bekki , k. 1998 , , 504 , 42 stanford , s. a. , eisenhardt , p. r. , & dickinson , m. 1998 , , 492 , 461 tinsley , b. m. 1980 , fundamentals of cosmic physics , 5 , 287 trager , s. c. , faber , s. m. , worthey , g. , gonzalez , j. j. 2000 , , 120 , 165 tytler , d. , & vidal , n. v. 1978 , mnras , 182 , 33 van dokkum , p. g. , & franx , m. 1996 , mnras , 281 , 985 van dokkum , p. g. , franx , m. , kelson , d. d. , illingworth , g. d. 1998a , , 504 , l17 van dokkum , p. g. , franx , m. , kelson , d. d. , illingworth , g. d. , fisher , d. , & fabricant , d. 1998b , , 500 , 714 van dokkum , p. g. , franx , m. , fabricant , d. , kelson , d. d. , & illingworth , g. d. 1999 , , 520 , l95 van dokkum , p. g. , franx , m. , fabricant , d. , illingworth , g. d. , kelson , d. d. 2000 , , 541 , 95 white , s. d. m. , & frenk , c. s. 1991 , , 379 , 52 worthey , g. 1994 , , 95 , 107 worthey , g. , trager , s. c. , & faber , s. m. 1996 , in asp conf . 86 , fresh views on elliptical galaxies , ed . a. buzzoni , a. renzini , & a. serrano ( san francisco : asp ) , 203

we introduce the notion of a centroidal locating set of a graph , that is , a set of vertices such that all vertices in are uniquely determined by their relative distances to the vertices of . a centroidal locating set of of minimum size is called a centroidal basis , and its size is the centroidal dimension . this notion , which is related to previous concepts , gives a new way of identifying the vertices of a graph . the centroidal dimension of a graph is lower- and upper - bounded by the metric dimension and twice the location - domination number of , respectively . the latter two parameters are standard and well - studied notions in the field of graph identification . we show that for any graph with vertices and maximum degree at least 2 , . we discuss the tightness of these bounds and in particular , we characterize the set of graphs reaching the upper bound . we then show that for graphs in which every pair of vertices is connected via a bounded number of paths , , the bound being tight for paths and cycles . we finally investigate the computational complexity of determining for an input graph , showing that the problem is hard and can not even be approximated efficiently up to a factor of . we also give an-approximation algorithm .

1406.7490

a large body of work has evolved concerning the problem of identifying an `` intruder '' vertex in a graph . as examples , one might seek to identify a malfunctioning processor in a multiprocessor network , or the location of an intruder such as a thief , saboteur or fire in a graph - modeled facility . in this paper , we introduce the new model of _ centroidal detection _ as such a graph identification problem . an early model considered the case where one could place detection devices like sonar or loran stations at vertices in a graph ; each detection device could determine the distance to the intruder s vertex location . as introduced independently in slater @xcite and harary and melter @xcite , vertex set @xmath8 is a _ locating set _ ( also called _ resolving set _ in the literature ) if for each vertex @xmath9 , the ( ordered ) @xmath10-tuple @xmath11 of distances between the detector s locations and the intruder vertex @xmath12 uniquely determines @xmath12 . a minimum cardinality locating set is called a _ metric basis _ ( also called _ reference set _ in the literature ) , and its order is the _ metric dimension _ of @xmath0 , denoted by @xmath13 . other studies involving metric bases include for example @xcite . carson @xcite and , independently , delmas , gravier , montassier and parreau @xcite ( for the latter authors , under the name of _ light @xmath14-codes _ ) considered the case in which each detection device at @xmath15 can only detect an intruder at distance at most @xmath14 . in another model , the presence of any edge @xmath16 indicates that a detection device at @xmath17 is able to detect an intruder at @xmath12 . let us denote by @xmath18 and @xmath19 $ ] the open and the closed neighbourhood of vertex @xmath17 , respectively . a set @xmath20 is a _ dominating set _ if @xmath21=v(g)$ ] . clearly , in the latter model , if every possible intruder location must be detectable , the set of detector locations must form a dominating set . the concepts of locating and dominating set were merged in slater @xcite : when a detection device at vertex @xmath17 can distinguish between there being an intruder at @xmath17 or at a vertex in @xmath18 ( but which precise vertex in @xmath18 can not be determined ) , then we have the concept of a _ locating - dominating set_. more precisely , a set @xmath22 of vertices is locating - dominating if it is dominating and every vertex in @xmath23 is dominated by a distinct subset of @xmath22 . the minimum cardinality of a locating - dominating set of graph @xmath0 is denoted @xmath24 . when one can only decide if there is an intruder somewhere in @xmath19 $ ] , one is interested in an _ identifying code _ , as introduced by karpovsky , chakrabarty and levitin @xcite . haynes , henning and howard @xcite added the condition that the locating - dominating set or identifying code not have any isolated vertices . when a detection device at @xmath17 can determine that an intruder is in @xmath18 , but will not report if the intruder is at @xmath17 itself , one is interested in _ open - locating - dominating sets _ , as introduced for the hypercube by honkala , laihonen and ranto @xcite and for all graphs by seo and slater @xcite . a bibliography of related papers is maintained by lobstein @xcite . in what follows , we will denote the path and the cycle on @xmath3 vertices by @xmath25 and @xmath26 , respectively . in this paper , we introduce the study of _ centroidal bases_. in this model , we assume that detection devices have unlimited range , as for metric bases . however , exact distances to the intruder are not known , but for detection devices @xmath27 , if there is an intruder at vertex @xmath28 , then the presence of the intruder in the graph is determined earlier by @xmath17 than by @xmath12 when @xmath17 is closer to @xmath28 than @xmath12 , that is , when @xmath29 . when @xmath30 we assume that @xmath17 and @xmath12 report simultaneously . for example , consider @xmath31 ( see figure [ fig : p8 ] ) . for vertex @xmath32 , the order in which the detectors of @xmath33 report is @xmath34 because @xmath35 . for vertex @xmath36 , we have @xmath37 , hence the order of reporting is @xmath38 . the smallest size of a set @xmath39 for which the order of reporting uniquely identifies each vertex is , in fact , four . in 1869 , jordan @xcite showed that each of the center and the ( branch weight ) centroid of a tree either consists of one vertex , or of two adjacent vertices . eccentricity _ of vertex @xmath17 is the maximum distance from @xmath17 to another vertex in graph @xmath0 , @xmath40 , and the _ center _ @xmath41 of @xmath0 is the set of vertices of minimum eccentricity . for a tree @xmath42 , the _ branch weight _ of @xmath17 , @xmath43 , is the maximum number of edges in a subtree with @xmath17 as an endpoint . the _ branch weight centroid _ of @xmath42 is the set @xmath44 of vertices with minimum branch weight in @xmath42 . in 1964 , hakimi @xcite considered two facility location problems , one involving the center . the second one involved the _ distance _ @xmath45 , measuring the total response time at @xmath17 ( this notion was called _ status _ of @xmath12 by harary in 1959 @xcite ) . the _ median _ of @xmath0 , @xmath46 , is the set of vertices of minimum distance in @xmath0 . in 1968 , zelinka @xcite showed that for any tree @xmath42 , @xmath47 , which seemed to imply that the median would be a good generalization of the branch weight centroid of a tree to an arbitrary graph . however ( with details in slater @xcite and slater and smart @xcite ) , note that components of @xmath48 of the same order , one being a path and the other a star , contribute the same value of a branch weight , but have much different distances . in trying to keep closer to the spirit of what the branch weight centroid measures , the _ centroid _ @xmath49 of an arbitrary graph @xmath0 was defined in terms of competitive facility location @xcite . for facilities located at vertices @xmath17 and @xmath12 , a customer at vertex @xmath28 is interested in which of the facilities is the closer . as defined in slater @xcite , the set @xmath50 is the set of vertex customer locations strictly closer to @xmath17 than to @xmath12 . then , @xmath51 rates how well @xmath17 does as a facility location , in comparison to @xmath12 . letting @xmath52 , the _ centroid _ of a graph @xmath0 is the set @xmath53 . when @xmath0 is a tree , the centroid and branch weight centroid are easily seen to coincide . interestingly , there are graphs @xmath0 for which @xmath54 for all @xmath55 . note that in our context of centroidal bases , detectors located at @xmath17 and @xmath12 enable us to determine if an intruder is in @xmath56 , in @xmath57 , or in @xmath58 . let @xmath59 be a set of vertices of graph @xmath0 with detection devices located at each @xmath15 . as noted , we will assume that each detection device has an unlimited range an intruder entering at any vertex @xmath28 will , at some point , have its presence noted at each @xmath15 . simply the presence will be noted , with no information about the location of the intruder . in particular , unlike in the setting of metric bases , @xmath60 will not be known . however , the time it takes before @xmath15 detects the intruder at @xmath28 will be an increasing function of the distance @xmath60 . that is , @xmath15 will indicate an intruder presence before @xmath61 whenever @xmath62 ( in our previous terminology , @xmath63 ) . we will say that @xmath28 is _ located _ first by @xmath15 , and then by @xmath61 . thus , each vertex @xmath28 has a rank ordering @xmath64 of the elements of a partition of @xmath65 ( in fact , @xmath64 is an _ ordered partition _ of @xmath65 , that is , an ordered set of disjoint subsets of @xmath65 whose union is @xmath65 ) . this ordering lists all the elements of @xmath65 in non - decreasing order by their distance from @xmath28 , with ties noted . note that the number of ordered partitions of a set @xmath65 of @xmath10 elements is the _ @xmath10-th ordered bell number _ , denoted @xmath66 ( see the book of wilf ( * ? ? ? * section 5.2 , example 1 ) ) . for @xmath67 in path @xmath68 ( see figure [ fig : p8 ] ) , @xmath69 , @xmath70 , @xmath71 , @xmath72 , @xmath73 , @xmath74 , @xmath75 and @xmath76 . vertex set @xmath77 is called a _ centroidal locating set _ of graph @xmath0 if @xmath78 for every pair @xmath79 of distinct vertices . a _ centroidal basis _ of @xmath0 is a centroidal locating set of minimum cardinality . the _ centroidal dimension _ of @xmath0 , denoted @xmath2 , is the cardinality of a centroidal basis . in our example , @xmath67 is the unique centroidal basis of path @xmath68 , and @xmath80 . observe that every graph @xmath0 has a centroidal locating set , for example , @xmath81 : each vertex @xmath28 is the only vertex having @xmath28 as the first element of @xmath64 . a useful reformulation of the definition of a centroidal locating set is as follows : [ obs : otherdef ] a set @xmath65 of vertices of a graph @xmath81 is a centroidal locating set if and only if for every pair @xmath79 of distinct vertices of @xmath81 , there exist two vertices @xmath82 in @xmath65 such that either @xmath83 but @xmath84 , or @xmath85 and @xmath86 ( in other words , @xmath87 but not @xmath28 , or @xmath88 , but not @xmath89 ) . we start in section [ sec : prelim ] by stating some preliminary observations and lemmas , and by giving bounds on parameter @xmath90 involving the order , the diameter , and other parameters of graphs ; in particular , we show that @xmath91 when @xmath0 has @xmath3 vertices and maximum degree at least 2 . in section [ sec : tightness ] , we discuss the tightness of the two aforementioned bounds by constructing graphs with small centroidal dimension , and by fully characterizing the graphs having centroidal dimension @xmath92 . in section [ sec : paths - cycles ] , we give a lower bound @xmath93 when @xmath0 has @xmath94 edges and every pair of vertices is connected by a small number , @xmath10 , of paths . we show that the bound is tight ( up to a constant factor ) for paths and cycles . finally , in section [ sec : complex ] , we discuss the computational complexity of finding a centroidal basis ; we show that for graphs with @xmath3 vertices , it is -hard to compute an @xmath95-approximate solution , and describe an @xmath96-approximation algorithm . we also remark that the problem is fixed - parameter - tractable when parameterized by the solution size . in this section , we give a series of preliminary lemmas and bounds for parameter @xmath90 that will prove useful later on , and also help the reader become familiar with some of the aspects of the problem . we state a few lemmas that will prove very useful in the study of centroidal locating sets . let @xmath0 be a graph . the following statements are true : + ( a ) if @xmath17 is a vertex of degree 1 , then any centroidal locating set of @xmath0 contains @xmath17.[lemma : deg1 ] + ( b ) if @xmath17 is a vertex of degree 1 having a neighbour @xmath12 of degree 2 , then any centroidal locating set of @xmath0 contains either @xmath12 or a neighbour of @xmath12 other than @xmath17.[lemma : deg1 - 2 ] + ( c ) if @xmath27 are two vertices with @xmath97 or @xmath19=n[v]$ ] , then any centroidal locating set of @xmath0 contains at least one of @xmath17 and @xmath12.[lemma : twins ] ( a ) : otherwise , @xmath17 and its neighbour are not distinguished . \(b ) and ( c ) : otherwise , @xmath17 and @xmath12 are not distinguished . [ lemma : deg2 ] let @xmath33 be a set of vertices of a graph @xmath0 such that for each @xmath98 , @xmath99 and such that for each @xmath100 , @xmath101 . then @xmath102 is a centroidal locating set of @xmath0 . let @xmath103 . every vertex of @xmath65 is first located by itself , while every vertex of @xmath33 is first located by a distinct set of at least two vertices of @xmath65 . note that in particular , lemma [ lemma : deg2 ] shows that for any vertex @xmath17 of degree at least 2 in a graph @xmath0 , @xmath104 is a centroidal locating set of @xmath0 . we now provide some lower and upper bounds for the value of parameter @xmath90 . [ thm : lb ] let @xmath0 be a graph on @xmath3 vertices with maximum degree at least 2 . then @xmath105 lemma [ lemma : deg2 ] immediately implies the upper bound . for the lower bound , assume that @xmath65 is a centroidal basis of size @xmath106 , and @xmath0 has @xmath3 vertices . then , to each vertex of @xmath0 , one can assign a distinct ordered partition of @xmath65 . it is known that the number of ordered partitions of a set of @xmath10 elements , the ordered bell number @xmath66 , is approximated by @xmath107 see wilf ( * ? ? ? * section 5.2 , example 1 ) . it is clear that @xmath108 . let us assume that @xmath65 is a centroidal basis of @xmath0 of size @xmath10 , with @xmath109 : for large enough @xmath3 , we have @xmath110 for some constant @xmath111 . taking the logarithm on both sides we get @xmath112 . by using stirling s approximation @xmath113 , we obtain : @xmath114 hence , @xmath115 ; again taking the logarithm , we get : @xmath116 merging equalities and [ eq : lnk ] , we get : @xmath117 since in @xmath0 , there can not be any smaller centroidal locating set than @xmath65 , the bound follows . the considerations of theorem [ thm : lb ] can be strengthened if we assume that the distances of a vertex to vertices in @xmath65 are bounded . the following result was already known in the context of the metric dimension , see e.g. khuller , raghavachari and rosenfeld @xcite or chartrand , eroh , johnson and oellermann @xcite . [ prop : diam ] let @xmath0 be a graph with a centroidal locating set @xmath65 of size @xmath10 and let @xmath22 be an integer . if for every vertex @xmath118 and for every vertex @xmath119 , @xmath120 , then @xmath121 and hence @xmath122 . in particular , this holds if @xmath0 has diameter @xmath22 . let @xmath118 be a vertex . since every vertex of @xmath65 is at distance at most @xmath22 from @xmath17 , @xmath123 contains at most @xmath22 sets . there are exactly @xmath124 different ordered partitions of @xmath65 into at most @xmath22 sets ( this number is equal to the number of words of length @xmath10 over an alphabet of size @xmath22 ) , hence there can be at most @xmath124 vertices in @xmath125 , and hence @xmath126 vertices in @xmath0 . the bound of proposition [ prop : diam ] was improved by hernando , mora , pelayo , seara and wood @xcite : [ thm : diam ] let @xmath0 be a graph on @xmath3 vertices with diameter @xmath127 and @xmath128 . then @xmath129 . we improve the bound in theorem [ thm : diam ] for @xmath130 : [ thm : diam3 ] let @xmath0 be a graph on @xmath3 vertices with diameter @xmath131 and let @xmath132 . if @xmath133 and @xmath134 , @xmath135 . if @xmath136 and @xmath137 , @xmath138 . let @xmath65 be a centroidal locating set of @xmath0 . _ @xmath139 @xmath133 and @xmath134 . _ by theorem [ thm : diam ] we have @xmath140 . however , observe that for every vertex @xmath141 , @xmath142 . but there can only be @xmath143 distinct sets @xmath144 , and moreover if @xmath145 and @xmath146 we have @xmath147 . hence @xmath148 and we are done . _ @xmath139 @xmath136 and @xmath137 . _ since the diameter is 3 , for every vertex @xmath12 , @xmath149 has at most three components , unless @xmath150 , then it may have four . moreover , if @xmath141 , then @xmath149 either has one component ( then @xmath151 ) , or two ( then @xmath152 with @xmath153 ) , or three ( then @xmath154 with @xmath155 and @xmath156 ) . therefore , we have @xmath157 that is , @xmath158 plus the number of ways of choosing a nonempty subset @xmath33 of @xmath65 and a nonempty subset of @xmath159 . now , if for every pair @xmath160 in @xmath65 we have @xmath161 , then no vertex @xmath12 has four components in @xmath149 and so the claimed bound holds . now , assume that there is a pair @xmath160 in @xmath65 with @xmath162 . this implies that for any vertex @xmath12 , @xmath149 can not be of the form @xmath163 with @xmath33 a nonempty proper subset of @xmath164 since otherwise we must have @xmath165 and hence @xmath161 , a contradiction . this gives us at least @xmath166 forbidden triples of the form @xmath163 since @xmath42 can be chosen to be one of @xmath167 possible singletons and @xmath168 possible pairs . since @xmath137 , @xmath169 and we are done . the bounds of theorem [ thm : diam3 ] will be proved tight in section [ sec : tightness ] . we will now relate parameter @xmath90 with parameters @xmath170 and @xmath171 . [ lemm : ldset ] let @xmath0 be a graph with a locating - dominating set @xmath172 such that @xmath173 vertices from @xmath174 have a unique neighbour in @xmath172 . then @xmath175 . we construct a centroidal locating set @xmath176 from @xmath172 by adding at most @xmath173 vertices to @xmath172 . note that in the setting of a centroidal locating set and considering @xmath172 as a potential solution , each vertex of @xmath172 is located first by itself , and then by its neighbourhood within @xmath172 , while each vertex @xmath12 of @xmath174 is first located by @xmath177 . hence , any two vertices both in @xmath172 or both in @xmath174 are distinguished . however , if for some vertex @xmath12 of @xmath174 , @xmath178 , @xmath12 and @xmath179 might not be distinguished . in that case , it is enough to add @xmath12 to @xmath172 to solve this problem . this does not cause any other conflict since any superset of a centroidal locating set is also a centroidal locating set . repeating the process @xmath173 times completes the proof ( observing that any other vertex of @xmath172 is distinguished from all other vertices ) . using lemma [ lemm : ldset ] , we obtain the following theorem : [ thm : bounds - md - ld ] for any graph @xmath0 , @xmath180 . for the first inequality , note that any centroidal locating set @xmath65 is a locating set . indeed , if two vertices were at the same distance to each vertex of @xmath65 , then they would not be distinguished by their relative distances by @xmath65 , a contradiction . the second inequality is proved by lemma [ lemm : ldset ] by observing that for any locating - dominating set @xmath172 , @xmath181 : for each @xmath182 , if there were two vertices of @xmath174 having only @xmath179 as a neighbour in @xmath172 , they would not be distinguished by @xmath172 , a contradiction . for graphs of diameter 2 , one gets the following improvement : [ thm : bound - diam2 ] let @xmath0 be a graph of diameter 2 . then @xmath183 . the last two inequalities come from theorem [ thm : bounds - md - ld ] . for the first inequality , we show that any locating set @xmath1 is almost a locating - dominating set . each vertex @xmath12 of @xmath184 has distance 1 to all elements of its neighbourhood @xmath185 in @xmath1 , and distance 2 to all vertices of @xmath186 . in other words , vertices in @xmath1 only distinguish vertices they are adjacent to , from non - adjacent ones . since @xmath1 is a locating set , it follows that each vertex in @xmath184 has a distinct neighbourhood within @xmath1 . therefore , if @xmath1 is dominating , it is also locating - dominating . otherwise , there is at most one vertex that is not dominated ; adding it to set @xmath1 , we get a locating - dominating set of size @xmath187 . in this section , we discuss the tightness of some of the bounds from subsection [ subsec : bounds ] . for @xmath188 it is easy to construct graphs @xmath0 on @xmath3 vertices with @xmath132 and @xmath109 : for @xmath189 , @xmath190 and @xmath191 is the only answer ; for @xmath192 , @xmath193 and both @xmath194 are answers . for @xmath195 , @xmath196 ; the two graphs of figure [ fig : g13 ] have 13 vertices and centroidal dimension @xmath197 ( the black vertices form a centroidal basis ) . it holds that @xmath198 @xcite ; a more intricate construction for this case is presented in figure [ fig : g75 ] . we do not know such optimal examples for @xmath137 ( recall that @xmath66 grows very rapidly with @xmath10 : @xmath199 , @xmath200 @xcite ) . note that it is not possible to directly extend our example for @xmath201 to higher values by using the same idea ; indeed , every two vertices from the centroidal basis @xmath65 are at distance at most 3 from each other . but in our construction , the vertices whose vector is a permutation of @xmath65 have a neighbour in the basis . hence their vector @xmath14 can have length at most 4 , but these @xmath202 vertices need to have a vector of length @xmath10 . * bounded diameter . * we now discuss the tightness of the bounds for diameter 2 and 3 of theorem [ thm : diam3 ] . [ prop : diam2-construct ] for any @xmath203 , there is a graph @xmath0 of diameter 2 with @xmath204 vertices and @xmath132 . we construct @xmath0 in the following way . we let @xmath205 , where @xmath206 are disjoint sets , @xmath65 has @xmath10 vertices , and @xmath33 has @xmath207 vertices . we make sure that in the subgraph induced by @xmath65 , every vertex has a neighbor and a non - neighbor , which is possible since @xmath203 . the set @xmath33 induces a clique and the neighborhood of every vertex of @xmath33 within @xmath65 is distinct and nonempty . since @xmath0 has a vertex of degree @xmath92 ( in @xmath33 ) the diameter is 2 . for every vertex @xmath12 in @xmath65 , @xmath208)$ ] , and by our assumption on the structure of @xmath65 , these three sets are nonempty . for every vertex @xmath12 in @xmath33 , @xmath142 . by construction all these vectors are distinct . in fact , in the construction of proposition [ prop : diam2-construct ] , @xmath65 is a locating - dominating set ; similar constructions are well - known in this context , see for example slater @xcite . for diameter 3 , we give a more complicated construction : [ thm : diam3-construct ] for any @xmath203 , there is a graph @xmath0 of diameter 3 with @xmath209 vertices and @xmath132 . we construct @xmath0 as follows ( see figure [ fig : diam3 ] for an illustration ) . * let @xmath65 be an independent set of size @xmath10 , which will be the centroidal locating set of @xmath0 . * let @xmath210 be a clique containing , for every subset @xmath33 of @xmath65 with @xmath211 , a vertex @xmath212 that is adjacent to all vertices in @xmath33 . set @xmath210 has size @xmath213 . * let @xmath214 be an independent set containing , for every subset @xmath33 of @xmath65 with @xmath215 and for every proper nonempty subset @xmath42 of @xmath159 , a vertex @xmath216 ( note that @xmath217 ) . vertex @xmath216 is adjacent to all vertices of @xmath33 . moreover , if @xmath218 , @xmath216 is adjacent to @xmath219 . if @xmath220 has size 1 , let @xmath221 be an arbitrary size 2-subset of @xmath65 formed by @xmath222 and an arbitrary vertex of @xmath33 , and let @xmath216 be adjacent to @xmath223 . note that set @xmath214 has size @xmath224 . * let @xmath225 be a clique of size @xmath226 containing , for each subset @xmath33 of @xmath65 with @xmath227 , a vertex @xmath228 that is adjacent to the vertices in @xmath33 . . gray sets are cliques.,scaledwidth=55.0% ] the order of @xmath0 is @xmath229 furthermore , the diameter of @xmath0 is exactly 3 . it is at least 3 : consider some vertex @xmath230 . we have @xmath231 , and the vertices in @xmath42 are at distance 2 of @xmath89 ( via vertex @xmath219 ) . however , all vertices of @xmath232 are at distance 3 of @xmath89 , and since @xmath42 is a proper subset of @xmath159 , @xmath232 is nonempty . on the other hand , the diameter is at most 3 . if @xmath12 is any vertex of @xmath0 , @xmath12 is within distance 1 of some vertex @xmath233 in @xmath65 , and since @xmath210 and @xmath225 are cliques and @xmath233 has a neighbor in both @xmath210 and @xmath225 , every vertex in @xmath234 is within distance 2 of @xmath233 and within distance 3 of @xmath12 . moreover , every two vertices in @xmath65 are at distance at most 2 away since they all share a neighbor , @xmath235 . hence , any vertex @xmath236 , since it has a neighbor in @xmath65 , is at distance at most 3 from any vertex in @xmath65 . finally , any vertex in @xmath214 has a neighbor in @xmath210 , which is a clique ; hence any two vertices in @xmath214 have distance at most 3 from each other . it remains to check that @xmath65 is a centroidal locating set . for every @xmath119 , we have @xmath237 . for every vertex @xmath238 in @xmath210 , we have @xmath239 and @xmath240 . for every vertex @xmath241 in @xmath225 , @xmath242 if @xmath243 and @xmath244 . we have now realised all possible vectors with at most two components . finally , for any vertex @xmath230 in @xmath214 , @xmath245 and by the definition of @xmath214 , none of these three components is empty . this completes the proof . we leave the question of the tightness of theorems [ thm : lb ] and [ thm : diam ] open . [ quest : lb ] is the bound @xmath246 asymptotically tight , that is , can we find an infinite family of graphs @xmath0 with @xmath247 ? observe that , by proposition [ prop : diam ] , @xmath122 when @xmath0 has diameter @xmath22 and @xmath3 vertices . hence , in order to construct a graph with a centroidal locating set of size @xmath248 for some @xmath249 , @xmath0 should have diameter @xmath250 ; in particular , for @xmath251 , the diameter should be @xmath252 . it was proved by seb and tannier @xcite that for the hypercube @xmath253 with @xmath254 vertices , @xmath255 . in this regard , it would be interesting to determine whether the family of hypercubes is a positive answer to question [ quest : lb ] ; determining @xmath256 would be of independent interest . [ quest : lb - diam ] what is the maximum order of a graph with centroidal dimension @xmath10 and diameter @xmath257 ? a direct consequence of theorem [ thm : lb ] is that a graph has centroidal dimension equal to its order if and only if it has maximum degree at most 1 . we now fully characterize the set of graph with centroidal dimension of value the order minus one . for some @xmath258 , @xmath259 denotes the complete graph on @xmath3 vertices . for some @xmath260 , @xmath261 denotes the complete bipartite graph with parts of sizes @xmath262 and @xmath233 . let @xmath263 . we denote by @xmath264 , the graph obtained by joining @xmath265 to an independent set of @xmath266 vertices . we call @xmath267 the tree obtained from @xmath268 by attaching @xmath269 degree 1-vertices to one of the ends of @xmath268 . finally , we call @xmath270 the graph obtained from @xmath271 by attaching @xmath269 degree 1-vertices to one of the vertices of @xmath271 . in particular , @xmath272 is a star , @xmath273 is the cycle @xmath274 , @xmath275 is the diamond graph , and @xmath276 is the path @xmath277 . see figure [ fig : extremal ] for illustrations of these graphs ( black vertices belong to a centroidal basis ) . + let @xmath0 be a graph on @xmath278 vertices belonging to @xmath279 . then @xmath280 . by lemma [ lemma : deg2 ] applied on a single degree 2-vertex , it is easily seen that in each case , @xmath281 . if @xmath0 is isomorphic to @xmath259 , the lower bound is directly implied by lemma [ lemma : twins](c ) . if @xmath0 is isomorphic to @xmath272 , then by lemma [ lemma : deg1](a ) any centroidal locating set contains all @xmath92 leaves of @xmath272 . if @xmath0 is isomorphic to @xmath282 or to @xmath264 , by lemma [ lemma : twins](c ) at least @xmath269 vertices of degree 2 belong to any centroidal locating set , as well as one of the other two vertices . however , if no further vertex does belong to the centroidal locating set , then the vertex of degree 2 that is not in the set is not distinguished from its neighbour in the set , a contradiction . if @xmath0 is isomorphic to @xmath267 , by lemma [ lemma : deg1](a ) all @xmath266 vertices of degree 1 belong to any centroidal locating set . moreover , one further vertex also does according to lemma [ lemma : deg1 - 2](b ) . finally , if @xmath0 is isomorphic to @xmath270 , again by lemma [ lemma : deg1](a ) all @xmath269 vertices of degree 1 belong to any centroidal locating set , as well as one degree 2-vertex of the triangle by lemma [ lemma : twins](c ) . if no further vertex does belong to a centroidal locating set , then the two degree 2-vertices of the triangle are not distinguished , a contradiction . in fact , the graphs from figure [ fig : extremal ] are the only extremal graphs , as shown in the following theorem : let @xmath0 be a connected graph on @xmath278 vertices with @xmath280 . then @xmath0 belongs to @xmath283 @xmath284 . we assume by contradiction that @xmath280 but @xmath0 does not belong to the list . hence , @xmath285 since all connected graphs on three or four vertices are in the list . let @xmath17 be a vertex of @xmath0 of degree at least 2 . by lemma [ lemma : deg2 ] , @xmath104 is a centroidal locating set . first of all , observe that there is no vertex at distance 3 of @xmath17 : for contradiction , assume that @xmath286 is such a vertex and let @xmath222 be a neighbour of @xmath286 with @xmath287 that lies on a path from @xmath17 to @xmath286 . by lemma [ lemma : deg2 ] , @xmath288 is a centroidal locating set , a contradiction . we now assume that @xmath289 . let @xmath290 be the two neighbours of @xmath17 , and let @xmath79 be two other vertices in @xmath0 ( they exist since @xmath285 ) . since there is no vertex at distance 3 of @xmath17 , both @xmath79 are neighbours of at least one of @xmath290 . first , assume that @xmath28 is a neighbour of both @xmath12 and @xmath291 and that @xmath89 is only a neighbour of @xmath12 . then @xmath292 is a centroidal locating set , a contradiction . indeed , each vertex of @xmath65 is first located by itself , then by its neighbours in @xmath65 , while @xmath290 are first located by a set of at least two vertices . moreover , @xmath12 is ( in particular ) first located by @xmath89 , while @xmath291 is not . hence , either all vertices other than @xmath293 are adjacent to both @xmath290 , or none is . in the first case , if all common neighbours of @xmath290 form an independent set , @xmath0 is isomorphic to either @xmath282 or @xmath264 , a contradiction . hence , there is an edge between two common neighbours of @xmath290 , say between @xmath79 . then , @xmath294 is a centroidal locating set , a contradiction . indeed , each vertex of @xmath65 is first located by itself , then by its neighbours in @xmath65 , while @xmath295 are first located by a set of at least two vertices . moreover , @xmath28 is ( in particular ) first located by @xmath89 , while @xmath17 is not . we now have that each vertex other than @xmath293 is adjacent to exactly one of @xmath290 . let @xmath296 be the set of neighbours of @xmath12 other than @xmath297 and @xmath298 be the set of neighbours of @xmath291 other than @xmath27 . if both @xmath299 , by lemma [ lemma : deg2 ] , @xmath300 is a centroidal locating set , a contradiction . otherwise , either @xmath0 is isomorphic to @xmath267 if @xmath290 are non - adjacent , or to @xmath270 otherwise , a contradiction in both cases . by the previous discussion , @xmath0 has no degree 2-vertices . hence @xmath301 . suppose moreover that @xmath18 is an independent set . since @xmath0 is not isomorphic to @xmath272 , there are vertices at distance 2 of @xmath17 . hence @xmath17 has a neighbour , @xmath12 , with two such vertices as neighbours ( since @xmath302 ) : let @xmath79 be these vertices . then , @xmath303 is a centroidal locating set , a contradiction . indeed , all vertices but @xmath27 are located first by themselves only , whereas @xmath17 is located first by @xmath304 and @xmath12 by a set containing both @xmath79 . now , we assume that @xmath18 is a clique . since @xmath0 is not a complete graph , @xmath17 has a neighbour @xmath12 , having a neighbour @xmath28 with @xmath305 . then , by lemma [ lemma : deg2 ] , @xmath306 is a centroidal locating set of @xmath0 , a contradiction . hence , @xmath18 is neither an independent set , nor a clique : there is a vertex @xmath12 in @xmath18 with a neighbour @xmath291 and a non - neighbour @xmath28 , both being in @xmath18 . since no vertex in @xmath0 has degree 2 , @xmath12 has an additional neighbour , @xmath89 . but then by lemma [ lemma : deg2 ] , @xmath306 is a centroidal locating set of @xmath0 , a contradiction . we now study parameter @xmath90 for graphs in which every pair of vertices is connected via a bounded number @xmath10 of paths . we show that such graphs have centroidal dimension @xmath307 , where @xmath94 is the number of edges . in particular , this applies to paths and cycles ; for these graphs , we show that the lower bound is asymptotically tight . these cases are particularly interesting for the following reason : the metric dimension of a path or a cycle with @xmath3 vertices is easily seen to be constant ( 1 for any path , 2 for any cycle ) , whereas the location - domination number is linear ( roughly @xmath308th of the vertices @xcite ) . in contrast , the centroidal dimension is about the square - root of the order . but first , the following technical lemma will be useful when showing our lower bound . [ lemma : lbcycle - path ] let @xmath0 be a graph with @xmath27 two adjacent vertices , and let @xmath65 be a centroidal locating set . then , there are two vertices @xmath82 of @xmath65 and a path @xmath309 such that at least one of the following properties hold : 1 . @xmath310 and @xmath311 is the edge @xmath312 ; 2 . @xmath313 , @xmath314 ( or , symmetrically , @xmath315 , @xmath316 and @xmath311 contains a shortest path from @xmath17 to @xmath317 and a shortest path from @xmath12 to @xmath318 ; 3 . @xmath319 , @xmath320 , @xmath311 contains a shortest path from @xmath17 to @xmath317 and a shortest path from @xmath12 to @xmath318 , @xmath311 has odd length and @xmath312 is the middle edge of @xmath311 . if both @xmath27 belong to @xmath65 , we are in the first case and we are done . otherwise , since @xmath65 is a centroidal locating set , without loss of generality there are two vertices @xmath160 of @xmath65 such that @xmath321 and @xmath322 , or vice - versa . * case a : . * we show that the second case of the statement holds . if @xmath12 lies on a shortest path @xmath323 from @xmath17 to one of @xmath160 ( say , @xmath233 ) , we are done : then no shortest path from @xmath17 to @xmath324 can go through @xmath12 ( otherwise @xmath325 ) . hence , setting @xmath326 and @xmath327 , the concatenation of @xmath323 and any shortest path from @xmath17 to @xmath324 is a path @xmath311 satisfying the desired properties . hence , we assume that @xmath12 does not lie on a shortest path from @xmath17 to @xmath233 . since @xmath328 , we can assume without loss of generality that @xmath329 ( moreover since @xmath330 , we have @xmath331 ) . hence a shortest path @xmath332 from @xmath12 to @xmath324 does not go through @xmath17 . let @xmath323 be a shortest path from @xmath233 to @xmath17 . if the concatenation @xmath333 is a path , we are done by setting @xmath334 , @xmath335 and @xmath336 . therefore , assume that it is not a path , and let @xmath291 be the vertex closest to @xmath17 appearing in both @xmath323 and @xmath337 : @xmath333 contains a cycle going through @xmath293 . now , since no shortest path from @xmath12 to @xmath324 goes through @xmath17 , we have @xmath338 . also , since we assumed that @xmath12 does not lie on a shortest path from @xmath17 to @xmath233 , in particular @xmath339 . therefore , @xmath340 . this implies that @xmath341 ( indeed , we had @xmath331 and now @xmath342 ) . hence , @xmath343 . let @xmath344 be the path obtained from the concatenation of a shortest path from @xmath17 to @xmath291 and the subpath of @xmath337 from @xmath291 to @xmath324 , and let @xmath345 be a shortest path from @xmath12 to @xmath233 . then , @xmath345 does not contain any vertex of @xmath344 ( indeed , if there was such a vertex @xmath222 , then @xmath346 , a contradiction ) . hence , the concatenation @xmath347 is a path that has the desired properties . * case b : . * we show that the third case of the statement holds . since @xmath322 and @xmath27 are adjacent , @xmath348 and @xmath349 . hence @xmath350 and @xmath351 , and @xmath352 . we set @xmath353 and @xmath354 . observe that the concatenation @xmath311 of a shortest path from @xmath317 to @xmath17 , the edge @xmath355 , and a shortest path from @xmath12 to @xmath318 is a path from @xmath317 to @xmath318 ( if it were not a path , we would have @xmath356 ) . since @xmath311 has the desired properties , this completes the proof . in what follows , for a pair @xmath312 of vertices in a graph @xmath0 , we let @xmath357 and @xmath358 be the number of odd and even ( not necessarily disjoint ) paths connecting @xmath17 to @xmath12 , respectively . [ thm : bounded - nr - paths ] let @xmath0 be a graph on @xmath3 vertices and @xmath94 edges such that for every pair @xmath312 of vertices , @xmath359 for some integer @xmath10 . then , @xmath360 . in particular , for every tree @xmath42 , @xmath361 . let @xmath65 be a centroidal locating set of @xmath0 . to each pair @xmath27 of adjacent vertices in @xmath0 , we assign a triple @xmath362 of two vertices @xmath82 of @xmath65 and a path @xmath309 satisfying one of the three properties described in lemma [ lemma : lbcycle - path ] . the assignment is done as follows : 1 . if both @xmath27 belong to @xmath65 , set @xmath363 ( we say that the pair @xmath27 is of _ type 1 _ ) ; 2 . otherwise , if there is some pair @xmath364 of @xmath65 and the corresponding path @xmath311 such that the second property of lemma [ lemma : lbcycle - path ] holds , then set @xmath365 ( we say that the pair @xmath27 is of _ type 2 _ ) ; 3 . otherwise , let @xmath366 consist of an arbitrary pair @xmath82 of vertices of @xmath65 and the corresponding path @xmath311 satisfying the third property of lemma [ lemma : lbcycle - path ] ( we say that the pair @xmath27 is of _ type 3 _ ) . now , for a given pair @xmath82 of @xmath65 and a path @xmath311 from @xmath317 to @xmath318 , we will upper - bound the number of pairs of adjacent vertices @xmath27 such that @xmath365 . assume first that @xmath311 has odd length . if @xmath311 is the edge @xmath82 , there is only the pair @xmath310 of type 1 with @xmath365 . otherwise , assume @xmath312 is a pair of type 2 with @xmath365 . then by lemma [ lemma : lbcycle - path ] , @xmath367 and the subpath of @xmath311 from @xmath17 to @xmath317 has length @xmath368 . since the subpath from @xmath12 to @xmath318 is also a shortest path and @xmath369 , it has length @xmath370 with @xmath371 . the length of @xmath311 is @xmath372 . since it is an odd number , @xmath373 and @xmath312 is the middle edge of @xmath311 . if @xmath312 is of type 3 , by lemma [ lemma : lbcycle - path ] , @xmath312 is also the middle edge of @xmath311 . hence in total there is at most one pair @xmath27 with @xmath365 . now , assume that @xmath311 has even length . let @xmath312 be a pair with @xmath365 . then , by lemma [ lemma : lbcycle - path ] , @xmath27 can not be of type 1 or type 3 , so it must be of type 2 . by the same arguments as in the previous paragraph , the length of the subpath of @xmath311 from @xmath317 to @xmath17 is @xmath368 , and the length of the subpath of @xmath311 from @xmath12 to @xmath318 is either @xmath374 or @xmath375 . in both cases , one of @xmath27 is the middle vertex of @xmath311 . hence there can be at most two such pairs . to summarize , we proved that for each pair @xmath82 of @xmath65 , the number of pairs @xmath27 with @xmath365 for some path @xmath311 is at most @xmath376 since in total , there are @xmath94 pairs of adjacent vertices in @xmath0 and each such pair is associated to exactly one of the @xmath377 pairs of @xmath65 , we have @xmath378 , and the bound follows . when @xmath0 is a tree , there are @xmath92 edges , and there is a unique path between any pair of vertices , hence @xmath379 . we now show that the bound of theorem [ thm : bounded - nr - paths ] is tight up to a constant factor for paths and cycles . [ thm : paths - cycles ] let @xmath278 . + ( 1 ) if @xmath3 is even , @xmath380 . if @xmath3 is odd , @xmath381 . in both cases , @xmath382 . if @xmath383 for some integer @xmath173 , then @xmath384 . + ( 2 ) @xmath385 . if @xmath386 for some integer @xmath173 , then @xmath387 . * lower bounds . * the bounds follow from theorem [ thm : bounded - nr - paths ] since paths are trees , and by observing that cycles have @xmath3 edges . for even cycles , for any pair @xmath312 of vertices , @xmath388 and @xmath389 and @xmath390 , hence @xmath391 . for odd cycles , @xmath392 and @xmath393 , hence @xmath394 . * upper bounds for cycles . * we first prove that for any @xmath395 , if @xmath396 , then : @xmath397 assuming that @xmath383 , and setting @xmath398 and @xmath399 , inequality yields the claimed bound @xmath400 . let @xmath401 be the vertex set of @xmath26 . let us divide @xmath26 into @xmath402 portions of @xmath403 consecutive vertices each : for @xmath404 , @xmath405 . for each @xmath406 , we further define two subsets of @xmath407 as follows : @xmath408 and @xmath409 , where for @xmath410 , @xmath411 and @xmath412 . in other words , @xmath413 contains @xmath414 vertices from @xmath407 , starting from the second one , and @xmath415 contains the @xmath414 last vertices of @xmath407 . observe that the first and the @xmath416-nd vertices from @xmath407 neither belong to @xmath413 , nor @xmath415 . we now define a set @xmath417 , which we claim , will be our centroidal locating set . we let @xmath418 , where for @xmath419 , @xmath420 ( that is , @xmath421 contains each second vertex of @xmath422 ) . we let @xmath423 , where for @xmath404 , @xmath424 ( that is , @xmath425 contains the first vertex of each set @xmath407 ) . an illustration of sets @xmath426 is given in figure [ fig : cycleub ] . .,scaledwidth=90.0% ] observe that @xmath65 has @xmath427 elements . it remains to show that @xmath65 is a centroidal locating set . first of all , notice that for any @xmath404 , @xmath428 is the unique vertex that is first located by itself , and later by @xmath429 at the same time . for @xmath430 , the @xmath416-nd vertex of @xmath407 , @xmath431 , is the only one that is located _ first _ by @xmath432 at the same time . similarly , @xmath433 is the unique vertex first located by the vertices of @xmath421 ( in some order ) , and then by @xmath434 at the same time . if @xmath435 , if @xmath436 , @xmath28 is the only vertex located first by itself only ; if @xmath437 , @xmath28 is the only vertex located first by its two neighbours . hence , all the previously considered vertices are distinguished from any other vertex in @xmath26 . now , let @xmath27 be a pair of vertices not yet proved to be distinguished . if @xmath438 ( @xmath439 ) , then @xmath440 but @xmath441 . if @xmath442 ( @xmath443 ) , then @xmath444 but @xmath445 . the case where @xmath446 is symmetric . if @xmath447 ( @xmath443 ) , if @xmath448 , then @xmath449 while @xmath450 . otherwise , @xmath451 while @xmath452 . it remains to prove that for @xmath439 , any two vertices from @xmath413 are distinguished ( the case of two vertices of @xmath415 would follow by symmetry ) . to see this , let @xmath453 . if @xmath454 , @xmath455 , and no other vertex of @xmath413 has this property . similarly , if @xmath456 , @xmath457 . this completes the proof of validity of @xmath65 . in order to prove the bound for all cycles , if @xmath3 is not of the form @xmath458 , let @xmath459 be the integer of this form that is closest to @xmath3 and such that @xmath460 : @xmath461 for some @xmath10 , and @xmath462 . a construction similar to the previous one can be done . letting @xmath398 , @xmath399 , we construct @xmath423 with @xmath424 as previously ; however this time @xmath425 does not include any vertex from @xmath463 . instead , we let @xmath464 : @xmath421 contains the first @xmath403 vertices ( except @xmath465 ) , together with the @xmath10 last vertices . it is clear that the construction works the same way than in the previous proof we omit the details as a formal proof would be tedious . are probably not needed . ] in total , @xmath417 has size at most @xmath466 . since @xmath467 , @xmath468 and @xmath469 . furthermore @xmath470 , and @xmath471 . hence @xmath472 , proving the bound . * upper bounds for paths . * a similar construction than the one for cycles can be done for the case of paths : for any @xmath395 , if @xmath473 , then @xmath474 . the idea of the construction is again to divide the vertex set @xmath401 into @xmath402 portions of size @xmath403 each ( vertex @xmath475 does not belong to any such portion ) . for @xmath476 , @xmath424 , and @xmath425 contains all vertices of the form @xmath477 . @xmath421 is defined in the same way as for our construction for cycles , but now we also consider a set @xmath478 similar to @xmath421 , but on the other end of the path . we have @xmath479 , and similar arguments than for the construction for cycles show that @xmath480 is a centroidal locating set . hence , assuming @xmath386 for some @xmath481 and setting @xmath482 and @xmath399 , we get that @xmath483 . for the general bound , once again we do not optimize the constant . assume that @xmath3 is not of the form @xmath484 , and let @xmath485 be the integer of this form that is closest to @xmath3 and @xmath460 : we have @xmath461 for some @xmath10 , and @xmath486 . let @xmath482 and @xmath399 . now , @xmath425 is selected as before among the first @xmath94 vertices ; @xmath425 has @xmath402 elements . @xmath421 contains the first @xmath403 vertices ( except @xmath465 ) , and @xmath478 contains the last @xmath487 vertices ( except @xmath488 ) . in total , @xmath489 has @xmath490 vertices . since @xmath491 , @xmath492 this implies @xmath493 . since @xmath494 , @xmath495 ; hence , @xmath496 . let us now turn our attention to the computational complexity of finding a small centroidal locating set , that is , the computational complexity of the following problem : + we have seen in theorem [ thm : bound - diam2 ] that for any graph of diameter 2 , @xmath497 . we get the following corollary , showing that centroidal dimension is computationally very hard , even from the approximation point of view ( recall that an @xmath498-approximation algorithm for problem @xmath311 is a polynomial - time algorithm for @xmath311 which always outputs a solution of size no greater than @xmath498 times the size of an optimal solution ) . [ cor : complexity ] centroidal dimension is -hard to approximate within any factor @xmath95 for graphs on @xmath3 vertices ( even for graphs with a vertex adjacent to all other vertices , and hence diameter 2-graphs ) . for graphs of diameter 2 , it has an @xmath499-approximation algorithm . since @xmath497 and the bounds are constructive , any @xmath498-approximation algorithm ( @xmath500 ) for minimum locating - dominating set can be transformed into an approximation algorithm of factor @xmath501 for centroidal dimension for graphs of diameter 2 , and vice - versa . indeed , given an @xmath498-approximate locating - dominating set @xmath22 of @xmath0 , we construct a centroidal locating set of @xmath0 of size at most @xmath502 . we have @xmath503 . for the converse , the reasoning is similar . this also implies that if minimum locating - dominating set is -hard to @xmath498-approximate for graphs of diameter 2 for some @xmath504 , then centroidal dimension is -hard to approximate within factor @xmath505 for graphs of diameter 2 . moreover , it follows from a reduction for minimum identifying code in the first author s thesis ( * ? ? ? * section 6.4 ) and a lemma from gravier , klasing and moncel @xcite ( see also foucaud @xcite ) that minimum locating - dominating set is -hard to approximate within a factor of @xmath95 for graphs having a vertex adjacent to all other vertices . this proves the non - approximability bound . note that corollary [ cor : complexity ] fully determines the computational complexity of centroidal dimension in graphs of diameter 2 from the approximation point of view . it was recently proved by hartung and nichterlein that the related problem metric dimension remains -hard to approximate within a factor of @xmath95 even for subcubic graphs @xcite . in general , it would be interesting to extend the result of corollary [ cor : complexity ] to other families of graphs . as it is often the case with domination or identification problems in graphs , a good way of reformulating our problem is to represent it as an instance of minimum set cover , which is well - known to be @xmath506-approximable ( see johnson @xcite ) : + for example , metric dimension for a graph @xmath0 can be expressed in this way by constructing a hypergraph @xmath510 on vertex set @xmath511 with a hyperedge @xmath512 for each vertex @xmath9 containing all pairs @xmath513 of vertices with @xmath514 . then @xmath510 has a set cover of size @xmath10 if and only if @xmath0 has a locating set of size @xmath10 , as shown by khuller , raghavachari and rosenfeld @xcite . hence metric dimension is @xmath499-approximable . using observation [ obs : otherdef ] , finding a centroidal locating set is equivalent to finding a set of _ pairs _ of vertices which identifies each pair of vertices in @xmath0 where a pair @xmath82 identifies @xmath79 if @xmath83 but @xmath84 , or @xmath515 and @xmath86 . let @xmath0 be a graph on @xmath3 vertices . we define the hypergraph @xmath516 on vertex set @xmath517 . for each pair @xmath513 of vertices of @xmath0 , @xmath518 has a hyperedge @xmath519 that contains all pairs that are identified by @xmath513 . let @xmath172 be a set cover of @xmath518 . using our previous observation , one can construct a centroidal locating set @xmath520 of @xmath0 by taking the union of all elements in the pairs that correspond to hyperedges in @xmath172 : @xmath521 . indeed , every pair of vertices of @xmath0 is identified by a pair corresponding to a hyperedge of @xmath518 . hence we have : latexmath:[\[\begin{aligned } for the other direction , given a centroidal basis @xmath65 of @xmath0 , one can construct a set cover of @xmath518 consisting of all @xmath377 pairs of vertices of @xmath65 : each pair @xmath27 in @xmath81 is identified by some pair @xmath79 in @xmath65 , hence vertex @xmath312 in @xmath518 is covered by the corresponding hyperedge @xmath519 . denoting by @xmath523 the size of an optimal set cover of @xmath518 , this implies : @xmath524 now , in order to approximate centroidal dimension , we construct @xmath516 from @xmath0 , and apply the standard approximation algorithm for minimum set cover @xcite to get a set cover @xmath172 of @xmath518 of size @xmath525 . by inequalities and , we get a centroidal locating set @xmath520 of size at most @xmath526 . we note that the quadratic dependence between @xmath523 and @xmath2 is necessary for inequality in the reduction of theorem [ thm : sqrtn - approx ] . indeed , when we considered the case of cycles in section [ sec : paths - cycles ] , we had @xmath3 special pairs to distinguish using other pairs @xmath82 from @xmath65 , but any pair @xmath82 could only distinguish a small ( constant ) number of these @xmath3 pairs . however , we could build a centroidal locating set @xmath65 of size @xmath531 , meaning that a large fraction of the pairs from @xmath65 were indeed necessary to distinguish the pairs . hence this would lead to a set cover of @xmath518 of size @xmath532 in our reduction . this suggests that one can not improve the approximation ratio from theorem [ thm : sqrtn - approx ] for centroidal dimension by using our reduction . hence we ask the following question : we close the section with a remark on the parameterized complexity of centroidal dimension : this problem is fixed - parameter - tractable with parameter @xmath10 , the size of the solution , that is , it admits an algorithm of running time @xmath533 for some computable function @xmath534 : we know that the order @xmath3 of a graph with a centroidal locating set of size @xmath10 is at most @xmath66 , hence , if the input has more thatn @xmath66 vertices , we answer no . otherwise , we enumerate all @xmath535 subsets of vertices of size @xmath10 to check if one of them is a centroidal locating set ( checking whether a given set is centroidal locating can be done in time @xmath536 ) . this algorithm has running time @xmath537 , which is computable and only depends on @xmath10 . in contrast to proposition [ prop : fpt ] , it was proved by hartung and nichterlein @xcite that deciding whether there is a solution of size @xmath10 for metric dimension is highly unlikely to be solvable by an algorithm of running time @xmath538 . * acknowledgements . * we thank the anonymous referees for carefully reading the paper , especially for detecting a flaw in the original proof of theorem [ thm : bounded - nr - paths ] and in an erroneous construction that was replaced by theorems [ thm : diam3 ] and [ thm : diam3-construct ] ( which were suggested by one of the referees ) . d. i. carson , on generalized location domination , graph theory , combinatorics and applications : proc 7th quadrennial international conference on the theory and applications of graphs , vol . 1 , 1995 , pp . 161179 . f. foucaud , combinatorial and algorithmic aspects of identifying codes in graphs , phd thesis , universit bordeaux 1 , bordeaux , france , 2012 . available online at http://tel.archives-ouvertes.fr/tel-00766138 . f. foucaud , the complexity of the identifying code problem in restricted graph classes , proc 24th international workshop on combinatorial algorithms , lecture notes in computer science , vol . 8288 , springer , 2013 , pp . 150163 . f. foucaud , on the decision and approximation complexities for identifying codes and locating - dominating sets in restricted graph classes , manuscript ( 2013 ) . available online at http://hal.archives-ouvertes.fr/hal-00923356 .

the restricted numerical range of an operator acting on a-dimensional hilbert space is defined as a set of all possible expectation values of this operator among pure states which belong to a certain subset of the of set of pure quantum states of dimension . one considers for instance the set of real states , or in the case of composite spaces , the set of product states and the set of maximally entangled states . combining the operator theory with a probabilistic approach we introduce the restricted numerical shadow of a normalized probability distribution on the complex plane supported in . its value at point is equal to the probability that the expectation value is equal to , where represents a random quantum state in subset distributed according to the natural measure on this set , induced by the unitarily invariant fubini study measure . studying restricted shadows of operators of a fixed size we analyse the geometry of sets of separable and maximally entangled states of the composite quantum system . investigating trajectories formed by evolving quantum states projected into the plane of the shadow we study the dynamics of quantum entanglement . a similar analysis extended for operators on dimensional hilbert space allows us to investigate the structure of the orbits of and quantum states of a three qubit system . = 1

1201.2524

recent studies on quantum entanglement , a crucial resource in the theory of quantum information processing , contributed a lot to our understanding of this deeply non - classical phenomenon ( see _ e.g. _ @xcite and references therein ) . in particular , some progress has been achieved in elucidating the geometry of quantum entanglement @xcite , but several questions concerning this topic remain still open @xcite . it was also suggested that a geometric approach is useful to classify and quantify quantum entanglement @xcite . the phenomenon of quantum entanglement , non - classical correlations between individual subsystems , may arise in composite physical systems . consider then the simplest composite system , which consists of two parts and can be described in a hilbert space with a tensor product structure , @xmath13 . any product pure state , @xmath14 , is called _ separable _ , while all other pure states are called _ entangled_. the set @xmath15 of all separable pure states has the structure of the cartesian product of the complex projective spaces @xcite , @xmath16 . among entangled pure states of a bipartite system one distinguishes the set @xmath17 of _ maximally entangled _ states , such that the partial trace of the corresponding projector is proportional to the identity matrix . the set @xmath17 contains the generalized bell state , @xmath18 , and all states obtained from it by a local unitary transformation , @xmath19 . the set of maximally entangled states is thus equivalent to @xmath20 . the structure of the set @xmath17 and other sets of locally equivalent entangled states of a bipartite system was studied in @xcite . in this work we propose to analyse the geometry of the set of entangled and separable quantum states using the algebraic concepts of the numerical range and numerical shadow of an operator . for any operator @xmath1 acting on the complex hilbert space @xmath21 one defines its _ numerical range _ @xcite as a subset of the complex plane which contains expectation values of @xmath1 among arbitrary normalized pure states , @xmath22 due to the classical toeplitz - hausdorff theorem the set @xmath23 is convex see e.g. @xcite . the differential topology and projection aspects of numerical range were investigated in @xcite the standard notion of numerical range , often used in the theory of quantum information @xcite , can be generalized in several ways @xcite . for instance , for an operator acting on a composite hilbert space one defines the _ product _ numerical range @xcite ( also called _ local _ numerical range @xcite ) and a more general class of numerical ranges _ restricted _ to a specific class of states @xcite , @xmath24 here @xmath25 denotes a selected subset of the set of pure quantum states of a given size @xmath2 . for instance , one may consider the set of all real states , or , in the case of composite spaces , the set of complex product states or the set of real maximally entangled states . for any operator @xmath1 acting on @xmath21 one defines a probability density @xmath26 on the complex plane @xcite , supported in the numerical range @xmath23 , @xmath27 here @xmath28 denotes the unique unitarily invariant measure on the set @xmath29 of @xmath2dimensional quantum pure states , also called fubini - study ( fs ) measure . in other words the shadow of @xmath1 at a given point @xmath30 characterizes the likelihood that the expectation of @xmath1 among a random pure states is equal to @xmath6 . sometimes it is convenient to treat the numerical shadow as a probability measure @xmath31 , on a complex plane . for any measurable set @xmath32 it reads @xmath33 . if the operator @xmath1 is normal , @xmath34 , its shadow @xmath26 can be interpreted as a projection of the set of classical states the @xmath2dimensional regular simplex of probability distributions into a two - plane @xcite . in the more general case of a non normal @xmath1 its shadow can be associated with a probability distribution obtained by projecting the set @xmath29 of quantum pure states of size @xmath2 into a plane . thus choosing a matrix @xmath1 to be analysed one fixes the relative position of the set @xmath29 and determines the direction , along which it is projection on the plane . hence investigating all possible shadows of various operators of a given size @xmath2 one gathers information about the structure of the set @xmath29 . combining the ideas of the restricted numerical range ( [ rest_range ] ) with the numerical shadow ( [ shadow ] ) one is led in a natural way to the definition of a _ restricted numerical shadow _ , @xmath35 where @xmath36 is the selected subset of @xmath37 . in particular , choosing the appropriate subsets of the set of pure states we define the numerical shadow restricted to _ real _ states or @xmath38coherent states . in the case of a composite hilbert space one defines the shadow with respect to _ separable _ states or _ maximally entangled _ states . the restrictions can be combined so one can consider the shadow restricted e.g. to real separable states , or in the case of operators acting on @xmath39 dimensional space , which describes a three qubit system , one may study the shadow with respect to real @xmath11 states or complex @xmath12states . it will be convenient to use simplified terms , so for brevity we will slightly abuse the notation and write about separable shadow , real entangled shadow or complex ghz shadow of a given operator . using the notion of probability measure we will denote this restricted by @xmath40 . on one hand , for a given matrix @xmath1 one may study its restricted numerical shadows @xmath41 determined by a given set @xmath3 of quantum states . alternatively , investigating numerical shadows of various matrices of a fixed dimension @xmath2 with respect to a concrete subset @xmath42 one may analyse its geometry . in this work we study in this way the structure of the set of real , separable and maximally entangled quantum states for a two and three qubit system . this paper is organized as follows . in section 2 some basic properties of the numerical range and the numerical shadow are reviewed . in section 3 we discuss numerical shadow restricted to real states . in the case of operators acting on the hilbert space with a tensor product structure , @xmath43 , one defines classes of separable and maximally entangled states . numerical ranges restricted to these sets are analysed in section 4 and 5 respectively . in section 6 we show that investigations of trajectories formed by evolving quantum states projected into the plane of the shadow contributes to the understanding the dynamics of quantum entanglement . in section 7 we discuss the simplest case of composite system consisting of three parts and described in the hilbert space @xmath44 . in the set @xmath45 of pure states acting on this space one defines two classes of entangled states called @xmath11 and @xmath12 . for any operator acting on @xmath46 it is then natural to introduce the shadow restricted to states locally equivalent to @xmath11 or @xmath12 , and these are investigated in section 6 . elements of a hilbert space are used as basic objects of the quantum theory . a physical system with @xmath2 distinguishable states can be described by an element @xmath7 of the complex hilbert space @xmath21 . it is assumed that such a _ pure quantum state _ is normalized , @xmath47 , so it belongs to the hypersphere of dimension @xmath48 . one identifies any two states , which differ only by a global phase , @xmath49 . the set of all pure quantum states @xmath29 acting on @xmath21 is therefore equivalent to the complex projective space , @xmath50 ( see _ e.g. _ @xcite ) . any convex combination of projectors onto pure states forms a _ mixed quantum state _ , @xmath51 with @xmath52 and @xmath53 . any such state @xmath54 , also called a density operator , is hermitian , positive and normalized by the trace condition , tr@xmath55 . thus the set @xmath56 of all density matrices of order @xmath2 forms a convex set of @xmath57 real dimensions . the projectors corresponding to the extremal points of the set of density operators form the set of pure quantum states , @xmath58 . in the one qubit case , @xmath59 , the set of pure quantum states forms the _ bloch sphere _ , @xmath60 , which in this case is equivalent to the boundary of the bloch ball @xmath61 . for a larger dimension the @xmath62dimensional set of pure states @xmath29 forms only a zero measure subset of the @xmath63 dimensional boundary of the set @xmath56 . note that the definitions of the set of pure and mixed quantum states are unitarily invariant , so they can be formulated without specifying any basis in the hilbert space . on the other hand , among quantum states , one distinguishes also the set of _ classical states _ , which are formed by the density matrices diagonal in a certain basis . hence any classical state @xmath64 is represented by a normalized probability vector , @xmath65 such that @xmath66 and @xmath67 . the set of classical states forms thus the probability simplex @xmath68 . each of @xmath2 corners of the simplex represents a classical pure state , and their convex hull is equivalent to @xmath69 . in the one qubit case , @xmath59 , the set @xmath70 of classical states forms an interval which joins two poles of the bloch sphere and traverses across the interior of the bloch ball of one qubit mixed quantum states . the definition ( [ range1 ] ) of the numerical range of a matrix @xmath1 of order @xmath2 specifies @xmath23 as the set of expectation values @xmath71 for normalized pure states @xmath7 . in fact the compact set @xmath23 in the complex plane can be considered as a _ projection _ of the set @xmath56 of quantum states onto a two plane . the following facts were recently established in @xcite . [ prop : classical ] consider the set of classical states of size @xmath2 , which forms the regular simplex @xmath69 in @xmath72 . then for each normal matrix @xmath1 of dimension @xmath2 there exists an affine rank @xmath73 projection @xmath74 of the set @xmath69 whose image is congruent to the numerical range @xmath23 of the matrix @xmath1 . conversely for each rank @xmath73 projection @xmath74 there exists a normal matrix @xmath1 whose numerical range @xmath23 is congruent to image of @xmath69 under projection @xmath74 . [ prop : quantum ] let @xmath56 denote the set of quantum states of dimension @xmath2 embedded in @xmath75 with respect to euclidean geometry induced by hilbert - schmidt distance . then for each ( arbitrary ) matrix @xmath1 of dimension @xmath2 there exists an affine rank @xmath73 projection @xmath74 of the set @xmath56 whose image is congruent to the numerical range @xmath23 of the matrix @xmath1 . conversely for each rank @xmath73 projection @xmath74 there exists a matrix @xmath1 whose numerical range @xmath23 is congruent to image of @xmath56 under projection @xmath74 . thus by fixing the dimension @xmath2 and selecting various operators @xmath1 of this order and analysing their numerical ranges @xmath23 we may gather information about the projections of the set of quantum states . for instance , for @xmath59 the projections of the bloch sphere form ellipses ( which could form circles or may reduce an intervals ) , and so the numerical range of any operator @xmath1 of size @xmath59 may be the result . similarly the numerical range of operators of size @xmath76 can be viewed as projections of the set @xmath77 of single qutrit quantum states , while the numerical range for matrices of order @xmath78 can be associated with projections of the set @xmath79 @xcite . the haar measure on the unitary group @xmath80 induces on the set @xmath29 of quantum pure states the unitarily invariant fubini - study measure @xmath81 . in the simplest case of @xmath59 this measure corresponds to the uniform distribution of points on the bloch sphere @xmath82 . for any operator @xmath1 of size @xmath2 we may compute its expectation value for random pure state @xmath83 chosen with respect to measure @xmath81 . in this way for any operator @xmath1 we define its _ numerical shadow _ a probability distribution ( [ shadow ] ) on the complex plane . note that by construction the numerical shadow @xmath26 is supported on the numerical range @xmath23 . furthermore , the numerical shadow is unitarily invariant : @xmath84 . for any normal matrix @xmath1 , which commutes with its adjoint , the numerical shadow @xmath26 covers the numerical range @xmath23 with the probability corresponding to a projection of a regular simplex @xmath69 of classical states embedded in @xmath72 ) onto a plane . in general , for a non normal operator @xmath1 acting on @xmath21 , its shadow covers the numerical range @xmath23 with the probability corresponding to an orthogonal projection of the complex projective manifold @xmath50 onto a plane . in this work we will analyse the numerical range restricted @xcite to a certain subsets @xmath3 of pure quantum states and the corresponding _ restricted _ numerical shadows , with probability density determined by random states distributed on the subset @xmath3 according to the measure induced by the fubini study measure on @xmath29 . consider a diagonal matrix of size @xmath2 , namely @xmath85 . let @xmath86 be an arbitrary fixed pure state in @xmath21 . then the entire set of random pure states can be obtained as @xmath87 , where @xmath88 is a random unitary matrix distributed according the haar measure . thus the expansion coefficients of the state @xmath7 read @xmath89 . the _ numerical shadow _ of the operator @xmath90 is defined as the density distribution of random numbers @xmath91 , where @xmath7 is a random state defined by the random unitary matrix @xmath92 . therefore @xmath93 note that the vector of coefficients @xmath94 with @xmath95 belongs to the @xmath96 dimensional simplex @xmath69 . so the complex random variable @xmath6 can be treated as a scalar product , @xmath97 where the complex vector @xmath98 is given by the diagonal of the operator @xmath90 , while the probability vector @xmath99 is random . its distribution inside the simplex depends on the distribution in the space of unitary matrices @xmath80 . in cases of interest for us this distribution belongs to the class of _ dirichlet distributions _ parametrized by a real number @xmath100 , @xmath101 in particular , we are interested in two situations . * _ standard ( complex ) numerical shadow _ @xcite , generated by the haar measure on @xmath80 . then the vector @xmath99 is distributed uniformly with respect to the flat ( lebesgue ) measure on @xmath69 , i.e. the dirichlet measure with the dirichlet parameter @xmath102 @xcite . * _ real _ numerical shadow , generated by the haar measure on the group @xmath103 of orthogonal matrices . then the vector @xmath64 is distributed with respect to the _ statistical measure _ , @xmath104 , i.e. the dirichlet measure with the dirichlet parameter @xmath105 . in both cases the shadow covers the entire numerical range of @xmath90 , equal to the convex hull of the spectrum @xmath106 with a density determined by the appropriate dirichlet distribution . the case of real numerical shadow , i.e. the shadow with respect to the real pure states is treated in more detail in the subsequent section . quantum states belonging to a complex hilbert space form a standard tool of the quantum theory . however , in some cases it is instructive to restrict attention to real states only . on one hand the set of the real states is easier to analyse than the full set of complex states of larger dimensionality . for instance , the phenomenon of quantum entanglement can be studied for the case of real states of a four level systems , sometimes referred to as a pair of rebits ( i.e. real bits ) @xcite . on the other hand is some physical applications it is easier to use real orthogonal rotation matrices to construct elements of the entire set of real density matrices . therefore in this section we will study the shadow of an operator with respect to the set of quantum states with real coefficients . such an investigation allows us to improve understanding of the structure of the set of real mixed states and its natural subsets . consider the subset of pure quantum states of size @xmath2 , that can be represented by real expansion coefficients in a given basis . this set forms a @xmath108 dimensional real projective space @xmath109 . substituting into the definition ( [ rest_range ] ) of a restricted numerical range @xcite the set @xmath109 as the subset @xmath3 of @xmath29 we arrive at the numerical range restricted to real states . in general the restricted numerical range needs not to be convex or simply connected . similarly , restricting the integration in ( [ rshadow ] ) to the set @xmath110 of real quantum states we obtain the shadow of the operator @xmath1 with respect to real states . for brevity we will also use a shorter expression , the _ real shadow _ of @xmath1 . in the case of @xmath59 the set of real states forms the real projective space equivalent to a circle , @xmath111 . thus the real shadow of a generic , non - normal matrix of size two forms a singular probability distribution supported on an elliptic curve in the complex plane , and not inside its interior . the real shadows presented in fig . [ fig : shadreal ] are obtained for illustrative operators of size @xmath59 @xmath112 , \label{a2}\ ] ] @xmath76 , @xmath113 , \quad a_{3b } = \left[\begin{array}{ccc } 0 & 1 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 2{\mathrm{i}}\end{array}\right ] , \label{a3}\ ] ] and @xmath78 @xmath114 , \quad a_{4b } = \left[\begin{array}{cccc } 0 & 1 & 1 & 1 \\ 0 & 1 & 1 & 1 \\ 0 & 0 & { \mathrm{i } } & 1 \\ 0 & 0 & 0 & 1+{\mathrm{i}}\end{array}\right ] . \label{a4}\ ] ] these probability distributions can be thus interpreted as shadows of real projective spaces @xmath115 , @xmath116 and @xmath117 on a plane . in the case [ fig : real - s3a3a ] and [ fig : real - s3a4a ] the shadow is supported on a real line , so we plot the corresponding probability distribution @xmath118 . in the other cases [ fig : real - s3a2 ] , [ fig : real - s3a3b ] and [ fig : real - s3a4b ] the real shadow is supported on the complex plane , so the density is encoded in the grey scale . the circle drawn by red dotted line represents the image of the sphere of dimension @xmath57 , in which the set of quantum states can be inscribed . the blue dotted line represents the boundary of the standard numerical range of an operator . the restricted shadow is supported in this set or its subset . + the numerical shadow carries also some information about the higher rank numerical range @xcite of an operator . for instance , the darker area of the shadow corresponding to a larger probability @xcite allows one to recognize in fig . [ fig : stars ] the _ numerical range of rank @xmath73 _ , written @xmath119 of selected unitary matrices . in the case of @xmath78 shown in fig . [ fig : stars ] it is equal to a single point at which the two diagonals of the quadrangle cross , while for the case @xmath120 shown in panel the numerical range of rank two is represented by the inner pentagon located inside the numerical range in this case numerical shadow appears to be uniform in the set @xmath121 . in the case @xmath122 shown in the panel one can see darker areas between segments connecting every other eigenvalue . given @xmath123 , the real shadow is taken to mean the distribution of @xmath124 when @xmath125 is a unit vector randomly chosen from the uniform distribution on @xmath126 . if @xmath127 itself is real , we have @xmath128 so that the distribution of @xmath124 is the same as that of the symmetric @xmath129 . since the uniform distribution on @xmath130 is invariant under orthogonal transformations , we are free to diagonalize this symmetric matrix . in other words , the real shadow of @xmath131 is that of @xmath132 , where @xmath133 are the ( real ) eigenvalues of @xmath129 . thus we need the distribution of @xmath134 it is known that @xmath98 uniform on @xmath130 implies that @xmath135 has a dirichlet(1/2 ) distribution on the simplex @xmath136^n:\sum_1^n p_k=1\}$ ] , _ ie . _ its density is proportional to @xmath137 . the distribution of ( [ eqn : real - john - e1 ] ) is known in general @xcite , so for @xmath138 we have @xmath139 } { u\prod_{j=1}^n[1+(a_j - t)^2u^2]^\frac14}\,du.\ ] ] as an example , let us consider the jordan nilpotent @xmath140 for the real shadow of @xmath141 , we first note that it is the same as that of @xmath142 . rescaling for convenience , we compute the real shadow of @xmath143 . the density @xmath144 of @xmath145 is 0 outside @xmath146 $ ] , satisfies @xmath147 , and , in view of the dirichlet(1/2 ) distribution on @xmath148 , @xmath144 is proportional to @xmath149 for @xmath150 . in particular , @xmath151 . in @xcite the case @xmath152 is treated in terms of hypergeometric functions and we can compute @xmath153 for @xmath154 . for a general @xmath123 , one may write @xmath155 with @xmath156 , and by diagonalizing @xmath157 and @xmath158 we may understand the distributions of @xmath159 and @xmath160 separately , using the techniques above . usually the _ joint _ distribution will be obscure , but we may have some hope of understanding it in certain cases . if , for example , @xmath127 is normal and presented in diagonal form , it may be possible to see how the horizontal and vertical distributions knit together . + fig . [ fig : john - example ] shows various shadows of @xmath127 when @xmath127 is a @xmath161 unitary ( in diagonal form ) . the sort of 1d shadow density seen in the left panel of fig . [ fig : complex_and_real_d3 ] seems typical ; notably , the real density of a @xmath161 hermitian is constant between the two middle eigenvalues . presumably , the vertical density has a similar form . [ fig : unitary5real ] presents two examples of real shadow of generic unitary matrices of dimension five . support of the shadow does not coincide with numerical range of those matrices because a generic unitary matrix can not be diagonalized using only orthogonal matrices . since methods based on moments of the shadow distribution were so effective in the complex case , we may try to mimic them for the real shadow . it is not hard to adapt the method outlined at the beginning of the proof of proposition 5.1 from @xcite to show that , for @xmath162 @xmath163 unless each component of @xmath164 is even , and that @xmath165 here we use the shifted factorial notation : @xmath166 , with the convention @xmath167 ; also @xmath168 denotes @xmath169 . in principle , we can use eq . ( [ eqn : real - john - e5 ] ) to evaluate the moments of the real shadow density @xmath170 of @xmath132 @xmath171 in the complex case we were able to progress beyond this point , obtaining such effective relations for the moments as the _ determinant relation _ ( see : eq . ( 10 ) in @xcite ) . as a test case , we may try to use eq . ( [ eqn : real - john - e6 ] ) to find the moments of @xmath144 , the real shadow density of @xmath143 , discussed in the previous example . we have @xmath172 and finally we obtain that @xmath173 it can be shown , that for even @xmath174 we have @xmath175 to see this we start with equation eq . ( [ eqn : real - john - e7 ] ) . changing the index of summation from @xmath176 to @xmath177 shows that the sum equals @xmath178 times itself , hence equals zero for odd @xmath179 . now suppose that @xmath180 . the sum can be written in hypergeometric form@xmath181 kummer s summation formula ( see @xcite ) implies ( later we set @xmath182 ) @xmath183 now we set @xmath184 and use the terminating form of gauss s sum ( the chu - vandermonde formula ) to obtain@xmath185 thus , setting @xmath186 we obtain , that the integral equals@xmath187 we use the variance of a complex random variable to give quantitative insight into figs . [ fig : stars]-[fig : unitary5real ] . the calculations leading to the following formulae are based on the values of integrals of monomials over the real or complex unit spheres . suppose @xmath1 is a normal matrix , written in the form @xmath188 where @xmath189 is unitary and @xmath2 is diagonal with entries @xmath190 . thus the columns of @xmath189 are eigenvectors of @xmath1 . let @xmath90 denote the random variable @xmath191 and let @xmath192 . if @xmath193 is a random vector in the unit sphere in @xmath194 or @xmath195 then @xmath196 ( where @xmath197 denotes the expectation ) . denote the _ complex variance _ @xmath198 , a measure of ( 2-dimensional ) spread of @xmath90 . for the complex shadow one finds @xmath199 for the real shadow and when @xmath1 can be diagonalized by a real orthogonal matrix , that is , @xmath189 is orthogonal ( @xmath200 ) then @xmath201 which is larger than the variance of the complex shadow ( see fig . [ fig : complex_and_real_vphi3 ] ) . in the general real case @xmath189 is not orthogonal , then set @xmath202 for @xmath203 . thus @xmath204 is a symmetric unistochastic matrix , and its eigenvalues all lie in @xmath205 $ ] . by straightforward computations we find@xmath206 a positive quadratic form in @xmath207 . the eigenvalues of @xmath208 lie in @xmath209 $ ] , and all equal @xmath73 when @xmath189 is real orthogonal . as example consider the unitary fourier matrix whose entries are primitive @xmath210 roots of unity : @xmath211 . if @xmath213 otherwise @xmath214 . the eigenvalues of @xmath208 are @xmath73 with multiplicity @xmath215 and @xmath216 with multiplicity @xmath217 . the matrix @xmath204 also contains information about approximating the eigenvectors of @xmath1 by points on @xmath130 : indeed @xmath218 . in particular if some @xmath219 is an extreme point in the convex hull of @xmath220 then @xmath219 is in the real numerical range of @xmath1 if and only if @xmath221 . this is illustrated by fig . [ fig : unitary5real ] . consider the shadow restricted to the set of pure product states . more formally , we assume that @xmath222 , apply the definition ( [ rshadow ] ) and take @xmath223 , where @xmath224 , while @xmath225 and these states are normalized . the set @xmath226 of separable ( product ) pure states has the structure of the cartesian product @xmath227 . the simplest case of @xmath228 corresponds to the two qubit case . the set of separable ( product ) pure states has then a form of the cartesian product of two spheres @xmath229 . in other words this set forms a _ segre embedding _ , @xmath230 . one may also consider the shadow with respect to real separable states . in the two - qubit case this shadow corresponds to a projection of the product of real projective spaces @xmath231 , which forms a torus @xmath232 . such a structure can be recognized on some plots shown in fig . [ fig : shadsep ] . the matrices used to obtain projections are following @xmath233 , \quad & b_{4b } = \left [ \begin{array}{cccc } 1 & 0 & 0 & 1 \\ 0 & { \mathrm{i } } & 0 & 0 \\ 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & -{\mathrm{i}}\end{array } \right ] , \quad\\ b_{4c } = \left [ \begin{array}{cccc } 1 & 0 & 0 & 0 \\ 0 & { \mathrm{i } } & 0 & 1 \\ 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & -{\mathrm{i}}\end{array } \right ] & b_{4d } = \left [ \begin{array}{cccc } 1 & 0 & 0 & 1 \\ 0 & { \mathrm{i } } & 1 & 0 \\ 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & -{\mathrm{i}}\end{array } \right ] , \quad\\ b_{4e } = \left [ \begin{array}{cccc } 1 & 1 & 1 & 1 \\ 0 & { \mathrm{i } } & 1 & 1 \\ 0 & 0 & -1 & 1 \\ 0 & 0 & 0 & -{\mathrm{i}}\end{array } \right ] , \quad & b_{4f } = \left[\begin{array}{cc } { \mathrm{i}}&\frac{1}{2}\\ 0&\frac{1}{2}{\mathrm{i}}\end{array}\right]\otimes { \mathds{1}}+ { \mathds{1}}\otimes \left[\begin{array}{cc } 0&2\\ 1&{\mathrm{i}}\end{array } \right].\end{aligned}\ ] ] one can note that the numerical ranges presented in fig . [ fig : shadsep ] have a particular structure . matrix @xmath234 is permutation equivalent to a simple sum of two matrices so its numerical range forms a convex hull of two ellipses one with focal points @xmath235 and the other @xmath236 . as these ellipses do intersect their convex hull contains four interval segments . a similar situation occurs for the matrix @xmath237 where two ellipses one with focal points @xmath238 and the other @xmath239 do not intersect and the convex hull has only two flat lines . for the matrix @xmath240 the numerical range is a convex hull of an ellipse with focal points at eigenvalues @xmath238 and the line segment between two eigenvalues @xmath239 . a similar situation arises for the matrix @xmath241 . consider a particular case of an operator @xmath90 with the tensor product structure , @xmath242 . then its product numerical range is equal to the minkowski product of numerical ranges @xmath243 - for more information see @xcite . is in this case the shadow of @xmath244 restricted to product states can be expressed by the numerical shadows of both operators , @xmath245 here @xmath31 and @xmath246 denote the probability measures related to the numerical shadows of @xmath1 and @xmath247 respectively , while @xmath248 is a measurable subset of the complex plane . in the case of separable shadow of matrix @xmath249 it is possible to obtain explicit expressions for the mean and the variance . we have @xmath250 and @xmath251 formulas involve the partial traces @xmath252 and @xmath253 , follows from more general fact given in appendix [ appen:2 ] . these expressions imply directly the following result . @xmath254 consider a diagonal matrix @xmath90 defined on a composite hilbert space , @xmath255 of dimension @xmath256 . its diagonal elements forming the spectrum @xmath106 can be also represented by two indices , @xmath257 with @xmath258 and @xmath259 . let @xmath260 be an arbitrary fixed pure product state in @xmath21 , so the set of random separable pure states can be obtained as @xmath261 and @xmath262 are independent random unitary matrices distributed according to the haar measure . thus the expansion coefficients of the product state @xmath7 read @xmath263 . the _ separable numerical shadow _ of the diagonal operator @xmath90 is defined as the density distribution of random numbers @xmath91 , where @xmath7 is a separable random state defined by random unitaries @xmath92 and @xmath189 . in this case one has @xmath264 where @xmath265 is a real probability vector of size @xmath256 . it can be considered as a tensor product of two probability vectors @xmath266 and @xmath267 , since its components read @xmath268 with with @xmath269 and @xmath259 . thus the separable numerical shadow of a diagonal operator can be considered as a projection of the _ cartesian product _ of classical probability simplices , @xmath270 . in the simplest case of @xmath271 the cartesian product of two intervals ( 1-simplices ) forms a square , which lives inside the tetrahedron of the @xmath272dimensional probability vectors . as in the previous case we can distinguish two probability measures in the space of unitary matrices . they lead to * _ complex separable shadow _ , generated by the haar measure on @xmath273 and @xmath274 , for which both probability vectors @xmath64 and @xmath99 are distributed uniformly with respect to the lebesgue measure on the simplices @xmath275 and @xmath276 , respectively . * _ real separable shadow _ , generated by the haar measure on the orthogonal groups @xmath103 and @xmath277 , which lead to the statistical measure ( dirichlet measure with @xmath105 ) in both simplices . note that in these case the separable shadow of @xmath90 is supported on its _ product numerical range _ @xcite , which in general forms a proper subset of the convex hull of the spectrum . the product structure of the classical probability vector @xmath265 in ( [ prod1 ] ) , generalized for a multiple tensor product structure , is consistent with the parametrization of the product numerical range described in prop . 12 in @xcite . consider an operator @xmath90 acting on a hilbert space with a tensor product structure , @xmath278 . for simplicity let us assume that the dimensions of both subspaces are equal to @xmath279 so the total dimension reads @xmath280 . among all pure states of the @xmath281 system one distinguishes the set @xmath17 of maximally entangled states . it contains the states equivalent with respect to a local unitary operation @xmath282 to the generalized bell state , @xmath283 , where @xmath284 is the discrete permutation group @xcite , ( * ? ? ? choosing @xmath285 for the set @xmath3 in ( [ rshadow ] ) we define the shadow @xmath286 of an operator @xmath90 with respect to the maximally entangled states . the corresponding probability measure will be denoted as @xmath287 . in the simplest case of @xmath289 hilbert space the set @xmath17 of maximally entangled states has the structure @xmath290 see @xcite . hence the numerical shadow of an operator @xmath1 of order four with respect to the complex maximally entangled states can be considered as a projection of the real projective space on the plane see the shadow for some illustrative operators presented in fig . [ fig : shadent ] . if one considers a further restriction and studies the shadow with respect to _ real _ maximally entangled states , the result can be interpreted as an image of the space @xmath291 observe that the illustrative shadows obtained in this case and presented in fig . [ fig : shadent ] show indeed projections of a circle onto the complex plane . in the special case of a diagonal operator @xmath247 of size four its shadow with respect to complex maximally entangled states can be identified with a standard shadow of a reduced operator @xmath292 of size @xmath73 . this fact is formulated in the following proposition , proved in appendix [ appen:1 ] . [ prop:2x2ent ] consider an diagonal matrix of order four , @xmath293 which acts on a composite hilbert space @xmath278 and the reduced matrix @xmath294 . then the numerical shadow @xmath286 of @xmath90 with respect to complex maximally entangled states @xmath17 is equal to the standard numerical shadow @xmath295 of the reduced matrix @xmath296 of order two . in this general case the set @xmath17 of maximally entangled states forms a manifold of @xmath298 real dimensions with the structure of @xmath20 @xcite . in the case of the shadow with respect to real maximally entangled states the space under consideration reads @xmath299 . to analyse the entangled shadow of a diagonal matrix we can perform initial steps used to prove proposition [ prop:2x2ent ] . it will be convenient to use a slightly different notation and consider a diagonal matrix @xmath90 of size @xmath280 acting on a tensor product space @xmath278 with entries @xmath300 where @xmath301 . consider a local unitary matrix @xmath302 where @xmath303 . the entangled unit state @xmath304 has entries @xmath305 where @xmath306 . for purposes of numerical shadow the integration with respect to the haar measure over both matrices @xmath307 and @xmath308 can be replaced by a single integration over the @xmath279 dimensional random unitary matrix @xmath92 . observe @xmath309 is a real automorphism of @xmath310 invariant for the haar measure . taking a diagonal matrix @xmath311 we find its expectation value for random pure state @xmath304 @xmath312 here we reshape the diagonal matrix @xmath90 of order @xmath313 to get a matrix @xmath314 of order @xmath279 with entries @xmath315 , while @xmath247 stands for a unistochastic matrix , @xmath316 for @xmath317 . the case @xmath318 studied above relied on the simple nature of the set of unistochastic matrices of order two , equivalent to an interval . it is known that the structure of unistochastic matrices for @xmath319 is complicated and interesting @xcite . thus we are not able to formulate a direct generalization of proposition [ prop:2x2ent ] for the @xmath281 problem . for the case of @xmath320 matrices consider the function@xmath321 where the ( variable ) unistochastic matrix reads@xmath322{ccc}b_{1 } & b_{2 } & 1-b_{1}-b_{2}\\ b_{3 } & b_{4 } & 1-b_{3}-b_{4}\\ 1-b_{1}-b_{3 } & 1-b_{2}-b_{4 } & b_{1}+b_{2}+b_{3}+b_{4}-1 \end{array } \right ] .\ ] ] given the matrix @xmath314 let @xmath323 , @xmath324 for @xmath325 , and @xmath326 . for simplification we set@xmath327 one can observe , that the row and column sums of @xmath328 $ ] are zero and @xmath329{cc}4b_{1}+2b_{2}+2b_{3}+b_{4}-3 & 2b_{1}+4b_{2}+b_{3}+2b_{4}-3\\ 2b_{1}+b_{2}+4b_{3}+2b_{4}-3 & b_{1}+2b_{2}+2b_{3}+4b_{4}-3 \end{array } \right ] \left [ \begin{array } [ c]{cc}\gamma_{11 } & \gamma_{12}\\ \gamma_{21 } & \gamma_{22}\end{array } \right ] + \frac{1}{9}c_{\cdot\cdot } \ .\end{aligned}\ ] ] without loss of generality we can assume @xmath330 . powers of @xmath331 can be integrated using the formula from ( * ? ? ? 3.3):@xmath332 where @xmath333 denotes the haar measure on @xmath334 . for example@xmath335 there is an interesting special case when one of the variables @xmath336 , say @xmath337 , does not appear in @xmath331 . the triple @xmath338 is a point in the pyramid with square base @xmath339 and vertex @xmath340 . the induced measure ( from @xmath341 ) on the pyramid is @xmath342 . the numerical range of @xmath1 is an affine image of the pyramid , hence is the convex hull of the images of the vertices of the pyramid , @xmath343 , that is , a convex polygon . for three arbitrary complex numbers @xmath344 let@xmath345{cc}\gamma_{11 } & \gamma_{12}\\ \gamma_{21 } & \gamma_{22}\end{array } \right ] = \left [ \begin{array } [ c]{cc}-6z_{1}-6z_{2 } & 3z_{1}+3z_{3}\\ 3z_{2}+3z_{3 } & -3z_{3}\end{array } \right ] , \ ] ] then @xmath346 . the vertices of the pyramid are mapped to @xmath347 . it is possible that these points form a ( not regular ) pentagon . for example let @xmath348 then the range is the convex hull of @xmath349 . it may not be easy to find the shadow measure explicitly , but one expects a higher density in the neighbourhood of @xmath350 , the image of @xmath351 . fix @xmath352 and consider the set @xmath353 . the normalized volume of @xmath354 is @xmath355 and the @xmath356-measure of @xmath357 is @xmath358 , so the relative density is @xmath359 . it is possible to get explicit expressions for the mean and the variance of the entangled shadow @xmath360 of matrix @xmath90 acting on @xmath281 hilbert space @xmath278 . the following results @xmath361 and @xmath362 which involve the partial traces @xmath252 and @xmath253 , are derived in appendix [ appen:2 ] from more general fact . these expressions imply directly the following result . the expected squared distance from the mean with respect to the entangled shadow @xmath360 reads @xmath363 let us apply these formulae in the special case of a diagonal matrix @xmath364 . in this case the necessary ingredients of ( [ distestshad ] ) simplify considerably , e.g. @xmath365 @xmath366 @xmath367 for the easy case @xmath318 we find @xmath368 , in agreement with the previous calculations . in previous sections we analysed the entire set of quantum states with its subsets and their projections onto a plane . in this section we specify a concrete quantum dynamics ( in general non unitary ) , choose an initial quantum state @xmath369 , and following @xcite we analyse its trajectory projected on the plane of a shadow of a selected non - hermitian operator @xmath1 . in particular we will be interested in dynamics of quantum entanglement , so the separable shadow of @xmath1 will be used as a background for the trajectory obtained from the expectation values , @xmath370 . investigation of the dynamics of quantum entanglement was initiated in @xcite , in which the evolution of certain measures of entanglement in time was studied for a model non unitary dynamics of a two qubit system and several qualitatively different scenarios of behaviour of entanglement in time were identified . in particular , revivals of entanglement in time and an effect of sudden decay of quantum entanglement was reported . the latter effect was later called _ entanglement sudden death _ by yu and eberly @xcite and the dynamics of entanglement was studied by several authors in various setups @xcite . + in general , all these dynamical effects can be explained in a simple geometric manner , if one takes into account the structure of the set of separable states analysed in @xcite . as the convex set @xmath371 of separable mixed states of a bipartite system occupies the central part of the set @xmath372 of all states of size @xmath2 , entanglement revivals occur if the unitary dynamics moves the initial state several times across the separability boundary . on the other hand , entanglement sudden death effect takes place , if the decoherence is so strong that the initially pure state gets mixed in such a pace that it crosses the separability boundary only once . these possible scenarios are shown on a schematic sketch , used in a conference talks for several years see fig . [ fig : dynentaglement - concept ] . making use of the technique of numerical shadow we are now in position to observe similar behaviour of entanglement for a concrete choice of quantum dynamics and initial states . consider the following discrete - time dynamic quantum process of a system consisting of two qubits @xcite . the system is assumed to be initially described by a maximally entangled pure state , @xmath373 . the discrete time evolution of the system is given by one step unitary evolution expressed by pauli matrices , @xmath374 , followed by an action of the depolarising channel acting locally on the second qubit , @xmath375 so that @xmath376 . as the initial state is chosen to be pure , the trajectory begins at the boundary of the set of mixed states . as time @xmath377 increases the trajectory plunges into the set of mixed states periodically crossing the set of entangled states . artists impression of such processes is depicted in fig . [ fig : dynentaglement - concept ] where the outer oval corresponds to the set of the pure states , its interior corresponds to the set of all mixed states , ellipse corresponds to the set of separable states and spirals depict the trajectory . let us fix the parameters of the discussed process , by setting the interaction strength @xmath378 and decoherence rate @xmath379 . we arbitrarily chose two matrices @xmath380 \text { and } x_2= \left [ \begin{array}{cccc } 1 & 0 & 0 & { \mathrm{i}}\\ 0 & -1 & 0 & 0 \\ 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & 1 \\ \end{array } \right],\ ] ] which allow us to project the trajectory @xmath381 onto the complex plane . we can calculate the images of the trajectory @xmath382 and superimpose them on the separable shadows of matrices @xmath383 . the resulting images are shown in fig . [ fig : dynentaglement ] panels and . the red crosses indicate entangled states and blue circles indicate separable states . it can be easily seen that by choosing an appropriate observation matrix it is convenient to observe the dynamics of entanglement in the process . in the general case for any given trajectory , it is hardly possible to find such an observation matrix whose product numerical range contains only images of separable states , but images of entangled states always lie outside of product numerical range . therefore product numerical range and separable shadow of a matrix are useful tools to visualize dynamics of entanglement , the effects of entanglement sudden death and entanglement revival . it is natural to ask about the properties of the numerical shadow in the case when one aims to study composite quantum systems consisting of more than two subsystems . let us consider the simplest case of a multipartite quantum system , _ i.e. _ a system composed of three qubits . in this case @xmath384 . as an example we take an unitary matrix @xmath385 of size eight written in the standard computational basis @xmath386 @xmath387 + the product numerical range of this operator is not simply connected @xcite , so it is instructive to study the shadow of @xmath385 with respect to the space of ) all pure states ( standard shadow ) ; product states ( product shadow ) , @xmath388 ; ghz entangled states , @xmath389 ; so - called @xmath12entangled states , @xmath390 . here @xmath391 are independent random unitary matrices taken from @xmath392 with respect to the haar measure . [ fig : thomas - nonsimply - example ] presents the shadows generated by matrix @xmath385 with respect to those classes of states . observe that a generic operator acting on the three - qubit system leads to different shadows , if they are taken with respect to @xmath11 states and the @xmath12 states . this is a consequence of the different topology of the orbits with respect to local unitary transformations produced by these two classes of entangled states . study of numerical shadows restricted to certain classes of entangled states can contribute to a better understanding of the geometry of the manifold of locally equivalent states . investigations of the restricted shadows of operators acting on multipartite systems lead to a wide class of interesting problems which in general are difficult to solve . however , some results obtained in the previous sections for the bipartite setup can be generalized for multipartite systems . consider an operator acting on the composite hilbert space describing @xmath393partite system of dimensions @xmath394 , respectively . its shadow with respect to separable states @xmath395 , induced by the dirichlet measures on @xmath393 simplices of the dimension @xmath396 with @xmath397 . in this work we propose to combine the notion of restricted numerical range with the numerical shadow of an operator which is a probability measure on the complex plane . on one hand the numerical shadow can be investigated for a given operator @xmath90 . on the other hand , one may analyse the shadows of all operators of a fixed dimension @xmath2 , which can be considered as projections of the set of all pure states of size @xmath2 onto a plane . in a similar way it could be interesting to study the shadow of a given operator @xmath90 with respect to various sets of pure states . for instance , in this work , we analysed the standard shadow with respect to complex states and the shadow restricted to real states . for operators acting on composite hilbert spaces one can additionally study the shadow with respect to separable or maximally entangled states , complex or real . note that in general the restricted numerical range is not convex @xcite , which implies that the restricted numerical shadow can be supported on non convex sets . following the complementary strategy one may take the set of all operators of a given dimension and analyse their shadows restricted to a certain class of states . these probability distributions on the complex plane convey some information about the structure of these specific subsets of the set of all quantum states . consider for instance the simplest hilbert space with a tensor product structure @xmath398 . then the standard shadow carries information about the set of all pure states @xmath399 which forms the complex projective space @xmath400 , while the shadow restricted to real states corresponds to the real projective space @xmath116 . in a similar way , the shadows of matrices of order four with respect to separable states illustrate the projections of the product of two spheres @xmath401 , while the real separable shadow corresponds to projections of a torus @xmath402 . the shadow with respect to maximally entangled states ( also called briefly the _ entangled shadow _ ) represents the set @xmath403 also equivalent to real projective space @xmath116 . in the case of real entangled shadow we observe projections of the set @xmath404 equivalent to the circle @xmath405 . the notion of separable shadow is useful to analyse the dynamics of quantum entanglement . for a given initial quantum state @xmath369 and a certain dynamics we select a non - hermitian matrix @xmath1 and study trajectories on the complex plane formed by the time evolution of its expectation value , @xmath406 . investigations of such a trajectory superimposed on the separable shadow of @xmath1 contribute to our understanding of the dynamics of quantum entanglement and allow us to visualize the effects of entanglement sudden death and entanglement revivals . the notion of restricted numerical range is easily formulated for operators acting on hilbert space with multiple tensor product structure @xcite , which correspond to multipartite quantum systems . therefore it is natural to define restricted numerical shadow for various classes of quantum states of multipartite systems . for instance , in the simplest case of a three - qubit system , described in the hilbert space @xmath407 one may distinguish two classes of maximally entangled states called @xmath11 and @xmath12 , which can not be locally converted in any direction @xcite . studying numerical shadows of matrices of order @xmath39 , restricted to @xmath11 states or @xmath12 states , we are thus in position to investigate the differences between the structure of these two important classes of three - qubit quantum entangled states . in conclusion , we have introduced the notion of the restricted numerical shadow of an operator and established its basic properties . on one hand we believe that this topic is interesting from the mathematical point of view , as it relates operator theory and probability . moreover , we are tempted to expect that further investigations of the restricted numerical shadow will contribute to a better understanding of the geometry of quantum entanglement , so that they become directly applicable to the theory of quantum information . it is a pleasure to thank e. gutkin for fruitful discussions . work by j. holbrook was supported in part by an nserc of canada research grant . work by p. gawron was supported by the polish national science centre under the grant number n n516 481840 , z. puchaa was supported by polish national science centre under the research project n n514 513340 , j.a . miszczak was supported by polish ministry of science and higher education under the research project ip2011 036371 , while k. yczkowski acknowledges support by the polish ministry of science and higher education grant number n202 090239 . numerical calculations presented in this work were performed on the ` leming ` server of the institute of theoretical and applied informatics , polish academy of sciences . to analyse properties of the shadow with respect to complex maximally entangled states for an operator acting on the @xmath289 hilbert space let us analyse the structure of a local unitary transformation @xmath302 acting on @xmath278 . consider two generic elements of @xmath408 @xmath409{cc}e^{{\mathrm{i}}\left ( \psi_{1}+\phi_{1}\right ) } \cos\theta_{1 } & -e^{{\mathrm{i}}\left ( \psi _ { 1}-\phi_{2}\right ) } \sin\theta_{1}\\ e^{{\mathrm{i}}\left ( \psi_{1}+\phi_{2}\right ) } \sin\theta_{1 } & e^{{\mathrm{i}}\left ( \psi _ { 1}-\phi_{1}\right ) } \cos\theta_{1}\end{array } \right ] , \ \ \ u^{\left ( b\right ) } = \left [ \begin{array } [ c]{cc}e^{{\mathrm{i}}\left ( \psi_{2}+\phi_{3}\right ) } \cos\theta_{2 } & -e^{{\mathrm{i}}\left ( \psi _ { 2}-\phi_{4}\right ) } \sin\theta_{2}\\ e^{{\mathrm{i}}\left ( \psi_{2}+\phi_{4}\right ) } \sin\theta_{2 } & e^{{\mathrm{i}}\left ( \psi _ { 2}-\phi_{3}\right ) } \cos\theta_{2}\end{array } \right ] , \end{aligned}\ ] ] where @xmath410 and the haar measure ( for @xmath411 ) is@xmath412 taking tensor products of corresponding columns of @xmath411 and @xmath413 and adding them we obtain a parametrization of an entangled state , @xmath414{c}u_{11}^{\left ( a\right ) } u_{11}^{\left ( b\right ) } + u_{12}^{\left ( a\right ) } u_{12}^{\left ( b\right ) } \\ u_{21}^{\left ( 1\right ) } u_{11}^{\left ( 2\right ) } + u_{22}^{\left ( a\right ) } u_{12}^{\left ( b\right ) } \\ u_{11}^{\left ( 1\right ) } u_{21}^{\left ( 2\right ) } + u_{12}^{\left ( a\right ) } u_{22}^{\left ( b\right ) } \\ u_{21}^{\left ( 1\right ) } u_{21}^{\left ( 2\right ) } + u_{22}^{\left ( a\right ) } u_{22}^{\left ( b\right ) } \end{array } \right].\ ] ] consider an diagonal matrix of order four , @xmath293 so that its expectation value for a maximally entangled state reads @xmath415 observe that @xmath416 . compare this situation to the complex shadow of the reduced operator @xmath417 @xmath418{cc}\frac{1}{2}\left ( d_{1}+d_{4}\right ) & 0\\ 0 & \frac{1}{2}\left ( d_{2}+d_{3}\right ) \end{array } \right ] , \\ x^{\dag}y x & = \frac{1}{2}\left ( d_{1}+d_{4}\right ) \left\vert x_{1}\right\vert ^{2}+\frac{1}{2}\left ( d_{2}+d_{3}\right ) \left\vert x_{2}\right\vert ^{2},\end{aligned}\ ] ] where @xmath98 belongs to the unit sphere @xmath419 in @xmath420 . we know that @xmath421 to identify the two shadows it only remains to show@xmath422 in fact it suffices to show @xmath423 . the integration over the angles @xmath424 can be combined into one @xmath425 , due to the rotational invariance . write @xmath426 now integrate with @xmath427 over @xmath428 , then over the @xmath429 part:@xmath430 to complete the proof@xmath431 where @xmath432 is pochhammer symbol . this used the chu - vandermonde sum ( terminating @xmath433 ) and @xmath434 . consider a general matrix @xmath90 acting on a @xmath281 composite hilbert space @xmath278 with entries written in a four index notation @xmath435 , where @xmath436 . upper pair of indices determines the row of the matrix , while the lower pair determines its column . we use the standard operations on matrices , which in this notation read @xmath437 and introduce reduced matrices of size @xmath279 obtained by a partial trace over a single subsystem , @xmath438 by the formula in collins and niady @xcite we have @xmath439 where @xmath333 denotes the haar measure on @xmath440 . let @xmath441 where @xmath442 are stochastically independent random unitary matrices of size @xmath279 , distributed with haar measure and @xmath443 is an arbitrary vector . we are interested in obtaining mean and variance of @xmath6 , thus we will calculate @xmath444 and @xmath445 . we have @xmath446 and @xmath447 to obtain the mean we calculate @xmath448 where we have used a convention , that we split indices @xmath449 , and @xmath450 has values from @xmath451 . from the above , we have that for any normalized @xmath443 @xmath452 let us now we calculate the second moment . let @xmath453 we have @xmath454 using formula ( [ eqn : int - formula-2-idx ] ) for the integral we have that @xmath455 , where @xmath456 similarly @xmath457 next we have @xmath458 finally @xmath459 the integrals of the above type , for fixed dimension , can be calculated with the use of computer algebra program ` intu ` @xcite . let us consider any pure state @xmath443 from @xmath281 composite hilbert space @xmath278 with fixed schmidt numbers @xmath460 , thus we have @xmath461 for some orthogonal bases @xmath462 . for simplicity we can take computational bases . we have the following identities @xmath463 thus we have , @xmath464 note that , the above depends only on a purity of reduced state . in the case of maximally entangled state ( @xmath465 for @xmath466 ) the second moment is given by @xmath467 the above implies the result ( [ distestshad ] ) for the variance with respect to the entangled shadow . d. c. brody , a. c. t. gustavsson , l. p. hughston , entanglement of three - qubit geometry , j. physics : conf . * 67 * , 012044 ( 2007 ) . a. sawicki , a. huckleberry , m. ku , symplectic geometry of entanglement , _ commun . phys . _ * 305*,441 - 468 ( 2011 ) v. vedral , _ introduction to quantum information science _ , oxford university press , oxford 2007 t. schulte - herbrggen , g. dirr , u. helmke , and s. j. glaser . the significance of the @xmath471-numerical range and the local _ c_-numerical range in quantum control and quantum information , _ linear and multilinear algebra _ * 56 * , 3 - 26 ( 2008 ) p. gawron , z. puchaa , j. a. miszczak , . skowronek , m .- d . choi , and k. yczkowski , restricted numerical range : a versatile tool in the theory of quantum information , _ j. math . phys . _ * 51 * 102204 ( 24pp ) ( 2010 ) . g. dirr , u. helmke , m. kleinsteuber , and t. schulte - herbrggen . c_-numerical ranges for applications in quantum control and quantum information . _ linear and multilinear algebra _ * 56 * , 27 - 51 ( 2008 ) . dunkl , p. gawron , j.a . holbrook , j. miszczak , z. puchaa and k. yczkowski , _ numerical shadow and geometry of quantum states _ , j. phys . a : math . theor . 33 ( 2011 ) , p. 335301 _ preprint _

we present results on the star - formation rate ( sfr ) versus stellar mass ( ) relation ( i.e. , the `` main sequence '' ) among star - forming galaxies at 1.37.61 using the mosfire deep evolution field ( mosdef ) survey . based on a sample of 261 galaxies with h and h spectroscopy , we have estimated robust dust - corrected instantaneous sfrs over a large range in ( ) . we find a correlation between log(sfr(h ) ) and log( ) with a slope of 0.65.08 ( 0.58.10 ) at 1.4.6 ( 2.1.6 ) . we find that different assumptions for the dust correction , such as using the color - excess of the stellar continuum to correct the nebular lines , sample selection biases against red star - forming galaxies , and not accounting for balmer absorption can yield steeper slopes of the log(sfr)-log( ) relation . our sample is immune from these biases as it is rest - frame optically selected , h and h are corrected for balmer absorption , and the h luminosity is dust - corrected using the nebular color - excess computed from the balmer decrement . the scatter of the log(sfr(h))-log( ) relation , after accounting for the measurement uncertainties , is 0.31dex at 2.1.6 , which is 0.05dex larger than the scatter in log(sfr(uv))-log( ) . based on comparisons to a simulated sfr- relation with some intrinsic scatter , we argue that in the absence of direct measurements of galaxy - to - galaxy variations in the attenuation / extinction curves and the imf , one can not use the difference in the scatter of the sfr(h)- and sfr(uv)- relations to constrain the stochasticity of star formation in high - redshift galaxies .

1507.03017

star - forming galaxies show a relatively tight correlation between their star - formation rate ( sfr ) and stellar mass . this relation , commonly known as the star - forming _ `` main sequence '' _ @xcite , has been intensively studied over the past decade ( e.g. , at @xmath8 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . the main sequence is generally assumed to be a linear relation between log(sfr ) and log(@xmath0 ) with an associated scatter due to observational uncertainties and intrinsic scatter . the slope of the relation tells us about star formation efficiency as a function of stellar mass @xcite . the intrinsic scatter of the log(sfr)-log(@xmath0 ) relation is predicted to be due to variations in the gas accretion histories of different galaxies @xcite and indicates the level of burstiness in star formation history . the common physical interpretation of the tightness of the sfr-@xmath0 relation is that the bulk of the stellar mass in star - forming galaxies is built in a relatively steady process , as opposed to a rapid starburst mode , as occurs in major mergers . the merger - driven starburst galaxies with high specific sfrs tend to lie above the main sequence relation @xcite and the quiescent galaxies that have little ongoing star formation , identified by their red near - infrared colors @xcite , populate a region below the sequence @xcite . it has been argued that the majority of galaxies spend most of their lifetime on the main sequence @xcite . therefore , studying the parameters of the main sequence and comparing them with the galaxy formation models can shed light on our understanding of processes that govern galaxy evolution , such as stellar and agn feedback @xcite , gas accretion rates @xcite , and gas inflows and outflows @xcite . the sfr-@xmath0 relation can be used as a test for galaxy evolution simulations to see how well the simulations can reproduce the observed universe ( e.g. , * ? ? ? determining the slope and intrinsic scatter of the log(sfr)-log(@xmath0 ) relation can be hindered by sample selection effects , the diagnostic used to infer the sfr , and the method used to correct observed sfr for dust attenuation ( for a summary see , * ? ? ? the uv - selected samples and the uv - inferred sfrs are biased against massive and dusty galaxies where the bulk of star formation is obscured and the uv slope is decoupled from extinction . moreover , as sed- and uv - based sfrs are dependent on the same stellar population models used to infer the stellar masses , the scatter in the log(sfr)-log(@xmath0 ) relation may be underestimated when using sed or uv based sfr tracers @xcite . in order to overcome this issue , and more robustly measure sfrs for dustier galaxies , some studies have used sfr(ir)+sfr(uv ) as a proxy for bolometric sfr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? however , estimating the sfr(ir ) based solely on _ spitzer_/mips 24@xmath9 m imaging has its shortcomings and could artificially tighten the main sequence if the scatter in l(ir ) at a given l(@xmath10 m ) luminosity is not taken into account ( e.g. , * ? ? ? * ; * ? ? ? we improve upon previous high - redshift studies by utilizing a representative sample of galaxies with measurements of the h@xmath3 and h@xmath4 emission lines . these lines offer the advantage of probing star formation on shorter timescales than the uv and ir , they are largely independent of the stellar population , and thus can be used to more accurately assess the intrinsic scatter in the log(sfr)-log(@xmath0 ) relation . the star - formation activity was at its peak at @xmath8 @xcite , making that redshift a critical epoch to study the evolution of galaxies . obtaining rest - frame optical spectra of large samples of galaxies at @xmath8 has been challenging until recently due to the high terrestrial background in the near - ir and the lack of multi - object near - ir spectrographs on 10-meter - class telescopes . the newly - commissioned mosfire spectrograph @xcite on the 10 m keck i telescope has enabled us to build a large sample of galaxies at @xmath8 with coverage in both the h@xmath3 and h@xmath4 lines . in this study , we use a large sample of 261 rest - frame optically selected and spectroscopically confirmed galaxies at @xmath11 , which was obtained as part of the mosfire deep evolution field ( mosdef ) survey @xcite . to study the sfr-@xmath0 relation , we use sfrs based on the h@xmath3 luminosity that are robustly corrected for dust attenuation using the balmer decrement ( l(h@xmath3)/l(h@xmath4 ) ) . we demonstrate how different sfr diagnostics and dust correction recipes can alter the slope and scatter of the log(sfr)-log(@xmath0 ) relation , and thus potentially affect our conclusions regarding the way in which galaxies build up their stellar mass at high redshift . the outline of this paper is as follows . in section [ sec : sample ] , we present our sample properties and our measurements including sfrs and stellar masses . in section [ sec : ms_analysis ] , we describe the methodology we used to derive the log(sfr)-log(@xmath0 ) relation parameters . we compare the scatter of the log(sfr)-log(@xmath0 ) relation for different sfr indicators in section [ sec : scatter ] . in section [ sec : slope ] , we discuss the slope of the log(sfr)-log(@xmath0 ) relation and the potential biases arising from sample selection , dust correction , and balmer absorption correction . we compare our findings to previous studies in section [ sec : compare ] , and finally , the results are summarized in section [ sec : summary ] . throughout this paper , a @xcite initial mass function ( imf ) is assumed and a cosmology with h@xmath12 = 70 km s@xmath13 mpc@xmath13 , @xmath14 = 0.7 , and @xmath15 = 0.3 is adopted . the mosdef survey is a large multi - year survey with the mosfire multi - object spectrometer on the keck i telescope @xcite . the aim of the survey is to obtain rest - frame optical spectra of @xmath16 _ h_-selected galaxies to study their stellar , gaseous , metal , dust , and black hole content . the mosdef observations are conducted in five fields of the cosmic assembly near - ir deep extragalactic legacy survey ( candels ; * ? ? ? * ; * ? ? ? * ) , consisting of aegis , cosmos , goods - n , goods - s , and uds , targeting three redshift bins : @xmath17 , @xmath18 , and @xmath19 . targets for spectroscopy are prioritized by their _ h_-band magnitude ( using the 3d - hst catalogs , @xcite ) and the availability of spectroscopic , grism , and photometric redshifts , down to _ h _ = 24.0 , 24.5 , and 25.0 magnitude , for each redshift bin respectively . details of the survey strategy , observations , data reduction , and characteristics of the full galaxy sample are described in @xcite . in this study , we use the data accumulated from the first two years of the survey . out of the full mosdef sample we selected a subsample of 342 galaxies that have secure redshifts and coverage of both the h@xmath3 and h@xmath4 lines . these criteria limited our sample to the low and middle redshift bins that correspond to @xmath17 and @xmath18 , respectively . there were 49 agn that were identified and removed from the sample based on their ir and x - ray properties , as well as high [ nii]/h@xmath3 ratios ( [ nii]/h@xmath3@xmath20 , @xcite ) . in order to have secure sfr measurements , we further limited our sample to objects where the h@xmath3 line was detected with signal - to - noise ( s / n ) @xmath21 . this selection resulted in a sample of 264 objects . among the h@xmath3-detected galaxies , the h@xmath4 fluxes of 19% ( 48 out of 264 ) were below @xmath22 . out of these 48 h@xmath4-undetected objects , 40 ( 83% ) had a bright sky line close to the h@xmath4 line , while for the detected h@xmath4 sample the fraction was only 34% . the higher frequency of skyline contamination in the h@xmath4-undetected sample contributes to the lower h@xmath4 detection rate among these galaxies . to assess the relation between sfr and @xmath0 for star - forming galaxies , we proceeded to remove all those galaxies that were deemed to be `` quiescent '' . in particular , we used a rest - frame _ u__v _ versus _ v__j _ color selection @xcite to identify the quiescent galaxies . there was only one object in the h@xmath3-h@xmath4 detected sample and two in the h@xmath4 undetected sample that were identified as quiescent and thus removed from the sample , leaving a final sample of 261 galaxies , with 215 detected in both lines . the redshift distribution is shown in figure [ fig : zhist ] . detections and coverage of the h@xmath4 line . there are 88 objects at @xmath23 and 173 at @xmath24 . ] [ fig : zhist ] for each object , an initial redshift was estimated based on the emission line with the highest s / n ratio . the line fluxes were then measured by fitting gaussian functions to the line profiles . the h@xmath3 and [ nii]@xmath256548,6584 doublet were fit using three gaussian functions . to estimate the uncertainty in the line fluxes , the spectra were perturbed within the error spectra and all emission lines were re - fit . the dispersion among perturbed spectra was taken as the uncertainty in the line fluxes @xcite . an important aspect in determining the balmer emission line fluxes is to correct for the underlying absorption that is produced in the atmosphere of primarily a - type stars . the balmer absorption fluxes were determined from the best - fit sed models to the multi - wavelength photometry @xcite , and were added to the emission line fluxes to get the total , absorption - corrected fluxes . for objects with h@xmath3 ( h@xmath4 ) detected at @xmath26 level , ratio of the average h@xmath3 ( h@xmath4 ) balmer absorption flux to the average h@xmath3 ( h@xmath4 ) observed flux is @xmath27 . the uncertainty in the absorbed fluxes is @xmath28 @xcite , which is negligible compared to the typical uncertainty of our emission line fluxes @xmath29 , and we decided not to include it in our flux uncertainty estimations . the stellar masses were determined through a @xmath30 minimization sed fitting procedure assuming @xcite models for solar metallicity , a chabrier imf , the @xcite dust attenuation curve with @xmath31 , rising star - formation histories relation , as the scatter in the ratio of stellar masses inferred from a constant star - formation history modeling to the stellar masses derived assuming an exponentially rising star - formation history is @xmath32 , which is smaller than the typical relative error of the stellar masses ( @xmath33 , see section [ sec : method ] ) . ] and a 50myr lower limit on the age of the galaxies , assumed to be the typical dynamical time scale at this redshift ( see , * ? ? ? * ) . the sed - fitting was performed using photometry from the 3d - hst catalogs @xcite that has been corrected for the strongest emission lines measured in the mosdef survey , including [ oii]@xmath253727,3730 , [ oiii]@xmath254960,5008 , h@xmath4 , and h@xmath3 @xcite . errors in the sed parameters were estimated through monte carlo simulations ; each photometric flux was randomly perturbed assuming a gaussian distribution centered at the measured value with a standard deviation equal to the photometric error . this procedure was repeated many times and each photometric realization was refit to find the best - fit sed model . the standard deviation of the simulated sed parameters was taken as the error in the respective parameter . we converted the h@xmath3 luminosity to sfr using the @xcite relation modified for a @xcite imf , and corrected for dust attenuation using the nebular @xmath34 derived from the balmer decrement ( h@xmath3/h@xmath4 ) , assuming a galactic extinction curve taken from ( * ? ? ? * ccm ) . to determine the standard deviations in the corrected sfrs , we perturbed the measured h@xmath3 and h@xmath4 fluxes according to their uncertainties by assuming a gaussian distribution centered at the measured value with a standard deviation equal to the measurement error and recalculated the corrected sfrs based on each flux realization . we also accounted for spectroscopic slit losses by adding an additional uncertainty of 16% ( 20% ) to the h@xmath3 ( h@xmath4 ) measurement uncertainties in quadrature @xcite . where h@xmath4 is undetected , a @xmath22 upper limit on the h@xmath4 flux was used to estimate a lower limit on the nebular @xmath34 and the corresponding dust - corrected sfr . we used the ancillary 3d - hst broad- and medium - band photometry at rest - wavelengths 1268 2580 to find the uv luminosity at 1600 and the uv slope by fitting a power law to the photometry ( @xmath35 ) . the uv slope was used to correct the uv luminosity for dust attenuation , following the @xcite relation . we did not use the sed - inferred @xmath34 because it is correlated with the stellar mass as both quantities are determined through sed fitting . , the sfr(sed ) and the sfr(uv ) that is dust corrected by the @xmath34 inferred from the sed modeling show tighter and steeper sfr-@xmath0 relations compared to that of sfr(h@xmath3 ) and sfr(uv ) corrected by the uv slope , indicating a stronger correlation with @xmath0 . we should note that sfr(uv ) corrected with the uv slope is still not completely uncorrelated with @xmath0 as the uv slope translates to an @xmath34 . ] the luminosities were converted to sfrs using the @xcite relation , adjusted for a chabrier imf . to further investigate the sfr-@xmath0 relation and the role of the h@xmath4-undetected objects , we employed spectral stacks constructed in the following manner . individual spectra were converted from flux density to luminosity density , normalized by the h@xmath3 luminosity , and interpolated to a rest - frame grid with wavelength spacing of 0.5 . the composite spectrum was then created by taking the average of the individual spectra , weighted by inverse variance of the error spectra . the composite error spectrum was estimated as the square root of the variance of the weighted mean . the composite h@xmath3 and h@xmath4 lines were fit with a gaussian profile . the errors on the line fluxes are derived using the standard deviation of the distribution of line fluxes by perturbing the stacked spectrum according to the composite error spectrum and recalculating the line fluxes 1000 times . the composite line fluxes were corrected for balmer absorption using the mean absorption , inversely weighted by flux error of the individual galaxies contributing to the bin ( the individual balmer absorptions were inferred from the best - fit stellar models ) . the normalized composite spectrum was used to derive the h@xmath3/h@xmath4 ratio . to assess the total flux of the lines , we stacked the individual spectra without normalizing them by the individual h@xmath3 luminosities . ) as a function of stellar mass for star - forming galaxies at @xmath2 = 1.37 2.61 , known as the star - forming main sequence . the individual points show galaxies with detected h@xmath3 and h@xmath4 and the triangles are those with undetected h@xmath4 , for which @xmath22 lower limits of the dust corrected sfrs are plotted . the dashed line is the best linear fit to the individual detected galaxies with @xmath36 . the larger circles are stacks in five mass bins , taking into account only the detected galaxies ( the light blue circles ) and including the h@xmath4 non - detections ( the red circles ) . the shaded regions and numbers on the bottom represent the fraction of galaxies with robust redshifts and h@xmath3 detection with respect to the targeted objects in each mass bin . the @xmath22 limits on the sfr at @xmath37 and @xmath38 are calculated using the line sensitivities in _ h _ and _ k _ bands , respectively @xcite , and are shown by dashed lines . ] the correlation between stellar mass and sfr has been studied intensively since @xcite . as a first step in studying this correlation in our h@xmath3- and h@xmath4-detected sample , we calculated the spearman s rank correlation coefficient ( @xmath39 ) between the log(sfr(h@xmath3 ) ) and log(@xmath0 ) . we obtained @xmath40 at @xmath41 , indicating a significant positive correlation between the two quantities . the sfrs versus stellar masses of individual galaxies are plotted in figure [ fig : stack ] . the stacked sfrs in five stellar mass bins ( the first bin is @xmath42 and the rest are divided by 0.5dex widths , section [ sec : stack ] ) are also plotted on top of the individual points . in this section , we describe our low - mass completeness , the fitting methodology , and discuss the effect of h@xmath4 nondetections and agn on the sfr-@xmath0 relation . @xcite in the goods - n field . the fluxes and uncertainties are from the 3d - hst photometric catalog @xcite . the orange histogram denotes galaxies with photometric redshifts that fall outside of the spectroscopic redshift range ( bad photo - z ) . the blue histogram shows galaxies with photometric redshifts in the same range as the spectroscopic redshift ( good photo - z ) . the galaxies with bad photo - z s have on average bluer colors and larger uncertainties , indicative of a weaker balmer/4000 break compared to the good photo - z galaxies . _ bottom : _ stellar mass distribution of the same samples described above . the orange / blue numbers indicate fraction of galaxies with bad / good photo - z s in each mass bin relative to the total number of galaxies in the corresponding bin . the histograms in these figures show the fractions , not the actual numbers . therefore , the relative bin heights do not correspond to the percentage numbers on the bottom and should not be confused with each other . , title="fig : " ] @xcite in the goods - n field . the fluxes and uncertainties are from the 3d - hst photometric catalog @xcite . the orange histogram denotes galaxies with photometric redshifts that fall outside of the spectroscopic redshift range ( bad photo - z ) . the blue histogram shows galaxies with photometric redshifts in the same range as the spectroscopic redshift ( good photo - z ) . the galaxies with bad photo - z s have on average bluer colors and larger uncertainties , indicative of a weaker balmer/4000 break compared to the good photo - z galaxies . _ bottom : _ stellar mass distribution of the same samples described above . the orange / blue numbers indicate fraction of galaxies with bad / good photo - z s in each mass bin relative to the total number of galaxies in the corresponding bin . the histograms in these figures show the fractions , not the actual numbers . therefore , the relative bin heights do not correspond to the percentage numbers on the bottom and should not be confused with each other . , title="fig : " ] [ fig : bias ] as it is evident from figure [ fig : stack ] , the stacked sfr of the first mass bin is elevated compared to the best - fit line . in this section , we discuss the completeness of our sample and address the possible causes of incompleteness at low stellar masses . the redshifts of targets in the mosdef parent sample were adopted primarily from the 3d - hst grism measurements or other available spectroscopic campaigns ( see , * ? ? ? * ) . when a spectroscopic redshift was not available , which was the case for 67% of the whole sample and 73% of the objects with @xmath43 , redshifts were estimated using the 3d - hst photometry and the eazy sed - fitting code @xcite . objects with secure photometric redshifts may be biased against those objects with weak balmer/4000 breaks ( and hence younger ages and lower masses ) . to assess the magnitude of this bias , we cross correlated our sample with the @xcite spectroscopic sample of uv - selected galaxies in the goods - n field . rest - frame uv - selected samples are immune to the bias mentioned above as their selection criteria are not dependent on the strength of the balmer or 4000 break . there were 81 galaxies at @xmath44 that were in common between the mosdef and the uv - selected sample . we divided the common galaxies into two sets : those that have photometric redshifts in the same range as their spectroscopic redshifts of @xmath45 ( good photo - z , 61 objects ) , and those with photometric redshifts outside of this range ( bad photo - z , 20 objects ) . the bad photo - z objects are not scattered symmetrically around @xmath44 ; their photometric redshifts are systematically lower than our target redshift range , i.e. , @xmath46 . the two filters that bracket the balmer break are _ f160w _ and _ hst_/acs _ f850lp _ , which were available for all 81 galaxies examined here . as expected , the galaxies with incorrect photometric redshifts had on average bluer _ f850lp _ - _ f160w _ colors , with larger uncertainties . the distributions of the flux ratios and the associated uncertainties are plotted in figure [ fig : bias ] . were it not for prior spectroscopic redshifts , those galaxies with bad photo - z s would have been missing from our sample . the conclusion is further supported by the two - sided kolmogorov - smirnov test , which indicates a low probability of @xmath47 that the _ f850lp _ - _ f160w _ color distribution of the good and bad photo - z sets are drawn from the same parent distribution . the stellar - mass distributions of the two sets of galaxies are shown in the bottom panel of figure [ fig : bias ] , with the fraction of galaxies in each mass bin written on the bottom of the plot . the distributions indicate that the majority of galaxies with stellar masses below @xmath48 would have been missed from the mosdef sample if we had relied on their photometric redshifts . therefore , to avoid any biases we limited our analysis to those with @xmath49 . there are 128 and 185 galaxies above @xmath48 at @xmath45 and @xmath50 , respectively . the dashed line in figure [ fig : stack ] is the linear regression fit to the individual galaxies with @xmath49 . the low mass limit we adopted is also the limit above which the inferred log(sfr)-log(@xmath0 ) slope ( see section [ sec : method ] ) deviates by greater than @xmath51 from the slope over the full stellar mass range , including the low mass galaxies . the other possible bias at the low - mass end of our sample is a malmquist bias , due to the fact that our requirement of an h@xmath3 detection may bias against objects with faint h@xmath3 emission . to test this possibility , we calculated the fraction of galaxies that were detected in h@xmath3 to the number of galaxies targeted for spectroscopy , in bins of stellar mass . the fractions are listed on the bottom of figure [ fig : stack ] . the detection fraction of the first bin ( the lowest mass bin ) is similar to that of the higher mass bins , indicating that by limiting our sample to the h@xmath3-detected objects we did not miss a significantly higher fraction of galaxies at low masses . the rest - frame optically selected sample of the mosdef survey is less prone to missing highly obscured galaxies at high stellar masses , compared to the uv - selected samples . the effect of this potential bias of uv - selected samples versus our sample will be discussed in section [ sec : slope_sample ] . cccccc & h@xmath3 & @xmath52 & @xmath53 & @xmath54 & @xmath55 + ( n = 185 ) & uv@xmath56 & @xmath57 & @xmath58 & @xmath59 & @xmath60 + & sed & @xmath61 & @xmath62 & @xmath63 & @xmath64 + & uv@xmath65 & @xmath66 & @xmath67 & @xmath68 & @xmath69 + & & & & & + @xmath70 & h@xmath3 & @xmath71 & @xmath72 & @xmath55 & @xmath73 + ( n = 128 ) & uv@xmath56 & @xmath74 & @xmath75 & @xmath73 & @xmath64 + & sed & @xmath76 & @xmath77 & @xmath78 & @xmath64 + & uv@xmath65 & @xmath79 & @xmath80 & @xmath63 & @xmath81 [ tab : param ] there are various linear regression methods that can be used to characterize a correlation between two variables . in most studies of the sfr-@xmath0 relation , the correlation between sfr and @xmath0 has been determined using the ordinary least - squares regression ( ols ) . in the ols method , one variable ( here , the sfr ) is assumed as the dependent variable and the regression line is defined to be that which minimizes the sum of the squares of the vertical distances between the data points and the regression line . the ols method should be used in science cases where there is an independent variable that is estimated confidently and will be used to predict the other variable . as over a galaxy s lifetime the stellar mass is built relatively smoothly as opposed to the sfr , which can be stochastic and vary to a greater extent , we are more interested in determining the scatter and mean of sfr as a function of stellar mass . for this purpose , the ols method is a valid statistical procedure to be used for characterizing the sfr-@xmath0 relation . we determined the best estimates for the ols regression slope and intercept by perturbing the data according to the stellar mass and sfr uncertainties 1000 times . the median of the slope and intercept distributions are taken as the best - fit parameters . the slope and intercept uncertainties were calculated based on the ols method error estimates in @xcite . @xcite estimated the errors of regression parameters for linear relations where there is an intrinsic scatter . the average of the uncertainties obtained for the individual realizations was adopted as the uncertainty in the regression coefficients . we found a slope of @xmath52 for sfr(h@xmath3 ) at [email protected] , and @xmath71 for the [email protected] sample only , with a lower mass limit of @xmath48 . the relation is presented in figure [ fig : stack ] and table [ tab : param ] . the sfr-@xmath0 relation was not investigated separately for the sample at @xmath83 due to the small number of galaxies at this redshift range . in the ols method , the scatter is defined as the scatter in sfr at a certain stellar mass . we estimated the total observed scatter by calculating the 3-@xmath84 clipped dispersion in the distribution of vertical distances from the best - fit line . we propagated the mean measurement errors of mass and sfr ) . ] and subtracted them ( in quadrature ) from the total observed scatter to obtain the measurement - subtracted scatter . we found a measurement - subtracted scatter of 0.31 dex for sfr(h@xmath3 ) at @xmath85 . the measurement - subtracted scatter still includes galaxy - to - galaxy variations in the assumed dust attenuation curve ( section [ sec : scatter ] ) . an alternative statistical procedure to the ols method is a regression method that treats the two variables symmetrically , such as the ols bisector method . the ols bisector is recommended to be applied in cases where an underlying functional relation between two independent variables is studied @xcite . the scatter in this method is the scatter about the best - fit line to the relation . for comparison , we reported the ols bisector values in table [ tab : param_bisec ] . we should emphasize that as we are most interested in determining the scatter and mean of sfr at a given stellar mass , the ols method is the one on which we based our main conclusions . cccccc & h@xmath3 & @xmath86 & @xmath87 & 0.33 & 0.30 + ( n = 185 ) & uv@xmath56 & @xmath88 & @xmath89 & 0.29 & 0.26 [ tab : param_bisec ] it is important to investigate whether excluding the h@xmath4 nondetections from the main analysis causes any biases . in other words , do the h@xmath4-undetected galaxies represent different physical characteristics or are they undetected due to observational shortcomings ? we stacked the spectra of the h@xmath4-undetected objects as described in section [ sec : stack ] , and derived the dust - corrected sfr(h@xmath3 ) using the stacked spectrum . in the composite spectrum , the stacked h@xmath4 line is detected at @xmath90 . the stacked sfr and median stellar mass of the undetected - h@xmath4 objects are plotted on top of the sfr-@xmath0 relation in figure [ fig : scatter ] as the red star . the star lies below the linear fit by 2.5@xmath84 , but is still within the sfr(h@xmath3 ) scatter of 0.36dex , implying that these objects are not on average intrinsically different from the detected objects and excluding them should not systematically bias the sfr-@xmath0 relation . as mentioned in section [ sec : hasample ] , this conclusion is further supported by the fact that the spectra of a large fraction of these galaxies are contaminated by skylines . as mentioned before in section [ sec : sample ] , agn were identified through x - ray , ir , and optical line observations and removed from the sample . the agn mostly reside on the high mass end of the main sequence and it is interesting to examine how they would change the slope and scatter of the log(sfr)-log(@xmath0 ) relation . here , we assumed that all of the rest - frame uv continuum emission in agn host galaxies is coming from the stellar population ; we determined the sfrs from the uv luminosity at 1600 , dust corrected using the uv slope , and estimated the masses through the sed modeling . including the agn does not change the slope significantly ( @xmath91 ) but results in a larger scatter ( 0.41dex ) that is mainly occurring at the massive end , as most of the agn have larger stellar masses than the typical stellar mass of the star - forming objects . this is a well - understood selection effect for agn that they are easier to detect in more massive galaxies @xcite . in this section , we investigate the intrinsic scatter in the log(sfr)-log(@xmath0 ) relation by using different sfr indicators and considering variations in the dust attenuation curve . the observed and measurement - subtracted scatter in sfr(h@xmath3 ) versus @xmath0 and sfr(uv ) versus @xmath0 , calculated as described in section [ sec : method ] , are presented in table [ tab : param ] and figure [ fig : scatter ] . for a fair comparison , we limited our uv sample to the same sample as the h@xmath3 detected galaxies . the measurement - subtracted scatter reported here is affected by uncertainties such as galaxy - to - galaxy variations in imf and dust attenuation curve . in order to test the effect of dust attenuation variations , we simulated an sfr-@xmath0 relation with 1000 galaxies and an intrinsic scatter of 0.2dex , assuming the log(sfr)-log(@xmath0 ) slope and intercept derived based on our mosdef data ( section [ sec : method ] ) . we attenuated the intrinsic sfrs to derive the observed ( unobscured ) sfrs following the relation , log(sfr@xmath92)@xmath93log(sfr@xmath94@xmath34 . in this equation , the @xmath34 was calculated using a linear correlation between the nebular @xmath34 and the intrinsic sfr with a scatter of 0.5dex , based on the mosdef data . the reddening value , @xmath95 , was selected randomly according to a valid range of the most commonly used attenuation curves in the literature ; @xmath96 varied between the mosdef @xcite and the @xcite curves reddening values , and @xmath97 varied between the milky way and the smc @xcite reddening values . the dust - corrected sfrs were then reproduced by assuming a _ single _ @xmath95 ( the calzetti value for the sfr(uv ) and the cardelli value for the sfr(h@xmath3 ) ) . the scatter in the new dust - corrected log(sfr(h@xmath3))-log(@xmath0 ) relation was @xmath98dex larger than the intrinsic simulated scatter , while the dust - corrected log(sfr(uv))-log(@xmath0 ) scatter increased by @xmath99dex . the larger increase in the sfr(uv ) scatter is followed by larger variations of the attenuation curves at shorter uv wavelengths compared to the variations of the attenuation curves at optical wavelengths does not change as a function of luminosity or stellar mass , but we know that the scatter in @xmath97 is larger at higher uv luminosities ( e.g. , * ? ? ? * ; * ? ? ? this effect would increase the recovered scatter in log(sfr)-log(@xmath0 ) relation even to a greater degree . ] . this simple simulation indicates that without knowing the attenuation curve variations , it is not trivial to interpret the difference in the sfr(h@xmath3 ) and the sfr(uv ) scatter solely as the sfr stochasticity due to different timescales . variations in the assumed imf and metallicity would also affect the sfr(h@xmath3 ) and sfr(uv ) scatter to different degrees , again making it difficult to conclude about the burstiness of the star formation based on the difference in the scatter of sfr(h@xmath3 ) and sfr(uv ) . it is worth pointing out that picking a random star - formation history between a constant and an exponentially rising star - formation history affects the sed inferred stellar masses by @xmath100dex , which is negligible compared to the scatter in the log(sfr)-log(@xmath0 ) relation . the calculated measurement - subtracted scatter in the log(sfr(h@xmath3))-log(@xmath0 ) relation is 0.06dex larger than that of the log(sfr(uv))-log(@xmath0 ) relation . as the two sfr indicators trace sfr on different timescales luminosity traces sfr on shorter timescales compared to the uv luminosity , as h@xmath3 is sensitive to the most massive ( @xmath101 ) and short - lived ( ages @xmath102myr ) stars , while uv traces stars with @xmath103 and ages @xmath104myr , @xcite . ] , the difference in sfr scatter may be attributed to the stochasticity of sfr . it has been suggested in the galaxy formation simulations that for a massive galaxy ( @xmath105 halo mass of @xmath106 ) with a typical merger history , the dispersion in sfr averaged over timescales of @xmath107 years could be @xmath108dex smaller than the scatter of sfr averaged over shorter timescales of @xmath109 years @xcite . on the other hand , in our study the h@xmath3 light is captured through a slit while the uv light is obtained from imaging data , which may introduce additional scatter to our measurements . considering these uncertainties and poorly constrained effects of the variations in dust attenuation curve and imf , as discussed earlier in this section , we conclude that the 0.06dex difference in scatter is not significant and can not be attributed solely to the sfr stochasticity in these galaxies . according to @xcite , the intercept of the log(sfr)-log(@xmath0 ) relation changes as @xmath110 , where @xmath111 is the age of the universe in gyr . this change corresponds to an offset of 0.09dex from @xmath112 to 2.6 . the offset has a negligible effect on the 0.31dex scatter of log(sfr(h@xmath3))-log(@xmath0 ) relation derived here at @xmath113 , and subtracting it in quadrature will reduce the scatter to 0.30dex . prior to the use of multi - wavelength estimates of the sfr ( e.g. , sfr(ir)+sfr(uv ) ) , it was common to assess the sfr-@xmath0 relation using sed - inferred sfrs ( e.g. , * ? ? ? * ; * ? ? ? * ) . to investigate how sfr(sed ) would affect the relation , we also calculated the scatter using sed - inferred sfr and uv sfr , where the latter was corrected by the sed - derived @xmath34 rather than the uv slope . the scatter in sfr is reduced to 0.25 and 0.20dex for sfr(sed ) and sfr(uv ) , respectively ( the last two rows in table [ tab : param ] ) . these values clearly underestimate the true scatter in the log(sfr)-log(@xmath0 ) relation , because sfr(sed ) and sed - inferred @xmath34 are highly correlated with the sed - derived stellar masses . galaxy formation simulations suggest various drivers for the scatter in log(sfr)-log(@xmath0 ) relation . @xcite argued that the source of the small scatter they derived ( 0.12dex ) is the variance in mass accretion histories for halos of a given virial mass , but @xcite suggested that the more complex relation between the gas inflow rate and the dilution time , which shows how quickly galaxies can return to equilibrium after being perturbed away from the main sequence also influences the scatter . our results show that the larger scatter we found ( @xmath114dex ) is not only due to the measurement uncertainties , which indicates that the gas accretion and star - formation histories are not as smooth as predicted by the some simulations . also , by assuming an average attenuation curve and a single imf and metallicity for all the galaxies contributing to the sfr-@xmath0 relation , the measurement - subtracted scatter is a lower limit on the intrinsic scatter in sfrs at a given stellar mass . in this section , we focus on the slope of the log(sfr)-log(@xmath0 ) relation for the sfr(h@xmath3 ) sample . the inferred slope ( see section [ sec : method ] for the methodology ) is @xmath71 for the @xmath24 sample and when we include the @xmath23 sample the slope is @xmath52 , at @xmath49 ( refer to section [ sec : mass_completeness ] for the discussion on the low - mass limit ) . the slope of the full sample ( i.e. , @xmath115 ) is @xmath116 . there are many studies in the literature that parameterize the sfr-@xmath0 relation and their reported slopes span in a wide range of @xmath117 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? various effects could lead to discrepancies in the slope estimates . in this section , we try to address three of the most significant effects by investigating them in our sample . ) -@xmath0 relation , where the sfr is corrected for dust attenuation by different methods and attenuation curves . _ top : _ the blue dots show sfr(h@xmath3 ) dust corrected by the nebular @xmath34 ( derived from the observed balmer decrement ) and assuming a calzetti curve . the red dots represent the sfr(h@xmath3 ) corrected by the sed inferred @xmath34 multiplied by 2.27 , again assuming a calzetti curve . the latter is a method that is commonly used in high - redshift studies to correct the h@xmath3 luminosities without having access to the balmer decrement . this method overestimates the corrected sfrs and results in a steeper slope ( @xmath118 ) compared to the nebular corrected sfrs ( slope = @xmath74 ) . the dashed lines show the linear fits to galaxies with @xmath36 . the yellow line is the linear fit to galaxies corrected by nebular @xmath34 but assuming a cardelli galactic extinction curve , which we used as a default in this study . _ bottom : _ the sfr-@xmath0 relation for the same set of galaxies , corrected with the nebular @xmath34 assuming three different dust curves : the calzetti curve , the mosdef curve , and the cardelli curve . the best - fit lines are shown with the relevant colors . all three slopes are consistent with each other within their uncertainties . , title="fig : " ] ) -@xmath0 relation , where the sfr is corrected for dust attenuation by different methods and attenuation curves . _ top : _ the blue dots show sfr(h@xmath3 ) dust corrected by the nebular @xmath34 ( derived from the observed balmer decrement ) and assuming a calzetti curve . the red dots represent the sfr(h@xmath3 ) corrected by the sed inferred @xmath34 multiplied by 2.27 , again assuming a calzetti curve . the latter is a method that is commonly used in high - redshift studies to correct the h@xmath3 luminosities without having access to the balmer decrement . this method overestimates the corrected sfrs and results in a steeper slope ( @xmath118 ) compared to the nebular corrected sfrs ( slope = @xmath74 ) . the dashed lines show the linear fits to galaxies with @xmath36 . the yellow line is the linear fit to galaxies corrected by nebular @xmath34 but assuming a cardelli galactic extinction curve , which we used as a default in this study . _ bottom : _ the sfr-@xmath0 relation for the same set of galaxies , corrected with the nebular @xmath34 assuming three different dust curves : the calzetti curve , the mosdef curve , and the cardelli curve . the best - fit lines are shown with the relevant colors . all three slopes are consistent with each other within their uncertainties . , title="fig : " ] the method used for dust correction affects the derived slope of the log(sfr)-log(@xmath0 ) relation . in high - redshift studies , a commonly used dust correction method for the nebular lines is to multiply the @xmath34 derived for the stellar continuum by 2.27 factor and assume the @xcite attenuation curve . it has been shown in some previous studies that this recipe overestimates the corrected h@xmath3 sfrs for uv - selected galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in the top panel of figure [ fig : sfrha_dust ] we compared the h@xmath3 sfrs corrected by @xmath119 , computed from the balmer decrement , with those corrected by @xmath120 . we emphasize that this practice was done solely for the purpose of demonstrating how the commonly used dust correction recipes could potentially affect the parameterization of the sfr-@xmath0 relation in the absence of balmer decrement measurements . the sfrs corrected with @xmath120 assuming the calzetti curve yield a main sequence slope of @xmath121 , while the slope of those corrected by the @xmath119 with the calzetti curve is @xmath74 . the steeper slope is expected as the color - excess ( @xmath34 ) increases with stellar mass ( e.g. , * ? ? ? * ) and multiplying @xmath34 by a factor ( here , 2.27 ) causes a more significant difference at higher masses . adopting a value between 1 and 2.27 @xcite will make the slope shallower than 0.90 but still steeper than the 0.71 slope derived by the @xmath119 with the calzetti curve we also investigated the potential effect of the assumed attenuation curve on the best - fit slope of the log(sfr)-log(@xmath0 ) relation . we corrected the observed h@xmath3 sfrs assuming different attenuation curves : the calzetti curve @xcite , the attenuation curve derived from galaxies in the mosdef sample @xcite , and the ccm curve @xcite . the first two are stellar attenuation curves while the latter is a line of sight extinction curve that we used to correct the h@xmath3 luminosities throughout this paper . the results are presented in the bottom panel of figure [ fig : sfrha_dust ] . the nebular color excess ( i.e. , @xmath119 ) does not change significantly between the three curves , the reason being that the shapes of these three curves in the rest - frame optical are very similar . the main difference arises from normalization of the curves , affecting the total attenuation correction . @xcite showed that assuming the calzetti curve for the nebular lines as opposed to the ccm curve results in higher sfrs , where the difference increases with increasing sfr ( see figure 21 in @xcite ) . we see the same trend here in figure [ fig : sfrha_dust ] , namely that the discrepancy between corrected sfrs is larger at higher sfrs . despite the change in sfr , the three main sequence slope estimates are consistent with each other within their uncertainties . and h@xmath4 detected sample at @xmath122 in the rest - frame uvj diagram . the solid line separates blue and red star - forming galaxies . _ sfr(h@xmath3 ) versus @xmath0 for the blue and red star - forming galaxies . the log(sfr)-log(@xmath0 ) slope of the blue star - forming sample is steeper than the slope of the whole sample , including the red star - forming galaxies . , title="fig : " ] and h@xmath4 detected sample at @xmath122 in the rest - frame uvj diagram . the solid line separates blue and red star - forming galaxies . _ bottom : _ sfr(h@xmath3 ) versus @xmath0 for the blue and red star - forming galaxies . the log(sfr)-log(@xmath0 ) slope of the blue star - forming sample is steeper than the slope of the whole sample , including the red star - forming galaxies . , title="fig : " ] selection effects can have a major effect on the derived log(sfr)-log(@xmath0 ) slope . the samples that are selected based on rest - frame uv colors ( such as lyman - break selection , @xcite ) are biased towards less dusty galaxies . adding the redder and more dust - attenuated _ star - forming _ galaxies to these samples results in shallower slopes @xcite . to test this effect on our sample , we divided the star - forming galaxies into a blue bin and red bin , based on their rest - frame uv and vj colors . previously , we used the uvj diagram to remove the quiescent galaxies from our sample ( section [ sec : hasample ] ) , and now we set another criterion on the star - forming part of the diagram to separate the blue and red star - forming galaxies . the selection regions are shown in figure [ fig : uvj ] . after the removal 21 galaxies that were classified as red star - forming , the slope of the relation increases to @xmath123 . these galaxies are both more dust - attenuated and older compared to the rest of the sample . a steeper slope is expected as the red star - forming galaxies dominate the lower right corner of the main sequence ( those with lower specific sfrs ) . in conclusion , the red star - forming galaxies should be included in a complete sample , otherwise the slope of the log(sfr)-log(@xmath0 ) relation may be overestimated . considering that the mosdef sample is a near - ir selected sample and because the h@xmath3 luminosity is less affected by dust attenuation than the shorter - wavelength uv luminosity , this study is less prone to the bias associated with uv selection . relation for stacks of galaxies in five bins of stellar mass , the difference being in how the stacked h@xmath3 and h@xmath4 luminosities are corrected for the underlying balmer absorption . for illustrative purposes each set of colored points is slightly shifted in mass direction . _ blue symbols : _ the balmer absorption in each bin is estimated as the weighted average of the individual absorption fluxes . _ orange symbols : _ the same as the blue points , but the average is used instead of the weighted average . _ green symbols : _ no absorption correction is applied . _ bottom : _ ratio of the h@xmath4 absorption flux to the emission flux as a function of stellar mass . the absorption flux becomes stronger with stellar mass , which results in a more significant correction for galaxies with higher stellar masses . ] another factor that affects the sfr-@xmath0 parameterization in studies that use the hydrogen balmer lines is the balmer absorption correction . for the star - forming galaxies considered in this study , the difference between the absorption corrected sfrs and those that are not corrected becomes larger at higher masses because the absorption flux , which is produced in the atmosphere of primarily a - type stars , increases as a function of stellar mass ( see the bottom panel in figure [ fig : balmabs ] ) . the galaxies with higher stellar masses have more mature stellar populations and hence higher balmer absorption equivalent widths . in figure [ fig : balmabs ] , we show the stacks of sfr(h@xmath3 ) in bins of stellar mass , corrected for the balmer absorption in two different ways as well as the stacks without any corrections . the absorption correction affects both the value of the observed sfr(h@xmath3 ) and the dust correction ( i.e. , the ratio of h@xmath3 to h@xmath4 ) . we used the following correction method in this study . in each stellar mass bin , the balmer absorption of individual galaxies inferred from the stellar population models were averaged , weighted by the inverse variance calculated from the errors in the individual balmer emission line luminosities , and added to the stacked emission line luminosities . these are blue symbols in figure [ fig : balmabs ] . on the other hand , as presented in figure [ fig : balmabs ] , not correcting for the balmer absorption ( green symbols ) results in values that are inconsistent with the blue linear fit by as large as @xmath124 at @xmath125 . the absorption corrections reduce the inferred dust - corrected sfrs by values that are typically larger at larger masses . ( here , by factors of 1.3 , 2.1 , and 2.5 at the stellar masses of @xmath126 , @xmath127 , @xmath128 . ) as a result , the sfrs at higher masses are lower once the balmer absorption correction is applied . in other words , not correcting for the underlying balmer absorption results in higher sfrs at high masses and hence , a steeper slope . in conclusion , we emphasize the importance of accounting for the balmer absorption when using the balmer emission lines to quantify the sfr-@xmath0 relation . using the balmer lines that are not corrected for the underlying absorption results in higher sfr values , especially at the large masses where the balmer absorption is stronger , and affects the parameterization of the sfr-@xmath0 relation . the stacked points in bins of stellar mass , for which the weighted averaged balmer absorption corrections are applied , are in the best agreement ( within @xmath129 ) with the @xmath130 linear regression fitted to the individual galaxies ( the blue line in figure [ fig : balmabs ] ) . relation for various studies . the blue points and the blue line are our @xmath45 results , using sfr(h@xmath3 ) corrected for dust according to the balmer decrement . the gray squares are from @xcite narrow - band selected star - forming sample , where they used sfr(h@xmath3 ) dust corrected based on the local mass - dependent correction relation of @xcite . the @xcite and @xcite relations are based on sfr(uv)+sfr(ir ) . we also plotted our results based on the sfr(uv ) and sfr(sed ) values . the relations from the literature are converted to a chabrier imf , when necessary . ] the sfr-@xmath0 relation has been studied extensively . the parameterization of the sfr-@xmath0 relation depends on the sfr indicator and the typical timescale over which it traces star formation . first we consider studies that also use h@xmath3 to quantify sfrs and therefore provide the most direct comparison . @xcite , @xcite , and @xcite used h@xmath3 narrow - band selected samples at @xmath8 . such studies , however , tend to be biased towards galaxies with larger h@xmath3 equivalent widths . these three studies also did not measure h@xmath4 , and relied on a mass - dependent dust correction from @xcite . the @xcite relation was derived based on a local sample of galaxies but its validity has not been verified for @xmath8 galaxies . the gray symbols in figure [ fig : compare ] are the data points from @xcite . at the low mass end , @xmath131 , the high equivalent width bias mentioned above and the effect of the improper dust correction are evident . when we applied an average mass - dependent dust correction based on our balmer decrement measurements to the @xcite observed sfrs , we see a better agreement with our results at low masses . the @xcite main sequence is tighter compared to ours , mainly because the scatter in the dust extinction@xmath0 relation is not taken into account . @xcite is another h@xmath3 study , where they used _ near - ir grism spectroscopy for a sample of emission line galaxies at @xmath132 . they used the h@xmath3 , h@xmath4 , [ oii]@xmath1333727 , and [ oiii]@xmath1335008 nebular emission lines as the sfr indicators , dust corrected in bins of stellar mass . the log(sfr)-log(@xmath0 ) slope they found was @xmath134 at @xmath24 , though they also identified a malmquist bias as being a potential issue in their analysis . the sfr-@xmath0 relation was also investigated by @xcite using a sample 114 star - forming galaxies at @xmath8 with h@xmath3 spectra corrected for dust extinction by the sed @xmath34 and the calzetti curve . @xcite used the @xcite sample to fit a linear relation to the median sfrs in bins of stellar masses . similar to the other h@xmath3 studies , they also found a shallow slope of @xmath135 . we also considered sfr-@xmath0 studies where the total sfr was estimated as the sum of uv and ir - based sfrs , plotting the sfr-@xmath0 relations of @xcite and @xcite on top of our points in figure [ fig : compare ] . the log(sfr)-log(@xmath0 ) slopes of these studies are generally steeper than our slope . as we used the balmer decrement to correct the observed sfr(h@xmath3 ) for dust extinction , as opposed to using the ir data , it is probable that we missed optically thick star - forming regions , particularly in massive galaxies with high specific sfrs . this bias may be the reason of our shallower slope compared to the studies with the ir data . there are several important differences between the @xcite and @xcite studies and ours , such as the samples , the sfr diagnostics , and the statistical methods used for the fit . at this point , it is not trivial to reconcile these discrepancies and discuss the agreement or disagreement between these studies and ours . a more informative comparison will become possible once we incorporate ir data into our analysis . a few studies found indications of a non - linear sfr-@xmath0 relation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? they suggested that the log(sfr)-log(@xmath0 ) slope flattens at high masses so that a single power law can not explain the sfr-@xmath0 relation . according to these studies , at higher redshifts the high - mass flattening becomes less prominent and at @xmath8 the turning point shifts to @xmath136 . a turning point in the sfr-@xmath0 relation is not obvious from the individual detections in our sample , nor from the stacks ( figure [ fig : stack ] ) . however , because of the large uncertainty in the most massive stacked bin of our galaxies ( see the blue and orange points in figure [ fig : balmabs ] ) , we can not confidently rule out the flattening at the high - mass end . as part of the mosdef survey , we used a sample of 185 star - forming galaxies with h@xmath3 and h@xmath4 spectroscopy at @xmath49 to study the star - forming main sequence relation at @xmath8 . the parent catalog is _ h_-band ( i.e. , rest - frame optically ) selected and is accompanied by ancillary multi - band photometry @xcite , from which the uv - based sfrs were measured . the stellar masses were derived by comparing the emission - line corrected photometric seds with stellar population models . after taking into account the measurement uncertainties in the sfr and stellar mass , we found a measurement - subtracted scatter of 0.31dex in the log(sfr(h@xmath3))-log(@xmath0 ) relation at @xmath137 , which is 0.06dex larger than what we found based on the uv sfrs ( see , table [ tab : param ] ) . although in theory the time - scale variations in sfr can be traced by using different sfr diagnostics , such as uv and h@xmath3 , in the sfr-@xmath0 relation @xcite , we argue that in the absence of direct measurements of galaxy - to - galaxy variations in the attenuation curves and the imf , the difference in the scatter of the log(sfr(h@xmath3))-log(@xmath0 ) relation and the scatter of the log(sfr(uv))-log(@xmath0 ) relation can not be used to constrain the stochasticity of star formation . given these variations , the sfr(uv ) scatter could be as large or greater than the sfr(h@xmath3 ) scatter . theorists trying to reproduce the observed scatter in the log(sfr)-log(@xmath0 ) relation should proceed with caution , based on the results shown in this paper . the scatter in the log(sfr)-log(@xmath0 ) relation is tighter when the sfr(uv ) is corrected for extinction by @xmath34 derived from the sed modeling or when the sed - inferred sfrs are used , because the uv slope regulates the color - excess in the sed fitting from which the stellar masses were derived . in addition to the variations in the dust attenuation , imf , and metallicity , the scatter in the log(sfr)-log(@xmath0 ) relation may be due to other variables such as geometry , size , sfr surface density , and age ( e.g. , * ? ? ? * ; * ? ? ? . the significance of the effect of these variables on the scatter will be investigated in future studies . although in agreement with some simulations @xcite , the measurement - subtracted scatter we found is larger than some of the other works ( e.g. , * ? ? ? * ; * ? ? ? the discrepancy may indicate that the galaxy accretion and feedback processes are not as smooth as predicted by the simulations . we found a constant slope of @xmath71 for the @xmath138 sample and @xmath52 when we used the whole sample at @xmath139 , for @xmath140 . galaxy evolution simulations generally find a close - to - unity slope for the log(sfr)-log(@xmath0 ) relation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which translates to a stellar mass - independent specific sfr ( i.e. , sfr/@xmath0 ) . our shallower slope indicates higher ( lower ) sfr at low ( high ) stellar masses . as the slope of the log(sfr)-log(@xmath0 ) represents the star - formation efficiency , one explanation for close - to - unity slope in the galaxy simulations is that star formation is too inefficient at the low stellar - mass end due to feedback processes . the steep slope in the simulations may also be caused by a lack of quenching feedback that causes a dearth of red star - forming galaxies at the high - mass end of the sfr-@xmath0 relation . the slope of the log(sfr)-log(@xmath0 ) is influenced by various observational and measurement effects that could explain discrepancies in the slopes reported in different observational studies . we demonstrated three of the main effects here : dust correction , sample biases , and stellar absorption correction ( the last one is specific to the emission line studies ) . correcting the h@xmath3 emission line with the sed inferred @xmath141 , with the assumption of the calzetti curve , leads to a log(sfr)-log(@xmath0 ) slope of close to unity . a bias against dusty galaxies with low specific sfrs , as might be the case for uv - selected samples , may also result in steeper slopes of the log(sfr)-log(@xmath0 ) relation . finally , we showed that the balmer absorption correction has a significant effect on the dust - corrected sfrs and not applying the correction to the balmer emission lines could lead to an overestimation of the log(sfr)-log(@xmath0 ) slope . the evolution of the slope , normalization , and the scatter of the log(sfr)-log(@xmath0 ) relation is essential to our understanding of the processes that govern galaxy growth during cosmic time . once the whole mosdef sample becomes available , we have access to a large sample of rest - frame optical spectra for galaxies at @xmath142 , which makes it possible to study the sfr-@xmath0 evolution in redshift , using instantaneous and accurately - measured sfrs in a representative sample . furthermore , incorporating ir data into our analysis will allow us for a more detailed comparison with the existing studies . + we thank the referee for the constructive comments . funding for the mosdef survey is provided by nsf aag grants ast-1312780 , 1312547 , 1312764 , and 1313171 and grant ar-13907 , provided by nasa through a grant from the space telescope science institute . the authors thank the mosfire instrument team for building this powerful instrument , and for taking data for us during their commissioning runs . we are grateful to marc kassis at the keck observatory for his many valuable contributions to the execution of this survey . we also acknowledge the 3d - hst collaboration , who provided us with spectroscopic and photometric catalogs used to select mosdef targets and derive stellar population parameters . is thanks shoubaneh hemmati and mostafa khezri for useful discussions and feedback on the manuscript . we thank david sobral for providing the hizels h@xmath3 catalog that was used as a comparison to our results , and james aird for his help in matching agn selection catalogs . support for is is provided through the national science foundation graduate research fellowship dge-1326120 . nar is supported by an alfred p. sloan research fellowship . mk acknowledges support from the hellman fellows fund . alc acknowledges funding from nsf career grant ast-1055081 . we wish to extend special thanks to those of hawaiian ancestry on whose sacred mountain we are privileged to be guests . without their generous hospitality , most of the observations presented herein would not have been possible . this work is also based on observations made with the nasa / esa hubble space telescope ( programs 12177 , 12328 , 1206012064 , 1244012445 , 13056 ) , which is operated by the association of the universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 .

we propose a wavelet - based approach to construct consistent estimators of the pointwise hlder exponent of a multifractional brownian motion , in the case where this underlying process is not directly observed . the relative merits of our estimator are discussed , and we introduce an application to the problem of estimating the functional parameter of a nonlinear model . + * * keywords : * * pointwise hlder exponent ; multifractional process ; wavelet coefficients ; parametric estimation

1512.05054

multifractional brownian motion ( mbm ) is considered as one of the most natural extensions of fractional brownian motion ( fbm ) . nowadays applications of mbm are numerous and growing . similar to fbm , mbm has been used in such diverse areas as geology , image analysis , signal processing , traffic networks and mathematical finance . for instance , we refer to lvy - vhel ( 1995 ) , bertrand et al . ( 2012 ) and bianchi et al . ( 2013 ) . in this brief introduction , we focus on applications to mathematical finance , which we know best . since generally neither fbm nor mbm are semi - martingales , rogers ( 1997 ) pointed out that there would be arbitrage in a market where stocks are modelled by fbm . however , cheridito ( 2003 ) showed that , if one relaxes the definition of arbitrage , fbm is an excellent candidate to model long - term memory in stock markets . bayraktar et al . ( 2006 ) obtained fbm as the limit of the stock price in an agent - based model where investors display inertia . moreover , unlike stock prices , several processes , like stochastic volatility , exchange rates , or short interest rates do not need to be semi - martingales for a mathematical model to be arbitrage - free in a strict sense . for each of these processes , there is empirical evidence of long - term memory . we refer to corlay et al . ( 2014 ) for stochastic volatility , xiao et al . ( 2010 ) for exchange rates , and ohashi ( 2009 ) for interest rates . making the hurst parameter time - dependent allows to model different regimes of the stochastic process of interest . for example , in times of financial crisis , asset volatility rises significantly . likewise , empirical evidence shows that there has been periods of different volatility in either exchange rates or interest rates . this phenomena motivates one to introduce mbm into finance , since unlike fbm , the local regularity of volatilities driven by mbm allows to change via different periods . let @xmath0 denote an mbm with hurst function @xmath1 . we consider a general model @xmath2 with @xmath3 , @xmath4)$ ] . in this paper we are interested in estimating @xmath1 , starting from the observations of @xmath5 . an advantage of our model and methodology is that the functions @xmath6 and @xmath7 do not need to be known a priori . this is for instance the case of stochastic volatility , where the volatility @xmath8 is an unknown @xmath9 class function of @xmath10 . we define the mbm @xmath11}$ ] through its harmonizable representation ( see benassi et al . 1997 ) : for @xmath12 $ ] , @xmath13 where : @xmath14 : : the hurst functional parameter @xmath15 is a @xmath16-hlderian function with some @xmath17 ; @xmath14 : : the complex - valued stochastic measure @xmath18 is defined as the fourier transform of the real - valued brownian measure @xmath19 . more precisely , for all @xmath20 , @xmath21 where @xmath22 denotes the fourier transform of @xmath23 : @xmath24 statistically , the most significant feature of mbm , its local hlder regularity , can be measured by its pointwise hlder exponent . recall that for a continuous nowhere differentiable process @xmath25 , its pointwise hlder exponent @xmath26 is a stochastic process defined by : for each @xmath27 , @xmath28:~\limsup\limits_{\epsilon\rightarrow0}\frac{|y(t_0+\epsilon)-y(t_0)|}{|\epsilon|^{\alpha}}=0\big\}.\ ] ] ayache and lvy - vhel ( 2004 ) show that for each @xmath29 , the pointwise hlder exponent of @xmath30 at each point @xmath27 , @xmath31 , is with probability @xmath32 equal to its hurst functional parameter @xmath33 : @xmath34 since generating the scenarios of fbm and mbm relies only on their hurst parameters ( see ayache and lvy - vhel 2004 ) , the problem of estimation of the hurst parameter has significant interests . several results have already been obtained recently on the estimation of the mbm s hurst functional parameter . we refer to rosenbaum ( 2008 ) , coeurjolly ( 2005 , 2006 ) , bardet ( 2002 ) , bardet and surgailis ( 2013 ) and bertrand et al . ( 2013 ) . rosenbaum ( 2008 ) and bardet ( 2002 ) used wavelet - based methods to study inference problem on fbm , i.e. , the hurst parameter is constant ; coeurjolly ( 2005 , 2006 ) and bertrand et al . ( 2013 ) studied the estimation of @xmath35 by using respectively generalized variations ( see also chan et al . 1995 ) and increment ratio statistic method , where a discretized sample path of the mbm is assumed to be observed . bardet and surgailis ( 2013 ) developed a nonparametric estimation method ( based on the increment ratio ) for evaluating the local hurst function of a multifractional gaussian process ( whose increments are asymptotically a multiple of an fbm ) which extends mbm , starting from a discrete sample path of this process . in this work we consider a model more general than mbm but slightly less general than the one considered in bardet and surgailis ( 2013 ) . we study a different statistical setting ( either the wavelet coefficients of @xmath5 or a discrete path of @xmath36 points is observed ) , by applying a wavelet - based approach . note that the main advantage of this statistical setting is that , it could be applied to inferential problems when only a set of wavelet coefficients are available ( we refer to delbeke and van ( 1995 ) , abry et al . ( 2002 ) , abry and conalvs ( 1997 ) and the references therein ) . by a study on the fine regularity property , our central limit theorem provides an explicit form for the limit covariance matrix . it is worth noting that , the construction of our estimator relies on a sharp estimation of the covariance structure of the mbm s wavelet coefficients . the main technical difficulty arises due to the fact that the covariance structure of mbm is much more complicated than that of fbm . the techniques used to identify such covariance structure is not obvious and thus has its specific interests in statistical inferences on multifractional processes as well as on stochastic analysis . consider the following system : for @xmath12 $ ] , @xmath37 where @xmath14 : : @xmath38 is an mbm defined in ( [ mbm ] ) . assume that its hurst functional parameter @xmath1 belongs to @xmath39)$ ] and @xmath40}h(t),\sup\limits_{t\in[0,1]}h(t)\big]\subset(0,1);\ ] ] @xmath14 : : the closed forms of the deterministic functions @xmath6 and @xmath7 are unknown . however we assume that + @xmath41 ; ; @xmath3 , @xmath42 almost everywhere and there exist two constants @xmath43 such that @xmath44 for all @xmath45 . @xmath46 ; ; @xmath47)$ ] and @xmath48 almost everywhere . @xmath14 : : suppose that a discrete trajectory of @xmath5 : @xmath49 is observed for some @xmath50 large enough . our major goal is to evaluate the functional parameter @xmath15 . as in coeurjolly ( 2005 , 2006 ) , we introduce a pointwise estimation method , namely , the function @xmath15 is estimated pointwisely for any @xmath29 . once the time @xmath27 is fixed , the problem becomes a parametric estimation problem . peng ( 2011a ) studied this problem when @xmath8 is some stationary increment process ( with @xmath51 ) , where the optimal convergence rate @xmath52 is obtained by using the observations @xmath53 . however , when @xmath1 varies via time , the estimation of @xmath33 only relies on the sample size of the observed data in the neighborhood of @xmath54 and this neighborhood s radius convergence speed . hence the convergence speed of the corresponding estimator would be reasonably slower than @xmath52 ( see e.g. coeurjolly ( 2005 , 2006 ) for a particular case when @xmath5 is mbm ) . in this work , heuristically speaking , since the sample size of the neighborhood data of @xmath54 for estimating each @xmath33 is about @xmath55 , it is then believed that a good estimator should have its convergence rate near @xmath56 . subject to this statistical setting , we try to get the `` optimal '' rate of convergence estimator of @xmath33 by using wavelet basis . let the integer @xmath57 and let us pick any mother wavelet @xmath58)$ ] whose first @xmath59 moments are vanishing : @xmath60 and @xmath61 @xmath62 is called the cancellation order of @xmath63 . fix an integer @xmath64 , with @xmath65 being a subsequence of @xmath66 . the wavelet coefficients of @xmath8 are given by : for @xmath67 , @xmath68 we introduce a set of indices corresponding to a neighborhood of each @xmath29 : @xmath69 where the radius of this neighborhood @xmath70 satisfies @xmath71 and @xmath72 as @xmath73 . then the quadratic form estimator of @xmath74 is given as : @xmath75 for showing the consistency of our estimator and existence of central limit results , we need to impose the following technical assumptions to @xmath76 , @xmath62 and @xmath15 : * ( a0 ) : * : : @xmath77 . this condition leads to @xmath78}h(t)>1 $ ] . we remark that the latter inequality still holds when @xmath79 and @xmath80}h(t)<3/4 $ ] , as a consequence our results are still valid . however , we wo nt consider the latter assumptions , because practitioners in data analysis would nt expect the unknown functional parameter @xmath15 should be valued less than @xmath81 . * ( a1 ) : * : : a lower bound of the convergence rate of @xmath70 toward 0 is given as : @xmath82 * ( a2 ) : * : : an upper bound of the convergence rate of @xmath70 toward 0 is described as : for any @xmath83 arbitrarily small , @xmath84 * ( a3 ) : * : : there exists a constant @xmath85 such that @xmath86 . * ( a4 ) : * : : this condition is stronger than * ( a1 ) * : @xmath87 through this paper we assume that * ( a0 ) * is always satisfied . we will successively show that our estimator of @xmath33 is weakly consistent under assumption * ( a1 ) * ; it is strongly consistent under assumptions * ( a1)-(a2 ) * ; and it has an asymptotic gaussian behavior subject to assumptions * ( a2)-(a4)*. before stating these main results , we make some notation conventions : ( a ) : : a sequence @xmath88 of real - valued random variables is said to be bounded almost surely if and only if @xmath89 it is said to be bounded in probability if and only if @xmath90 let @xmath91 be a sequence of strictly positive real numbers and let @xmath88 be a sequence of positive random variables . the notations @xmath92 mean respectively that the sequence @xmath93 is bounded almost surely and in probability . ( b ) : : we say two sequences of strict positive values @xmath91 and @xmath94 are equivalent as @xmath95 tends to infinity if and only if there exist two constants @xmath96 such that @xmath97 we denote the relation of equivalence by @xmath98 . ( c ) : : we use the notations `` @xmath99{a.s.}$ ] '' , `` @xmath99{\mathbb p}$ ] '' and `` @xmath99{dist}$ ] '' to respectively denote convergence @xmath100-almost surely , convergence in probability and convergence in law . it is also useful to briefly introduce the steps which lead to the construction of the estimators : * step 1 : * : : identification of @xmath101 starting from the observations @xmath49 . in order to estimate @xmath101 , it suffices to make an identification of @xmath102 given in ( [ d ] ) . such an identification can be naturally obtained by discretization of integrals to the sum as @xmath103 by lvy s modulus of continuity theorem for mbm ( see e.g. theorem 1.7 in benassi et al . 1997 ) , @xmath4)$ ] , ( [ supx ] ) and the fact that @xmath104}h(u)}=\mathcal o\big(|s - t|^{\max\{h(s),h(t)\}}\big)~\mbox{as}~|s - t|\rightarrow0,\ ] ] there exists a positive - valued random variable @xmath105 all of whose moments are finite , such that for all @xmath106 $ ] with @xmath107 , @xmath108 then , the following important relations hold ( the proofs are given in the appendix ) : @xmath109 which show that @xmath110 is a good estimator of @xmath102 . next we set @xmath111 and show that @xmath112 satisfies @xmath113 please see the appendix for the proof of ( [ widev ] ) . * step 2 : * : : identification of @xmath33 starting from @xmath114 . we use similar computations appeared in peng ( 2011a ) . the main result is the following ( see proposition [ varvx ] for a more explicit formula ) : @xmath115 where @xmath116 is a constant which does not depend on @xmath117 ; @xmath118 denotes the cardinal of @xmath119 . it is observed that @xmath120 therefore subject to feasible choices of the sequence @xmath121 , the following convergence can accordingly take place in probability or almost surely : @xmath122 where @xmath116 is some constant not depending on @xmath117 . * step 3 : * : : identification of @xmath33 starting from @xmath123 . + under assumption * ( a1 ) * ( resp . * ( a1)-(a2 ) * ) , one has the following relation of equivalence ( see ( [ as1 ] ) , ( [ taylor2 ] ) , ( [ diffv_p ] ) and ( [ diffv_as ] ) ) : for @xmath124 , @xmath125 in probability ( resp . almost surely ) . as a consequence , @xmath126 is a consistent estimator of @xmath33 . we remark that the speed of convergence relies on the choice of @xmath127 . more details on the choice of @xmath127 will be discussed in theorem [ vydn ] . in this part we provide a sharp estimation of the covariance structure of wavelet coefficients of @xmath10 . for @xmath129 , define the wavelet coefficient @xmath130 of @xmath128 by @xmath131 for all @xmath132 - 1 \right\},$ ] with @xmath133 $ ] being the integer part function . + the following proposition provides a fine identification of @xmath134 s covariance structure . [ prop1 ] let @xmath135 satisfy that , there exist two constants @xmath136 such that @xmath137 . ] . let @xmath138 - 1\}$ ] , @xmath139 - 1\}$ ] . + @xmath41 if @xmath140 and @xmath141}|\frac{at - bs}{ak - bk'}|\leq1 $ ] , then we have @xmath142where the closed form of @xmath143 is given in @xmath144 and the remaining terms @xmath145 verify @xmath146 - 1,0\leq k'\leq [ b^{-1}]-1,ak\neq bk'}\!\!\!\!\!\!\!\!\!\!\!\!\!\!\big(t(k , k',q , a , b)\big)^2=\mathcal o((ab)^{1/2}\log a\log b),~\mbox{as $ a , b\rightarrow0$}.\ ] ] @xmath46 if @xmath147 , we have @xmath148 where the explicit form of @xmath149 is given in @xmath150 . the proof of proposition [ prop1 ] is given in the appendix . the following lemma is a straightforward consequence of proposition [ prop1 ] ( it suffices to take @xmath151 in proposition [ prop1 ] . ) that we will rely on heavily through the remaining context . [ lemma1 ] for @xmath152 and @xmath153 , we have @xmath154 where @xmath155)$ ] is defined as @xmath156 and @xmath157 where @xmath158)$ ] is defined as @xmath159 in ( [ c2 ] ) . in this section we first construct a consistent estimator of @xmath33 , when the wavelet coefficients of @xmath128 : @xmath161 can be straightforwardly observed ( i.e. @xmath162 ) . to this end we let @xmath163 the following proposition provides sharp identifications of @xmath114 s first order and second order moments . [ varvx ] for any @xmath29 , @xmath164 and @xmath165 where the constants @xmath166 and @xmath167 are given in ( [ computei00 ] ) and ( [ c2 ] ) respectively , and @xmath168 the proof of proposition [ varvx ] is given in the appendix . we remark that , since a sequence of random variables bounded almost surely is also bounded in probability , therefore the following identification of @xmath114 holds , thanks to proposition [ varvx ] , chebyshev s inequality and borel - cantelli s lemma . [ iv ] if @xmath169 , we have @xmath170 as a consequence , ( a ) : : if @xmath171 , then @xmath172 ( b ) : : if @xmath171 and @xmath173 for arbitrarily small @xmath83 , then @xmath174 for arbitrarily small @xmath83 . * proof . * by using chebyshev s inequality , ( [ varv ] ) and the condition that @xmath169 , we get for any @xmath175 , @xmath176 where @xmath116 is some constant which does not depend on @xmath117 . this implies @xmath177 then it follows from ( [ varv ] ) , ( [ iv5 ] ) and the fact that @xmath178 that @xmath179 ( [ iv1 ] ) has been proven . note that ( [ iv2 ] ) follows straightforwardly from ( [ iv1 ] ) . now we only need to show ( [ iv3 ] ) holds . from ( [ iv5 ] ) and chebyshev s inequality , we observe that there exists a constant @xmath116 which does not depend on @xmath180 nor on @xmath117 such that for any @xmath175 , @xmath181 since @xmath173 , then applying borel - cantelli s lemma leads to @xmath182 further observe that @xmath183 therefore ( [ iv3 ] ) follows . lemma [ iv ] has been proven . @xmath184 now we are ready to state the main results of this section . the following theorem constructs a consistent estimator of @xmath33 starting from the wavelet coefficients @xmath160 . [ vxd ] for @xmath29 , denote by @xmath185 ( a ) : : if @xmath171 , then @xmath186 ( b ) : : if @xmath171 and @xmath173 for any @xmath83 arbitrarily small , then @xmath187 * proof . * this theorem is a straightforward consequence of lemma [ iv ] , as we observe that , by construction , @xmath188 verifies @xmath189 then , by lemma [ iv ] , the fact that the logarithm function belongs to @xmath190 , continuous mapping theorem and the mean value theorem , we get theorem [ vxd ] . @xmath184 in theorem [ vxd ] , if we assume @xmath191 with some @xmath192 and @xmath193 , then by elementary computation we obtain the following corollary , which leads to choose @xmath194 to obtain the best rate of convergence of the estimator @xmath188 . this result is similar to the choice of @xmath195 in bardet ( 2002 ) in the setting of fbm . [ cor ] under assumption @xmath191 with @xmath192 and @xmath193 , ( a ) : : taking @xmath196 and @xmath197 , we have @xmath198 ( b ) : : for small @xmath83 , taking @xmath199 and @xmath200 , we obtain @xmath201 the second main result of this part describes an asymptotic gaussian behavior of the estimator of @xmath33 , where the limit covariance matrix is precisely given , which depends on @xmath33 . [ cltx ] for @xmath29 , if @xmath171 and assumptions @xmath202 are verified , then @xmath203{dist}\mathcal n\big(0,\tilde{c}(t_0)\big),\ ] ] where the constant @xmath204 with @xmath205 given in @xmath206 and @xmath207 given in ( [ c3t0 ] ) . in order to prove theorem [ cltx ] , we rely heavily on the following proposition . [ prop : clt1 ] for any @xmath29 and any integer @xmath208 , denote by @xmath209 then @xmath210{dist}\mathcal n(0,\sigma),\ ] ] where @xmath211 with @xmath212 and @xmath213 * proof . * following the similar method as in bardet ( 2000 ) , we show that for the empirical average of @xmath214 , a multivariate central limit theorem holds thanks to a lindeberg s condition . more precisely , for @xmath117 big enough and for @xmath215 , @xmath216 , denote by @xmath217 a lindeberg s condition thus can be deduced from the following relations : * for @xmath218 , @xmath219 * for @xmath220 , @xmath221 * for @xmath222 , @xmath223 next we present the following classical result ( see e.g. oehlert 1992 ) . [ deltamethod ] let the estimators @xmath242 ( valued in @xmath243 ) of @xmath244 satisfy the following central limit theorem : @xmath245{dist}\mathcal n(0,\sigma),\ ] ] where @xmath246 denotes the covariance matrix of the limit distribution and @xmath247 is a sequence of positive numbers tending to infinity . let @xmath248 ( @xmath249 or @xmath250 ) belong to @xmath251 , then the following convergence in law holds : @xmath252{dist}\mathcal n\big(0,\nabla g(\theta_1,\theta_2)^t\sigma\nabla g(\theta_1,\theta_2)\big),\ ] ] where @xmath253 denotes the transpose of the gradient of @xmath254 on @xmath244 . note that if @xmath255 in proposition [ deltamethod ] , the gradient of @xmath254 becomes a jacobian matrix . + * proof of theorem [ cltx ] . * for simplifying notation we denote by @xmath256therefore the following decomposition holds : @xmath257 since the fact that @xmath258 implies @xmath259 and also by taking @xmath260 in ( [ vnv2 ] ) , we have @xmath261 by ( [ lemme12 ] ) ( [ dec1 ] ) , ( [ jensen ] ) , ( [ cardident ] ) and ( [ das ] ) , we then obtain @xmath262 it follows from markov s inequality that there exists @xmath116 such that @xmath263 the assumption * ( a2 ) * then allows us to apply borel - cantelli s lemma to obtain @xmath264{a.s.}0.\ ] ] therefore , it follows from proposition [ prop : clt1 ] and continuous mapping theorem that @xmath265{dist}\mathcal n(0,\sigma).\ ] ] since @xmath266 , using slutsky s theorem , we get @xmath267{dist}\mathcal n(0,\sigma).\ ] ] this is in fact equivalent to @xmath268{dist}\mathcal n(0,\widetilde{\sigma}),\ ] ] with @xmath269 . then by applying proposition [ deltamethod ] to ( [ dec2 ] ) with @xmath270 , @xmath271 , @xmath272 and @xmath273 , we get @xmath274{dist}\mathcal n(0,\widetilde{\sigma}),\end{aligned}\ ] ] because the jacobian matrix @xmath275 . now we apply again proposition [ deltamethod ] to ( [ dec3 ] ) with @xmath276 , to get @xmath277{dist}\mathcal n\big(0,\tilde{c}(t_0)\big),\ ] ] where @xmath278 because @xmath279 . since in practice , it is more realistic to assume that a discretized trajectory of @xmath281 is observed , therefore in this section , we obtain a consistent estimator of @xmath33 starting from the high frequency @xmath280 is still available . recall that the assumptions on @xmath6 are given in section 1.1 . , then the key point leading to these results is to take a second order taylor expansion with integral remainder , to obtain for @xmath29 , @xmath282 this together with ( [ ah ] ) and the fact that @xmath283 for all @xmath45 yields @xmath284 note that ( [ ah ] ) can lead to @xmath285 it follows from ( [ taylor2 ] ) , the definition of @xmath101 and ( [ bound - dx ] ) that @xmath286 hence similar to lemma [ iv ] , we have [ ivy ] if @xmath169 , then @xmath287 as a consequence , ( a ) : : under assumption @xmath288 , @xmath289 ( b ) : : under assumptions @xmath290 , @xmath291 for @xmath83 arbitrarily small . here we note that , to show ( [ ivy2 ] ) and ( [ ivy3 ] ) hold , we must observe that under assumption @xmath288 , @xmath292 lemma [ ivy ] further implies the following : [ vyd ] for @xmath29 , denote by @xmath293 ( a ) : : under assumption @xmath288 , @xmath294 ( b ) : : under assumptions @xmath290 , @xmath295 for @xmath83 arbitrarily small . ( c ) : : under assumptions @xmath296 , @xmath297{dist}\mathcal n\big(0,\tilde{c}(t_0)\big),\ ] ] where the covariance matrix @xmath298 is given in ( [ tildec ] ) . * proof . * ( [ vyd2 ] ) and ( [ vyd3 ] ) are obvious by using lemma [ ivy ] and the fact that convergences in probability and almost surely are preserved under continuous transformations . in order to show ( [ clty1 ] ) , we only need to verify @xmath299{a.s.}0.\ ] ] equivalently , it suffices to show @xmath300{a.s.}0.\ ] ] we have , by using the mean value theorem on @xmath301 , @xmath302 where @xmath303 is some random variable valued in the open interval with ending points @xmath304 and @xmath32 . since @xmath305 tends to @xmath32 a.s . as @xmath73 , then according to ( [ vxy ] ) and * ( a1 ) * , the right - hand side of the above equation can be bounded by @xmath306 which converges to @xmath307 as @xmath73 , thanks to assumption * ( a4)*. @xmath184 the following results are of the most interests in this section . we construct consistent estimators of @xmath33 starting from the observations @xmath308 . [ vydn ] for @xmath29 , denote by @xmath309 ( a ) : : set @xmath310 $ ] , with @xmath311 . we let @xmath312 satisfy assumption * ( a1 ) * , then @xmath313 ( b ) : : let @xmath312 satisfy assumptions * ( a1)-(a2 ) * , then @xmath314 for any @xmath315 arbitrarily small . ( c ) : : under assumptions * ( a2)-(a4 ) * and @xmath316 , @xmath317{dist}\mathcal n\big(0,\tilde{c}(t_0)\big).\ ] ] * proof . * in order to prove ( [ vydn2 ] ) and ( [ vydn3 ] ) , we rely on the following relation : under assumptions * ( a1)-(a2 ) * , @xmath318 this is because , by using markov s inequality , cauchy - schwarz inequality , ( [ widev ] ) , ( [ ivy3 ] ) and the dominated convergence theorem , @xmath319 similarly to ( [ ediffv ] ) , we also obtain for any @xmath320 arbitrarily small , @xmath321 the fact that @xmath322 implies @xmath323 , then by borel - cantelli s lemma , @xmath324 therefore , ( [ vydn2 ] ) ( resp . ( [ vydn3 ] ) ) follows from the following 2 decompositions : @xmath325 @xmath326 and equations ( [ vyd2 ] ) , ( [ diffv_p ] ) ( resp . ( [ vyd3 ] ) , ( [ diffv_as ] ) ) . for showing ( [ cltyn1 ] ) , we only need to show @xmath327{\mathbb p}0.\ ] ] by using the same idea we took to prove ( [ diffh1 ] ) , we just need to verify @xmath328{\mathbb p}0.\ ] ] this is true since , according to ( [ ediffv ] ) and the fact that ( by assumption * ( a4 ) * ) @xmath329 , the left - hand side of the above term can be bounded in probability by @xmath330 the fact that @xmath316 entails @xmath331 . consequently , @xmath328{\mathbb p}0,\ ] ] hence ( [ cltyn1 ] ) holds . @xmath184 in theorem [ vydn ] , the choices of @xmath70 and @xmath16 depend on the target parameter @xmath33 . this is unacceptable from practical point of view . to overcome this inconvenience , we make assumptions * ( a1 ) , ( a4 ) * stronger so that the values of @xmath70 and @xmath16 do nt rely on @xmath33 : suppose the lower bound @xmath332}h(t)$ ] and the upper bound @xmath333}h(t)$ ] are known . * ( a1 ) * : : @xmath334 with @xmath335 ; * ( a4 ) * : : @xmath334 with @xmath336 and @xmath337 . without great effort we could see * ( a1 ) * implies * ( a1 ) * and * ( a4 ) * implies * ( a4 ) * for all @xmath29 . then from theorem [ vyd ] and theorem [ vydn ] , we easily derive the following results : [ cor2 ] ( a ) : : let @xmath338 satisfy assumption * ( a1 ) * , then @xmath339 ( b ) : : if @xmath338 satisfies assumption * ( a1 ) * , then it also satisfies * ( a2)*. as a result , @xmath340 for @xmath83 arbitrarily small . ( c ) : : under assumptions * ( a2 ) , ( a3 ) , ( a4 ) * , @xmath341{dist}\mathcal n\big(0,\tilde{c}(t_0)\big).\ ] ] [ cor1 ] ( a ) : : set @xmath310 $ ] , with @xmath311 . if @xmath312 satisfies assumption * ( a1 ) * , then @xmath342 ( b ) : : if @xmath312 satisfies assumption * ( a1 ) * , then @xmath343 for any @xmath315 arbitrarily small . ( c ) : : under assumptions * ( a2 ) , ( a3 ) , ( a4 ) * , @xmath344{dist}\mathcal n\big(0,\tilde{c}(t_0)\big).\ ] ] we conclude that the selection of @xmath345 can be made for each setting as follows : * in corollary [ cor2 ] ( a ) , for any @xmath346 , taking @xmath347 , we obtain the best rate of convergence in this setting : @xmath348 * in corollary [ cor2 ] ( b ) , similarly , taking @xmath349 obtains @xmath350 * in corollary [ cor1 ] ( a ) , the best choices of @xmath16 , @xmath351 are the ones such that @xmath352 and @xmath347 . consequently , @xmath353 * in corollary [ cor1 ] ( b ) we select @xmath354 and @xmath349 to obtain @xmath355 from the above discussion we conclude that our estimator has its best convergence rate approximately @xmath356 when the wavelet coefficients of @xmath5 are observed , and @xmath357 when a discrete sample path of @xmath5 is observed . this convergence rate is poorer than the case when @xmath128 is straightforwardly available . the estimation of the hidden pointwise hlder exponent also allows to solve stochastic volatility model s nonparametric estimation problem . for example , suppose some underlying mbm @xmath30 satisfies the following nonlinear model : for @xmath12 $ ] , @xmath359 where * @xmath360}$ ] is an mbm with unknown @xmath361)$ ] ; * @xmath4)$ ] , @xmath362 almost everywhere is an unknown real - valued deterministic function ; * only a discrete trajectory @xmath308 is available . from the statistical setting , the information of the hidden mbm remains unknown . however , the following result shows it is still possible to evaluate the parameter @xmath7 . the strategy is : first estimate the pointwise hlder exponent @xmath15 by @xmath363 given in theorem [ vydn ] , then the following result holds : [ prop : application ] fix @xmath29 and @xmath310 $ ] with @xmath17 . let @xmath364 then , ( a ) : : under assumption @xmath365 , @xmath366{\mathbb p}\theta(t_0)^2.\ ] ] ( b ) : : under assumptions @xmath365 and * ( a2 ) * , @xmath367{a.s.}\theta(t_0)^2.\ ] ] * proof of proposition [ prop : application ] . * we only show ( [ theta1 ] ) holds , since the way to obtain ( [ theta1 ] ) is similar . under assumptions * ( a1 ) * and * ( a2 ) * , on one hand , it follows from ( [ diffv_as ] ) and ( [ ivy3 ] ) that @xmath368{a.s.}2c_2(t_0).\ ] ] on the other hand , by using ( [ c2 ] ) and the fact that @xmath369 , we see @xmath370 since the functions @xmath371 and @xmath372 are both continuous over @xmath373 , therefore by combining ( [ ptheta1 ] ) , ( [ diffhath ] ) and ( [ vydn3 ] ) , we obtain proposition [ prop : application ] . @xmath184 bardet and surgailis ( 2013 ) established a pseudo - generalized least squares version of the localized increment ratio estimator and quadratic estimator of @xmath15 , denoted by @xmath374 and @xmath375 respectively . this work is so far the most achieved one on estimation of the mbm s pointwise hlder exponent . in this section we compare our model and approach to bardet and surgailis ( 2013 ) s and illustrate some simulation results . the main differences between our statistical setting and bardet and surgailis ( 2013 ) s are : 1 . bardet and surgailis ( 2013 ) considers observation of a multifractional gaussian process which has asymptotic self - similarity and `` tangent to '' an fbm with hurst parameter @xmath376 and scaled by @xmath377 at each time @xmath378 . this setting covers ours when @xmath379 . however , our model is different when @xmath6 is some other function . 2 . the estimators @xmath374 and @xmath375 are obtained in terms of observations of discrete sample path of the process . we have established estimators based on observations of discrete sample path of @xmath5 and wavelet coefficients of the process ( see @xmath380 in theorem [ vyd ] ) , respectively . the latter result is the main contribution of this paper . for more details on wavelet - based statistics we refer to delbeke and van ( 1995 ) , abry et al . ( 2002 ) , abry and conalvs ( 1997 ) . 3 . the estimators @xmath374 and @xmath375 apply to all @xmath381)$ ] with @xmath175 , however our estimators only work for @xmath382 . 4 . in both @xmath374 and @xmath375 , the length of the local window @xmath383 plays the role as @xmath384 in our setting , which partially explains the estimator s asymptotic behavior . for building up the convergence , both bardet and surgailis ( 2013 ) s and our model are subject to some technical assumptions . here we consider a particular case to compare the estimators rate of convergence . let @xmath385 , where @xmath128 has pointwise hlder exponent @xmath15 . from proposition 3 ( iii ) and ( iv ) in bardet and surgailis ( 2013 ) we see for @xmath386 and arbitrarily small @xmath387 , the estimators @xmath388 and @xmath389 have the best rate of convergence in probability and a.s . respectively : @xmath390 and @xmath391 where * * @xmath392**@xmath393 ir2 or qv2 . we set @xmath394 in corollary [ cor ] . then according to corollary [ cor ] , @xmath388 has a better rate of convergence in probability than @xmath395 , while @xmath395 outperforms @xmath389 by having a better rate of a.s . convergence . now we deliver algorithms to generate our estimators @xmath396 and @xmath397 . from theorem [ vyd ] , an algorithm to generate @xmath396 can be given as follows : + * algorithm i : estimation of @xmath398 starting from wavelet coefficients * + @xmath32 : * input : * a sample @xmath399 , @xmath332}h(t)$ ] and @xmath29 + @xmath250 : _ * initialize all the parameters : * _ + @xmath400 : @xmath401 ; @xmath402 ; + @xmath403 : _ * establish a set of indices corresponding to the neighbors of @xmath54 : * _ + @xmath404 : @xmath405 $ ] ; + @xmath406 : _ * compute `` partial '' sum of squares of the wavelet coefficients : * _ + @xmath407 : @xmath408 ; + @xmath409 : _ * compute estimate of @xmath33 : * _ + @xmath410 : @xmath411 ; + @xmath412 : * output : * @xmath396 + in algorithm i , line 3 proposes a data - driven procedure to set the parameters @xmath351 and @xmath413 . from line 7 we see that this algorithm requires at least @xmath414 operations . by this result and corollary [ cor2 ] it appears a trade - off between computational cost and precision of estimation : less value of @xmath415 reduces computational cost but leads to less precision of estimation . next we numerically study the estimator @xmath416 and compare the results with bardet and surgailis ( 2013 ) s . in view of corollary [ cor1 ] , we choose the parameters @xmath310 $ ] , @xmath417 with @xmath352 and @xmath418 . we also choose @xmath63 to be haar wavelet s mother wavelet function : @xmath419}(x)$ ] . below is an algorithm to simulate @xmath416 , according to theorem [ vydn ] . + * algorithm ii : estimation of @xmath33 starting from discretized trajectory * + @xmath32 : * input : * a sample @xmath420 , @xmath332}h(t)$ ] and @xmath29 + @xmath250 : _ * initialize all the parameters : * _ + @xmath400 : @xmath421 ; @xmath422 ; @xmath423 $ ] ; @xmath424 ; + @xmath403 : _ * establish a set of indices corresponding to the neighbors of @xmath54 : * _ + @xmath404 : @xmath425 $ ] ; + @xmath406 : _ * estimate the wavelet coefficients : * _ + @xmath407 : * for @xmath426 in @xmath427 , do : * + @xmath409 : @xmath428 ; + @xmath410 : * end for * + @xmath412 : _ * estimate `` partial '' sum of squares of the wavelet coefficients : * _ + @xmath429 : @xmath430 ; + @xmath431 : _ * compute estimate of @xmath33 : * _ + @xmath432 : @xmath433 ; + @xmath434 : * output : * @xmath416 + line 3 shows the procedure to set the parameters @xmath16 , @xmath351 , @xmath127 and @xmath312 . in line 8 , each item @xmath435 can be approximated by @xmath436 , when @xmath95 is large . hence from lines 7 - 9 of the algorithm we see that the algorithm requires at least @xmath437 operations . this fact together with the discussion on corollary [ cor1 ] reveals that the choice of @xmath415 has no impact on the computational cost , but only on the precision of the estimation . next we provide the empirical mean , the standard deviation and the quantile - quantile plots ( qq plots ) through a simulation study , where all the codes in matlab are available from authors upon request . we let @xmath438 , then choose 4 different types of @xmath6 and @xmath400 different hurst functional parameters @xmath376 , @xmath439 as follows : @xmath440 since @xmath441 for the above three @xmath15 , we then choose @xmath442 , @xmath443 , @xmath444 . we assume a single discrete trajectory : @xmath445 is available , and denote @xmath446 to be the estimate of @xmath447 : @xmath448 ( @xmath449 ) for short . by generating each estimator @xmath450 times , we present the corresponding empirical mean ( @xmath451 ) and standard deviation ( @xmath452 ) in the following table . .mean and std of the estimates @xmath416 with @xmath453 and @xmath454 . [ cols= " < , < , < , < , < , < , < " , ] note that in the function _ variair@xmath455mbm(eta , n , tot ) _ , the mbm was previously generated using the choleski decomposition of the covariance matrix , so that the sample size @xmath95 is limited to @xmath456 . however here we have generated the mbm using wood @xmath457 chan circulant matrix , some krigging and a prequantification ( see chan and wood 1998 ; barrire 2007 ) , the sample size @xmath95 can be thus taken as @xmath458 . the empirical comparison shows no significant difference between the performances of @xmath416 and @xmath459 , except that @xmath460 estimator has less variance . below we compare the probability distributions of @xmath416 versus @xmath459 , by displaying qq plots ( see fig . 1 ) . the first @xmath400 qq plots show whether @xmath416 and @xmath459 come from the same distribution for @xmath461 , @xmath462 , @xmath463 and @xmath464 ; the last one illustrates the asymptotically normal behavior of our estimator @xmath416 for @xmath463 and @xmath465 . versus @xmath459 for @xmath464 and @xmath416 versus standard normal for @xmath465 . ] the qq plots show that the probability distribution of our estimator for each @xmath33 are close to that of ir2 estimator when the observed signal process is mbm and it is asymptotically normally distributed . through the above simulation study we conclude that there is no significant difference among ir2 estimator provided in bardet and surgailis ( 2013 ) and our wavelet - based estimator . and no significant difference is observed among wavelet - based estimators corresponding to different @xmath3 . we also state that the bias and variance of @xmath466 are generally greater than ir2 estimator when the sample size is relatively small . this is because , the wavelet - based method generally provides estimators of slower convergence rate than ir2 estimator . first by using triangle inequality , we get @xmath467 recall that @xmath281 for @xmath12 $ ] . define the random variable @xmath468 to be @xmath469}|x(t)|.\ ] ] since @xmath7 is continuous and not equal to @xmath307 almost everywhere , then @xmath470}$ ] is a gaussian process with continuous trajectories , by applying dudley s theorem and borell s inequality ( more precisely , with the same arguments for the proof of @xmath471 on page 1445 - 1446 in rosenbaum ( 2008 ) . see also ledoux and talagrand ( 2010 ) ) , we can show that @xmath472 this means all of @xmath468 s moments are finite . hence , using the mean value theorem , we get @xmath473}|\phi'(s)|$ ] is a random variable . it follows from ( [ dnd1 ] ) and ( [ dnd2 ] ) that @xmath474 in order to get ( [ dnds ] ) , we need ( [ ah ] ) , from which we see there exists a positive random variable @xmath475 with all finite moments such that @xmath476 observe that for @xmath477 $ ] , @xmath478 and @xmath479 this together with the fact that @xmath480 for @xmath477 $ ] yields there exists a positive random variable @xmath481 with all finite moments such that , @xmath482 then ( [ dnds ] ) results from ( [ dnd3 ] ) and ( [ dnds1 ] ) . @xmath184 now we are going to prove ( [ dnd ] ) . for @xmath483 , we consider the @xmath484-order moment of @xmath485 in ( [ dnd3 ] ) . by applying the following two versions of jensen s inequalities : @xmath486 and cauchy - schwarz inequality , we obtain @xmath487}|\psi(s)|^r\big)2^{-j(r/2 - 1)}\nonumber\\ & & ~~\times\sum_{l=0}^{2^{n - j}-1}\int_{\frac{l}{2^n}}^{\frac{l+1}{2^n}}\big(\mathbb e(c_1^{2r})\big)^{1/2}\big(\mathbb e|x(l2^{-n}+k2^{-j})-x(t+k2^{-j})|^{2r}\big)^{1/2}{\,\mathrm{d } } t. \end{aligned}\ ] ] note that by lemma 2.12 ( i ) in ayache et al . ( 2011 ) , there exists a constant @xmath488 which does not depend on @xmath489 and @xmath1 such that @xmath490 therefore by using again ( [ neighbor ] ) and the fact that @xmath480 , there exists some constant @xmath116 such that @xmath491 using ( [ dnd5 ] ) and similar computations as in ( [ computex ] ) , we obtain there exists @xmath492 such that @xmath493 by using the fact that all the moments of gaussian variable are equivalent , we get there exists some constant @xmath494 ( only depending on @xmath484 ) such that @xmath495 finally it results from ( [ dnd4 ] ) and ( [ dnd6 ] ) that @xmath496 where @xmath497}|\psi(s)|^r\big)\big(c_3\mathbb e(c_1^{2r})\big)^{1/2}$ ] . therefore ( [ dnd ] ) has been proven . @xmath184 it follows from ( [ jensen ] ) , cauchy - schwarz inequality , ( [ card ] ) and the fact that @xmath500 that @xmath501 roughly speaking ( and it can be proven without efforts ) , since the trajectory @xmath502 is at least as smooth as @xmath503 , then for @xmath483 , there exists a constant @xmath504 ( only depending on @xmath484 ) such that , @xmath505 then it results from ( [ vnv1 ] ) , ( [ vnv2 ] ) , ( [ dnd ] ) and ( [ card ] ) that @xmath506 in view of the equivalence relation between @xmath507 and @xmath33 as @xmath215 and @xmath73 , ( [ widev ] ) finally results from ( [ vnv3 ] ) . @xmath184 by using the fact that @xmath508 , @xmath509 are zero - mean gaussian random variables , fubini s theorem , the isometry property of mbm s harmonizable presentation and a change of variables , we get @xmath510 since the pointwise hlder exponent of @xmath30 in the neighborhood of @xmath511 behave locally asymptotically like those of fractional brownian motions with hurst parameters @xmath512 : @xmath513 on @xmath514 and @xmath515 on @xmath516 , we can thus consider a taylor expansion of @xmath1 respectively on @xmath514 and @xmath517 . to be more explicit , let s fix @xmath518 and define @xmath519 , since @xmath23 belongs to @xmath520)$ ] , we take the second order taylor expansion for @xmath23 respectively on @xmath514 and @xmath521 there exist @xmath522 and @xmath523 such that @xmath524where we denote , for @xmath525 , @xmath526 , @xmath527 and @xmath528 thus we rewrite ( [ cov1 ] ) as @xmath529 where @xmath530we note here @xmath531 s and @xmath532 s are notations in short for ( [ a ] ) and ( [ a ] ) . by using ( [ cov1 ] ) , it suffices to make an identification of all the terms @xmath533 in order to estimate the covariance structure of the wavelet coefficients . we consider different cases according to the values of @xmath534 . the key to these identifications is to observe the following : * first , observe that for @xmath535 , @xmath536 , we have @xmath537 where for @xmath538 , @xmath539 , with @xmath540 being the binomial coefficient . * secondly , for @xmath541 , @xmath57 , @xmath542 , @xmath536 and @xmath543 , a @xmath544 order taylor expansion of @xmath545 on @xmath546 yields : @xmath547 where the integral remainder @xmath548 of @xmath549 s @xmath550-th order taylor expansion ( see e.g. apostol 1967 ) is given as : for @xmath551 , @xmath552 for @xmath553 , @xmath554 and the function @xmath555 here we note that for @xmath556 and @xmath557 , @xmath558 case ( i ) : : @xmath559 . + in this case we have @xmath560 let @xmath561 , @xmath562 and @xmath563 in ( [ general ] ) . it follows @xmath564then by the assumption @xmath565}\vert \frac{at - bs}{ak - bk^{\prime } } \vert \leq 1 $ ] , we can thus take @xmath566 in ( [ computei ] ) and obtain @xmath567 we finally obtain @xmath568where the function @xmath569 case ( ii ) : : @xmath570 + by definition , @xmath571 equals @xmath572 since @xmath573)$ ] , then by the triangle inequality the above item can be expressed as @xmath574 let @xmath561 and @xmath249 ( respectively corresponding to @xmath575 and @xmath32 of the above expression ) , @xmath576 and @xmath577 in ( [ general ] ) , @xmath578 in ( [ computei ] ) , we obtain @xmath579 since @xmath580 , thus we get at the meanwhile , @xmath581 using the fact that @xmath582 , we conclude @xmath583 case ( iii ) : : @xmath584 + observe that @xmath585 can be expressed as @xmath586 + let @xmath587 respectively , @xmath562 and @xmath588 in ( [ general ] ) and @xmath589 in ( [ computei ] ) , we get @xmath590 case ( iv ) : : @xmath591 + by using the fact that @xmath592)$ ] , @xmath593 can be expressed as @xmath594 let @xmath587 respectively , @xmath562 and @xmath595 in ( [ general ] ) and notice that @xmath596 depends on @xmath378 , we get @xmath597 + then similarly to ( [ computei ] ) , using a @xmath62 order taylor expansion of @xmath598 respectively on @xmath599 and on @xmath600 , and also use the fact that for @xmath601 , @xmath602 ( since @xmath603 ) , we obtain + @xmath604 + by using symmetric property , @xmath605 case ( v ) : : @xmath606 . + since the computations are quite similar as in the previous cases , we present the results without proof . @xmath607 @xmath608 @xmath609 now denote by @xmath610 it remains to show that for @xmath77 and @xmath582 , @xmath611 - 1}\sum_{k'=0}^{[b^{-1}]-1}\big(t(k , k',q , a , b)\big)^2=\mathcal o((ab)^{1/2}\log a\log b),~\mbox{as $ a , b\rightarrow0$}.\ ] ] according to ( [ tkk ] ) , it suffices to prove for any @xmath612 , @xmath613 - 1}\sum_{k'=0}^{[b^{-1}]-1}(ab)^{-h(ak)-h(bk')-1}\big(\mathcal{i}_{l , l'}(k , k',a , b)\big)^2=\mathcal o((ab)^{1/2}\log a\log b).\ ] ] to prove ( [ boundi ] ) holds we only take @xmath614 as an example , since the computations of the other items are similar . recall that @xmath615 remember that , the fact that @xmath77 yields @xmath616 , and @xmath617 $ ] . these facts together with @xmath582 imply @xmath618 - 1}\sum_{k'=0}^{[b^{-1}]-1}(ab)^{-h(ak)-h(bk')-1}\big(\mathcal i_{1,0}(k , k',a , b)\big)^2\\ & & = \mathcal o\big(ab|\log(ab)|^2\sum_{k=0}^{[a^{-1}]-1}\sum_{|l|=1}^{+\infty}\frac{(\log|l|)^2}{|l|^{4q-4\sup_{t\in[0,1]}h(t)-2}}\big)\\ & & = \mathcal o\big(a^{-1}ab|\log(ab)|^2\big)=\mathcal o\big(b|\log(ab)|^2\big)=\mathcal o\big((ab)^{1/2}\log a\log b\big).~~\square\\\end{aligned}\ ] ] when @xmath619 , we still make a taylor expansion as in the proof of ( [ prop11 ] ) . for the first term @xmath620 , since @xmath621 ( some constant ) , then using the same formula in ( [ computei00 ] ) , we get@xmath622 where @xmath623 the remaining items of @xmath624 s are of higher order so we only give a bound of them . by means of a taylor expansion of @xmath625 on @xmath626 , then similar discussion shows that , for @xmath627 , @xmath628 fix @xmath29 . it follows from the definition of @xmath114 , ( [ vxv ] ) , ( [ lemme12 ] ) and the fact that @xmath629 that @xmath630 it remains to show , for @xmath215 , @xmath631 for this purpose , for fixed @xmath632 we define @xmath633)$ ] as for @xmath634 $ ] , @xmath635 then by using the taylor expansion of @xmath636 around @xmath637 and the facts that @xmath638 , @xmath639)$ ] , @xmath361)$ ] , we obtain @xmath640 where @xmath641 is some value in @xmath642 . therefore , by using ( [ varv1 ] ) , ( [ varv2 ] ) and the fact that @xmath643 we get @xmath644 thus ( [ varv ] ) follows . now we are going to show ( [ varv ] ) . first observe @xmath645 then we recall the fact that if @xmath646 is a centered gaussian random vector , then ( see e.g. lemma 5.3.4 in peng 2011b ) @xmath647 it yields @xmath648 ( [ varv1 ] ) together with lemma [ lemma1 ] implies @xmath649 now we study the second line of ( [ varv2 ] ) . by the facts that @xmath629 , @xmath650 , @xmath651 , we obtain @xmath652 for the fourth line of ( [ varv2 ] ) , by using cauchy - schwarz inequality and proposition [ prop1 ] , we get @xmath653 therefore ( [ varv2 ] ) is equivalent to @xmath654 in order to make an identification of the dominating part of ( [ varv5 ] ) , for fixed @xmath655 we define @xmath656),~k_3\in c^1([0,1]^2)$ ] as , for @xmath657 $ ] , @xmath658 then it follows from the taylor expansion and the fact that @xmath638 that @xmath659 where @xmath180 is some value in @xmath642 and @xmath660 where @xmath661 are some values satisfying @xmath662 and @xmath663 . observe that , by a change of variable , @xmath664 then ( [ varv21 ] ) together with ( [ varv5 ] ) , ( [ varv6 ] ) , ( [ varv7 ] ) entails that @xmath665 observe that , since @xmath77 , @xmath666}^{+\infty}l^{4h(t_0)-4q}+\frac{1}{card(\nu_{t_0,2^j})}\sum_{|l|\in\nu_{t_0,2^j},l\neq0}|l|^{4h(t_0)-4q+1}\big).\nonumber\\\end{aligned}\ ] ] on one hand , we recall that if @xmath23 is a monotonic decreasing function and the series @xmath667 is convergent , then for @xmath668 , @xmath669 it yields @xmath670}^{+\infty}l^{4h(t_0)-4q}&=&\mathcal o\big(\int_{[2^j\epsilon_j]}^{+\infty}x^{4h(t_0)-4q}{\,\mathrm{d } } x+ ( 2^j\epsilon_j)^{4h(t_0)-4q}\big)\nonumber\\ & = & \mathcal o\big((2^j\epsilon_j)^{4h(t_0)-4q+1}\big).\end{aligned}\ ] ] on the other hand , since @xmath671 , then it is easy to see @xmath672 finally , @xmath673 it follows by ( [ varv8 ] ) and ( [ varv10 ] ) that @xmath674 we thus can conclude @xmath675 where @xmath676 is a constant only depending on @xmath27 . proposition [ varvx ] has been proven . @xmath184 ayache a , shieh n r , xiao y ( 2011 ) multiparameter multifractional brownian motion : local nondeterminism and joint continuity of the local times . h. poincar probab . 47 ( 4 ) : 1029 - 1054 barrire o ( 2007 ) synthse et estimation de mouvements browniens multifractionnaires et autres processus rgularit prescrite : dfinition du processus autorgul multifractionnaire et applications . dissertation , ecole centrale de nantes delbeke l , van a w ( 1995 ) a wavelet based estimator for the parameter of self - similarity of fractional brownian motion . proceedings of the 3rd international conference on approximation and optimization in the caribbean , 14 pp . ( electronic ) , benemrita univ . autn .

in chaotic reaction - diffusion systems with two degrees of freedom , the modes governing the exponential relaxation to the thermodynamic equilibrium present a fractal structure which can be characterized by a hausdorff dimension . for long wavelength modes , this dimension is related to the lyapunov exponent and to a reactive diffusion coefficient . this relationship is tested numerically on a reactive multibaker model and on a two - dimensional periodic reactive lorentz gas . the agreement with the theory is excellent . * key words:*reaction - diffusion systems ; hydrodynamic modes ; fractals ; microscopic chaos . = msbm10

cond-mat0204264

recently , many studies have been devoted to fractal structures induced by the chaotic dynamics in the phase - space of non - equilibrium statistical systems and , in particular , to the fractal character of the hydrodynamic modes governing the exponential relaxation of the system to the thermodynamic equilibrium @xcite . this fractal character has been related to the entropy production in the approach of equilibrium , for a diffusive multibaker model @xcite . previous works had shown similar results for the entropy production in non - equilibrium steady states @xcite . on the other hand , the fractal dimension characterizing the hydrodynamic modes of diffusion has been related to the diffusion coefficient and to the positive lyapunov exponent by a relation , first obtained for diffusive multibaker maps @xcite , and proved for general chaotic systems with two degrees of freedom @xcite . the construction of the hydrodynamic modes in refs . @xcite gives us a new approach to relate macroscopic transport coefficients and microscopic chaotic properties . it establishes a link between the irreversible transport processes and the reversible microscopic dynamics . two other approaches are the so - called escape rate formalism and the thermostated system approach . the first approach can be applied to open systems with absorbing boundaries @xcite . the escape of trajectories leads to the formation of a fractal repeller , consisting in the set of orbits forever trapped within the absorbing boundaries . under such conditions , the transport properties can be related to the positive lyapunov exponents and the kolmogorov - sinai entropy per unit time as well as to fractal dimensions @xcite . in the second approach , an external field is applied to the system , and in order to keep the kinetic energy constant , a special force acts on the particles as a heat pump @xcite . the system is still time - reversible but does not preserve phase - space volumes anymore . the trajectories converge to a fractal attractor . in thermostated systems , the transport properties have been related to the sum of lyapunov exponents @xcite . in the new approach of refs . @xcite , the advantage is that the fractal curve considered is directly related to the hydrodynamic modes of relaxation towards thermodynamic equilibrium . it does not require absorbing boundaries neither a thermostat . in the present paper , we extend the new approach of refs . @xcite to reaction - diffusion processes . reaction - diffusion processes are of fundamental importance to understand self - organization in physico - chemical systems @xcite . in chemical systems , the microscopic mechanisms of the relaxation toward the thermodynamic equilibrium are still poorly understood already for simple reactions . for the purpose of contributing to this important question , we consider here a simple reactive process of isomerization . the particle is supposed to perform a deterministic motion among static scatterers and to carry a color @xmath0 or @xmath1 . when colliding on some special scatterers , playing the role of catalysts , its color changes instantaneously with a given probability @xmath2 , @xmath3 for this class of systems , we obtain an expression of the hausdorff dimension of the reactive modes of relaxation in terms of the reactive dispersion relation and of a function @xmath4 generalizing the ruelle topological pressure @xmath5 . in the long wavelength limit , we then infer a relation between the hausdorff dimension of the modes , the reactive diffusion coefficient , the reaction rate , and two derivatives of the function @xmath4 . in the limit @xmath6 , we recover a relation previously derived for the diffusive case @xcite . this new expression relates the macroscopic transport and reaction processes to the microscopic underlying dynamics . the paper is organized as follows . in section [ theory ] , we generalize the work of refs . @xcite to the reactive case considered here . in section [ triadic ] , we test our relation on a reactive multibaker model . we next study the case of a two - dimensional periodic reactive lorentz gas in section [ lorentz ] . conclusions are drawn in section [ conclusions ] . we consider a system such as a lorentz gas in which the particles move independently of each other along trajectories of a deterministic dynamical system given by @xmath7 the variables of this system are for instance the positions and velocities of the particle : @xmath8 . the vector field @xmath9 is defined in this space . the system is spatially periodic in the positions @xmath10 . the particle moves in a periodic lattice of scatterers given by a certain potential of interaction . moreover , each particle carries a color which is either @xmath0 or @xmath1 . the color changes when the particle interacts with some of the scatterers , called the catalytic scatterers according to the reaction ( [ scheme ] ) . we notice that we may also consider that the particle has a spin one - half which flips upon collision on special scatterers . we can model the reactive events by a change of color with probability @xmath2 when the trajectory meets a certain hypersurface @xmath11 defined in the space of the variables @xmath12 . this hypersurface surrounds each catalytic scatterer and constitutes the locus where the reaction occurs . the introduction of the color has for consequence that the phase space is composed of two copies of the space of the variables @xmath12 , which are glued together along the hypersurface @xmath13 . this hypersurface can be viewed as the gate from copy corresponding to the color @xmath0 to the one corresponding to color @xmath1 and vice versa . these two copies now form two subspaces of the whole phase space of the reactive system . in this framework , a reactive event corresponds to the crossing of the hypersurface @xmath13 from one color subspace to the other . a statistical ensemble of particles of both colors is described by two probability densities @xmath14 and @xmath15 , to find a particle at the point @xmath12 with color @xmath16 or @xmath1 , at the current time @xmath17 . the time evolution of these probability densities is ruled by two coupled liouville equations : @xmath18 = - p_0\ , \xi \left[\sigma({\bf x})\right ] \left [ q({\bf x},a , t)-q({\bf x},b , t)\right]\ , \cr \partial_t \ ; q({\bf x},b , t ) + { \rm div}\left[{\bf f}({\bf x } ) \ ; q({\bf x},b , t ) \right ] = + p_0\ , \xi \left[\sigma({\bf x})\right ] \left [ q({\bf x},a , t ) -q({\bf x},b , t)\right]\ . \cr } \label{liouville}\ ] ] where @xmath19 is a special distribution defined as @xmath20 \ h({\bf x } ) \ \equiv \ \lim_{\epsilon \to 0^+ } \ \left|{\bf f}({\bf x})\cdot \frac{\partial\sigma}{\partial{\bf x } } ( { \bf x})\right| \ \delta\left[\sigma({\bf x})\right ] \ x}-\epsilon { \bf f}({\bf x } ) \right ] \label{distrib}\ ] ] @xmath21 being the dirac distribution . the distribution ( [ distrib ] ) expresses the fact that the particles change their color upon crossing of the hypersurface @xmath13 . accordingly , the probability densities just after the hypersurface @xmath13 in the direction of the vector field @xmath22 are the following linear combinations of the densities just before @xmath13 : @xmath23 where @xmath24 with @xmath25 and an arbitrarily small @xmath26 . these linear combinations can be expressed in terms of the transmission matrix @xmath27 we denote by @xmath28 the trajectory of the differential equation ( [ diffeq ] ) from the initial condition @xmath29 . moreover , we gather both probability densities in the vector @xmath30 the solution of the coupled liouville equations ( [ liouville ] ) from the initial probability densities @xmath31 can then be written in the form of the following frobenius - perron operator : @xmath32 with the matrix ( [ transitionmatrix ] ) and where @xmath33 is the number of crossings through the hypersurface @xmath13 performed by the segment of trajectory from @xmath34 to @xmath28 . if the flow of eq . ( [ diffeq ] ) is volume preserving , we have that @xmath35 which implies that @xmath36 , which is the case for the systems considered in the following . the system of liouville s equations ( [ liouville ] ) defines a random process and it is important to characterize the dynamical randomness generated by this process . for this purpose , we calculate its kolmogorov - sinai entropy per unit time which is the amount of information produced per unit time by the process . the kolmogorov - sinai entropy is calculated by partitioning the whole phase space into small cells and by computing the probability for a trajectory to visit successively different cells of the partition @xcite . if we suppose that the differential equation ( [ diffeq ] ) defines a chaotic dynamical system with lyapunov exponents @xmath37 , the dynamical instability of the deterministic motion is a first contribution to the kolmogorov - sinai entropy given by the sum of positive lyapunov exponents according to pesin s theorem @xcite . a second contribution comes from the random change of color upon crossing of the hypersurface @xmath13 . at each crossing , the particle jumps from one color subspace to the other with probability @xmath2 . if the number of crossings is equal to @xmath38 during the lapse of time @xmath17 , the frequency of @xmath13 crossings is equal to : @xmath39 where @xmath40 denotes the average with respect to the equilibrium state . at each crossing , we have an independent dichotomous random variable with probabilities @xmath41 , which defines a bernoulli process with the rate @xmath42 . accordingly , the overall kolmogorov - sinai entropy per unit time of the process ( [ liouville ] ) is given by @xmath43 . \label{ks}\ ] ] pesin s expression of the kolmogorov - sinai entropy is recovered in the deterministic limits @xmath44 and @xmath45 . for @xmath46 , we remark that the kolmogorov - sinai entropy ( [ ks ] ) remains positive and finite as for a fully deterministic dynamical system . the way the reaction is modelled here is therefore compatible with a deterministic dynamics that would take place near each catalytic scatterer on a time scale much shorter than the free flight between each scatterer . we can imagine that the reaction follows a complicated trajectory on such a short time scale in a complicated potential , which is here simplified by capturing the complexity of the reaction into a sole transition probability @xmath2 . still the so - defined random process has a finite kolmogorov - sinai entropy contrary to stochastic processes which have an infinite kolmogorov - sinai entropy per unit time . indeed , it has been shown elsewhere @xcite that stochastic processes such as the birth - and - death processes or the linearized boltzmann processes have a dynamical randomness on arbitrarily small scales @xmath47 with an @xmath47-entropy per unit time increasing as @xmath48 . stochastic processes defined by langevin stochastic differential equations have an even larger dynamical randomness with an @xmath47-entropy per unit time increasing as @xmath49 . the kolmogorov - sinai entropy is the supremum of the @xmath47-entropy in the limit @xmath50 and is therefore infinite . in contrast with such stochastic processes , the present random process has a finite kolmogorov - sinai entropy . this dynamical randomness is comparable with the randomness of the classical dynamics on graphs which has been recently characterized @xcite . for the classical graphs , the motion is one - dimensional so that the only source of randomness is the random bifurcations at the vertices of the graph . here , there is an extra source of randomness which is the dynamical instability due to the chaotic motion of the particle in the multi - dimensional phase space of eq . ( [ diffeq ] ) . the overall dynamical randomness ( [ ks ] ) induces a relaxation of the probability densities toward the thermodynamic equilibrium . in order to study this relaxation we have to find the relaxation rates which are given by the eigenvalues of the frobenius - perron operator ( [ fpmatrix ] ) . it turns out that the liouville equations ( [ liouville ] ) can be decoupled if we introduce the probability density @xmath51 to find a particle at the point @xmath12 no matter which is its color , and the difference @xmath52 between both color densities : @xmath53 these new functions are gathered in the vector @xmath54 which is related to the previous vector ( [ qaqb ] ) by the relation @xmath55 in terms of the matrix @xmath56 this matrix diagonalizes the transition matrix ( [ transitionmatrix ] ) into @xmath57 accordingly , the liouville equations ( [ liouville ] ) decouple into @xmath58 = 0\ , \cr \partial_t \ ; g({\bf x},t ) + { \rm div}\left[{\bf f}({\bf x } ) \ ; g({\bf x},t ) \right ] = -2p_0\ , \xi \left[\sigma({\bf x})\right ] g({\bf x},t)\ . \cr } \label{newliouville}\ ] ] which shows that the probability density @xmath51 obeys the deterministic liouville equation while the difference @xmath52 solely controls the reaction . similarly , the frobenius - perron operator ( [ fpmatrix ] ) splits in the two following operators : @xmath59 an eigenvalue problem can now be considered separately for both operators ( [ opdiff ] ) and ( [ opreac ] ) . if we denote them generically by @xmath60 , we are looking for eigenstates given by some distributions which are quasiperiodic in the position space @xmath10 with a wavenumber @xmath61 : @xmath62 where @xmath63 is periodic in @xmath10 . the wavenumber @xmath61 describes spatial inhomogeneities of wavenumber @xmath64 . the distribution ( [ eigen ] ) is supposed to be a solution of the eigenvalue problem @xmath65 associated with the operator @xmath60 . the quantity @xmath66 is the relaxation rate of the eigenstate ( [ eigen ] ) . the idea is that these eigenstates can be obtained in principle by applying the operator @xmath60 over a long time to some expression ( [ eigen ] ) and by renormalizing the result in order to compensate the relaxation according to @xmath67 \label{limit}\ ] ] even starting from some arbitrary function @xmath68 . this renormalization - semigroup idea has been developed in detail for diffusion in refs . @xcite . for chaotic dynamical systems ( [ diffeq ] ) , such eigenstates turn out to be singular distributions without a density function but with cumulative functions defined by @xmath69\ ] ] where @xmath70 is a parameter ( such as an angle ) of a curve @xmath71 defined in the phase space . the hydrodynamic modes of diffusion as well as the reactive modes can be identified as the eigenstates corresponding to the slowest relaxation rates and they can be constructed thanks to the above idea of renormalization semigroup , as we shall show below . in order to construct a diffusive mode of wavenumber @xmath72 we consider the frobenius - perron operator ( [ opdiff ] ) and an initial distribution @xmath73 of particles of initial positions @xmath74 . we integrate the trajectories by using the equations of motion ( [ diffeq ] ) in order to obtain the position @xmath75 after a long time @xmath17 . as a consequence of the aforementioned idea of renormalization semigroup , the relaxation rate of the diffusive mode is given by the decay rate of the van hove intermediate incoherent scattering function @xcite as @xmath76\rangle \label{vanhove}\ ] ] where @xmath77 is an average over initial conditions distributed according to some distribution @xmath68 @xcite . in a similar way , the cumulative function of the diffusive mode can be obtained as @xmath78\ } } { \int_0^{2\pi}d\theta'\ \exp\{i{\bf k}\cdot[{\bf r}_t(\theta')- { \bf r}_0(\theta')]\ } } \label{cumul}\ ] ] where the initial conditions are distributed according to a distribution @xmath68 which is here taken as a uniform distribution along a one - dimensional line parametrized by the angle @xmath79 @xcite . for two - degree - of - freedom chaotic systems with one positive lyapunov exponent @xmath80 , the cumulative function ( [ cumul ] ) forms a fractal curve in the complex plane and its hausdorff dimension is given in terms of the diffusion coefficient and the lyapunov exponent according to @xmath81 as shown in refs . the derivation of ref . @xcite is here summarized because we need it for the generalization to the reactive modes . we assume that we can use the result by sinai , bowen and ruelle that averages can be performed in terms of unstable trajectories covering the phase space within a certain resolution @xcite . the probability weight given to each of these trajectories is inversely proportional to their stretching factor @xmath82 which characterizes their dynamical instability . accordingly , eq . ( [ vanhove ] ) can be transformed into the condition @xmath83\}\sim_{t \to \infty}\ 1 \ . \label{ratesum}\ ] ] if the sum in eq . ( [ ratesum ] ) is restricted to the trajectories issued from initial conditions in the interval @xmath84 $ ] , we obtain at time @xmath17 a polygonal approximation of the cumulative function ( [ cumul ] ) : the integral @xmath85 is an average over trajectories with initial conditions in @xmath84 $ ] and the denominator is proportional to the factor @xmath86 . at time @xmath17 , the curve @xmath87 is thus approximated by a polygon of sides given by the small complex vectors @xmath88\right\ } \;.\ ] ] each side has the length @xmath89 in the limit @xmath90 , this polygon converges to a fractal curve , characterized by a hausdorff dimension given by @xmath91 . accordingly , the hausdorff dimension of the hydrodynamic mode should satisfy the condition @xmath92 on the other hand , ruelle s topological pressure is defined in dynamical systems theory by @xmath93 where the average @xmath77 is carried out over the equilibrium invariant measure @xcite . the mean positive lyapunov exponent of the system is given by @xmath94 and in this system without escape , @xmath95 . equation ( [ press ] ) can be transformed into @xmath96 \sim_{t\to\infty } 1 \ ; \label{press2}.\ ] ] comparing with eq . ( [ ratesum ] ) , we obtain @xmath97 if the wavenumber @xmath72 vanishes , the cumulative function ( [ cumul ] ) becomes @xmath98 , which forms a straight line in the complex plane . in this equilibrium limit , the relaxation rate vanishes , @xmath99 , so that @xmath100 . it means for this system without escape that @xmath101 , as it should be for a straight line . for a non - vanishing but small wavenumber , the hausdorff dimension is expected to deviate from unity . inserting @xmath102 and the dispersion relation @xmath103 in eq . ( [ result ] ) , we can expand both sides in powers of the wavenumber , using the aforementioned properties of the topological pressure . this straightforward calculation shows that the hausdorff dimension of the hydrodynamic mode is given by eq . ( [ diff ] ) @xcite . we are now going to extend this result to the reactive modes of the isomerization process ( [ scheme ] ) . the modes of interest here are the ones describing the evolution of the difference between the two colors @xmath0 and @xmath1 , a quantity which relax to zero when the system reaches equilibrium . in this case , we consider the operator ( [ opreac ] ) which differs from the diffusive frobenius - perron operator ( [ opdiff ] ) by the extra factor @xmath104 . accordingly , this factor should multiply the exponential @xmath105 in the expression ( [ vanhove ] ) in order to obtain the relaxation rate of the reactive modes , which is thus given by @xmath106\rangle \label{vanhovereac}\ ] ] where @xmath38 is the number of collisions performed on catalysts during the time @xmath17 @xcite . the cumulative function of the reactive modes are similarly given by @xcite @xmath107\}}{\int_0^{2\pi}d\theta'\ ( 1 - 2p_0)^{n_t(\theta ' ) } \exp\{i{\bf k}\cdot[{\bf r}_t(\theta')-{\bf r}_0(\theta')]\ } } \ ; . \label{cumulreac}\ ] ] we notice that eqs . ( [ vanhovereac ] ) and ( [ cumulreac ] ) consistently reduce to eqs . ( [ vanhove ] ) and ( [ cumul ] ) in the purely diffusive limit @xmath108 . as for the diffusive case , eq . ( [ vanhovereac ] ) can be transformed into the condition @xmath109\}\sim_{t \to \infty}\ 1 \ ; . \label{ratesumreac}\ ] ] the curve @xmath87 is approximated at time @xmath17 by a polygon of sides given by @xmath110 in the limit @xmath90 , this polygon converges to a fractal curve , characterized by a hausdorff dimension given by @xmath111 . accordingly , the hausdorff dimension of the reactive mode should satisfy the condition @xmath112 by analogy with eq . ( [ press2 ] ) , we can define a function @xmath4 by @xmath113\ ; \sim_{t\to\infty } 1 \ ; \label{q2}\ ] ] which can be rewritten as @xmath114 comparing eq . ( [ q2 ] ) with eq . ( [ ratesumreac ] ) , we obtain @xmath115 the relation between this new function and the topological pressure is given by @xmath116 if the wavenumber @xmath72 vanishes , the cumulative function ( [ cumulreac ] ) is reduced to @xmath117 this is not a straight line as in the diffusive case : each initial condition is here weighted by a factor @xmath104 . provided that @xmath118 , this means that the contribution of a given initial condition to @xmath119 will be smaller if the corresponding trajectory meets many catalysts , and vice versa . but the hausdorff dimension of @xmath119 is anyway equal to unity . indeed , for @xmath120 , the relaxation rate is given by @xmath121 which is also equal to @xmath122 according to the definition ( [ qdef ] ) . using these equalities , we can deduce from ( [ resultreac ] ) that @xmath101 for vanishing wavenumber . again the hausdorff dimension will deviate from unity for a non - vanishing wavenumber . the reactive dispersion relation is given by @xmath123 . if we compare it with ( [ s0 ] ) at vanishing wavenumber , we obtain @xmath124 using this equality , the dispersion relation and the expression @xmath125 , eq . ( [ resultreac ] ) can be expanded in powers of the wavenumber , giving an expression of the hausdorff dimension of the reactive mode @xmath126 an expression for the two partial derivatives of @xmath127 can be directly obtained from eq . ( [ qdef ] ) , provided that @xmath128 : @xmath129 and @xmath130 similarly , the reactive diffusion coefficient is given by @xmath131 where @xmath132 is the dimension of the space of positions @xmath10 where diffusion occurs . let us now consider the limit case @xmath133 . the three terms of the denominator appearing in eq . ( [ hausdim ] ) can be expanded in powers of @xmath2 . the reaction rate is given by @xmath134\nonumber\\ \iff 2 \kappa&= & 2 \ ; \nu\ ; p_0 + 2 \ ; ( \nu-\sigma^2)\ ; p_0 ^ 2 + 4 ( \frac{2}{3}\ ; \nu\ ; - \ ; \sigma^2 \ ; + \tau ) \ ; p_0 ^ 3 + { \mathcal o}(p_0 ^ 4 ) \label{reacrate2}\end{aligned}\ ] ] in terms of the frequency ( [ freq ] ) of collisions on the catalysts @xmath42 and of the quantities @xmath135 the two further terms are given by @xmath136\nonumber \\ \iff \frac{\partial q}{\partial \alpha}(1,1)&=&-2\kappa+2\;\sigma^2\;p_0 ^ 2 + 4\;(\sigma^2\;-2\;\tau)\;p_0 ^ 3 + { \mathcal o}(p_0 ^ 4 ) \label{qa}\end{aligned}\ ] ] and @xmath137 \nonumber \\ & + & 4\;p_0 ^ 3\;\lim_{t \to \infty}\;\frac{1}{t}\left[\left(\frac{2}{3 } + 2\langle n_t \rangle+2\;\langle n_t \rangle^2 -\langle n_t^2 \rangle\right)\;c_{\kappa \lambda}(t)\right.\nonumber \\ & & \left . -(1+\langle n_t \rangle)\;d_{\kappa \lambda}(t ) + \frac{1}{3}e_{\kappa \lambda}(t)\right ] + { \mathcal o}(p_0 ^ 4 ) \nonumber \\ \iff \frac{\partial q}{\partial \beta}(1,1)&=&-\lambda + \alpha_{\kappa \lambda}p_0+\beta_{\kappa \lambda}p_0 ^ 2 + \gamma_{\kappa \lambda}p_0 ^ 3+{\mathcal o}(p_0 ^ 4 ) \label{qb}\end{aligned}\ ] ] where @xmath138 \label{beta}\\ \gamma_{\kappa \lambda}&=&4\;\lim_{t \to \infty}\;\frac{1}{t } \left[\left(\frac{2}{3}+2\langle n_t \rangle+2\;\langle n_t\rangle^2 -\langle n_t^2 \rangle\right)\;c_{\kappa \lambda}(t)\right . \nonumber \\ & & \left . n_t \rangle)d_{\kappa \lambda}(t ) + \frac{1}{3}e_{\kappa \lambda}(t)\right]\ ; \label{gkl}.\end{aligned}\ ] ] for these coefficients , the index @xmath139 refers to the reaction , related to the number of collisions on a catalyst @xmath38 , while the index @xmath80 refers to the dynamical instability , described by the stretching factor @xmath140 and the lyapunov exponent @xmath80 . the reactive diffusion coefficient ( [ reacdiffcoeff ] ) can also be expanded in powers of the reaction probability @xmath2 as @xmath141 in terms of the standard diffusion coefficient @xmath142 and of the coefficient @xmath143 \label{delta}\ ] ] which characterizes the correlation between the transport by diffusion and the reaction . we notice that , in the limit @xmath108 , the reaction rate ( [ reacrate2 ] ) vanishes while the reactive diffusion coefficient ( [ reacdiffcoeff ] ) tends to the diffusion coefficient ( [ diffcoeff ] ) , as expected . using ( [ qa ] ) and ( [ qb ] ) , the hausdorff dimension ( [ hausdim ] ) becomes @xmath144 } + { \mathcal o}(k^4 ) \nonumber \\ & = & 1+\frac{{\cal d}^{\rm ( r)}k^2}{\lambda}\;\left[1 + \frac{\alpha_{\kappa \lambda}}{\lambda}\;p_0 + 2\;\frac{\sigma^2}{\lambda}\;p_0 ^ 2 + \frac{\beta_{\kappa \lambda}}{\lambda}\;p_0 ^ 2 + \frac{\alpha_{\kappa \lambda}^2}{\lambda^2}\;p_0 ^ 2 + 4\;\frac{\sigma^2}{\lambda}\;p_0 ^ 3 -8\;\frac{{\tau}}{\lambda}\;p_0 ^ 3 \right.\nonumber \\ & & \left . + \frac{\gamma_{\kappa \lambda}}{\lambda}\;p_0 ^ 3 + 2\;\frac{\alpha_{\kappa \lambda}\beta_{\kappa \lambda}}{\lambda^2}\;p_0 ^ 3 + 4\;\frac{\alpha_{\kappa \lambda}\sigma^2}{\lambda^2}\ ; p_0 ^ 3 + \frac{\alpha_{\kappa \lambda}^3}{\lambda^3}\ ; p_0 ^ 3 + { \mathcal o}(p_0 ^ 4)\right]+{\mathcal o}(k^4 ) \ ; . \label{hausdim2}\end{aligned}\ ] ] or @xmath145\;k^2 + { \mathcal o}(k^4)\ ; .\ ] ] we immediately see that for @xmath6 , eq . ( [ hausdim2 ] ) reduces to the expression ( [ diff ] ) obtained for the diffusive case . we are now going to test this formula for two examples : a reactive triadic multibaker and a reactive two - dimensional periodic lorentz gas . the multibaker map is a simple model used to mimic a diffusive motion @xcite . in the case of the triadic symmetric multibaker , we consider an infinite one - dimensional chain of squares , @xmath146 where @xmath147 is the label of the square . a point particle performs jumps from square to square , according to the transition rule @xmath148 if @xmath149 , @xmath150 if @xmath151 and @xmath152 if @xmath153 . in order to introduce a reaction , the point particle carries a color @xmath0 or @xmath1 , and the central part @xmath154\frac{1}{3},\frac{2}{3}]$ ] of one square over @xmath155 is catalytic : if the point particle is in this portion of the square , it changes its color with a probability @xmath2 ( see fig . [ multi ] ) . the map of the model is thus given by @xmath156 where @xmath157 is a discrete variable taking the values @xmath158 or @xmath159 when the color of the point particle is respectively @xmath0 or @xmath1 , and @xmath160 we consider the probability density @xmath161 which evolves under the frobenius - perron equation @xmath162\ ; q_t\left(\frac{x+1}{3},3y-1,l , c\right)+p_0(l)\ ; q_t\left(\frac{x+1}{3},3y-1,l ,- c\right)\ ; , & $ 1/3 < y \le 2/3$\cr q_t\left(\frac{x+2}{3},3y-2,l-1,c\right)\ ; , & $ 2/3 < y \le 1$\cr}\ ] ] where @xmath163 the kolmogorov - sinai entropy per iteration ( [ ks ] ) of the reactive triadic multibaker is given by @xmath164 because the frequency ( [ freq ] ) is equal to the ratio of the area of a reactive rectangle to the area of @xmath155 squares : @xmath165 . we notice that the result ( [ ksbak ] ) can also be obtained by using a generating partition which establishes the equivalence between the reactive multibaker map ( [ bak ] ) and a markov chain ( see ref . @xcite ) . if we define the difference between the densities of the colors @xmath0 and @xmath1 , @xmath166 we obtain the following equation for its evolution @xmath167\ ; g_t\left(\frac{x+1}{3},3y-1,l\right)\ ; , & $ 1/3 < y \le 2/3 $ \cr g_t\left(\frac{x+2}{3},3y-2,l-1\right)\ ; , & $ 2/3 < y \le 1 $ \cr } \label{bakop}\ ] ] which is the form of the reactive operator ( [ opreac ] ) in the case of the present multibaker model . we restrict our description to a unit cell of length @xmath155 , @xmath168 being the catalytic square , and we choose quasiperiodic boundary conditions , assuming that the solution of the equation ( [ bakop ] ) is quasiperiodic on the chain with a wavenumber @xmath169 . moreover , we suppose exponentially decaying solutions , with decaying factor @xmath170 where @xmath171 or @xmath172 @xmath173 the following operator @xmath174 governs the time evolution of @xmath175 @xcite : @xmath176 we now want to obtain the eigenvalues and the eigenstates of this operator @xmath177 we suppose that the leading eigenstates are uniform along the unstable direction @xmath178 @xmath179 if we now cumulate over @xmath178 and @xmath180 , we get @xmath181 where @xmath182 is the reactive cumulative function ( [ cumulreac ] ) for the multibaker . we replace @xmath175 by @xmath183 and @xmath184 by @xmath185 in eq . ( [ operator ] ) and we integrate over the interval @xmath186 $ ] . we obtain @xmath187\ ; , & $ 1/3 < y \le 2/3$\cr \frac{1}{3\chi}\;[c(1,1)+(1 - 2p_0)c(1,0)+e^{-ikl}c(3y-2,l-1 ) ] \ ; , & $ 2/3 < y \le 1$\cr } \nonumber\\ c(y,1 ) & = & \cases { \frac{1}{3\chi}\;c(3y,2 ) \ ; , & $ 0\le y \le 1/3$\cr \frac{1}{3\chi}\;[c(1,2)+c(3y-1,1 ) ] \ ; , & $ 1/3 < y \le 2/3$\cr \frac{1}{3\chi}\;[c(1,2)+c(1,1)+c(3y-2,0 ) ] \ ; , & $ 2/3 < y \le 1$\cr } \nonumber\\ & \vdots & \nonumber \\ c(y , l-2 ) & = & \cases { \frac{1}{3\chi}\;c(3y , l-1 ) \ ; , & $ 0\le y \le 1/3$\cr \frac{1}{3\chi}\;[c(1,l-1)+c(3y-1,l-2 ) ] \ ; , & $ 1/3 < y \le 2/3$\cr \frac{1}{3\chi}\;[c(1,l-1)+c(1,l-2)+c(3y-2,l-3 ) ] \ ; , & $ 2/3 < y \le 1$\cr } \nonumber\\ c(y , l-1 ) & = & \cases { \frac{e^{ikl}}{3\chi}\;c(3y,0 ) \ ; , & $ 0\le y \le 1/3$\cr \frac{1}{3\chi}\;[e^{ikl}\;c(1,0)+c(3y-1,l-1 ) ] \ ; , & $ 1/3 < y \le 2/3$\cr \frac{1}{3\chi}\;[e^{ikl}\;c(1,0)+c(1,l-1)+c(3y-2,l-2 ) ] \ ; , & $ 2/3 < y \le 1$\cr}. \label{cequ}\end{aligned}\ ] ] the eigenvalues are obtained by setting @xmath188 in eq . ( [ cequ ] ) , which leads to the eigenvalue equation @xmath189 we thus have to solve the determinant of an hermitian matrix to find @xmath190 . for this purpose , we assume a solution of the form @xmath191 after substitution in the system ( [ eigenequ ] ) , we find that @xmath192 and @xmath193 for the present reactive system , the relaxation rate @xmath194 depends on the wavenumber @xmath169 as @xmath195 , where @xmath139 is the reaction rate and @xmath196 is the reactive diffusion coefficient . solving eq . ( [ determinant ] ) for its leading root , we can thus obtain the reaction rate and the reactive diffusion coefficient either in powers of the transition probability @xmath2 : @xmath197 or in the asymptotic limit @xmath198 : @xmath199 in the limit @xmath108 , eq . ( [ ratep0 ] ) shows that the reaction rate vanishes linearly with the transition probability and eq . ( [ diffreacp0 ] ) that the reactive diffusion coefficient tends to the purely diffusive one @xmath200 , as expected . on the other hand , eq . ( [ ratel ] ) shows that the reaction rate decreases as @xmath201 when the catalytic sides are diluted in the limit @xmath198 , as expected for a diffusion - controled reaction in a one - dimensional system @xcite . moreover , the vanishing of the reactive diffusion coefficient ( [ diffreacl ] ) as @xmath202 for @xmath198 is consistent with the results previously obtained in ref . @xcite . we notice the remarkable property that , in the diffusion - controled regime of the reaction , the leading term of the asymptotic expression for the reaction rate ( [ ratel ] ) does not depend on the transition probability @xmath2 , which only appears in the next terms . therefore , the leading term of ( [ ratel ] ) no longer expresses the property that the reaction rate vanishes in the limit @xmath108 , as shown by eq . ( [ ratep0 ] ) . this is consistent with the fact that the reaction rate is largely independent of the local reaction rate at the catalysts in the diffusion - controled regime , as already noticed in refs . @xcite . let us consider more explicitly the simple case @xmath203 . in this case , eq . ( [ eigenequ ] ) reduces to @xmath204 which gives the solution @xmath205 for @xmath206 , @xmath190 is then given by @xmath207 expanding ( [ chil1 ] ) in powers of @xmath169 up to the second degree , we obtain @xmath208 which allows us to identify @xmath209 and @xmath210 . the dependence of the reactive diffusion coefficient @xmath196 on @xmath2 is shown in fig . [ drvsl ] for different values of the chain length @xmath155 . the value of the diffusion coefficient @xmath211 is indicated for comparison . we now want to study the cumulative functions @xmath212 , solutions of eq.([cequ ] ) , and more specifically to calculate their dimension . these functions are represented in the complex plane in the case @xmath203 in figs . [ modevsp ] and [ modevsk ] . [ modevsp ] shows their dependence on the reaction probability @xmath2 , the wavenumber being fixed , @xmath213 . their dependence on the wavenumber for @xmath214 is shown in fig . [ modevsk ] . as explained in section [ theory ] , these complex cumulative functions can be approximated by a polygon at a given time @xmath17 , which means at a certain iteration step @xmath215 in this discrete case . this concept will here be applied using a symbolic dynamics . the symbolic dynamics is constructed by partitioning the phase space of the multibaker maps depicted in fig . [ multi ] . each of the @xmath155 squares of the chain is divided in three parts for @xmath216 $ ] , @xmath217\frac{1}{3},\frac{2}{3}]$ ] or @xmath217\frac{2}{3},1]$ ] . these @xmath218 rectangles are successively assigned the symbols @xmath219 from left to right along the chain . at the @xmath220 step of the construction described in subsection [ reacmodes ] , the cumulative function is a polygon in the complex plane with its sides corresponding to a subrectangle associated with a sequence of @xmath215 symbols @xmath221 . moreover , we notice that the sequence of symbols also determines the square @xmath222 where the point is located along the chain . at the next steps of the construction , the cumulative function keeps its values already obtained at the borders of each of the aforementioned subrectangles . we can thus denote the values of the cumulative function at the lower and upper borders @xmath223 of one of the subrectangles in the square @xmath222 more shortly by @xmath224.\ ] ] the complex vector defining the side of the polygon at the @xmath220 step of the construction is thus given by @xmath225 the succession of two symbols @xmath226 corresponds to a transition in the multibaker phase - space , with which a certain probability is associated . using ( [ cequ ] ) , we can construct a @xmath227 transition matrix @xmath228 of elements @xmath229 describing the factor connecting the portion of square corresponding to @xmath230 to the one corresponding to @xmath231 . using this matrix , we can describe the time evolution of the intervals @xmath232 as @xmath233 we define @xmath234 where the matrix @xmath235 is composed of the elements : @xmath236 supposing that this matrix admit the spectral decomposition , @xmath237 , eq . ( [ gamma ] ) can be rewritten as @xmath238 where @xmath239 denotes the vector of unit elements @xmath240 and @xmath241 the vector with the elements @xmath242 . according to the definition of the hausdorff dimension , for @xmath243 , the quantity ( [ regamma ] ) must converge to a finite non - vanishing value if @xmath244 . this will only be the case if @xmath245 . @xmath246 is thus solution of the equation obtained by imposing @xmath247 as eigenvalue of the matrix @xmath248 . let us consider explicitly the case @xmath203 . we have to calculate the @xmath249 determinant @xmath250 this gives the equation @xmath251 we will solve ( [ bakdh ] ) in the limit case @xmath6 , using the expression of @xmath252 in the case @xmath203 ( [ chil1 ] ) . for @xmath44 , we have @xmath253 . on the other hand , the hausdorff dimension takes the unit value for @xmath254 . substituting the expansion @xmath255 in eq . ( [ bakdh ] ) and expanding in powers of the wavenumber @xmath169 , we can solve for the unknown coefficient @xmath256 and obtain @xmath257 now expanding in powers of the transition probability @xmath2 , we finally obtain @xmath258 we can again compare this expression with the general result ( [ hausdim2 ] ) of section [ theory ] . for the triadic multibaker , the lyapunov exponent is given by @xmath259 . moreover , the stretching factor @xmath260 is constant in the triadic multibaker , which implies that the coefficients @xmath261 , @xmath262 , @xmath263 and therefore @xmath264 , @xmath265 , @xmath266 defined in eqs . ( [ ckl])-([gkl ] ) are vanishing . whereupon , eq . ( [ hausdim2 ] ) is here reduced to @xmath267 + { \mathcal o}(k^4)\ ; . \label{hausdim3}\ ] ] using the expansion of the reactive diffusion coefficient up to the third order in @xmath2 @xmath268\ ; , \ ] ] eq . ( [ hausdim3 ] ) becomes @xmath269 the second term of the right members of ( [ bakdim ] ) and ( [ hausdim4 ] ) are equal . they correspond to @xmath270 . let us now compare the other corrections in @xmath2 . the only correction in @xmath271 in ( [ hausdim4 ] ) is @xmath272 . we can evaluate @xmath273 in the case of the triadic reactive multibaker @xmath203 . indeed here , colliding on a catalyst corresponds to being in the region @xmath274\frac{1}{3},\frac{2}{3}]$ ] . the mean number of collisions on a catalyst per unit time is then @xmath275 and @xmath276 which finally gives us @xmath277 the coefficient of the correction in @xmath271 is then equal to @xmath278 , which is exactly the coefficient obtained in ( [ bakdim ] ) . the correction in @xmath279 in ( [ hausdim4 ] ) contains three different contributions : @xmath280 . again we can evaluate @xmath281 @xmath282 we obtain @xmath283 , which is the coefficient expected for the correction in @xmath279 . accordingly , we have confirmed that eq . ( [ hausdim2 ] ) for the hausdorff dimension of the reactive modes is valid for the reactive multibaker map . the validity of the approach is furthermore tested in fig . [ dhvsk2 ] , where the hausdorff diemansion @xmath246 is plotted versus the square of the wavenumber @xmath284 . the symbols are obtained numerically by solving eq . ( [ bakdh ] ) and compared with the expression ( [ bakdim ] ) . the agreement is good up to @xmath214 . we notice that , beyond this value , further terms would be required in the taylor expansion ( [ bakdim ] ) . we are now going to consider as second example a two - dimensional reactive periodic lorentz gas . this model has been introduced in ref . @xcite . in this system , a point particle , carrying a color @xmath0 or @xmath1 undergoes elastic collisions on hard disks fixed in the plane and forming a regular triangular lattice characterized by the lattice fundamental vectors @xmath285 and @xmath286 . the radius of the disks is assumed to be equal to unity and @xmath132 is the distance between their centers . we shall work in the finite horizon regime , @xmath287 , for which the diffusion coefficient is finite . some of the disks play the role of catalysts : when the point particle collides on one of them , it changes its color instantaneously with a probability @xmath2 . the catalysts form a regular triangular superlattice over the disks lattice , with as fundamental vectors @xmath288 and @xmath289 , where @xmath215 is an integer parameter controlling the density of catalysts : in the direction of @xmath290 and @xmath291 , one disk over @xmath215 is a catalyst . globally , one disk over @xmath292 is a catalyst . the configurations with @xmath293 and @xmath294 are depicted in fig . [ fundamental ] . our goal here is to study the cumulative reactive eigenfunctions of this system . as done in ref . @xcite , we will construct the frobenius - perron operator governing the time evolution of the color density describing the reactive process in this system . we will then perform a spectral analysis of this operator . the variables needed to describe the point particle in this system are its position , its velocity and its color , i.e. @xmath295 , where @xmath157 is a discrete variable taking the values @xmath158 or @xmath159 when the particle color is @xmath0 or @xmath1 . however , the collisions on the disks being elastic , the energy of the point particle is conserved . the magnitude of its velocity is a constant of motion , that we suppose equal to one , @xmath296 . the velocity can therefore be described by only one coordinate , which can be for example the angle between the velocity and the @xmath178-axis . this system is periodic , which means that instead of describing the dynamics in the full space , we can restrict our description to an elementary cell of the superlattice of catalysts , containing @xmath297 disks ( see fig . [ fundamental ] ) . moreover , the relevant information concerning the dynamics is contained in the collision dynamics : between two collisions , the point particle performs a free flight . the three - dimensional flow dynamics can therefore be reduced to a poincar - birkhoff map , describing the dynamics from collision to collision . we will use the birkhoff coordinates @xmath298 , @xmath299 , @xmath300 , @xmath301 , where @xmath302 is the index of the disk of the elementary cell on which the collision takes place , @xmath303 is an angle giving the position of the impact on this disk and @xmath304 is the sine of the angle between the outgoing velocity and the normal at the impact . in these coordinates , the mapping of the reactive lorentz gas is given by @xmath305 @xmath306 is the time of the @xmath220 collision . @xmath307 is the first - return time function . the vector @xmath308 gives the cell of the superlattice visited at time @xmath306 and @xmath309 is a vector giving the jump carried out on the superlattice during the free flight from @xmath310 to @xmath311 . the change of color is controlled by the function : @xmath312 we notice that the dynamical randomness of the reactive lorentz gas is characterized by the following kolmogorov - sinai entropy per unit time @xmath313\ ] ] where @xmath80 is the unique positive lyapunov exponent of the two - dimensional lorentz gas and @xmath42 is the frequency of encounters with a catalytic disk . we now want to obtain the frobenius - perron operator associated with this map . we first start from a description of the particle in the full space . the current state and position of the particle are given by a suspended flow in terms of the coordinates @xmath314 , where @xmath315 is the time elapsed since the last collision , @xmath316 is a vector of the superlattice and @xmath317 . the phase - space probability density depending on these variables @xmath318 evolves in time under the frobenius - perron operator defined by eq . ( [ fpmatrix ] ) @xcite . this continuous - time operator can be reduced to an operator describing the dynamics from collision to collision under the mapping ( [ map ] ) . the details are presented in ref . the operator for the modes of wavenumber @xmath72 is given by @xmath319\\ & & \times \ \{[1-p(\bphi^{-1}{\bf x } ) ] \nonumber \tilde{q}_{{\bf k},s}(\bphi^{-1}{\bf x},c ) \ + \ p(\bphi^{-1}{\bf x})\ \tilde{q}_{{\bf k},s}(\bphi^{-1}{\bf x},-c)\}\end{aligned}\ ] ] where @xmath320 the operator @xmath321 is similar to a frobenius - perron operator of the mapping @xmath322 up to important factors which take into account the varying time of flight between the collisions , the spatial modulation by the wavenumber @xmath61 , as well as the color change on a part of the domain of definition of the poincar - birkhoff mapping . if we now consider the total phase - space density @xmath323 and the difference of phase - space densities @xmath324 , their time evolutions are decoupled and we can write an operator governing the evolution of @xmath325 , describing the reactive modes of wavenumber @xmath72 in this system @xcite @xmath326\ \exp[-st(\bphi^{-1}{\bf x } ) -i{\bf k}\cdot{\bf a}(\bphi^{-1}{\bf x } ) ] \ g(\bphi^{-1}{\bf x } ) \ , \label{react - op}\ ] ] which we call the reactive operator . to identify the eigenmode of the reactive operator ( [ react - op ] ) which controls the slowest relaxation of the color , we consider the eigenvalue problem for @xmath327 and its adjoint @xmath328 imposing @xmath247 as eigenvalue @xmath329 assuming the normalization condition @xmath330 a formal solution for @xmath331 ( respectively @xmath332 ) can be obtained by applying successively @xmath333 ( respectively @xmath334 ) to the unit function : @xmath335 \ \exp[-s_{\bf k}t(\bphi^{-m}{\bf x})-i{\bf k}\cdot{\bf a } ( \bphi^{-m}{\bf x } ) ] \ , \label{eigenfunction}\\ \tilde{\psi}_{\bf k}({\bf x})&=&\lim_{n\to\infty}\prod_{m=0}^{n-1 } [ 1 - 2p(\bphi^m{\bf x})]\ \exp[-s_{\bf k}^*t(\bphi^m{\bf x})+i{\bf k}\cdot { \bf a } ( \bphi^{m}{\bf x})]\ .\end{aligned}\ ] ] the normalization condition gives an equation to obtain @xmath336 : @xmath337 \ \exp[-s_{\bf k}t(\bphi^{-m}{\bf x})-i{\bf k}\cdot{\bf a } ( \bphi^{-m}{\bf x } ) ] \rangle \ .\label{norm}\ ] ] we notice that if we consider segments of trajectories of time @xmath17 from initial conditions @xmath74 we have that @xmath338\ , \label{corr3 } \end{aligned}\ ] ] so that eq . ( [ norm ] ) can be transformed into eq . ( [ vanhovereac ] ) for the reaction rate . we are interested in the smallest relaxation rate @xmath66 which dominates at long times @xmath17 . for the reactive case , the dispersion relation is given by @xmath339 where @xmath139 is the reaction rate and @xmath196 is the reactive diffusion coefficient . solving eq . ( [ norm ] ) allows us to obtain the values of these two coefficients . [ drlor ] shows the dependence of the reactive diffusion coefficient @xmath196 on the reaction probability @xmath2 , for two different configurations of catalysts , where @xmath293 and @xmath294 . the value of the diffusion coefficient @xmath211 is indicated for comparison . we notice the remarkable result that the behavior is similar to the one observed for the multibaker model ( see fig [ drvsl ] ) . the explanation is the following . the case @xmath293 , with a single catalytic disk in each cell where the motion is ballistic , is similar to the multibaker map for @xmath203 ( see fig . [ multi ] ) . in this case , the reactive diffusive coefficient rapidly increases with the transition probability @xmath2 in both the reactive multibaker model with @xmath203 and the reactive lorentz gas with @xmath293 . in contrast , the reaction starts to be controled by the diffusion as soon as @xmath340 in the reactive lorentz gas @xcite and , similarly , for the reactive multibaker with @xmath341 . in these cases , the reactive diffusion coefficient has a milder dependence on the transition probability @xmath2 and becomes even smaller than the diffusion coefficient for large values of @xmath2 . the cumulative function of the reactive eigenmode is obtained by integration of ( [ eigenfunction ] ) over the angle @xmath70 between the initial position and the @xmath178-axis @xmath342\ \exp[-s_{\bf k } \ t(\bphi^{-m}{\bf x}')-i{\bf k } \cdot { \bf a}(\bphi^{-m}{\bf x } ' ) ] \ , \label{cumullor}\ ] ] which is equivalent to the formula ( [ cumulreac ] ) according to the correspondences ( [ corr1])-([corr3 ] ) . [ lorvsk ] shows these functions in the complex plane and as a function of the wavenumber @xmath61 , in the case @xmath293 , @xmath343 and @xmath214 . the hausdorff dimension of these curves can be computed by a box - counting algorithm for different values of the wavenumber @xmath72 . the values obtained are plotted versus @xmath284 in fig . [ lorsq ] , in the case @xmath293 , @xmath343 , @xmath344 and @xmath214 . we have numerically computed the coefficients @xmath264 , @xmath265 and @xmath345 , using the expressions ( [ alpha ] ) , ( [ beta ] ) and ( [ sigma2 ] ) . we have obtained @xmath346 , @xmath347 and @xmath348 . the solid lines in fig . [ lorsq ] are given by eq . ( [ hausdim2 ] ) , considering the corrections up to the second order in @xmath2 . we observe the remarkable agreement between the theoretical prediction and the numerical results . in this paper , we have studied in detail simple reaction - diffusion systems in which an isomerization is induced by chaotic dynamics . we have constructed the reactive modes of relaxation toward the thermodynamic equilibrium . these modes are given by singular distributions having a cumulative function at small values of the wavenumber . the cumulative functions form fractal curves in the complex plane because of the chaotic dynamics and in spite of the randomness introduced by the transition probability @xmath2 of the color in the isomerization . we explain this remarkable result by the fact that the randomness of the system is of a mild type since its kolmogorov - sinai entropy per unit time ( [ ks ] ) remains finite . we have here derived a formula given by eq . ( [ hausdim ] ) for the hausdorff dimension of the reactive relaxation modes in chaotic reaction - diffusion systems with two degrees of freedom ( [ resultreac ] ) . we have defined a function @xmath127 generalizing the ruelle topological pressure @xmath5 . for small values of the wavenumber , we obtain a relationship between the hausdorff dimension @xmath246 , the reactive diffusion coefficient @xmath196 , the reaction rate @xmath139 , and two derivatives of @xmath127 , namely @xmath349 and @xmath350 . in the limit of a small reaction probability @xmath2 , the hausdorff dimension is related to @xmath196 and to the lyapunov exponent @xmath80 , with corrections in powers of @xmath2 . for @xmath44 , we recover the relationship ( [ diff ] ) obtained in ref . @xcite for the diffusive hydrodynamic modes . we have shown that the reaction rate , the reactive diffusion coefficient and the hausdorff dimension can be expressed in terms of quantities which characterize the statistical correlations between the number @xmath38 of encounters with the catalysts , the distance @xmath351 travelled in the diffusive motion across the system , and the stretching rates @xmath352 characterizing the dynamical instability and the chaos . our formula ( [ hausdim ] ) thus establishes a new relationship between macroscopic reaction - diffusion processes and the microscopic dynamical chaos . we have tested our results on two reaction - diffusion systems with a simple isomerization reaction : a triadic multibaker map and a two - dimensional reactive periodic lorentz gas . in both cases , we study the limit of small wavenumbers and small reaction probabilities . the agreement with the theory is excellent . moreover , the study confirms and clarifies results previously obtained for similar models of reaction - diffusion @xcite . in particular , our results confirm that the reaction - diffusion process will obey on a large spatial scale and a long time scale the macroscopic reaction - diffusion equations : @xmath353 for the color densities @xmath354 and @xmath355 . according to results ( [ drp0 ] ) and ( [ delta ] ) , the diffusion coefficients of eqs . ( [ eqb ] ) are given by @xmath356 which explains that the coefficients of cross - diffusion can be small if the transition probability @xmath2 is small . ( [ delta ] ) shows that this cross - diffusion depends on the correlation between the number @xmath38 of catalysts encountered and the square @xmath357 of the distance travelled by the particle in its diffusive motion . moreover , figs . [ drvsl ] and [ drlor ] and eqs . ( [ diffreacp0 ] ) and ( [ diffreacl ] ) for the multibaker model show that the coefficient of cross - diffusion can become significant for high transition probability @xmath2 and if the density of catalysts is very high , in the case @xmath203 for the multibaker and @xmath293 for the lorentz gas . we may here say that the cross - diffusion , i.e. the coupling between the diffusions of colors @xmath0 and @xmath1 , is induced by the reaction . in conclusion , the modes of relaxation toward equilibrium turn out to display unexpected fractal properties which have their origin in the chaotic dynamics of the microscopic motion of particles , even if this motion is time - reversal symmetric and volume - preserving . this result has been previously obtained for diffusive processes and , here , we have shown that this result extends to reaction - diffusion systems where it allows us to understand the relaxation to equilibrium at the level of the microscopic motion . * acknowledgements . * we thank professor g. nicolis for support and encouragement in this research . the authors are financially supported by the national fund for scientific research ( f. n. r. s. belgium ) . this research is supported , in part , by the interuniversity attraction pole program of the belgian federal office of scientific , technical and cultural affairs , by the royal academy of belgium , and by the f. n. r. s. . 30 t. gilbert , j. r. dorfman and p. gaspard , phys . * 85 * , 1606 ( 2000 ) . p. gaspard , j. stat . phys . * 88 * , 1215 ( 1997 ) . t. gilbert , j. r. dorfman and p. gaspard , nonlinearity * 14 * , 339 ( 2001 ) . p. gaspard , i. claus , t. gilbert and j. r. dorfman , phys . lett . * 86 * , 1506 ( 2001 ) . p. gaspard and g. nicolis , phys . lett . * 65 * , 1693 ( 1990 ) . p. gaspard and f. baras , phys . e * 51 * , 5332 ( 1995 ) j. r. dorfman and p. gaspard , phys . e * 51 * , 28 ( 1995 ) . p. gaspard and j. r. dorfman , phys . e * 52 * , 3525 ( 1995 ) . i. claus and p. gaspard , phys . e * 63 * , 036227 ( 2001 ) . b. moran and w. hoover , j. stat . * 48 * , 709 ( 1987 ) . d. j. evans , e. g. d. cohen and g. morriss , phys . rev . a * 42 * , 5990 ( 1990 ) . n. i. chernov , g. l. eyink , j. l. lebowitz , and ya g. sinai , phys . * 70 * , 2209 ( 1993 ) . g. nicolis and i. prigogine , _ self - organization in nonequilibrium systems _ ( wiley , new york , 1977 ) . g. nicolis , _ introduction to nonlinear science _ ( cambridge university press , cambridge uk , 1995 ) . eckmann and d. ruelle , rev . phys . * 57 * , 617 ( 1985 ) . p. gaspard and x .- j . wang , phys . rep . * 235 * , 291 ( 1993 ) . p. gaspard , _ chaos , scattering , and statistical mechanics_(cambridge university press , cambridge uk , 1998 ) . f. barra and p. gaspard , phys . e * 63 * , 066215 ( 2001 ) . l. van hove , phys . rev . * 95 * , 249 ( 1954 ) . j .- boon and s. yip , _ molecular hydrodynamics _ ( dover , new york,1980 ) . ya . g. sinai , russian math . surveys * 27 * , 21 ( 1972 ) . r. bowen and d. ruelle , invent . math . * 29 * , 1181 ( 1975 ) . d. ruelle , _ thermodynamic formalism _ ( addison - wesley , reading ma,1978 ) . p. gaspard , phys . e * 53 * , 4379 ( 1996 ) . p. gaspard and r. klages , chaos * 8 * , 409 ( 1998 ) . i. claus and p. gaspard , j. stat . * 101 * , 161 ( 2000 ) .

observations of the high - mass star forming region afgl 2591 reveal a large abundance of co , a molecule known to be enhanced by far uv ( fuv ) and x - ray irradiation . in chemical models assuming a spherically symmetric envelope , the volume of gas irradiated by protostellar fuv radiation is very small due to the high extinction by dust . the abundance of co is thus underpredicted by orders of magnitude . in a more realistic model , fuv photons can escape through an outflow region and irradiate gas at the border to the envelope . thus , we introduce the first 2d axi - symmetric chemical model of the envelope of a high - mass star forming region to explain the co observations as a prototypical fuv tracer . the model assumes an axi - symmetric power - law density structure with a cavity due to the outflow . the local fuv flux is calculated by a monte carlo radiative transfer code taking scattering on dust into account . a grid of precalculated chemical abundances , introduced in the first part of this series of papers , is used to quickly interpolate chemical abundances . this approach allows to calculate the temperature structure of the fuv heated outflow walls self - consistently with the chemistry . + synthetic maps of the line flux are calculated using a raytracer code . single - dish and interferometric observations are simulated and the model results are compared to published and new jcmt and sma observations . the two - dimensional model of afgl 2591 is able to reproduce the jcmt single - dish observations and also explains the non - detection by the sma . we conclude that the observed co line flux and its narrow width can be interpreted by emission from the warm and dense outflow walls irradiated by protostellar fuv radiation .

0906.0588

forming high - mass o and b stars with a surface temperature larger than @xmath1 k emit most of their radiation in far uv ( fuv ) wavelengths . after a not yet well defined point of the evolution , also x - rays are emitted . when still deeply embedded in their natal cloud , most of this high - energy radiation is absorbed by the large column density of gas and dust towards the protostar . it is thus not available to direct observation but does influence the composition of the molecular envelope through heating and photoionization . an essential ingredient of star formation are bipolar outflows transporting the excess angular momentum outwards . when the fast , low - density gas of molecular outflows or jets expand into the surrouding molecular cloud , large cavities may result ( e.g. @xcite ) . along these outflow cavities , fuv radiation may escape and irradiate the high density material at the border of the cavity . + a surprisingly large amount of co@xmath0 has been detected in the high - mass star forming region afgl 2591 by @xcite . their radiative transfer calculations indicate fractional abundances of order @xmath2 , much higher than @xmath3 predicted by dark cloud models . co@xmath0 has previously not been detected towards envelopes of young stellar objects ( ysos ) , but has been seen in photo dominated regions ( pdrs , e.g. @xcite or @xcite ) . strong far uv ( fuv ) radiation in these regions heats the gas and drives a peculiar chemistry which enhances the abundance of co@xmath0 . the detection of co@xmath0 in envelopes of ysos is thus strong evidence for the feedback of the protostar on the envelope by high - energy irradiation . in this work , we will study co@xmath0 as a prototypical species enhanced by fuv irradiation in high - mass ysos . + chemical models solve for the evolution of the abundances of molecules and atoms . they simulate a network of species reacting with each other . the network consists of different types of chemical reactions . in dark clouds for example , all fuv radiation is shielded by a large column density of dust and the chemistry is dominated by cosmic ray ionization and reactions between neutral and ionized species ( e.g. @xcite ) . in the innermost part of a yso envelope , x - rays may dominate the ionization ( @xcite ) . they influence the chemistry mainly by secondary fast photoelectrons hitting molecules and atoms . x - ray induced chemistry is very similar to cosmic ray induced chemistry ( @xcite ) . in regions with a strong fuv irradiation , direct photoionization of species with ionization potential @xmath4 ev drive the chemistry . the influence of fuv radiation on the envelope of ysos has been studied by @xcite . + so far , a spherically symmetric structure of the envelope has been assumed for chemical models of envelopes of high - mass star forming regions ( e.g. @xcite , @xcite , @xcite ) . while the abundances derived using this spherical model agree for many species with observations of specific sources , they fail to explain the amount of co@xmath0 . due to the attenuation by dust , fuv radiation can not escape the innermost few hundred au and the volume of gas where co@xmath0 may be formed is too small in spherical models to reproduce the observations . another possible formation mechanism of co@xmath0 is x - ray irradiation . unlike fuv photons , x - rays are mainly attenuated by geometrical dilution ( @xmath5 ) due to their smaller absorption cross - section proportional to @xmath6 . x - ray radiation can thus penetrate much deeper into the envelope on scales of a few thousand au . protostellar x - ray luminosities of up to @xmath7 erg s@xmath8 in the 1 - 100 kev band are observed ( @xcite ) . for this luminosity however , the x - ray flux is too low to enhance co@xmath0 sufficiently to match with observations . + the interaction zone between outflow and infall has been proposed before as sites of anomalous chemistry and molecular excitation . @xcite have explained the strong co(6 - 5 ) emission in narrow lines observed towards many low - mass ysos by heating of the cavity wall through protostellar fuv photons . compact emission of hco@xmath0 associated with the outflow wall has been detected by @xcite in the class 0 object l 1527 . a possible explanation is mixing of partially ionized gas from the outflow with infalling material leading to a thin layer of dense and ionized gas , where species with a high dipole moment ( e.g. hco@xmath0 or hcn ) can be excitet efficiently by collisions with electrons ( @xcite ) . detailed 3d radiative transfer studies by @xcite on the other hand prefer a scenario with an enhanced abundance of hco@xmath0 due to shock liberation and photoprocessing of molecular material stored in ice mantles . observational evidence for such an interaction layer in a high - mass star forming region are found by @xcite by studying the line profiles of different molecular species . recently , @xcite find evidence for uv photons escaping through the outflow cones and impacting the walls for the low - mass protostar hh 46 . + in this work , we will introduce a chemical model that implements a non - spherical density structure taking an outflow cavity into account . a proper treatment of the fuv irradiated outflow walls requires two - dimensional chemical models with high spatial resolution . this is however computationally excessively expensive , especially if the temperature structure is calculated self - consistently with the chemical abundances . in the first paper of this series ( @xcite ) , we have introduced a new method for fast chemical modelling . a precalculated grid of chemical abundances depending on different physical parameters like the density , temperature or fuv flux is calculated and the abundances can then be obtained by interpolation in this database . using different benchmark tests , we found very good agreement between fully calculated chemical abundances and interpolated values , while the interpolation approach is more than 5 orders of magnitudes faster . this speed - up allows to quickly construct detailed chemical models with a complex geometry . + we construct a detailed two - dimensional chemical model of the high - mass star forming region afgl 2591 . the goal is to explain the high abundance of co@xmath0 measured by the jcmt single - dish telescope ( @xcite ) as well as the non - detection by the submillimeter array ( sma ) interferometer ( @xcite and new data , reported in sect . [ sec : compsma ] ) . in this paper , we assume a fixed geometry and consider only co@xmath0 . the next part of this series of papers is dedicated to a study of the influence of geometry for a larger sample of species . + the paper is organized as follows : in the first section of the paper , we briefly discuss the chemistry of co@xmath0 . the next section introduces the two - dimensional model of afgl 2591 : we describe the modelling process , the radiative transfer of the fuv radiation and the calculation of the temperature structure . in section [ sec : results ] , we present results of the chemical model . modeled fluxes and synthetic maps are compared to jcmt and sma observations . the fractional abundance of co@xmath0 depending on the temperature is given in fig . [ fig : plot_param_co+ ] for different gas densities , x - ray fluxes [ erg s@xmath8 @xmath9 and fuv irradiation in units of the interstellar radiation field ( isrf ) . a parcel of gas has been modeled for this figure using the chemical model described in the first paper of this series ( @xcite ) . the chemical model solves for the evolution of the abundances of a network of chemical species starting from initial abundances . we assume a chemical age of @xmath10 yr as suggested by @xcite . chemical time - scales are very short in regions with strong fuv or x - ray irradiation , where a significant amount of co@xmath0 is produced . equilibrium conditions are thus reached very quickly and the chemical age does not influence the resulting abundances . photodissociation and ionization rates depend on the attenuating column density between the fuv / x - ray emission and the modeled parcel of gas . we assume an optical depth of @xmath11 ( fuv ) and a hydrogen column density of @xmath12 @xmath13 ( x - rays ) to derive the irradiating spectra . + in gas with strong fuv irradiation , co@xmath0 is efficiently produced by the reaction c@xmath0 + oh @xmath14 co@xmath0 + h ( @xcite , @xcite ) . direct photoionization of carbon atoms accounts for c@xmath0 , while oh is produced by the photodissociation of water . at temperatures below 100 k , h@xmath15o is frozen - out and the amount of oh reduced . between 100 k and about 250 k , the reaction o + h@xmath15 @xmath14 oh + h does not proceed due to an activation barrier and oh is destroyed through photodissociation . the amount of oh increases with higher temperature and thus co@xmath0 is produced . at very high fuv fields , co@xmath0 is photodissociated and its abundance again decreases . + in x - ray irradiated gas , ionization of co through secondary electrons accounts for the production of co@xmath0 . the abundance of co@xmath0 is thus constant with density and the fractional abundance @xmath16 inversely proportional to the total density . in low density gas with high x - ray irradiation , the production of co@xmath0 through c@xmath0 + oh can also be important as figure [ fig : plot_param_co+ ] shows at a density of 10@xmath17 @xmath18 and an x - ray flux of 10 erg @xmath13 s@xmath8 . in this regime , the fractional abundance of co@xmath0 increases with temperature . + co@xmath0 is destroyed by h@xmath15 leading to hoc@xmath0 and hco@xmath0 with equal branching ratios . in regions with very high fuv irradiation , h@xmath15 is photodissociated to atomic hydrogen and co@xmath0 is destroyed by the reaction co@xmath0 + h @xmath14 co + h@xmath0 . + to produce the observed fractional abundances of order @xmath2 ( @xcite ) by x - rays , a flux higher than 10 erg @xmath13 s@xmath8 is needed for a density of 10@xmath19 @xmath18 . for an x - ray luminosity of 10@xmath20 erg s@xmath8 as found by @xcite for afgl 2591 and in the extreme case of no attenuation , this flux can only be achieved in the innermost 60 au . in a spherical model of afgl 2591 , x - rays may enhance the abundance of co@xmath0 up to a few times @xmath21 , and @xcite have thus suggested the molecule as a tracer for x - rays . x - rays are however not able to enhance the abundance to @xmath2 on larger distances to the source . the fuv flux of a young b star on the other hand is still sufficient at a distance of almost 10000 au to produce a fractional abundance exceeding 10@xmath22 if no dust attenuates the fuv field . we thus consider co@xmath0 to be a tracer for fuv irradiation rather than for x - rays . + a short note concerning the combined influence of x - rays and fuv radiation : in the vicinity of a strong fuv source , co is dissociated and x - rays can not produce co@xmath0 efficiently . for example , the calculation of the parameter study including x - rays at a density of 10@xmath17 @xmath18 has been repeated in the presence of an fuv field of 10@xmath17 isrf . for this condition , the fractional abundance does not exceed 10@xmath23 and we thus expect the fractional abundance in a low density and x - ray / fuv irradiated outflow region to be very low . figure [ fig : flowchart ] summarizes the process for an axi - symmetric model of afgl 2591 . we first assume a density structure in sect . [ sec : dens2dafgl ] . this density distribution is used in the following section to calculate the local fuv and x - ray field at every position of the model . the high - energy flux together with the density structure then enters the thermal balance calculation , which solves for the temperature structure . since cooling and heating depend upon the composition of the gas , the temperature must thus be calculated self - consistently with the chemical abundances which are read out of the grid of chemical models presented by @xcite . the abundance of co@xmath0 is then obtained using the same approach . synthetic maps are calculated using a raytracer and convolved to the appropriate telescope beam for comparison with single - dish observations . to compare model results to interferometric observations , the synthetic maps are converted into visibility amplitudes and reduced in the same way as interferometric observations . one goal of this work is to study the difference in the modeled co@xmath0 line strength using a spherically symmetric 1d and an axi - symmetric 2d model . for the 2d model we thus assume a density distribution following the power - law of the spherically symmetric 1d model , except for an outflow - region ( fig . [ fig : out_2d_flared]a ) . we implement @xmath24 with @xmath25 @xmath18 and @xmath26 au , as proposed by @xcite . + for the separation between the outflow region and the envelope , we use the function @xmath27 where @xmath28 and @xmath29 denote coordinates along the outflow axis and perpendicular to the outflow ( in units of au ) . the full opening angle @xmath30 at @xmath31 au is assumed to be @xmath32 , approximately in agreement with the high resolution mid - ir observations by @xcite . the full opening angle of the outflow cone at @xmath33 au is about 40 degrees . the choice of a parabolic outflow cavity is supported by theoretical predictions ( @xcite ) and observations ( @xcite ) . it is also compatible with the mid - ir observations of afgl 2591 . + vibrationally excited co at a velocity of @xmath34 km s@xmath8 , emitted in outflow gas has been observed by @xcite . assuming an infall velocity in the envelope of order 10 km s@xmath8 , pressure equilibrium ( @xmath35 ) requires a density ratio of @xmath36 between the two regions . in the outflow region , the density of the power - law distribution is thus assumed to be reduced by this ratio . as in the spherical models of doty et al . and stuber et al . , we do not model the innermost @xmath37250 au . we make use of the symmetry and only model positive values of @xmath29 and @xmath28 . in this quadrant , the model consists of 300 cells along the outflow ( @xmath28-axis ) and 450 cells parallel to the midplane ( @xmath29-axis ) thus a total of @xmath38 cells . the resolution is thus about 100 au @xmath39 67 au . + high - energy radiation with photon energy in the range of 6 - 13.6 ev ( fuv radiation ) and 0.1 - 100 kev ( x - rays ) penetrates molecular gas , ionizes molecules and heats the gas . the local fuv and x - ray field at every position are thus required to calculate the temperature structure and the chemical abundances . + for the local x - ray intensity , we take into account the distance @xmath29 and the total hydrogen column density @xmath40 [ @xmath9 between the x - ray source and the local position . for an x - ray luminosity @xmath41 [ erg s@xmath8 ] , the local intensity is thus proportional to @xmath43 , where @xmath44 [ @xmath45 is the photoionization cross - section at a photon energy @xmath46 [ ev ] . we follow @xcite and assume an x - ray emission by a thermal plasma at a temperature of @xmath47 k and a protostellar x - ray luminosity of 10@xmath20 erg s@xmath8 . in their spherically symmetric models , the attenuating column density in the innermost 250 au is a free parameter found to be about @xmath48 @xmath13 . for the 2d model , the same value is adopted in the midplane ( @xmath49 ) . since density in the outflow is a factor of 400 lower , the column density between source and outflow walls can be of order 10@xmath50 @xmath13 , and photons with energies between 0.1 and 1 kev are not absorbed . unlike @xcite , we thus take into account these photons in the 2d model . + the central fuv source is assumed to emit a blackbody spectrum with a temperature of @xmath51 k and a luminosity of @xmath52 l@xmath53 ( @xcite ) . at a distance of 200 au , the fuv flux in absence of any attenuation corresponds to @xmath54 times the isrf . unlike elastic compton scattering for x - rays ( @xcite ) , scattering of fuv photons on dust grains can be important due to the similar wavelength of fuv radiation and the dust size . a monte carlo simulation in the vein of @xcite calculates the local fuv strength ( appendix [ sec : mc_uvimplement ] ) . photon packages starting at the central source are traced through the density model , while scattering and attenuation by dust is taken into account . for simplicity , we do not calculate the fuv spectra explicitly , but deal with the scaling factor @xmath55 in units of the isrf . to read out chemical abundances from the grid of chemical models , the visual extinction @xmath56 [ a@xmath57 is needed . we obtain the local extinction from @xmath58 where @xmath59 is the local flux for a photon package @xmath60 . the unattenuated flux @xmath61 is only geometrically diluted . brackets denote averaging over all photon packages passing through a cell . we implement the cross - section on dust given in @xcite for average milky way dust with @xmath62=3.1 and c / h=56 ppm in pahs . this choice however has a minor effect on the model results , as photionization no longer dominates in regions , where a higher value of @xmath63 is expected . since the adopted approach corresponds to the calculation at one fuv wavelength , we use the dust properties at a photon energy of 9.8 ev , in the middle of the 6 - 13.6 ev fuv band . for the conversion @xmath64 @xmath13 ( @xcite ) and an extinction cross - section of @xmath65 @xmath66 per h - atom at 9.8 ev , we obtain a conversion factor between the fuv and visual extinction of @xmath67 . + for the density profile of sect . [ sec : dens2dafgl ] , the calculated fuv flux @xmath68 and extinction @xmath69 are given in fig . [ fig : out_2d_flared]b and fig . [ fig : out_2d_flared]d . the attenuated flux shows the region where fuv radiation influences the chemistry , as this value enters the heating rates and the photodissociation rates of the form @xmath70 , with @xmath71 and @xmath72 fitting constants ( e.g. @xcite ) and @xmath55 the fuv flux in units of the isrf . note that the unattenuated flux @xmath73 has to be used together with the attenuation @xmath74 to read out the abundances from the chemical grid . + traditional pdr models ( e.g. @xcite ) find the photoelectric effect of fuv photons on small dust grains and pahs to be the dominant heating process in regions with strong fuv irradiation . in the dense outflow walls of the 2d model , this process can easily heat the gas to temperatures above 250 k. for temperatures between 250 k and several hundred k , the fractional abundance of co@xmath0 is a strong function of temperature ( sect . [ sec : chemcoplus ] ) . it is thus important to carry out a detailed calculation of the temperature profile in the outflow wall . input parameters are the gas density @xmath75 [ @xmath18 ] , the fuv field given by @xmath55 [ isrf ] and @xmath56 [ a@xmath57 and the energy deposition per density by x - ray photons h@xmath77 [ erg s@xmath8 ] . + to obtain the gas temperature @xmath78 , the thermal balance has to be solved . we assume steady state conditions . thus the equilibrium between the total heating ( @xmath79 ) and cooling ( @xmath80 ) rate yields the temperature . different physical processes contribute to the individual rates @xmath81 and @xmath82 , where @xmath60 denotes processes . the inner energy per parcel of gas @xmath83 [ erg @xmath18 ] is given by @xmath84 important examples for heating processes are h@xmath15 collisional de - excitation following pumping through fuv photons and the photoelectric effect . the cooling rate consists of gas - grain collisional cooling and atomic / molecular line emission ( e.g. the [ cii ] line at 158 @xmath85 m ) . an overview of the implemented cooling and heating processes is given in appendix [ sec : app_thermal ] . + heating and cooling rates depend on the composition of the gas . for example the c@xmath0 atomic line cooling depends on the fractional abundance @xmath86 of ionized carbon . to obtain a self - consistent solution , the composition of the gas is read out from the grid of chemical models . because the temperature also enters as a parameter for the interpolation of the fractional abundances , the equation for the thermal balance has to be solved iteratively . + molecular or atomic line emission contributes to the cooling only if the photons can escape . a radiative transfer calculation is thus required to obtain proper cooling rates . for the atomic fine structure lines , we implement an escape probability code . as a simplification , the radiative transfer is not carried out in full 2d , but from the modeled point to the closest point of the outflow wall ( fig . [ fig : plot_1d1d ] ) . the probability @xmath87 for a photon to escape from the outflow wall to the cavity is thus obtained from the molecular / atomic column density ( e.g. @xmath88 for co ) along this line . + a fully self - consistent temperature model requires a two - dimensional radiative transfer calculation of the dust continuum radiation to obtain the dust temperature @xmath89 . as a simplification , we use the dust temperature of the spherically symmetric model of @xcite for the region where the fuv radiative transfer calculation predicts an optical depth larger than 10 . in regions with @xmath90 , the dust temperature is obtained from the analytical expression given by @xcite . using this approximation , the temperature structure remains consistent with the spherically symmetric model for temperatures below 100 k. in the particular case of co@xmath0 , the relevant temperature range is @xmath91 k , and the interpolation to the spherically symmetric temperature profile is thus unimportant . the adopted separation between envelope and outflow ( @xmath92 ) yields a `` flared '' geometry that allows direct irradiation of the dense outflow walls by the central fuv source . this results in a much larger fuv irradiated volume compared to a `` linear '' separation ( @xmath93 ) where fuv photons can only penetrate the outflow walls if they are scattered on dust in the outflow ( fig . [ fig : fuvcomic ] ) . scattering is however relatively inefficient . + what is the size scale of the outflow walls ? we define the width of the outflow wall by the region with @xmath94 along a cut of constant @xmath28 . for example at @xmath95 au , this corresponds to a thickness of about 1130 au . the outflow cavity thus allows long - range streaming of fuv radiation to large distances from the central source . photon packages do not necessarily follow cuts of constant @xmath28 . the straight ray path from the fuv source to the far end of the outflow wall at @xmath95 au propagates for @xmath96 au in the outflow wall . scattering thus extends the outflow wall . for comparison , the monte carlo code was re - run with scattering switched off . this run yielded a width of the outflow wall of only about 470 au at @xmath95 , about 2.5 times smaller than the width calculated including scattering effects . this demonstrates the importance of scattering for the strengths of the local fuv field . + the width of the outflow wall with @xmath94 , 10 or 100 isrf depending on @xmath28 is given in fig . [ fig : width_outflow ] along with the error due to the finite resolution of the model . an fuv - enhanced layer along the outflow with a width of a few hundered au is thus produced by the protostellar fuv radiation . while the width defined by @xmath97 increases with @xmath28 , the layer with @xmath98 has an approximately constant width of @xmath99 au . this is a consequence of the adopted geometry : for the @xmath98 layer , scattering is less important than for the @xmath100 region . unscattered radiation at the surface layer of the outflow wall travels a longer way in the high density region for higher values of @xmath28 due to the larger impact angle . this effect compensates with the flux at the surface of the outflow wall @xmath101 , where @xmath102 is the distance from the fuv source to the surface of the outflow wall , and the attenuation decreasing with @xmath103 for the implemented density profile . only a small fraction of the envelope in a thin layer on the outflow wall is heated by fuv photons to temperatures above 100 k. the temperature along cuts of fixed values of @xmath28 is shown in fig . [ fig : plot_co+_cut_1e32 ] . for @xmath104 au , the temperature at the surface exceeds 1000 k , consistent with the surface temperatures obtained from pdr models for similar conditions ( e.g. @xcite ) . + the width of the region with @xmath105 k in the outflow wall is approximately constant with @xmath28 ( fig . [ fig : width_outflow ] ) as the @xmath106 layer . in the region with @xmath107 k , the heating is dominated by the photoelectric effect @xmath108 , with @xmath75 being the density . the main coolant is atomic oxygen which is approximately thermalized and the cooling rate thus @xmath109 . as found in the calculation , the gas temperature is thus approximately independent of density and thus follows @xmath55 . + an overview of the mass and volume of different regions of the axi - symmetric 2d model compared to the spherical 1d model is given in table [ tab : vol_afgl ] . indeed , the volume heated to @xmath110 k is more than two orders of magnitude larger in the 2d model , while the mass of the same region is only @xmath37 8 times larger due to the adopted density gradient . similarly , the volume irradiated by @xmath94 isrf is almost 4 orders of magnitude larger in the 2d model , while the mass of the gas is about 2 orders of magnitude larger . we note that size of the region with @xmath94 isrf in the spherical 1d model heavily depends on fuv attenuating column density in the innermost 250 au . for table [ tab : vol_afgl ] the extreme case of no attenuation between the protostar and a radius of 250 au was assumed . + physical parameters like the cavity shape or the density ratio between infalling envelope and outflowing gas enter the modelling . a complete study is beyond the scope of this study and will be presented in the third paper of this series ( bruderer et al . , in prep . ) . here we discuss some general trends . a large surface directly irradiated by protostellar fuv radiation is required for a significant volume and mass of the dense and fuv heated material in the outflow wall . by changing the outflow angle @xmath30 by @xmath111 degrees or the separation between outflow region and envelope to @xmath112 with @xmath113 , the volume and mass with @xmath110 k vary only by a factor of about two . on the other hand , a higher density in the outflow region affects the model much more due to the higher extinction of fuv radiation . increasing the density ratio @xmath114 from @xmath36 to @xmath115 reduces the mass of the @xmath110 k region by a factor of 4 , while the volume decreases by about an order of magnitude . these results for the relatively insensitivity with respect to angle and shape , and relative sensitivity to density , underline one of the key results of this paper namely the importance of direct irradiation on the chemical composition of outflow cavity walls . + l c c region & mass & volume + & [ m@xmath53 ] & [ au@xmath117 + + outflow & 0.014 & @xmath118 + envelope & 44.7 & @xmath119 + envelope ( @xmath120 k ) & 0.53 & @xmath121 + envelope ( @xmath122 ) & 1.65 & @xmath123 + envelope ( @xmath122 and @xmath124 k ) & 1.1 & @xmath125 + + envelope & 50.3 & @xmath126 + envelope ( @xmath120 k ) & 0.07 & @xmath127 + envelope ( @xmath122 ) & 0.01 & @xmath128 + the calculation of the fuv flux and gas temperature in the axi - symmetric 2d model suggests a region with temperature @xmath129 k but fuv irradiation @xmath94 of potentially fuv irradiated ices . while a detailed discussion of these ices is beyond the scope of this paper , it is interesting to note two things . first , such a region exists due to the direct irradiation allowed by the concave outflow walls . second , the potential mass in such a region is 1.1 m@xmath53 . together , these facts suggest that further exploration of irradiated ices is warranted . + given the physical parameters for an axi - symmetric 2d model of afgl 2591 defined previously , the grid of chemical models ( @xcite ) can now be used to quickly read out the abundances of co@xmath0 . since no outflow features have been observed in lines of co@xmath0 by @xcite , we assume there is no co@xmath0 in the outflow cavity . also , the discussion in sect . [ sec : chemcoplus ] indicates a very low fractional abundance of co@xmath0 in the outflow . + the fractional abundance of co@xmath0 is given in fig . [ fig : plot_co+_cut_1e32 ] along with the temperature and the density for cuts parallel to the midplane of the model . in a thin layer at the outflow wall , co@xmath0 is enhanced . this layer matches with the fuv heated gas at temperatures above 250 k. the co@xmath0 layer in the midplane ( @xmath49 ) is considerably larger due to the scattering of fuv radiation from both outflows along the positive and negative z - axis . at the surface of the outflow wall , the fractional abundance of co@xmath0 exceeds @xmath2 for values of @xmath28 up to a few thousand au . at @xmath130 , the fractional abundance of co@xmath0 is still a few times @xmath131 . + since the chemical time - scales of co@xmath0 in regions with abundance larger than a few times @xmath131 are shorter than 100 yrs , the evolutionary age of the source does only affect the modelling results if the fuv production of the protostar has not set in yet or the outflow cavity is not large enough yet . while direct evidence for protostellar fuv radiation of the source can not be given , the mid - ir observations of the source by @xcite suggest an outflow cavity with size of at least @xmath132 au along the outflow axis . for the measured flow velocity of a few hundered km s@xmath8 ( @xcite ) , the expansion to @xmath132 au would require @xmath133 yrs much less than the infered age since yso formation estimated to be a few times @xmath134 yrs ( @xcite ) . + for comparison , results of the spherical 1d model are given in the top left of fig . [ fig : plot_co+_cut_1e32 ] . in this model , fuv radiation of the central source can not escape the innermost part and the abundance of co@xmath0 is thus very small at fractional abundances below @xmath21 . outflow walls in the 2d model increase the surface irradiated by fuv and thus enhance the total amount of co@xmath0 . + mixing between warm and ionized outflow material with the envelope can also induce a particular chemistry in a thin layer along the outflow . charge - exchange reactions in this mixing layer enhance the abundance of co@xmath0 . the width of the mixing layer is however small due to the high electron abundance leading to short recombination time - scales . while a detailed study ( @xcite , @xcite or @xcite ) is beyond the scope of this work , a simple model is given in appendix [ sec : coplusmix ] . it predicts the amount of co@xmath0 produced in this mixing layer to be enhanced , however insufficiently to explain observations . two rotational lines of co@xmath0 with @xmath135 km s@xmath8 have been detected by @xcite at the center position of afgl 2591 using the jcmt . we follow their arguments and assume the lines to be detected . the observed line fluxes are given in table [ tab : coplus_molec ] along with molecular data of co@xmath0 taken from the jpl database ( @xcite ) . the einstein-@xmath136 coefficients are scaled to a dipole moment of 2.77 d ( @xcite ) . + c c c c cc transition & frequency & @xmath137 & @xmath138 & @xmath139 & line flux + & [ ghz ] & [ 10@xmath140 s@xmath8 ] & [ k ] & & [ k km s@xmath8 ] + @xmath141 - @xmath142 & 353.741 & 1.58 & 33.94 & 6 & 0.5 + @xmath143 - @xmath144 & 354.014 & 1.7 & 33.99 & 8 & 0.27 + synthetic spectra are calculated to compare the abundance modelling results to these observations . the excitation of co@xmath0 is needed to model the line flux . however , the mechanism is unclear since co@xmath0 is more more likely to be destroyed than excited in collision with h@xmath15 . however , electrons and atomic hydrogen are abundant at the surface of pdrs and can excite co@xmath0 as discussed in @xcite . possibly , co@xmath0 is formed in an excited state . in this work , we follow @xcite and assume a fixed excitation temperature @xmath145 . the effects of collisional excitation by electrons and atomic hydrogen and excited formation are being studied by @xcite . for the transitions discussed in this section and assuming an excited formation , they find a flux increase by a factor of only @xmath146 compared to the peak lte flux at an excitation temperature of @xmath147 k. + to constrain rotational temperatures , transitions for different upper levels are necessary . however , only tentative detections of co@xmath0 in ysos for @xmath148 are reported ( @xcite ) . we will thus follow @xcite and assume @xmath145=30 k , slightly higher than @xcite ( 20 k ) or @xcite ( 10 k ) . for an upper energy level @xmath149 , the level population is proportional to @xmath150 , with @xmath151 being the partition function . using @xmath152 k , the population of the @xmath153 levels is approximately maximized . the emission of optically thin lines scales with the population of the upper level . compared to @xmath152 k , the line flux is thus about about @xmath154 weaker for @xmath155 k and a factor of 3.5 for @xmath156 k or @xmath157 k. an excitation temperature of a few hundred k is expected for collisional excitation by electrons and atomic hydrogen ( @xcite ) . we conclude that a different excitation temperature , within the expected range or assuming an excited formation of co@xmath0 , changes the modeled line flux less than an order of magnitude . + synthetic line fluxes are obtained from the solution of the radiative transfer equation . the two lines considered in this work have a very low optical depth with @xmath158 for @xmath145=30 k at an assumed line width and molecular column density of 4 km s@xmath8 and @xmath159 @xmath13 , respectively . this column density corresponds to an emitting region with a diameter of 60000 au at a density of 10@xmath160 @xmath18 with a co@xmath0 fractional abundance of @xmath2 . for our application , we can thus neglect self absorption and the integrated flux is obtained from the velocity integrated radiative transfer equation @xmath161 with @xmath162 being the line frequency , @xmath163 the statistical weight of the upper level , @xmath164 the einstein - a coefficient [ s@xmath8 ] and @xmath165 the co@xmath0 density [ @xmath18 ] . the integral is taken along the line of sight ( los ) . to have comparable results to older literature , we assume a distance of 1 kpc to afgl 2591 , following @xcite and @xcite . a larger distance of 1.7 kpc as suggested by @xcite would however not change the main conclusions of this work . the line flux of the spherical 1d model calculated using this method agrees within a few percent with results from ` sky ` of the ratran radiative transfer code ( @xcite ) . + maps of the integrated line flux from the 2d model at different inclination angles are shown in fig . [ fig : plot_ray_fine ] . the angular resolution of the images is @xmath166 . we give only results for the @xmath143 - @xmath144 transition at 354.014 ghz in this section . the modeled flux of the other fine structure line can be obtained by scaling with 0.7 , the ratio of the einstein - a coefficients and the statistical weights . the modeled maps are convolved with a @xmath167 fwhm gaussian to account for the angular resolution of the jcmt . the effect of different inclination angles is clearly visible in unconvolved maps , but the convolved maps are almost independent of inclination angle . + the predicted line fluxes for the jcmt at the center position are given in table [ tab : coplus_sma ] . they only vary between 0.16 and 0.14 k km s@xmath8 for an inclination of 0@xmath168 and 90@xmath168 , respectively . this reflects a slightly larger fraction of the outflow walls in the telescope beam for low inclination angles . the difference is however small since most emission is from regions close to the center . indeed , about between @xmath169 and @xmath170 of the modeled jcmt flux stems from within a radius of @xmath171 from the center , corresponding to 1000 au at the adopted distance . an inclination angle of @xmath172 was suggested by @xcite and is thus displayed in this and the next section . + the line flux of the co@xmath0 line at 354.014 ghz as obtained from the spherically symmetric 1d model is with 1.1 @xmath39 10@xmath173 k km s@xmath8 about 3 orders of magnitude weaker than observed . on the other hand , the results of the axi - symmetric 2d model agree to within a factor of two which we consider as a good agreement given all uncertainties entering the modelling . the @xmath174 - @xmath175 line at 353.741 ghz however disagrees by about a factor of 5 , as the observed line ratio of 1.85 ( = [ 353.741 ghz ] / [ 354.014 ghz ] ) is larger than the ratio of 0.7 as predicted by the models . possible explanations are a different excitation mechanism of the fine structure levels or line blending of the 353.741 ghz line , e.g. by the @xmath176(@xmath177 ) line at 353.741 ghz ( @xcite ) . to explain the line ratio by a different excitation mechanism , a grossly different excitation temperature would however be required , as discussed earlier in this section . interferometric observations of the two co@xmath0 lines in afgl 2591 have been carried out using the submillimeter array ( sma ) . @xcite have used the extended configuration of the array with projected baselines covering the range between 38 - 214 k@xmath178 ( 32.2 m - 181.2 m ) . new observations have been carried out in the compact configuration on april 14 2006 . these data cover a projected baseline range of 10.6 - 82 k@xmath178 ( 9 m - 69.4 m ) . the frequency setting is the same as for the extended array observation , covering the range between 342.6 - 344.6 ghz in the lsb and 352.6 - 354.6 ghz in the usb . data of this observation will be discussed in a paper in preparation ( bruderer et al . + the rms noise per velocity bin of 0.5 km s@xmath8 is 0.46 jy beam@xmath8 ( extended array ) , 0.21 jy beam@xmath8 ( compact array ) and 0.18 jy beam@xmath8 when data of both arrays are combined using natural weighting . the better weather conditions and shorter baselines result in a higher sensitivity of the compact array observations . the detection limit for the integrated line can be estimated from @xmath179 , where @xmath180 is the rms noise per frequency bin , @xmath181 km s@xmath8 the expected linewidth and @xmath182 km s@xmath8 the velocity resolution ( @xcite ) . the factor 1.2 accounts for the uncertainty of the absolute calibration of about @xmath154 . for a 3@xmath183 detection , an integrated flux of 2.3 jy km s@xmath8 ( extended array ) and 0.91 jy beam@xmath8 km s@xmath8 ( compact and extended arrays combined ) is thus necessary . + the co@xmath0 line at 354.014 ghz has not been detected at a 3@xmath183 level in either configurations of the sma . the 353.741 ghz line is not detected in the extended configuration observations . in the data of the combined array however , a line is found at approximately 353.741 ghz . is the co@xmath0 line blended by @xmath176(@xmath177 ) at the same frequency ? several other lines of @xmath176 are within the observed frequency setting , but are not detected despite their presumably higher intensity due to a larger einstein - a coefficient , lower critical density and upper level energy . co@xmath0 is thus more likely blended by an unidentified line . the blended line has a velocity integrated line flux of 4.9 jy beam@xmath8 km s@xmath8 corresponding to a @xmath184 detection . for the predicted line ratio ( [ 353.741 ghz ] / [ 354.014 ghz ] ) of 0.7 and the ratio of 1.85 as observed by the jcmt , the co@xmath0 line at 354.014 ghz should have been detected . we thus conclude co@xmath0 is not detected by the sma observations . + is this non - detection consistent with the modeled line flux of the 2d model ? to answer that question , synthetic maps ( fig . [ fig : plot_ray_fine ] ) are converted to visibility amplitudes for the @xmath185 coverage of the sma observation using the miriad[multiblock footnote omitted ] task ` uvmodel ` . the simulated visibilities are then reduced in the same way as sma observations : they are inverted and the clean algorithm is applied on the resulting maps . simulated velocity integrated maps for inclination angles of 30@xmath168 and 90@xmath168 are presented in fig . [ fig : sma_obsmodel ] . the maximum flux of the simulated maps for the extended array is 0.17 jy beam@xmath8 km s@xmath8 at an inclination of @xmath186 . for the ( u , v ) coverage of the combined array , the peak flux is 0.47 jy beam@xmath8 km s@xmath8 at an inclination angle of @xmath187 . these integrated fluxes correspond to less than @xmath188 and the non - detection is thus consistent with the models . as the simulated sma maps in fig . [ fig : sma_obsmodel ] show , this non - detection is explained by the limited sensitivity of our observations rather than by over - resolving effects of the interferometer . + cccc inclination & sma ( ext . ) & sma ( both . ) & jcmt + @xmath189 $ ] & @xmath190jy beam@xmath8 km s@xmath191 $ ] & @xmath190jy beam@xmath8 km s@xmath191 $ ] & @xmath190k km s@xmath191 $ ] + + 0 & 0.05 & 0.23 & 0.16 + 10 & 0.06 & 0.25 & 0.16 + 20 & 0.05 & 0.26 & 0.16 + 30 & 0.06 & 0.30 & 0.15 + 40 & 0.06 & 0.38 & 0.15 + 50 & 0.11 & 0.44 & 0.15 + 60 & 0.16 & 0.47 & 0.14 + 70 & 0.17 & 0.45 & 0.14 + 80 & 0.16 & 0.44 & 0.14 + 90 & 0.16 & 0.43 & 0.14 + + & @xmath192 & @xmath192 & @xmath193 + observed : & @xmath194 & @xmath195 & 0.27 + we have used the grid of chemical models introduced by @xcite to construct a detailed two - dimensional model of the high - mass star forming region afgl 2591 . the spherically symmetric 1d models by @xcite and @xcite have been extended with a low - density outflow region allowing protostellar fuv radiation to escape from the innermost region . in the 2d model , fuv irradiates a larger surface and thus a larger volume compared to spherically symmetric models . the model is used to simulate the line fluxes of co@xmath0 as a prototypical fuv tracer . the main conclusions of this work are : 1 . the existence of concave outflow cavity walls allows the efficient , long - range streaming of fuv radiation to large distances from the central source ( sect . [ sec : outflowwalls ] ) . a thin fuv - enhanced layer in the outflow is produced , having a thickness of a few hundreds of au , and @xmath196 100 isrf ( sect . [ sec : outflowwalls ] ) . 3 . a large mass of fuv irradiated material more than a solar mass can reside in the outflow walls . this combined with the large extent can lead to a situation where the molecular emission from the outflow walls dominates the total line flux in single - dish observations ( sect . [ sec : outflowwalls ] , sect . [ sec : compobs ] ) 4 . detailed 2d modelling of co@xmath0 including these outflow walls is consistent with single - dish ( sect . [ sec : compobs ] ) and interferometric observations ( sect . [ sec : compsma ] ) , while the flux in a 1d model is orders of magnitude too low . this indicates the need for multidimensional chemical models for the interpretation of fuv sensitive molecules . the strong gradients in density and fuv flux require an accurate calculation of the gas temperature structure and inclusion of scattering of fuv radiation in the outflow walls , which can extend the width of the fuv enhanced region significantly ( sect . [ sec : outflowwalls ] ) . 6 . a region of high fuv ( @xmath196 1 isrf ) and low temperature ( @xmath197 100 k ) leads to the strong possibility of ice mantle processing by fuv photons in the outflow walls ( sect . [ sec : outflowwalls ] ) . 7 . the abundance of co@xmath0 is also enhanced in a mixing layer between the ionized and warm outflow and the envelope . mixing alone can however not explain the observed amount of co@xmath0 ( sect . [ sec : coplusabundance ] , appendix [ sec : coplusmix ] ) . this paper shows the application of the grid of chemical models ( @xcite ) for the construction of a multidimensional chemical model of a yso envelope . this interpolation method simplifies the construction of two or three dimensional chemical models considerably and the gain in speed allows the self - consistent calculation of the temperature structure with the chemical abundances ( sect . [ sec : temp2dafgl ] ) . our work shows that such detailed modelling including the influence of high - energy irradiation and the geometrical shape of the envelope can be necessary even to explain spatially unresolved single - dish observations . + some hydrides ( e.g. sh@xmath0 or nh@xmath0 ) will be observable for the first time using the upcoming hifi spectrometer onboard herschel . they are expected to be chemical tracers of fuv / x - ray radiation ( @xcite ) and further study will be carried out to investigate the influence of the geometry on these species for the interpretation of the herschel data . direct imaging of the outflow region will be available with the atacama large millimeter / submillimeter array ( alma ) . these high spatial resolution observations will deliver additional constraints on the geometry . we thank kaspar arzner , pascal stuber and jes jrgensen for useful discussions and an anonymus referee for his / her valuable comments . michiel hogerheijde and floris van der tak are acknowledged for use of their ratran code . this work was partially supported under grants from the research corporation and the nasa grant nnx08ah28 g ( sdd ) . astrochemistry in leiden is supported by the netherlands research school for astronomy ( nova ) and by a spinoza grant from the netherlands organization for scientific research ( nwo ) . the submillimeter work at eth is supported by the swiss national science foundation grant 200020 - 113556 . + _ facilities : _ , in the first appendix , we discuss the implementation of the radiative transfer calculation of the fuv radiation . unlike the work of @xcite , our calculation is implemented on a three dimensional cartesian grid and thus already able the solve three dimensional problems . to improve convergence , points with constant @xmath28 and @xmath198 are averaged for the axi - symmetric model in this work . to save memory , only cells with positve values of @xmath199 , @xmath200 and @xmath28 are stored , but photon packages can of course still travel from one quadrant to another . the flow of the calculation is 1 . a photon package is started from the fuv source at an arbitrary direction with a luminosity of @xmath201 [ erg s@xmath8 ] , where @xmath202 is the luminosity of the fuv source in the 6 - 13.6 ev band . @xmath203 is the number of photon packages in the simulation run . the factor 8 takes into account that only cells with positive values of @xmath199 , @xmath200 and @xmath28 are stored . as the photon package travels along a straight line , @xmath204 is attenuated by @xmath205 with @xmath206 , the local fuv optical depth . the optical depth depends on @xmath207 , the extinction cross - section , @xmath208 the local density and @xmath209 the step size of the code . the total extinction , that a photon package suffers from the source to the current position @xmath210 is stored . to avoid discretization errors , especially close to the source , we use the analytical expression for the local density . if the photon package passes the border of a cell , the local intensity is updated . the attenuated intensity is obtained from @xmath211 where the sum is taken over all photon packages passing the cell . the surface of the cell wall which the photon package just passed is given by @xmath136 [ @xmath212 and the directions of the photon package and the normal of the cell separation is given by @xmath213 and @xmath214 , respectively . the unattenuated intensity @xmath215 is calculated using the same expression as @xmath216 , except @xmath204 is replaced by @xmath217 . if a random number [ 0,1 ] is larger than @xmath218 , with @xmath219 , the scattering - optical depth , a new direction of the photon package is chosen following the implemented phase function @xmath220 . for our work , we use the tabulated function of @xcite for an average milky way dust with @xmath221 and c / h=56 ppm in pahs . 6 . for each photon package , steps 2 - 5 are repeated , until @xmath204 is smaller then a certain threshold or the package left the simulated area . the code propagates 10@xmath222 photon packages and then averages for the attenuated intensity @xmath216 in the axi - symmetric model . three subsequent solutions are used to calculate the signal - to - noise ratio ( snr ) , defined by @xmath223 only cells with an averaged intensity larger than 10@xmath173 isrf are considered . convergence is reached , if the snr exceeds 50 in all cells . this means less than @xmath224 difference to the average . the application in this work requires @xmath225 photon packages to achieve this accuracy . as a benchmark test , the code is re - run for the density distribution in sect . [ sec : dens2dafgl ] , but with scattering switched off . in this way , an analytical solution can be compared to the result of the code . for the analytical solution , only the distance @xmath29 and column density @xmath226 [ @xmath9 to the fuv source are needed to derive the local fuv flux [ erg s@xmath8 @xmath9 @xmath227 with the fuv luminosity @xmath228 erg s@xmath8 and the fuv optical depth @xmath229 } ) / 1.87 \times 10^{21}$ ] . the conversion factor of 1 isrf @xmath230 @xmath231 erg s@xmath8 @xmath13 yields the flux @xmath55 in units [ isrf ] . the agreement between the code and the analytical calculation is tested for points with @xmath232 . a very good agreement is found , with deviations less than @xmath233 . in this appendix , references for the heating and cooling rates used in sect . [ sec : temp2dafgl ] are given and the temperature balance calculation is benchmarked with pdr models . + _ photoelectric heating : _ the heating rate of photoelectrons from small dust grains and pahs is @xmath234 erg @xmath18 s@xmath8 , with the density @xmath75 , the attenuated fuv field @xmath55 and an efficiency @xmath83 . the efficiency depends on the ionization fraction of the grains , since the work function increases for ionized dust grains . @xcite give an analytical fit for @xmath83 as a function of the ratio @xmath72 between ionization and recombination rate ( @xmath235 , with the electron density @xmath236 and the temperature @xmath237 ) . h@xmath15 heating : _ different heating processes involving molecular hydrogen are considered : ( i ) collisional de - excitation of vibrationally pumped h@xmath15 by fuv photons ( fuv - pumping ) . ( ii ) formation heating : h@xmath15 forming on dust releases part of the binding energy to the gas . ( iii ) photodissociation of h@xmath15 heats the gas . the rates for process ( i ) and ( iii ) depend on the local fuv field . since line absorption is responsible for the pumping and dissociation of the molecule , we reduce the local fuv field by the self shielding factor @xmath238 ( @xcite ) . rates for ( i ) and ( ii ) are implemented by the analytical expression of @xcite . the rate of process ( iii ) is taken from @xcite . + _ c ionization : _ photoionization of atomic carbon , c + @xmath239 @xmath14 c@xmath0 + e@xmath240 , releases an energy of 1.06 ev to the gas and thus contributes to the heating . we follow @xcite for the implementation . + _ x - ray heating : _ fast photoelectrons produced by x - ray photons lose part of their energy through coulomb interaction with thermal electrons . the heating rate is given by @xmath241 , where h@xmath77 [ erg s@xmath8 ] is the energy deposition of x - rays in the gas ( c.f . @xcite ) per density . the efficiency factor @xmath242 depends on the h / h@xmath15 ratio and the ionization fraction . an analytical fit for @xmath242 to detailed calculations is given in @xcite . + _ cosmic ray heating : _ for low degrees of ionization @xmath243 , about 9 ev per primary ionization through a cosmic ray particle is used to heat the gas . the heating rate is thus @xmath244 erg @xmath18 s@xmath8 . _ atomic fine structure lines : _ forbidden fine structure lines of o , c and c@xmath0 are important coolants at the surface of the fuv irradiated zone . we consider the [ oi ] 63 @xmath85 m , [ oi ] 146 @xmath85 m , [ ci ] 369 @xmath85 m , [ ci ] 609 @xmath85 m and [ cii ] 158 @xmath85 m lines . the cooling rate is obtained from @xmath245 with the einstein - a coefficient @xmath164 [ s@xmath8 ] of the transition @xmath246 @xmath14 @xmath247 , @xmath248 [ s@xmath8 ] the line frequency , @xmath249 the optical depth , @xmath250 the escape probability and @xmath251 [ @xmath18 ] , the density of the species x in the excited level @xmath246 . an escape probability formalism ( e.g. @xcite ) is used to calculate the level population iteratively . collisional excitation by h , h@xmath15 and electrons is taken into account - the abundance of the collision partners is read out of the chemical grid . we use the same collision rate coefficients as @xcite . the optical depth @xmath56 is calculated using the molecular column density to the outflow cone ( sect . [ sec : temp2dafgl ] ) . + _ line cooling : _ vibrational lines of h@xmath15 can contribute to the cooling of the gas . due to the large gap between the ground state and the first excited state , corresponding to about 6000 k , we use the two level approximation given in @xcite as a simplification . again , self shielding is taken into account for the radiative pumping by fuv photons . + _ gas - grain cooling / heating : _ the difference in temperature between @xmath78 and the @xmath89 leads to a transfer of heat . this can be cooling close to the fuv irradiated zone , where @xmath252 , or heating deeper in the envelope . the rates are proportional to @xmath253 . we implement the results of @xcite for a minimal grain size of 10 @xmath254 . + _ cooling by co and h@xmath15o : _ molecules can contribute to the gas cooling by rotational lines at low temperature and ro - vibrational lines at higher temperature . we include line cooling by co and h@xmath15o using the fitted rates of @xcite ( for @xmath120 k ) and @xcite ( for @xmath255 k ) . the rates depend on the molecular column density , the temperature and the density of the collision partner . their work considered excitation by h@xmath15 . excitation by electrons and atomic hydrogen is taken into account by the approximation given in @xcite and @xcite . the molecular column density is obtained in the same way as for the cooling through atomic fine structure lines . cooling by isotopes ( @xmath256co , c@xmath257o , h@xmath258o ) can be important due to the smaller optical depth . isotope ratios by @xcite are used to scale the molecular density and the column density . + _ recombination : _ recombination of electrons with grains and pahs can cool the gas . we implement the fit - results by @xcite for our calculation . this rate also depends on the ratio @xmath72 between ionization and recombination rate , due to increased coulomb interaction in gas with a high ionization fraction . + _ ly@xmath30 emission : _ atomic hydrogen excited by electron collisions emit ly@xmath30 photons and thus contribute to the cooling . this process is only efficient at temperatures higher than a few thousand k. we use the cooling rate given in @xcite . we compare our calculation of @xmath78 to pdr codes , in a similar situation as model v4 in the pdr comparison study by @xcite . the test consists of a plane parallel slab at a density of 10@xmath259 @xmath18 and a fuv irradiation of @xmath260 . we assume an x - ray flux of about 0.04 erg s@xmath8 @xmath13 consistent to the afgl 2591 model at @xmath261 au assuming an x - ray luminosity of @xmath41 = 10@xmath20 erg s@xmath8 . differences in the resulting gas temperature , when x - rays are switched off , are however negligible . + figure [ fig : tempbench]a shows the dust and gas temperature obtained with the thermal balance calculation . as expected , the gas temperature is much higher than the dust temperature for low optical depth , while both temperatures agree well at high optical depth and density ( @xcite ) . the mean of the gas temperature obtained with different pdr codes is given by a dotted line , along with one standard deviation given by the gray shaded region . the results of our code agrees very well with the results of the pdr codes , while the calculation only takes a few seconds . + figure [ fig : tempbench]b gives the abundances of molecules involved in heating or cooling processes . as @xcite , we use a `` one - line '' approximation for the h@xmath15 and co self - shielding and are thus not able to reproduce the exact position of the h / h@xmath15 transition . notice the very low water abundance due to the destruction of water by x - rays ( @xcite ) . in fig . [ fig : tempbench]c , the rates of each heating process is given along with the total heating rate . at low optical depth , photoelectric heating and h@xmath15 -pumping are the main heating sources , while dust - gas coupling heats the gas at high optical depth ( @xmath262 in the example ) . x - ray heating does not contribute significantly to the heating rate , but the destruction the important coolant h@xmath15o by x - rays slightly enhances the temperature . [ fig : tempbench]d shows rates of different cooling processes . at the edge of the calculated region , h@xmath15 ro - vibrational lines govern the cooling . [ oi ] 63 @xmath85 m and molecular cooling is important at higher optical depth . due to the low abundance of water , the main molecular coolant is co. + as a toy - model for the mixing layer between the warm and ionized outflow and the envelope , we assume the outflow material to be in the form of electrons ( e@xmath263 ) and ionized hydrogen ( h@xmath0 ) . the chemical evolution of a parcel of gas consisting of a mixture of outflow and inflow material is modeled similarly to sect . [ sec : chemcoplus ] . free parameters are the temperature and the initial abundances of e@xmath263 and h@xmath0 . the peak fractional abundance of the temporal evolution of co@xmath0 is given in fig . [ fig : mixtoy ] together with the temporal evolution of the fractional abundance for selected temperatures and initial ionizations . + for temperatures above 60 k , formaldehyde ( h@xmath15co ) is evaporated from ice mantles . the reaction h@xmath15co @xmath14 co@xmath0 + h@xmath15 + h forms co@xmath0 and it is destroyed by the reaction with h@xmath15 . for very high initial ionization fractions ( @xmath264 ) , the dissociative recombination co@xmath0 + e@xmath240 @xmath14 c + o with rate coefficient @xmath265 becomes important and the peak fractional abundance is weakly temperature dependent . due to the short recombination time - scale of e@xmath263 with h@xmath0 and grains , the co@xmath0 abundance peaks at young chemical ages and then decreases quickly . + the column density along the mixing layer is @xmath266 , with the width of the mixing layer @xmath267 , the gas density @xmath75 and the peak fractional abundance @xmath268 . we set @xmath269 , with the alfvn velocity @xmath270 km s@xmath8 and the time @xmath271 , defined by the point of evolution where the abundance of co@xmath0 has dropped by 1 order of magnitude compared to the peak . the peak fractional abundance @xmath268 and the product @xmath272 are approximately independent of density as fig . [ fig : mixtoy]b shows . on the other hand , @xmath273 is roughly independent of temperature and peaks at an electron fraction of about @xmath274 , where @xmath275 s and @xmath276 for a density of @xmath277 @xmath18 . we obtain for the upper limit of the column denisty due to mixing @xmath278 @xmath13 . in the fuv model of sect . [ sec : afgl2dmod ] , the column density of co@xmath0 along constant @xmath28 is of order @xmath279 @xmath13 . we conclude that mixing alone can not reproduce the observed amount of co@xmath0 . , m. , abel , n. p. , bell , t. , bensch , f. , black , j. , ferland , g. j. , jonkheid , b. , kamp , i. , kaufman , m. j. , le bourlot , j. , le petit , f. , meijerink , r. , morata , o. , ossenkopf , v. , roueff , e. , shaw , g. , spaans , m. , sternberg , a. , stutzki , j. , thi , w .- f . , van dishoeck , e. f. , van hoof , p. a. m. , viti , s. , & wolfire , m. g. 2007 , , 467 , 187

to further our knowledge of the complex physical process of galaxy formation , it is essential that we characterize the formation and evolution of large databases of galaxies . the spectral synthesis starlight code of cid fernandes et al . ( 2004 ) was designed for this purpose . results of starlight are highly dependent on the choice of input basis of simple stellar population ( ssp ) spectra . speed of the code , which uses random walks through the parameter space , scales as the square of the number of basis spectra , making it computationally necessary to choose a small number of ssps that are coarsely sampled in age and metallicity . in this paper , we develop methods based on diffusion map ( lafon & lee , 2006 ) that , for the first time , choose appropriate bases of prototype ssp spectra from a large set of ssp spectra designed to approximate the continuous grid of age and metallicity of ssps of which galaxies are truly composed . we show that our techniques achieve better accuracy of physical parameter estimation for simulated galaxies . specifically , we show that our methods significantly decrease the age - metallicity degeneracy that is common in galaxy population synthesis methods . we analyze a sample of 3046 galaxies in sdss dr6 and compare the parameter estimates obtained from different basis choices . [ firstpage ] methods : data analysis methods : statistical methods : numerical galaxies : evolution galaxies : formation galaxies : stellar content

0905.4683

determining the physical parameters of large samples of galaxies is crucial to constrain our knowledge of the complicated physical processes that govern the formation and evolution of these systems . ultimately , we seek methods that quickly and effectively map observed data from @xmath0 galaxies to sets of parameters that describe the star formation and chemical histories of these galaxies . due to the complexity of the mechanisms of galaxy evolution , it is critical that we adopt a model that uses few assumptions about how galaxies evolve . moreover , full utilization of the available theoretical stellar models , avoiding ad hoc simplifications , is critical to obtaining accurate parameter estimates . there has been a large amount of work dedicated to estimating the physical parameters of databases of galaxies using different types of data . for example , spectral indices ( @xcite ) , emission features ( @xcite ) , and full high - resolution , broad - band spectra ( @xcite ) have all been utilized to estimate the star formation histories ( sfhs ) and metallicities of galaxies . recently , due to the abundance of high - quality , homogeneous databases of galaxy spectra such as the sloan digital sky survey ( sdss , @xcite ) , which provides high - resolution spectra for hundreds of thousands of galaxies , the approach of full high - resolution spectral fitting is now possible to infer the properties of large populations of galaxies . recent work has shown promise in the application of these methods to sdss galaxy spectra ( see @xcite and @xcite for reviews of results achieved by two such fitting methods ) . however , large - scale analyses , while possible , are not necessarily accurate , and can be computationally challenging . there is much room for improvement in accuracy of sfh parameter estimation . it is critical that we understand the shortcomings of the current models and improve their accuracy . at the same time , to achieve our goal of obtaining accurate sfh estimates for a large sample of galaxies , we can not sacrifice the computational efficiency of current methods . a common technique in the literature , called empirical population synthesis , is to model galaxies as mixtures of simple stellar populations ( ssps ) with known physical parameters . recent studies that have used this method are , e.g. @xcite , @xcite , @xcite , and @xcite . under this model , galaxy data are treated as linear combinations of the observed properties of ssps . historically , ssp parameters and observables were derived from observations of well - understood stellar systems . more recent studies have instead used model - produced ssps , such as those from the evolutionary population synthesis models of @xcite . the starlight spectral fitting code , introduced by @xcite , fits observed spectra with linear combinations of ssps from the models of @xcite . in @xcite , it was shown that sfh parameters of simulated galaxy spectra could be recovered by starlight in the absence of noise . however , their simulated spectra were generated and fit with the _ same basis _ of 45 ssps , rendering the results difficult to generalize to the expected performance on real galaxies , which can be composed of ssps on infinitely fine grids of age and metallicity . results of the starlight fitting code are highly dependent on the choice of basis of ssp spectra . the database of ssps available to us could theoretically be infinitely large , encompassing all combinations of age , metallicity , initial mass function , evolutionary prescription , etc . including too many ssps in our basis causes the code to be prohibitively slow and the solution to be degenerate due to the inclusion of ssps with almost identical spectra . if , however , we simply hand - select a few ssps on a coarse grid of age and metallicity with which to model our data , then we effectively ignore large subsets of our ssp parameter space , leading to suboptimal estimates for fits to databases of galaxies . using a hand - selected basis of ssps may also lead to the inclusion of multiple ssps whose spectra are essentially identical , which can lead to degeneracies . there is much room for improvement in the accuracy of sfh estimates by exploring numerical methods of selecting small bases of ssps . in this paper , we propose a method of choosing a small set of _ prototype _ spectra from a large database of ssp spectra based on their observable quantities . our method attempts to capture much of the variability of a large set of ssps in a few numerically chosen prototype ssps . the method is based on the diffusion map of @xcite and @xcite , a non - linear technique that seeks a simple , natural representation of data that are complex and high - dimensional , such as high - resolution astronomical spectra . we utilize the _ diffusion @xmath1-means _ method to numerically find a basis of prototype ssps with which we can fit galaxies to accurately estimate their sfhs . in a future study , we will consider the problem of computational speed in methods such as starlight , which takes @xmath2 minutes to fit each sdss galaxy spectrum for a basis of size 150 , and show how to quickly extend the sfh estimates obtained by using the methods in this paper to a large database of galaxies by taking advantage of the geometry of the manifold on which the data lie . the applicability of the methods introduced here extend beyond the problem of sfh estimation using high - resolution spectra . any astronomical task where observed data ( i.e. spectra , photometric data , spectral indices , etc . ) are fit as linear combinations of observable data from simpler systems can benefit from intelligent numerical selection of prototypes . the outline of the paper is as follows . in 2 we describe the starlight spectral synthesis code , discuss the drawbacks of ssp bases employed in the literature , and introduce the diffusion @xmath1-means method of prototype selection . we directly compare the basis of prototypes selected by our method to those used in the literature and those found using other methods . in 3 we fit several sets of realistic simulated galaxy spectra using the starlight code with different bases to directly compare the ability of each basis to accurately estimate galaxy parameters . in 4 we fit a set of sdss galaxy spectra and analyse physical parameter estimates obtained by different bases . we conclude with some remarks and discussion of future directions in 5 . we adopt the model of galaxy spectra as a mixture of simple systems introduced in @xcite : @xmath3 where @xmath4 is the model flux in wavelength bin @xmath5 . the component @xmath6 is the @xmath7th basis spectrum normalized at wavelength @xmath8 . the basis of ssp spectra @xmath9 is chosen before performing the analysis . the scalar @xmath10 $ ] is the proportion of flux contribution of the @xmath7th component at @xmath8 , where @xmath11 ; the vector @xmath12 with components @xmath13 is the _ population vector _ of a galaxy . these flux fractions can be converted to mass fractions , @xmath14 , using the light - to - mass ratios of each basis spectrum @xmath15 at @xmath8 . to describe the sfh of a galaxy , we will use time - binned versions of @xmath12 and @xmath16 , denoted @xmath17 and @xmath18 , respectively . using binned population vectors lets us avoid degeneracies that can occur in the estimation of the original population vectors . other components of the model in ( [ cfmodel ] ) are the scaling parameter @xmath19 , the reddening term , @xmath20 , and the gaussian convolution , @xmath21 . the reddening term describes distortion in the observed spectrum due to foreground dust , and is modeled by the extinction law of @xcite with @xmath22 . the gaussian convolution , @xmath21 , accounts for movement of stars within the observed galaxy with respect to our line - of - sight , and is parametrized by a central velocity @xmath23 and dispersion @xmath24 . to fit individual galaxies , we use the starlight fitting routine introduced by @xcite . the code uses a metropolis algorithm plus simulated annealing to find the minimum of @xmath25 ^ 2,\ ] ] where @xmath26 is the observed flux in the wavelength bin @xmath5 , @xmath4 is the model flux in ( [ cfmodel ] ) , and @xmath27 is the inverse of the noise in the measurement @xmath26 . the summation is over the @xmath28 wavelength bins in the observed spectrum . the minimization of ( [ chisq ] ) is performed over @xmath29 parameters : @xmath30 , and @xmath24 , where @xmath1 is the number of basis spectra in @xmath31 . the speed of the algorithm scales as @xmath32 . for the analysis of , e.g. @xmath33 galaxy spectra in the sdss database , it is crucial to pick a basis with a small number of spectra . in , for example , @xcite and @xcite , the starlight model ( [ cfmodel ] ) and fitting algorithm are tested using simulations . the simulated spectra are generated from the model in ( [ cfmodel ] ) using @xmath34 ssps from @xcite and @xmath35 ssps from @xcite , respectively . these authors fit the simulations using the same basis of ssps that was used to generate the simulations . from their analyses , they conclude that in the absence of noise , the algorithm accurately recovers the input parameters . in the presence of noise , the population vector @xmath12 is not recovered , and it is advised that a time - binned description of the sfh be adopted . although their use of the same basis for both generating and fitting simulated spectra is an appropriate test that the algorithm works , it is not a fair assessment of the expected performance of the methods for a database of real galaxy spectra . in [ models ] we discuss our concerns with their simulation assessments . in [ basis ] we present a new suite of methods that we test with several sets of simulated galaxy spectra . these simulations , which are described in [ simulations ] , are designed to be more representative of a large database of galaxy spectra , such as sdss . in reality , the population of all observed galaxies is not constrained to be mixtures of a small number of ssps on a discrete grid of age and metallicity . a more physically accurate representation of galaxies is as mixtures of an infinitely large basis of ssps on a continuous grid of age and metallicity and , depending on the complexity of the underlying physics , a possibly infinite grid of prescriptions for initial mass function and evolutionary track . in our simulations in [ simulations ] we simulate galaxies from a large database of @xmath36 ssps on a fine grid of age and metallicity to determine whether we can accurately describe their sfhs , metallicities , and kinematic parameters using an appropriately chosen , computationally tractable basis of @xmath35 or 150 spectra . we adopt a ssp database containing 1278 spectra . this set of spectra is used to both choose an appropriate basis ( [ basis ] ) and generate simulated spectra ( [ simulations ] ) . the database is meant to represent the large population of ssps from which observed galaxies can be composed . we use the spectra from the models of @xcite , computed for a @xcite imf , ` padova 1994 ' evolutionary tracks ( @xcite ) , and stelib library ( @xcite ) . theoretically , we could also vary the imf and evolutionary tracks across spectra and use the same methods developed in [ basis ] , but for simplicity decide to fix them in this study . the ssps in our database are generated on a grid of 213 approximately evenly log - spaced time bins from 0 to 18 gyrs and 6 different metallicities : @xmath37 = 0.0001 , 0.0004 , 0.004 , 0.008 , 0.02 and 0.05 , where @xmath37 is the fraction of the mass of a star composed of metals ( @xmath38 ) . this grid is designed to approximate a continuous variation across age and @xmath37 . the coarseness in the current @xmath37 grid is due to limitations in the current stellar population models . again , a finer grid of age and @xmath37 can be implemented with the techniques that we develop in [ basis ] . in previous studies using the starlight code , researchers hand - selected sets of ssp spectra to use in the model fitting . according to @xcite , `` the elements of the base should span the range of spectral properties observed in the sample galaxies and provide enough resolution in age and metallicity to address the desired scientific questions . '' in this same paper , they claim that their basis of @xmath34 ssps adequately recovers parameters of age and metallicity in their simulations . however , as discussed in [ specsyn ] , these simulations are unrealistic since they employ the same basis to both generate and fit the simulated spectra . in this section we present methods to numerically determine a basis @xmath31 of a small number of _ prototype spectra _ , @xmath15 , from a large set of ssp spectra . the resultant basis of prototype spectra will be designed to capture a large proportion of the variation of the set of ssps , which is chosen to span the range of observed spectral properties of a data set of galaxies at a high resolution in age and metallicity , such as the database of 1278 ssps described in [ models ] . the numerically chosen basis of prototype spectra , @xmath31 , should recover physical parameters better than any hand - chosen basis of the same size because hand - selected bases ignore subsets of the parameter space and may cause the solution to be degenerate by including multiple ssps with nearly identical spectra . in [ simulations ] we show that the bases computed from our numerical methods outperform hand - selected bases used in the literature in parameter estimation for simulated galaxies . diffusion map is a popular new technique for data parametrization and dimensionality reduction that has been developed by @xcite and @xcite , and was recently used by @xcite to analyse sdss spectra and @xcite to estimate photometric redshifts . as shown in these latter works , diffusion map has the powerful ability to uncover simple structure in complicated , high - dimensional data . these analyses demonstrate that this simple structure can be used to make precise statistical inferences about the data . in the problem at hand , we begin with a set @xmath39 of high - resolution broad - band ssp spectra @xmath40 , where @xmath41 can be very large and each @xmath42 has flux measurements in @xmath43 wavelength bins . from this set , our goal is to derive a small basis of @xmath1 prototype spectra , @xmath44 , each of length @xmath45 , that captures most of the variability in @xmath39 . in what follows , we briefly review the basics of the diffusion map technique and then show how it can be used to find an appropriate basis @xmath31 . the main idea of the diffusion map is that it finds a parametrization of a data set that preserves the connectivity of data points in a way that is dependent only on the relationship of each datum to points in a local neighborhood . by preserving the local interactions in a data set , diffusion map is able to learn , e.g. a natural parametrization of the of the spiral in fig . 1 of @xcite where other methods , such as principal components analysis , fail . starting with our set of ssp spectra , @xmath39 , in units of @xmath46/ per initial @xmath47 , we normalize the spectra at @xmath8 to ensure that we are not confused by absolute scale differences between individual spectra . we retain the scaling factors @xmath48 for later use to convert back to the original units . next , we select a discrepancy measure between ssps , i.e. @xmath49 which is the euclidean distance between two ssp spectra . the specific choice of @xmath50 is often not crucial to the diffusion map ; any reasonable measure of discrepancy will return similar results . it is important to note here the flexibility of the diffusion map method . for each problem , we can define a different discrepancy measure based on the type of data available to us and the types of information we want to incorporate . for example , we could define our measure @xmath51 by weighting more heavily those regions around absorption lines which are highly correlated with stellar age and metallicity . contrast this flexibility with the rigidity of principal components analysis , which is only based on the correlation structure of the data . from our discrepancy measure in ( [ distance ] ) , we construct a weighted graph on our data set @xmath39 , where the nodes are the individual ssp spectra and the weights on the edges of the graph reflect the amount of similarity between pairwise spectra . we define the weights as @xmath52 where @xmath53 should be small enough such that @xmath54 unless @xmath55 and @xmath56 are similar , but large enough so that the entire graph is connected ( i.e. every ssp has at least one non - zero weight to another ssp ) . in practice , we choose @xmath53 that produces a basis that achieves the best fits to galaxy spectra ( see [ simulations ] ) . by normalizing the rows of the weight matrix , @xmath57 , to sum to unity , we can define a markov random walk on the graph where the probability of going from @xmath55 to @xmath56 in one step is @xmath58 . we store all of these one - step probabilities in a matrix @xmath59 . by the theory of markov chains , the @xmath60-step transition probabilities are defined by @xmath61 . in the previous analyses in @xcite and @xcite , the authors fixed @xmath60 . in these previous papers , the choice of @xmath60 was unimportant because the task was linear regression on the diffusion coordinates , in which the multipliers @xmath62 of the diffusion map were absorbed into the linear regression coefficients @xmath63 . in the present work , our task is to determine a set of @xmath1 prototype spectra . the tool we use is @xmath1-means clustering in diffusion space , where results are dependent upon intra - point euclidean distances of the diffusion map , which in turn are dependent on @xmath60 . in this paper we introduce a diffusion operator that considers _ all _ scales @xmath64 simultaneously . we define the _ multi - scale _ diffusion map as @xmath65.\ ] ] where @xmath66 and @xmath67 are the right eigenvectors and eigenvalues of @xmath59 , respectively , in a bi - orthogonal decomposition . euclidean distance in the @xmath68-dimensional space described by equation ( [ dmap2 ] ) approximates the multi - scale diffusion distance , a distance measure that utilizes the geometry of the data set by simultaneously considering all possible paths between any two data points at all scales @xmath60 , in the markov random walk constructed above . an advantage to using the multi - scale diffusion map , besides elimination of the tuning parameter @xmath60 , is that it gives a more robust description of the structure of the data by considering the propagation of local interactions of data points through all scales . we also have three other tuning parameters : @xmath53 , @xmath68 ( dimensionality of the diffusion map ) , and @xmath1 ( number of prototypes ) over which to optimize the fits to galaxy spectra . for simplicity , we fix @xmath68 by choosing the dimensionality of the diffusion map to coincide with a 95% drop - off in the eigenvalue multipliers in the multi - scale diffusion map . this is reasonable because most of the information about the structure of the data will be in the few dimensions with large eigenvalue multipliers . simulations show that this choice of @xmath68 produces results comparable to optimizing over a large grid of @xmath68 . for the parameter @xmath1 , fits should improve as the basis gets larger . however , we want to keep @xmath1 small both for computational reasons and to avoid the occurrence of nearly identical spectra in our basis , which can cause degeneracies in our fits . in practice , we choose @xmath1=45 or 150 to directly compare our results with the bases used by @xcite ( cf05 ) and @xcite ( asa07 ) . this leaves us with only one tuning parameter , @xmath53 , over which to optimize the diffusion @xmath1-means basis in fits to galaxy spectra . finally , we determine a set of @xmath1 prototype spectra by performing @xmath1-means clustering in the @xmath68-dimensional diffusion map representation ( [ dmap2 ] ) of the ssp spectra , @xmath39 . the @xmath1-means algorithm is a standard machine learning method used to cluster @xmath41 data points into @xmath1 groups . it works by minimizing the total intra - cluster variance , @xmath69 where @xmath70 is the set of points in the @xmath7th cluster and @xmath71 is the @xmath68-dimensional geometric centroid of the set @xmath70 . as described in 3.2 of @xcite , for diffusion maps @xmath72 where @xmath73 is the trivial left eigenvector of @xmath59 . the @xmath1-means algorithm begins with an initial partition of the data and alternately computes cluster centroids and reallocates points to the cluster with the nearest centroid . the algorithm stops when no points are allocated differently in two consecutive iterations . the final centroids define the @xmath1 prototypes to be used in our basis . in fig . [ sspdmap ] , we plot the mapping of 1278 ssp spectra into the @xmath74 dimensional diffusion space described by ( [ dmap2 ] ) . the large black dots denote the @xmath1-means centroids for @xmath35 . individual ssps are coloured by cluster membership . the @xmath1 prototypes capture the variability of the ssp spectra along a low - dimensional manifold in diffusion space . note that the density of prototypes varies along different parts of the manifold . this is due both to the local complexity of the manifold and the sampling of the original base of ssps . this latter effect , that we tend to obtain more prototypes in areas where there are more ssps in our basis , can easily be corrected for by using a weighted @xmath1-means method , described in 2.2.3 , that accounts for varying density of ssps along the manifold . the last step of the diffusion @xmath1-means algorithm is to determine the @xmath1 protospectra and their properties . first , the normalization constant for the @xmath7th prototype spectrum is @xmath75 and the @xmath7th protospectrum is @xmath76 similarly , the age , metallicity , and `` stellar - mass fraction''the percent of the initial stellar mass still in stars of prototype @xmath7 are naturally defined as @xmath77 where @xmath78 , and @xmath79 are the age , metallicity and stellar - mass fraction of the @xmath80th ssp , respectively . equations ( [ eqn : protot])-([eqn : protosmf ] ) are natural definitions of the prototype parameters in the sense that the fit of this basis to a galaxy would yield exactly the same parameter estimates as the equivalent mixture of the original set of ssps , @xmath39 . in fig . [ protot ] we plot the 45 ssp spectra in cf05 and the 45 prototype spectra found by diffusion @xmath1-means . all spectra are normalized to 1 at @xmath8=4020 and are coloured by log age . the diffusion @xmath1-means prototypes spread themselves evenly over the range of spectral profiles , capturing a gradual trend from young to old spectra . on the other hand , the cf05 basis includes many similar spectra from younger populations and sparsely covers the range of spectral profiles of older populations . see algorithm 1 for a summary of the diffusion @xmath1-means algorithm . sample matlab and r code is available on the web at http://www.stat.cmu.edu/~annlee/software.htm . the diffusionmap r package is available at http://cran.r - project.org/. normalize ssp spectra , @xmath39 , at @xmath8 , fix @xmath81 compute similarity matrix , @xmath82 ( [ distance ] ) and weight matrix @xmath57 ( [ weights ] ) compute @xmath59 : @xmath58 decompose @xmath59 : @xmath83 project @xmath39 to the @xmath68-dimensional diffusion map , @xmath84 ( [ dmap2 ] ) set @xmath85=1 . fix @xmath1 . randomly partition data into sets @xmath86 @xmath87 for some @xmath7 @xmath88 compute cluster centroids , @xmath89 ( [ centroid ] ) partition the data so that @xmath90 @xmath1 protospectra ( [ eqn : protospec ] ) and corresponding parameters ( [ eqn : protot]-[eqn : protosmf ] ) in [ simulations ] , we compare the fits obtained by the diffusion @xmath1-means basis to those found by two other numerical methods : principal components ( pc ) @xmath1-means and standard @xmath1-means . each of these two methods is similar to diffusion @xmath1-means , but both have critical drawbacks . _ pc @xmath1-means _ works similarly to diffusion @xmath1-means ( algorithm 1 ) except that @xmath1-means is performed on the projection of the normalized ssp spectra into _ principal components _ ( pc ) space , not diffusion space . for an example of the application of principal components analysis to ssp spectra see @xcite . the main drawback to pc @xmath1-means is its assumption that the ssp spectra lie on a linear subspace of the original @xmath91 dimensional space . if the ssps actually lie on a non - linear manifold , then the @xmath1 prototypes may poorly capture the intrinsic variation of the original ssps because the non - linear structure will have been inappropriately collapsed on to a linear space by the principal components projection . _ standard @xmath1-means _ is also similar to algorithm 1 , except that @xmath1-means is performed in the original @xmath92 dimensional space ; i.e. , no reduction in dimensionality is done before running @xmath1-means . there are two obvious drawbacks to this procedure . first , the algorithm generally is slow because distance computations are cumbersome in high dimensions and @xmath1-means usually takes more iterations to converge . for comparison , the dimensionality of the spaces used by diffusion and pc @xmath1-means are each @xmath93 , a factor of 100 smaller than @xmath45 . second , and more importantly , appropriate prototype spectra are difficult to find by standard @xmath1-means because euclidean distances , used by @xmath1-means to define clusters , are only physically meaningful over short distances . diffusion @xmath1-means avoids this problem by clustering in diffusion space , in which euclidean distance approximates diffusion distance , a measure that has physical meaning on all scales . the result is that standard @xmath1-means inappropriately relates ssps that are not physically similar . in fig . [ prototz ] , we plot @xmath94 versus @xmath95 for @xmath1=150 prototypes in asa07 and diffusion , standard , and pc @xmath1-means . notably , diffusion @xmath1-means finds a much higher density of prototypes with high @xmath94 or high @xmath95 , reflecting the complicated manner in which ssps with those properties vary with respect to @xmath37 and @xmath60 . at the other extreme , the asa07 prototypes reside on a regular grid , and thus include many prototype spectra that are essentially identical and also exclude prototypes that have unique spectral properties . the standard and pc @xmath1-means prototypes also estimate more high @xmath94 and high @xmath95 prototypes than a regular grid . the methods introduced above use only the observable properties of ssps to choose representative prototypes . it may be the case that we want to incorporate other _ a priori _ information that we have about the ssps and their relationship with the galaxies we are fitting . for instance , we might know that a ssp with a particular age and metallicity is generally found in the types of galaxies we are trying to fit , and hence will want to include in our basis a prototype with characteristics closely matching those of this ssp . this information can easily be incorporated with the framework introduced above by defining an _ a priori _ weight , @xmath96 for each ssp in our database , where higher weights signify more importance of the ssp . by modifying the definition of the diffusion map geometric centroid ( [ centroid ] ) to be @xmath97 and altering the @xmath1-means algorithm to minimize @xmath98 instead of @xmath99 in ( [ kmeans ] ) , we choose a basis that reflects both the numerical observable properties of the ssps and our _ a priori _ knowledge . this weighted @xmath1-means algorithm can be used to correct for the tendency of the prototype ssps to depend on the specific set of ssps , @xmath39 , chosen . if , for example , we define @xmath100 as the inverse of the local density of ssp @xmath80 , then the prototype ssps in fig . [ sspdmap ] would more uniformly cover the manifold of ssps in diffusion space , and not tend to more heavily sample denser regions of the ssp manifold . in this manner , the prototype spectra will only depend on the spectral properties , and not the particular sampling of the set of ssps . we use simulations to assess the performance of the diffusion @xmath1-means method for selection of a basis of prototype ssp spectra compared to bases derived by other @xmath1-means methods and the bases of @xmath35 used by cf05 and @xmath101 used by asa07 . we analyse the performance of these methods on physically - motivated simulated spectra from two different prescriptions . all simulated galaxy spectra are generated as mixtures of a fine grid of 1278 ssp spectra from the model of @xcite , as described in [ models ] . we construct the simulated spectra with a fine grid of ssps because we expect real galaxies to consist of sub - populations of stars on a continuous grid of age and @xmath37 . this approach differs significantly from the simulation prescriptions of , for example , @xcite and @xcite as discussed in [ models ] . like @xcite , we simulate spectra over the range 3650 - 8000 . we bin all spectra to one measurement every 1 . for each simulated galaxy , we independently sample the reddening parameter , @xmath102 , and velocity dispersion , @xmath24 , from a realistic distribution based on fits to a large sample of sdss galaxies by the starlight code that was performed by the seagal ( semi empirical analysis of galaxies ) group and is available on the web at www.starlight.ufsc.br . we add gaussian noise following the normalized mean sdss error spectrum with specified s / n at @xmath103 . we mask out any spectral bins around emission lines , as done in @xcite . to choose the optimal value of @xmath53 for the diffusion @xmath1-means basis , we fit simulations to each basis from a grid of @xmath53 values and find the basis for which the fits achieve the smallest median @xmath104 value , ( [ chisq ] ) . although we choose the optimal @xmath53 by minimizing @xmath104 , our ultimate goal is accurate estimation of the sfh and physical parameters for each galaxy . however , we find that bases that achieve small @xmath104 values tend to have small errors in their physical parameter estimates . to compare results across different bases , we compare the errors in physical parameter and sfh estimation . our ultimate goal is to find the basis that most accurately describes the star formation and metallicity history of each galaxy . in this study , our main parameters of interest are @xmath105 @xmath102 , and @xmath24 . to test the ability of different bases to estimate these parameters , we compute the mean square error ( mse ) of the estimates across the simulated galaxies . to characterize the ability to recover sfh , we compute condensed population vectors by binning both @xmath12 and @xmath16 into 5 logarithmically spaced time bins and then comparing these true condensed vectors , @xmath17 and @xmath18 to the estimated condensed vectors , @xmath106 and @xmath107 . other techniques have proposed different schemes for compressing the sfh information retrieved by spectral fitting ( see @xcite for an overview of the different compression methods employed in current spectral synthesis codes ) . our choice of 5 age bins is an attempt to avoid the degeneracies that are likely if the population vector were gridded into very fine age bins while still retaining a sufficient number of age bins to detect significant recent starburst events . admittedly , the choice of 5 age bins is ad hoc . however , because we are only interested in comparing the ability of different bases of prototype spectra to recover sfh , any reasonable parametrization of the sfh will allow us to perform this analysis . in the first set of simulations , we choose a sfh with exponentially decaying star formation rate ( sfr ) : sfr @xmath108 , where @xmath60 is the age of the ssp in gyr and we use ssps up to a maximum of 14 gyrs . like @xcite , we set @xmath109 to have a sfh that is neither dominated by recent star formation nor too similar to a single old burst galaxy . in our simulations , we consider _ specific _ sfr ( ssfr ) as defined in @xcite by normalizing all simulated spectra to have the same mass . this is reasonable because in practice we are interested in how the sfh varies across many galaxies , and therefore wish to remove the absolute mass - scale dependence of each object s sfr . we have ssps from 196 time bins where each bin s metallicity is randomly chosen from 6 different values . we test our methods on 50 simulated spectra . across different spectra the sfh is the same , while @xmath110 and the random noise all vary . each spectrum has s / n=10 at @xmath8 . the following bases of prototype spectra are used to fit each simulated spectrum : 1 ) the cf05 basis , 2 ) diffusion @xmath1-means , 3 ) pc @xmath1-means , and 4 ) standard @xmath1-means . each basis contains @xmath35 prototype spectra . for diffusion @xmath1-means , we perform fits over a grid of values of @xmath53 to study how parameter estimates vary as a function of @xmath53 . in fig . [ toj : err ] we show the error in spectral fits using diffusion @xmath1-means for different values of @xmath53 . also plotted is the error in the standard @xmath1-means ( sk ) and pc @xmath1-means ( pc ) bases . all errors are plotted as the ratio of the @xmath104 statistic to the @xmath104 from the cf05 basis fit . in the vast majority of the simulated spectra , all three of the techniques considered produce better spectral fits than the cf05 basis . there are no significant differences between spectral fit errors for the diffusion @xmath1-means bases as @xmath53 varies . the minimum median @xmath104 for diffusion @xmath1-means is achieved at @xmath111 . we plot the mse of our estimates for @xmath112 , and @xmath24 in fig . [ toj : mse ] . for each parameter , the cf05 basis predictions are the worst in terms of mse . the performance of the diffusion @xmath1-means basis is not very dependent on the particular choice of @xmath53 . this is a nice property , as any reasonable choice of @xmath53 will give us good results , and we need not spend computing time searching for an optimal value of @xmath53 . diffusion @xmath1-means bases tend to find the best estimates of @xmath113 , @xmath102 , and @xmath24 , and performs competitively with the other @xmath1-means bases for @xmath114 estimation . we compare the condensed sfh estimates across bases . plots of the recovered @xmath106 and @xmath107 vectors in fig . [ toj : sfh1 ] show that the cf05 basis tends to significantly underestimate the contributions of the oldest population while overestimating the contribution of the second oldest population . all three @xmath1-means bases find estimates that are closer to the true @xmath106 , but tend to underestimate the contribution of the oldest @xmath106 component . all three @xmath1-means bases find very accurate estimates of @xmath107 . the bias in @xmath106 estimation occurs for at least two reasons . first , since the simulations are generated from ssps on a fine age grid , it is intrinsically difficult to estimate all of the contributions of the subpopulations . second , due to the exponential form of the sfr , older populations are the most important in the sfh , making it easy to underestimate their importance and almost impossible to overestimate their contribution . for this simple prescription of galaxy evolution , bases constructed using @xmath1-means methods find significantly better estimates of physical parameters and sfh than the cf05 basis . next , we test our methods on a slightly more complicated set of simulations and also compare bases of size 45 and 150 to characterize the advantage , if any , of using a larger basis and whether this advantage is worth the extra expenditure in computation time . in our second set of simulations , we use a prescription similar to @xcite to create a set of more realistic spectra on which to test our methods . the added sfh complexities over the simulations in [ sim : exp ] are : 1 . the time @xmath115 when a galaxy begins star formation is distributed uniformly between 0 and 5.7 gyr after the big bang , where the universe is assumed to be 13.7 gyr old . we again have sfr @xmath108 , but now allow @xmath116 to vary between galaxies . for each galaxy we draw @xmath116 from a uniform distribution between 0.7 and 1 @xmath117 . 3 . we allow for starbursts with equal probability at all times . the probability a starburst begins at time @xmath60 is normalized so that the probability of no starbursts in the life of the galaxy is 33% . the length of each burst is distributed uniformly between 0.03 and 0.3 gyr and the fraction of total stellar mass formed in the burst in the past 0.5 gyr is distributed logarithmically between 0 and 0.1 . the sfr of each starburst is constant throughout the length of the burst . as above , each galaxy spectrum is generated as a mixture of ssps of up to 196 time bins , with a randomly drawn metallicity in each bin . allowing for random starbursts over the exponential sfh increases the importance of younger populations in the simulated galaxies . we test our methods on 500 simulated spectra , 250 with s / n=10 and 250 with s / n=20 at @xmath8 . across each spectrum , sfh , @xmath118 and the random noise all vary . we fit bases of size 45 and 150 . for this set of simulations , we only look at results for the diffusion @xmath1-means @xmath53 that achieves the lowest @xmath104 fit error for each @xmath1 ; the optimal values are @xmath53=250 for the @xmath35 basis and @xmath53=500 for the @xmath101 basis . as before , results are not very sensitive to the specific choice of @xmath53 . note that the @xmath53=250 basis of size 45 also achieved good fits to the simpler simulations in [ sim : exp ] . in fig . [ wild : mse ] we plot the mse of the physical parameter estimates for each of the 4 bases for @xmath1=45 and 150 . the diffusion @xmath1-means bases perform significantly better than all other bases for mean age and @xmath102 estimation and achieve similar results as the other @xmath1-means bases for @xmath119 and @xmath24 estimation . the @xmath1-means bases heavily outperform the hand - chosen bases ( cf05 & asa07 ) for each parameter . the same trends are seen for error in estimation of time - binned star formation histories ( fig . [ wild : sfh ] ) . the error estimates for all bases , split into s / n=10 and s / n=20 galaxy spectra , are in table [ wild : tab ] . [ cols="^,^ , < , < , < , < , < , < , < , < , < , < , < , < " , ] an interesting trend appears in figs . [ wild : mse]-[wild : sfh ] . for the hand - chosen bases , we see considerable reduction in the estimation error of every parameter in going from the @xmath1=45 basis of cf05 to the @xmath1=150 basis of asa07 . however , in the @xmath1-means bases , and particularly for the diffusion @xmath1-means basis , there is no substantial lowering of the mse of any parameter in increasing from @xmath1=45 to @xmath1=150 . this tells us that even with as few as 45 prototypes , diffusion @xmath1-means provides us a basis that can successfully fit these particular simulations . moving to @xmath1=150 prototypes allows us no significant gains in parameter estimation while costing us @xmath12010 times the computational time . note , however , that although a basis of size 45 may be sufficient for this set of simulations , a set of real galaxy spectra may be more complicated and require a larger basis of prototypes to achieve good fits . a common difficulty in population synthesis studies is the so - called age - metallicity degeneracy , where young , metal - rich ssps are confused with old , metal - poor ssps . using a regular grid of ssp prototypes in population synthesis can exacerbate this problem because several ssps with nearly identical spectra ( but different ages and metallicities ) will likely be included , inducing degeneracy in the spectral model ( [ cfmodel ] ) , and causing the estimated sfh parameters for each galaxy to be highly variable . a diffusion @xmath1-means basis , on the other hand , picks ssp prototypes whose spectra more evenly span the range of spectral characteristics , minimizing the inclusion of very similar spectra and thus decreasing the degeneracy effects in the minimization of ( [ chisq ] ) . in fig . [ wild : agez ] we see that the age - z degeneracy is much smaller in diffusion @xmath1-means bases , quantified by the slope of the linear trend in @xmath121 versus @xmath122 . some degeneracy in age - metallicity still exists for the diffusion @xmath1-means bases , but the magnitude of the slope of the trend is half . one may note that both @xmath114 and @xmath123 are significantly biased for all 4 bases in fig . [ wild : agez ] . this probably occurs due to the simulation prescription , which includes mostly old ssps from a large , fine grid of age and z. one may be inclined to use these estimated biases to correct the estimates of real galaxy spectra . however , the amount of bias is dependent on the particular prescription of sfh in the simulations . since there is no general model that describes galaxy sfh for a wide range of galaxies , there is no straightforward way to estimate the bias , if any , that will occur . in this section , we apply the fitting techniques to a sample of sdss spectra to compare the results using different bases . we compare the physical parameters estimated from each of three different bases : asa07 , diffusion map with @xmath1=45 , and diffusion map with @xmath1=150 . we also compare parameter estimates to those estimated with the mpa / jhu data base . our goal in this section is to study how the choice of basis affects the final parameter estimates for a large set of real galaxy spectra . our data sample consists of spectra from ten arbitrarily chosen spectroscopic plates of sdss dr6 ( 0266@xmath1240274 inclusive , and 0286 ; @xcite ) that are classified as galaxy spectra . the goal is to test our methods on a sample that is large enough that we include galaxies with a broad range of star formation and metallicity histories but not so large that analysis is computationally cumbersome . in a future work , we will present results for the more than 700k galaxy spectra in sdss dr7 . we remove spectra from this sample by applying three cuts . the first , motivated by aperture considerations , is to analyse only spectra whose sdss redshift estimates @xmath125 0.05 . second , we determine the proportion of the 3850 wavelength bins that are flagged as bad . if this proportion exceeds 10% , we remove the spectrum from the sample ; if not , we retain the spectrum for further analysis . third , we select galaxy spectra with a sdss redshift confidence level @xmath125 0.35 . these cuts leave us with a sample of 3046 galaxy spectra . we bring all spectra to rest frame using the sdss redshift estimates . in the fitting routines , we exclude any spectral bins with sdss flag @xmath125 2 . this eliminates any sky residuals , bad pixels , and artefacts from the data reduction . we additionally mask the pixels in the regions within and around emission lines as in @xcite . results of the starlight spectral fits give a mean @xmath126 of 1.03 for each of the three bases used , close to a value of 1 that would be expected if wavelength bin fluxes were independently distributed gaussian random variables . note that this value varies considerably from the mean value of 0.78 found in the analysis of 50362 sdss spectra by @xcite . this difference can be attributed to our careful handling of spectral bin error estimates in our rebinning of the spectra to a dispersion of 1 measurement per . the starlight fitting routine is used to fit all 3046 sdss spectra with each of the three bases . each basis yields parameter estimates that are substantially different . in fig . [ sdss : tz ] we plot each galaxy s @xmath119 versus @xmath127 estimate for each basis . there are several notable differences in the @xmath128 plane for each basis . first , both diffusion map bases produce estimates that are tightly centered around an increasing linear trend while the asa07 estimates are diffusely spread around such a trend . moreover , the direction of discrepancy in the asa07 estimates from the linear trend corresponds exactly with the direction of degeneracy between old , metal - poor and young , metal - rich galaxies . this suggests that the observed variability along this direction is not due to the physics of these galaxies , but rather is caused by confusion stemming from the choice of basis . secondly , the slopes of the estimated trends produced by the diffusion map bases ( 0.146 for @xmath1=45 and 0.180 for @xmath1=150 ) are considerably smaller than the slope of the trend in the asa07 estimates ( 0.250 ) . these estimates have implications for the physical models that govern galaxy evolution . finally , the @xmath1=45 diffusion map basis estimates no young , metal - poor galaxies , showing a sharp decline from densely - populated regions of the parameter space to vacant regions . this suggests that this small number of prototypes is not sufficient to cover the parameter space ; particularly , young , metal - poor ssps have been neglected in the basis . next we compare kinematic parameter estimation . [ sdss : kin ] shows that the asa07 basis tends to consistently estimate slightly larger values of @xmath102 than either diffusion map basis . on average , asa07 estimates of @xmath102 are 0.045 larger than diffusion map @xmath1=45 estimates and 0.024 larger than diffusion map @xmath1=150 estimates . estimates of @xmath24 are consistent between all three bases . finally , we compare age - binned sfh estimates found by each basis . like @xcite , we condense the sfh to three parameters : old , @xmath129 ( @xmath130 yr ) , intermediate , @xmath131 ( @xmath132 yr ) , and young , @xmath133 ( @xmath134 yr ) . plots of diffusion map estimates of each component versus asa07 estimates are in fig . [ sdss : sfh ] . the estimated sfhs show a high degree of correspondence between the bases . on average , each diffusion map basis obtains slightly higher estimates of @xmath129 and slightly lower estimates of both @xmath131 and @xmath133 . we compare the sdss parameters estimated with the three different choices of starlight basis with parameters obtained using measured emission line and continua values from the mpa / jhu data base . the mpa / jhu collaboration has made their sdss spectral measurements publicly available on the mpa sdss website ( @xcite ) . although we do not expect any of the basis choices to produce estimates that perfectly correspond to the independent mpa / jhu estimates , we analyse the relationships between the estimates as a further test of the accuracy of each basis . we use the balmer emission - line ratio @xmath136 to estimate the nebular extinction , @xmath135 , of each galaxy . following @xcite , we assume case b recombination , a density of 100 @xmath137 and a temperature of 10@xmath138 k to compute @xmath135 according to ( [ aneb1])-([aneb2 ] ) : @xmath139 where we have used the same constants used in @xcite . [ jhu : a ] shows plots of the values of @xmath135 computed from the mpa / jhu data base versus the values of the stellar extinction , @xmath102 , computed by each of the three different bases in the starlight code . we have only plotted the 1755 sdss galaxies in our sample whose @xmath140 and @xmath141 emission line measurements both have signal - to - noise ratio @xmath142 . we find that @xmath135 and @xmath102 are loosely linearly related for each basis . the estimated spearman correlation is 0.51 for the asa07 basis and 0.52 for each diffusion map basis . the slope of the linear trend is slightly higher for the asa07 basis ( 1.21 ) than for the diffusion map bases ( 1.16 for each basis ) . all three bases obtain stellar extinction estimates that are consistent with the nebular extinction values @xmath135 , and there are no significant differences between the strength of the relationships between the estimates of each basis and the nebular extinction estimates . as in @xcite , we compare our spectral synthesis estimates to two age indicators from the mpa / jhu database : @xmath143 ( @xcite ) and @xmath144 ( @xcite ) . the value @xmath143 describes the break in the 4000 region of the galaxy spectrum ; higher values generally correspond to increasing ages of the galaxy s stellar subpopulations . larger strength of the @xmath145 absorption line , @xmath144 , indicates higher star formation rates in the past 0.1 - 1 gyr of a galaxy s lifetime . hence , we expect our estimates of @xmath123 to correlate positively with @xmath143 and negatively with @xmath144 . in fig . [ jhu : t ] , we see that the @xmath123 estimates show the expected relationships with both @xmath143 and @xmath144 for each of the three bases . correlations are strongly positive for the @xmath146 trend and strongly negative for the @xmath147 trend . we have plotted the 3031 galaxies in our sample that have reliable @xmath143 and @xmath144 measurements in the mpa / jhu catalogue . there is no significant difference in each relation between the bases . lastly , we compare our starlight estimates of @xmath148 with a measure of nebular metallicity based on the ratio of emission lines [ oiii]@xmath150 and [ nii]@xmath151 , provided by mpa / jhu . particularly , the quantity @xmath152 measures the oxygen abundance of a galaxy , and is defined as @xmath153 where o@xmath154n@xmath155\lambda 5007 / [ { \rm nii}]\lambda 6583)$ ] ( @xcite ) . all emission line fluxes are first corrected for extinction using the @xmath102 estimates found in starlight fits . we would expect @xmath156 to be positively related to @xmath152 , as the former measures stellar metallicity and the latter measures nebular oxygen abundance . in fig . [ jhu : z ] we see that all three bases find @xmath157 estimates that correlate positively with @xmath152 , and that the @xmath37=150 diffusion map basis estimates correlate more strongly with @xmath152 , with spearman correlation coefficient of 0.21 , compared to the asa07 basis estimates , which have a correlation of 0.12 with the @xmath152 estimates . furthermore , the best - fit linear trend of the @xmath152 versus diffusion map @xmath1=150 estimates has a significantly higher slope , 0.128 than the linear trend of @xmath158 vs. asa07 basis , 0.056 . this result suggests that the diffusion map basis obtains more reliable metallicity estimates than the asa07 basis . we have presented a method to choose an appropriate basis of ssp prototypes to be used in population synthesis models of observed galaxy spectra . particularly , we have used the starlight fitting routines of @xcite to study the ability of both literature and diffusion @xmath1-means bases to model both simulated and sdss galaxy spectra . on simulated galaxies , our diffusion @xmath1-means basis outperforms hand - chosen literature bases as well as bases constructed from standard @xmath1-means or pc @xmath1-means . for real sdss galaxy spectra , we show that the diffusion @xmath1-means basis finds parameter estimates that differ substantially from the estimates retrieved by the basis of @xcite . the diffusion @xmath1-means estimates seem to alleviate the age - metallicity degeneracy that has been detrimental to previous population synthesis studies of large data bases of galaxies . at this point , more discussion of the age - metallicity degeneracy is warranted . degeneracy arises because several ssps have very similar spectral characteristics , but different ages and metallicities . when degenerate ssps are included , estimates can be highly variable . this is shown in the first panel in fig . [ sdss : tz ] and in the starlight code itself , which in the presence of degeneracy tends to converge to different solutions due to the random walk taken through the parameter space by the metropolis algorithm . the diffusion @xmath1-means algorithm , on the other hand , includes only the mean spectrum and mean parameters of each set of degenerate spectra . this choice minimizes the statistical risk of the final solution by decreasing the variance and bias of the estimates ( our measure of statistical risk is the mse , which is the sum of the variance and squared bias of an estimate ) , as seen in simulations ( figs . our sfh estimates for sdss spectra are significantly less variable in the direction of the age - z degeneracy , as seen in fig . [ sdss : tz ] , while the relative amount of bias is impossible to quantify without knowledge of the true sfh of each galaxy . in the face of degeneracies and with no other outside information available , our proposed method ( of picking the mean of several degenerate solutions ) is the best one can do in terms of minimizing the statistical risk of the estimates . however , with more information about what features in the ssp spectra are most strongly related to their ages and metallicities , we could potentially improve our estimates by incorporating such information in the construction of the similarity measure @xmath50 in ( 3 ) . the weight matrix produced by an expert - produced similarity measure may produce a more complete coverage of the parameter space in our ssp prototypes and help to remove degeneracies in the parameter estimates . further study is required in this direction . the ultimate goal of this project is to be able to quickly and accurately estimate physical property and sfh parameters for a large data base of galaxies . as an extension of this work , we plan to study how to quickly and effectively extend estimates found for a small set of galaxies using the computationally intensive starlight routine to a large database of galaxies . to achieve this goal , we must exploit the low - dimensional structure of these data . this approach was taken by @xcite in galaxy redshift estimation . these methods must be further developed and refined to reliably estimate sfh parameters . the methods presented in this paper are applicable in a wide range of problems in astrophysical data analysis . diffusion map is a powerful tool to uncover simple structure in complicated , high - dimensional data , whether they be spectra , photometric colours , two - dimensional images , data cubes , etc . specifically , diffusion map is both more flexible and more widely applicable than pca , which relies on the linear correlation structure of the data and assumes that the data reside on a low - dimensional , linear manifold . the diffusion @xmath1-means method can be used in a variety of problems often encountered in astrophysics where a large set of observed or model data are described with a few prototype examples . future work will attempt to more concretely define criteria for the performance of algorithms that define @xmath1 prototypes from @xmath41 data points and to theoretically justify the use of diffusion @xmath1-means . we plan to explore more deeply the relationship between the @xmath1-means methods introduced here and to compare the methods introduced in this paper to other methods of basis selection , such as basis pursuit and matching pursuit . we also will investigate how meaningful standard errors can be attached to our parameter estimates using any choice of basis . we thank the reviewer for their helpful comments . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . 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 web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , cambridge university , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . adelman - mccarthy , j. k. , et al . 2008 , apjs , 175 , 297 alongi , m. , bertelli , g. , bressan , a. , chiosi , c. , fagotto , f. , greggio , l. , and nasi , e. , 1993 , a&as , 97 , 851 , n. v. , cid fernandes , r. , stasiska , g. , torres - papaqui , j. p. , mateus , a. , sodr , l. , schoenell , w. , and gomes , j. m. , 2007 , mnras , 381 , 263 balogh , m. l. , morris , s. l. , yee , h. k. c. , carlberg , r. g. and ellingson , e. , 1999 , apj , 527 , 54 bica , e. , 1988 , a&a , 195 , 76 bressan , a. , fagotto , f. , bertelli , g. , chiosi , c. , 1993 , a&as , 100 , 647 , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , and brinkmann , j. , 2004 , mnras , 351 , 1151 bruzual , g. , and charlot , s. , 2003 , mnras , 344 , 1000 cardelli , j. a. , clayton , g. c. , and mathis , j. s. , 1989 , apj , 345 , 245 chabrier , g. , 2003 , pasp , 115 , 763 chen , x. y. , liang , y. c. , hammer , f. , zhao , y. h. and zhong , g. h. , 2009 , a&a , 495 , 457 cid fernandes , r. , sodre , l. , schmitt , h. r. , leao , j. r. s. , 2001 , mnras , 325 , 60 cid fernandes , r. , gu , q. , melnick , j. , terlevich , e. , terlevich , r. , kunth , d. , rodrigues lacerda , r. , and joguet , b. , 2004 , mnras , 355 , 273 , r. , mateus , a. , sodr , l. , stasiska , g. , and gomes , j. m. , 2005 , mnras , 358 , 363 coifman , r. r. , & lafon , s. , 2006 , appl . , 21 , 5 fagotto , f. , bressan , a. , bertelli , g. , chiosi , c. , 1994a , a&as , 104 , 365 fagotto , f. , bressan , a. , bertelli , g. , chiosi , c. , 1994b , a&as , 105 , 29 freeman , p. e. , newman , j. a. , lee , a. b. , richards , j. w. , and schafer , c. m. , in preparation . gallazzi , a. , charlot , s. , brinchmann , j. , white , s. d. m. , and tremonti , c. a. , 2005 , mnras , 362 , 41 girardi , l. , bressan , a. , chiosi , c. , bertelli , g. , and nasi , e. , 1996 , a&as , 117 , 113 glazebrook , k. , et al . , 2003 , apj , 587 , 55 kauffmann , g. et al . , 2003 , mnras , 341 , 33 lafon , s. , & lee , a. , 2006 , ieee trans . pattern anal . and mach . , 28 , 1393 le borgne , j .- f . , et al . , 2003 , a&a , 402 , 433 leitherer , c. , robert , c. , heckman , t. m. , 1995 , apjs , 99 , 173 mas - hesse , j. m. & kunth , d. , 1991 , a&as , 88 , 399 mathis , h. , charlot , s. , and brinshmann , j. , 2006 , mnras , 365 , 385 moultaka , j. , boisson , c. , joly , m. , pelat , d. , 2004 , a&a , 420 , 459 ocvirk , p. , pichon , c. , lancon , a. , and thiebaut , e. , 2006 , mnras , 365 , 46 osterbrock , d. e. , 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) panter , b. , heavens , a. f. , and jimenez , r. , 2003 , mnras , 343 , 1145 panter , b. , jimenez , r. , heavens , a. f. and charlot , s. , 2007 , mnras , 378 , 1550 pelat , d. , 1997 , mnras , 284 , 365 pettini , m. and pagel , b. e. j. , 2004 , mnras , 348 , l59 reichardt , c. , jimenez , r. , and heavens , a. f. , 2001 , mnras , 327 , 849 richards , j. w. , freeman , p. e. , lee , a. b. , schafer , c. m. , 2009 , apj , 691 , 32 ronen , s. , aragn - salamanca , a. , and lahav , o. , 1999 , mnras , 303 , 284 strauss , m. a. et al . , 2002 , aj , 124 , 1810 tremonti , c. , 2003 , phd thesis , johns hopkins univ . tremonti , c. a. et al . , 2004 , apj , 613 , 898 tojeiro , r. , heavens , a. f. , jimenez , r. , and panter , b. , 2007 , mnras , 381 , 1252 worthey , g. , 1994 , apjs , 94 , 687 worthey , g. and ottaviani , d. l. , apjs , 111 , 377 york , d. g. et al . , 2000 , aj , 120 , 1579

we examine a classically - chaotic system consisting of two reservoirs of particles connected by a channel containing oscillating potential - energy barriers . we investigate whether such a system can preferentially pump particles from one reservoir to the other , a process often called `` quantum pumping . '' we show how to make a `` particle diode '' which under specified conditions permits net particle pumping in only one direction . then we examine systems having symmetric barriers . we find that if all initial particle energies are considered , a system with symmetric barriers can not preferentially pump particles . however , if only finite initial energy bands are considered , the system can create net particle transport in either direction . we study the system classically , semiclassically , and quantum mechanically , and find that the quantum description can not be fully understood without the insight gained from classical and semiclassical analysis . = 1

1510.02860

a ballistic atom pump is a system containing two or more reservoirs of neutral atoms or molecules and a junction connecting them containing a time - dependent potential . `` ballistic '' means that atoms move through the pump as independent particles . the theoretical description may be given by classical , semiclassical , or quantum theories . atom statistics may be bose , fermi , or boltzmann . we intend that the definition of `` ballistic atom pumps '' be interpreted broadly ( however , systems having intrinsically many - body phenomena such as viscosity should not be called `` ballistic ; '' if there are interactions among the particles , these interactions can be described by an average single - particle potential ) . particle transport is an ongoing topic of interest in a variety of systems from solid state circuitry to microfluidic devices to futuristic atomtronic components . since the advent of laser cooling , precise control and manipulation of neutral ultracold atoms has attracted attention to atomic systems that can mimic more challenging systems . one such phenomenon in electronic solid - state systems describes electronic transport through mesojunctions @xcite having time - dependent potential barriers , a phenomenon often called `` quantum pumping''@xcite . the choice of potential likewise emulates the turnstile quantum pump usually studied in mesoscopic electronics @xcite . although quantum pumping has been theorized for decades @xcite , there has only recently been an experimental realization of such a system due to the challenges of overcoming capacitive coupling and rectification effects in electronic systems @xcite . recent proposals have suggested bypassing these difficulties by simulating a quantum pump in a system of neutral cold atoms @xcite . neutral atom transport is also becoming increasingly important in its own right due to the ongoing development of atomtronics , which seeks to replicate critical tools of electronics in neutral atoms . analogues of batteries , diodes , transistors , and recently hysteresis @xcite have been explored in ultracold neutral atom systems . the motivation behind such devices is multifarious . unlike their electronic counterparts , these systems allow scientists to study analogous tools in well - controlled and idealized environments like optical lattices . additionally , long coherence times provide unique opportunities for quantum state preparation , storage , and readout , making atomtronic devices a serious competitor as a basis for quantum computers @xcite . finally , neutral atoms present degrees of freedom not available in their electronic counterparts , such as bosons , fermions , and scalable interactions . in this paper , we present a detailed study of the classical and quantum features of a ballistic atom pump which has potential applications such as a battery , diode , or rectifier in atomtronic circuits . the pumps we consider in this paper have two reservoirs and a pump which is effectively one - dimensional , so the hamiltonian is @xmath0 we choose @xmath1 to consist of two repulsive barriers oscillating with the same frequency @xmath2 , but not necessarily with the same amplitude or phase . we study rectangular barriers ( easiest theoretically ) and gaussian barriers ( easiest experimentally using optical forces ) . the questions we address are : can such systems pump atoms preferentially from one side to the other without an external bias , such as a difference in chemical potentials in the reservoirs ? in particular , can we make an atom `` diode '' that will allow atoms to pass through the pump in only one direction ? in order to understand the quantum features of such a pump , it is necessary to develop a clear understanding of classical scattering by a pair of oscillating potential barriers that function as a turnstile pump . we begin with a precise specification of the models we study . then we consider simple asymmetric pumps that rectify net particle transport , which we call `` particle diodes '' because they allow transport in only one direction for certain ranges of initial particle energy . these diodes have one barrier fixed and one oscillating barrier . then we consider pumps that are symmetric in the sense that the two barriers are identical , but their oscillations are not in phase with each other . we prove a symmetry theorem which shows that such pumps can give no net particle pumping if the behavior of the particles is classical and the initial phase - space distribution is uniform in both reservoirs . however , if the phase - space distribution is not uniform , then such pumps can produce net particle transport in either direction . we also show that if the two potential barriers are separated by a modest distance , atoms can get stuck in a `` complex '' or resonance zone between them , and the system is a nice model of chaotic transport @xcite . ( in a separate paper @xcite we have provided a topological description of this chaotic transport ) . the relationships among classical , semiclassical , and quantum descriptions for transport past a single oscillating gaussian barrier were discussed in detail in @xcite . consider the case that atoms enter the pump from one side with fixed momentum @xmath3 and kinetic energy @xmath4 . in the quantum description , because the barriers are oscillating with a fixed frequency , floquet theory tells us that after passing through the pump , the spectrum of transmitted energies is a set of narrow peaks at energies @xmath5 , where @xmath6 is an integer . the heights of these peaks can be computed numerically by solving the schroedinger equation , but in general no patterns are visible in those heights . in the classical description ( again assuming that particles enter with a fixed initial momentum @xmath3 but a range of positions @xmath7 ) , then the final momentum @xmath8 is a bounded periodic function of the initial position @xmath7 , @xmath9 . the upper and lower bounds of the range of this function define the classically allowed region . inside this classically allowed region , provided that @xmath10 is continuous , there must be an even number of trajectories leading to each final momentum . the distribution of final momenta is a smooth function except at extrema of @xmath10 , where the distribution has an integrable singularity . one finds that the floquet peaks obtained in the quantum description are large primarily in the classically allowed region , with small spillover past the boundaries ( momentum - space tunneling ) . still the heights of peaks are incomprehensible . finally , in semiclassical theory , for each final momentum one sums over the initial positions that give trajectories leading to that final momentum , and incorporates phases for each such orbit ( momentum - space action plus maslov indices ) . summing over one cycle of @xmath11 produces a smooth function , and the relative heights of the floquet peaks are discrete values of it . summing over many cycles of @xmath11 causes the peaks seen in the quantum description to emerge , with good agreement between the two methods ( see figs . [ gaussrects ] and [ gauss2osc ] ) . we show a few representative calculations of each type in this paper , but we concentrate on the classical description , with the understanding that semiclassical calculations can be carried out when desired , and that the semiclassical description agrees well with the quantum description . our atom pump consists of two repulsive potential barriers with amplitude oscillations that have the same frequency , but are not in phase with one another : @xmath12 in this paper we examine both rectangular and gaussian potentials . the rectangular barrier potentials are given by : @xmath13 where @xmath14 , @xmath15 , @xmath16 , @xmath17 , @xmath18 is the average height of each barrier , @xmath19 is the amplitude of oscillation of each , @xmath20 is the common frequency and @xmath21 is the period , @xmath22 is an additional phase term , and @xmath23 is width of each barrier . the left and right barriers are centered at @xmath24 and @xmath25 , respectively , and always have a center - to - center distance of @xmath26 . when the barriers touch , i.e. , have no separation , @xmath27 . if only the left - hand barrier is oscillating then @xmath28 . the gaussian potential barriers are given by @xmath29 where @xmath30 is the standard deviation of the gaussian . [ barrpar ] shows the parameters for the barriers . and have width @xmath23 . if @xmath31 , the barriers touch . ( b ) gaussian barriers centered at @xmath32 which have standard deviation @xmath30 . ] without loss of generality , we can choose units of mass , energy and time such that @xmath33 , @xmath34 , and @xmath35 . the remaining parameters are the barrier widths @xmath36 and @xmath37 , the barrier oscillation amplitudes @xmath38 and @xmath39 , and the phase difference between the barriers , @xmath22 . in this paper , we typically choose @xmath40 . in quantum and semiclassical mechanics one additional parameter arises , the value of @xmath41 , which we set as @xmath42 . the general way to apply such scaling principles is given in @xcite . the units used in this paper are theoretical , and are the same as those used in @xcite . namely , the choice of a theoretical unit convention based on @xmath42 and @xmath33 is equivalent to selecting an arbitrary time unit @xmath43 and a related length unit @xmath44 , with @xmath45 j@xmath46s . the corresponding energy unit is @xmath47 , while the mass unit is that of the particle , @xmath48 . we start with a distribution of particles far to the left of the barriers , far to the right , or both . for our classical calculations , the distribution has a single momentum ( i.e. it is a delta - function in momentum space centered at @xmath3 ) . the distribution in position space is uniform over a length @xmath49 where @xmath50 is the initial velocity of the particles ( i.e. uniform over a length corresponding to the distance the incident particles travel in one cycle of the barriers ) . in semiclassical and quantum calculations , we begin with a wave packet that is narrow in momentum space , centered at @xmath3 , and correspondingly wide in position space , @xmath51 . thus its magnitude is nearly uniform over the length @xmath52 corresponding to a cycle . the wave function in position space at the initial time is given by @xmath53 where @xmath54 is @xmath55the initial probability density is thus @xmath56 , which is a gaussian centered at @xmath57 with standard deviation @xmath58 . our quantum calculations are performed in the same fashion as in @xcite , and are based on propagating the wave packet with the time - dependent schrdinger equation via a split - step operator method @xcite . we determine the net particle transport in these systems by the following process : 1 ) for each initial momentum , launch particles toward the barriers from the left , and compute and record the fraction transmitted and reflected . also record the final momenta of transmitted and reflected particles . 2 ) launch particles with the same initial energy toward the barriers from the right , and compute the fraction transmitted and reflected , and their final momenta . 3 ) sum the results of each of these to obtain the net fraction of particles transmitted left to right ( which may be negative if more are transmitted from right to left ) . 4 ) if appropriate , average over initial momenta . we define the fractional transport of particles through the pump as @xmath59 where @xmath60 is the number of particles per cycle scattered to the right for each @xmath61 , and @xmath62 is the number of particles per cycle scattered to the left . the sum @xmath63 represents all particles for a given @xmath61 . @xmath64 is positive when more particles are scattered to the right for a given @xmath61 , and negative when more particles are scattered to the left . when equal numbers of particles scatter to the right and left , e.g. when all particles are reflected or transmitted , @xmath65 . a double - barrier particle pump can make a kind of diode , in which net particle pumping can only be in one direction if the initial particle energy is sufficiently small . this type of diode consists of one static barrier which is high enough to prevent transmission of particles incident from one direction , and one oscillating barrier which can lift particles approaching from the other direction over the static barrier . this is analogous to photon - assisted tunneling @xcite . when the incident energy of particles is greater than the height of the static barrier , net particle transport is only possible in the opposite direction . it is simplest if the two barriers are touching each other . let us simplify the description of the potentials to @xmath66 where @xmath67 is a periodic function of @xmath68 with period @xmath69 , and @xmath70 is a constant . suppose @xmath71 then particles incident from the right with kinetic energy @xmath72 can not pass over the right barrier . from the left , ( see fig . [ rectexample](a ) ) a stream of particles having fixed kinetic energy @xmath72 and density independent of position all file into the elevator when it is on the ground floor ( fig . [ rectexample](b ) ) , and then at @xmath73 they are lifted abruptly to the penthouse level on the roof , where the back door of the elevator opens ( fig . [ rectexample](c ) ) . the particles keep their kinetic energy in this process , and politely file out in line onto the roof ( fig . [ rectexample](d ) ) . at @xmath74 they are all lined up on top of the right - hand barrier , and one - by - one they slide down the edge of that barrier and escape to the right with kinetic energy @xmath75 ( fig . [ rectexample](e ) ) . meanwhile the door of the elevator has slammed again and it abruptly returns to ground level . , and particles on top of it gain enough energy to transmit past the right barrier , as seen in ( d ) and ( e ) . ] for this system , if the barrier width is @xmath76 and if the first particle arrives at the left edge of the left barrier at @xmath77 , half of the particles incident on the elevator from the left the ones that arrive for @xmath78(mod@xmath79 go over the barrier , and the other half arriving for @xmath80(mod@xmath81 are reflected by the left - hand barrier , so the transmitted fraction is half of the incident fraction . is this the theoretical maximum for transport ? we can not think of any other function @xmath67 that would improve the performance . however , we can get a larger fraction transmitted if on the elevator for @xmath82(mod@xmath81 a pusher shoves the passengers to the right so they exit the elevator more quickly . clearly , when the energy of incident particles is less than the amplitude of the static barrier , the only possible direction of fractional particle transport is left - to - right . this is because all particles approaching from the right are reflected , while some particles incident from the left may hop onto the oscillating barrier and gain enough energy from it to transmit over the static barrier . on the other hand , suppose the energy of incident particles is higher than the peak of the static barrier , and suppose that an equal number of particles approaches from the left and from the right . then the only possible direction of net fractional transport is the opposite direction , from right - to - left . in this regime , all particles approaching from the right transmit over both barriers , but particles incident from the left may lose energy while riding the oscillating barrier down , and can then be reflected from the static barrier . the pumping mechanism of a diode is easily pictured by thinking about rectangular barriers , but it also applies to smooth barriers with smooth time dependence . to keep the analysis simple , let us consider rectangular elevators with some smooth dependence on @xmath68 . again particles approach from the left with fixed kinetic energy @xmath83 , and uniform spatial density . let @xmath84 be the time that a particle arrives at the point @xmath85 . it is reflected if @xmath86 ; otherwise it jumps onto the elevator and moves across it with constant kinetic energy @xmath87 it reaches @xmath88 at time @xmath89 when its total energy is @xmath90 ( here the index @xmath91 does not mean `` initial , '' but rather `` when the particle arrives at @xmath88 . '' ) if @xmath92 , the particle is reflected by the right - hand barrier . otherwise it is transmitted , with kinetic energy @xmath93 at @xmath94 , its potential energy is converted to kinetic energy , and it escapes to the right with kinetic energy @xmath95 . summarizing , for @xmath96 and initial kinetic energy @xmath83 , we get transmission with final kinetic energy @xmath97 provided that i. ) @xmath98 , and ii . ) @xmath99 , where @xmath100 is given by eq . . each particle trajectory beginning at @xmath101 and ending near momentum @xmath102 contributes a term to the classical probability density @xmath103 , given by @xmath104 where @xmath105 is the jacobian for the @xmath106 trajectory ending near @xmath8 . summing over all trajectories gives a smooth result which diverges at extrema of @xmath11 . the `` primitive '' semiclassical wave function in momentum space is obtained via the same method as in @xcite , and is similar to the methods in @xcite . for each @xmath8 at the final time @xmath107 , we sum semiclassical terms @xmath108 where @xmath109 is the maslov index for the @xmath106 branch of the function @xmath110 , and @xmath111dt - \int e(t)dt\ ] ] is integrated over the classical path from initial to final time . the primitive semiclassical approximation in eq . applies only in classically - allowed regions , and it diverges at the boundaries of these regions . however , the divergences can be repaired and the function can be extended into classically - forbidden regions via the method in @xcite . , for the diode described by eq . with @xmath112 , @xmath113 , @xmath114 , and @xmath115 . the incident energy @xmath4 corresponding to each @xmath61 is shown on the right - hand axis . the fractional transport abruptly switches direction at @xmath116 . ] for rectangular barriers , particle momentum changes only at the barrier edges ( i.e. , @xmath88 and @xmath117 ) , so for particles which transmit past both barriers , @xmath118 -k_i t_{-b}\nonumber\\ & -\int^{t_0}_{t_{-b}}\left[k_l+q(t)\right]dt -e_0\left(t_f - t_0\right)\end{aligned}\ ] ] where @xmath119 we now examine a diode described by @xmath120 & , & -b&<x&<&0\\ u_r(x , t)&= \hat{u } & , & 0&<x&<&b,\label{1strec } \end{array}\end{aligned}\ ] ] with @xmath112 , @xmath113 , @xmath114 , @xmath115 . the left barrier oscillates between a minimum value of @xmath121 and a maximum value of @xmath122 , while the right barrier is static with a height of @xmath115 . [ rect_1stcurr ] shows fractional transport @xmath64 for this diode . in this example , when incoming particles have @xmath123 , @xmath65 below the energy at which particles incident from the left begin to gain enough energy from the oscillating barrier to transmit past the static barrier . when particles incident from the left begin to transmit , particles incident from the right are all reflected , @xmath124 , and there is left - to - right fractional transport . as the incident particle energy increases , fractional transport monotonically increases until @xmath125 , which is the threshold energy for particles approaching from the right to transmit past both barriers . at this point , fractional transport abruptly reverses direction to right - to - left ( @xmath126 ) . as @xmath4 increases , @xmath127 as particles incident from both sides transmit past both barriers . we now analyze the behavior of particles with an initial energy of @xmath128 ( initial momentum @xmath129 ) . particles incident from the right do not have enough energy to transmit past the right barrier , and are reflected with final energy @xmath130 ( final momentum @xmath131 ) . the initial wave packet approaching from the left has an envelope shape given by eq . centered at @xmath132 with @xmath133 . particles incident from the left all have enough energy to hop onto the left barrier ; approximately @xmath134 gain enough energy while traversing the left barrier to transmit past the right barrier , while the others are reflected from the right barrier . [ rectrect ] shows classical and semiclassical results for particles approaching from the left . fig [ rectrect](a ) shows initial position as a function of final momentum , @xmath135 . only a small portion of initial positions are shown . since the wave packet is wide in position space ( @xmath136 ) , there is a periodic relationship between final momentum and initial position . because the potential is not smooth , @xmath135 is discontinuous between transmitted and reflected portions . each branch of the function @xmath135 contributes a term @xmath137 to the primitive semiclassical wave function , given by eq . . the complete primitive wave function @xmath138 is obtained by summing eq . over all branches of @xmath135 . [ rectrect](b ) shows @xmath139 , the final primitive semiclassical probability density . fig [ rectrect](a ) shows that many trajectories end with any given @xmath8 inside the classically - allowed regions . the sharp peaks in @xmath140 arise from interference among all trajectories ending with any given @xmath8 . this calculation has not been extended into the classically - forbidden regions , so all peaks lie within the classically - allowed regions for both the transmitted and reflected portions . , @xmath113 , @xmath114 , and @xmath115 from the left with @xmath141 . ( b ) @xmath140 , the absolute square of the primitive semiclassical wave function , for the particles in ( a ) . ( c ) and ( d ) expansions of ( b ) . the dashed curve is the classical probability density , @xmath103 , for reflection or transmission with final momentum near @xmath8 . the smoothly - varying solid curve ( red online ) is a single - cycle primitive semiclassical probability , @xmath142 @xcite . the sharp peaks are the full primitive semiclassical probability summed over all cycles . these occur at momenta corresponding to floquet energies . ] since @xmath140 includes interference from a great number of trajectories , it is useful to differentiate between two distinct types of interference : i ) interference from within a single cycle of @xmath135 ( intracycle interference ) , and ii ) interference among all cycles ( intercycle interference ) . to view intracycle interference , we choose an arbitrary @xmath135 and sum the corresponding @xmath137 terms from within one cycle of the chosen @xmath135 to obtain @xmath143 . figures [ rectrect](c ) and ( d ) show the classical probability density @xmath103 ( dashed curve ) . note that the scales are different in figs . [ rectrect](c ) and ( d ) . we see that whereas classical theory gives a slowly - varying probability density @xmath103 , the primitive semiclassical single - cycle probability density @xmath144 ( thick solid curve , red online ) is oscillatory . the oscillations arise from interference among trajectories in the cycle that end with the same final momentum . the discontinuities seen in @xmath142 and @xmath103 in fig . [ rectrect](c ) at @xmath145 are due to the behavior of the branches in @xmath135 in fig . [ rectrect](a ) : for @xmath146 within the classically - allowed final momentum region of the reflected segments , there are two interfering branches per cycle , but for @xmath147 , there is only one branch per cycle . summing @xmath143 over all cycles yields the full primitive wavefunction @xmath138 , the square of which is @xmath140 , the sharply - peaked function in figs . [ rectrect](b ) , ( c ) , and ( d ) . this function has peaks at energies @xmath148 , consistent with floquet theory . in figs . [ rectrect](c ) and ( d ) , @xmath103 and @xmath142 are scaled ( multiplied by the same constant ) . when plotted in this fashion , one can see that the relative heights of the peaks in @xmath140 closely align with @xmath142 , i.e. , the relative heights of the floquet peaks are governed by the single - cycle probability . this occurs for any arbitrary @xmath135 chosen as the beginning of a cycle ; while different choices yield different @xmath142 , they all intersect at the locations of the floquest peaks . another interesting phenomenon arising from a similar elevator system is the quantum suppression of classical transmission . it may happen that the classical transmission probability is large , but the range of transmitted momenta is small so small that no floquet peaks lie in the classically - allowed range . then quantum interference ( we might better say semiclassical interference ) among trajectories from different cycles prevents transmission that is classically allowed . in such a case , a narrow initial wave packet ( in @xmath7 ) may allow transmission both classically and quantum - mechanically , not because it is broad in momentum space , but because it interacts with the barrier for only one ( or a few ) cycles . these phenomena occur for a diode described by @xmath149 & , & -b&<x&<&0\\ u_r(x , t)&= \hat{u } & , & 0&<x&<&b,\label{rect2supp } \end{array}\end{aligned}\ ] ] with @xmath150 , @xmath113 , @xmath114 , and @xmath115 . in this example , the left barrier oscillates between a minimum of @xmath151 and a maximum of @xmath152 , and particles approach the barriers from both sides with initial energy @xmath153 ( initial momentum @xmath154 ) . particles approaching from the right do not have enough energy to hop onto the right barrier , and are reflected with final energy @xmath130 ( final momentum @xmath131 ) . particles approaching from the left all have enough energy to hop onto the left barrier , and classically , more than one third ( approximately @xmath155 ) of these particles gain enough energy to transmit past the right barrier ( see fig . [ rectanti ] ) . these transmitted particles all end with @xmath8 inside a very small range . , @xmath113 , @xmath114 , and @xmath115 . all curves are as described in fig . [ rectrect ] . ] for the semiclassical calculation , we took an envelope given by eq . with @xmath132 and @xmath133 . [ rectanti](b ) shows the primitive semiclassical final momentum probability @xmath140 in the classically - allowed regions . in contrast to the classical result , we see no visible transmission . figures [ rectanti](c ) and ( d ) show the classical transmission probability and the single - cycle and final primitive semiclassical probabilities ( similar to fig . [ rectrect ] ) . the single - cycle primitive semiclassical calculation gives an even larger total transmission than the classical result , but the final semiclassical result is essentially zero . the explanation is that floquet peaks occur at energies @xmath156 , and the corresponding momenta for @xmath157 are @xmath158 , @xmath159 , and @xmath160 . none of these momenta lie inside the classically - allowed region of transmission . therefore , when summing interference from all cycles , this interference is destructive across the entire range of transmitted momentum , and at this level of approximation , there is no transmission . ( a uniform semiclassical approximation would extend into classically - forbidden regions , but the decay of the wave function in these regions combined with the floquet `` comb '' would yield small peaks , comparable to those seen outside the classically - allowed regions in fig . [ gauss2osc ] and in ref . @xcite ) . in this example , quantum interference suppresses the classical probability density for transmitted particles . a more realistic type of diode is one which has gaussian barriers described by eqs . and with @xmath161 . we examine one such case with barriers described by @xmath162 , @xmath163 , @xmath164 , @xmath165 ( full width at half maximum of @xmath166 ) , and @xmath167 . the right barrier has static height @xmath168 and the left barrier oscillates between zero and twice the height of the static barrier . [ gaussrects ] shows classical , semiclassical , and quantum calculations for particles incident on this diode from both directions with @xmath169 ( @xmath170 ) . particles incident from the right with this inital energy are all classically reflected , but approximately @xmath171 of particles incident from the left transmit , and there is left - to - right fractional transport of particles . [ gaussrects](a ) shows classical @xmath135 for particles incident from the left . classical trajectories are chaotic , as some particles are reflected from the left oscillating barrier , others directly transmit past both barriers , and others are temporarily trapped between the barriers before finally reflecting or transmitting . fig . [ gaussrects](a ) shows three periods of the function @xmath135 , and fig . [ gaussrects](b ) shows a zoom consisting of 10% of a period ( 30x magnification of ( a ) ) . extreme dependence on initial position is apparent , and there are a large number of trajectories ending with any classically - allowed @xmath8 . for the packet approaching a gaussian diode from the left . ( b ) zoom of ( a ) , showing the complexity of chaotic trajectories . ( c ) quantum ( downward , red ) and primitive semiclassical ( upward , blue ) final momentum probabilities . the semiclassical calculation only includes the contributions of slowly - varying branches of @xmath135 . ( d ) and ( e ) @xmath103 ( dashed curve ) , @xmath142 ( oscillatory curve , red ) , and @xmath140 ( sharply - peaked curve , blue ) . ( f ) quantum - mechanical final momentum probability for incoherent packets approaching the barriers from both sides . ] [ gaussrects](c ) shows quantum - mechanical ( plotted downward , red online ) and primitive semiclassical ( plotted upward , blue online ) final momentum probabilities for the packet incident from the left . the initial packet is described by eq . with @xmath133 and @xmath172 . the primitive semiclassical probability @xmath140 only includes contributions from slowly - varying branches of the function @xmath135 ( i.e. , regions of chaotic scattering are omitted ) . this rough approximation agrees reasonably well with the quantum probability @xmath173 except for peaks located at @xmath174 and @xmath175 . [ gaussrects](d ) and ( e ) show zooms of the primitive semiclassical approximation , along with the single - cycle momentum probability @xmath142 ( thick oscillatory curves , red online ) and classical momentum probability @xmath176 ( dashed curve ) . as before , the single - cycle probability governs the relative heights of the floquet peaks seen in the full primitive semiclassical probability . [ gaussrects](f ) shows the quantum - mechanical final momentum probability for particles approaching the barriers from both sides with @xmath169 . the largest probability is at @xmath177 , which is primarily caused by the reflection of particles incident from the right . particles incident from the left that are transmitted contribute only a small amount to this momentum state ( see fig . [ gaussrects](c ) ) ; these transmitted particles have a much higher probability of ending with @xmath178 with @xmath179 . the total probability in the quantum calculation for @xmath180 is approximately @xmath181 . the classical fractional transport @xmath182 corresponds to approximately @xmath183 of particles ending with @xmath180 , showing good agreement between classical and quantum theories . in the remainder of this paper , we consider pumps that are `` symmetric '' in the sense that @xmath184 , @xmath185 , and @xmath186 , so the barriers are identical , but not in phase with each other . intuitively one might have guessed the following behavior . suppose that we consider the case of the classic turnstile pump for which @xmath187 , so the barrier on the right oscillates a quarter - cycle behind the one on the left . then the two barriers together imitate a rightward - moving wave , @xmath188 . we might then expect that the system would preferentially pump particles from left to right . nothing of the sort happens , however . classically , if particles begin with a distribution that is uniform in _ both _ momentum and position ( _ i.e. _ the distribution includes all initial energies @xmath4 and is independent of @xmath4 ) then for every particle going from left to right , another goes from right to left there is no net pumping at all . this symmetry theorem can be violated if the initial distribution of particles is not uniform in phase space . for example , if the phase space distribution is constant only up to some maximum initial energy , then some net pumping is possible . more important , if particles begin from both sides with the same fixed energy , then there can be a net flow in one direction or the other . the amount and direction of this flow depends on that energy , so the apparent natural direction of the pump is an illusion . the critical step in proving this no - pumping result is choosing a reference phase of the oscillations , and then using a surface of section at integer number of cycles from this reference phase . we choose the reference phase to be when the two oscillating barriers have equal height , and the left barrier is going up and the right barrier is going down ( see fig . [ fig : nopumpingthm](a)(inset ) ) . if we take such a point in any cycle to be @xmath77 , then for any @xmath22 and all @xmath68 and @xmath189 we have @xmath190 let @xmath191 be the map that evolves a point @xmath192 forward one pumping period to @xmath193 , and let @xmath194 be the operator that reflects the position through the origin : @xmath195 . a trajectory on the left of the pump moving right sees the closest barrier going up , whereas a trajectory on the right moving left sees its closest barrier going down . a mirror image thus converts the upward moving barrier to downward and vice - versa , i.e. it reverses the time - dependence of the barriers , so the particle follows the time - reversed trajectory . consequently , @xmath196 will be a mirror image of @xmath197 . more formally , @xmath198 this relation is demonstrated in fig . [ fig : nopumpingthm](a ) . now define @xmath199 to be the set of points moving to the right that are reflected after one pumping period , i.e. @xmath200 we define @xmath201 similarly @xmath202 _ assuming a uniform initial distribution in phase space _ , the single - period net flux @xmath203 of rightward to leftward moving trajectories is thus @xmath204 we now show @xmath205 . ( see fig . [ fig : nopumpingthm]b . ) let @xmath206 be arbitrary . we then have @xmath207 and @xmath208 where @xmath209 . the point @xmath210 is then an arbitrary point of @xmath211 ; note @xmath212 . now , @xmath213 where the second equality follows from eq . ( [ r1 ] ) . since @xmath214 and @xmath207 , we find @xmath215 . hence , @xmath216 . the reverse inclusion follows similarly . since @xmath205 and @xmath217 and @xmath218 both preserve phase - space area , @xmath219 = @xmath220 . hence @xmath221 , i.e. there is no net flux pumped across the barrier . all of our numerical simulations of symmetric pumps have confirmed this theorem . we now consider symmetric turnstile pumps in which both barriers oscillate smoothly in time , with rectangular potential barriers described by eqs . and . we first examine the simplest pump of this type , which has no space between the barriers . in this case , any incident particle can either be reflected by the first barrier , hop onto the first barrier and be reflected from the second barrier , or transmit over both barriers . if the particle has enough energy to transmit over one or both barriers , it can gain or lose energy during the time it spends on top of the barrier(s ) . as in section [ recelev ] , since the barriers are rectangular , particles only experience acceleration at the boundaries of barriers , and have constant momentum everywhere else . a particle beginning to the left of the barriers is launched with momentum @xmath222 and arrives at the leftmost edge of @xmath223 at time @xmath84 , at which time the height of the left barrier is @xmath224 , and the total energy of the particle is @xmath225 . if @xmath226 , the particle is reflected from the first barrier with final momentum @xmath227 . otherwise , the particle is transmitted over the first barrier with momentum @xmath228 the time at which the particle reaches the opposite edge of the first barrier ( and therefore the first edge of the second barrier ) is @xmath229 the corresponding @xmath230 and @xmath231 are given by eqs . and , respectively , and @xmath232 . if @xmath233 , the particle is reflected from the second barrier with @xmath234 and spends another time interval @xmath235 going back over the first barrier , after which it falls off onto the left - hand side of the pump with final momentum @xmath236}.\end{aligned}\ ] ] if @xmath237 , the particle is transmitted over the second barrier with momentum @xmath238 the time at which the particle falls off the second barrier is @xmath239 at which time the height of the right - hand barrier is @xmath240 , and the final momentum is @xmath241 a similar algorithm is followed for particles beginning on the right of the pump with negative initial momentum . there is never more than one reflection of a particle . we calculate all particle trajectories using eqs . - to obtain each particle s final momentum @xmath8 . we examine the net particle fractional transport for mirrored sets of particle packets approaching the barriers from opposite directions with @xmath242 . classical computations shown in the remaining sections are done as follows . for a selected set of barrier parameters , we first choose a range of initial particle energies @xmath243 . each @xmath4 in this range has two corresponding momenta , @xmath242 . for each @xmath61 , we construct two incoming packets of particles : one starts to the left of the barriers with @xmath244 , and the other starts to the right of the barriers , with @xmath245 . the width in @xmath7 of each packet is @xmath246 ( recall @xmath33 ) . we start all trajectories at @xmath247 . the edge of each packet which is closest to the barriers is placed a distance @xmath248 away from the outer edge of the first barrier , which ensures that the first particle of each packet reaches the outer edge of the first barrier at @xmath249 . for these initial conditions , we define the time of arrival at the barrier as @xmath250 ; with this definition , the particle in each packet which starts closest to the barriers arrives at the outer edge of the first barrier at @xmath251 . our choice of packet width ensures that the last particle in each packet to arrive at the outer edge of the first barrier arrives at @xmath252 , which represents one full cycle of the barriers . referring to eq . , a particle that arrives at the left - hand edge of the left barrier at @xmath251 or @xmath252 encounters the barrier at its maximum , and one that arrives at @xmath253 encounters that barrier at its minimum . the numerical results for barriers with @xmath115 , @xmath254 , @xmath255 , @xmath256 and @xmath257 are shown in fig . [ rbnospace ] . figure [ rbnospace](a ) represents particles approaching the barriers from the left , and [ rbnospace](b ) represents particles approaching from the right . in both plots , individual particles are represented by their initial momenta @xmath61 and the time @xmath258 at which they arrive at the outer edge of the first barrier . the colors ( online ) in both plots correspond to the final momentum @xmath259 of each particle . blue ( online ) represents particles which scatter to the left ( @xmath260 ) of the barriers , and red ( online ) corresponds to particles which scatter to the right of the barriers @xmath261 . the intensity of the color corresponds to the magnitude of the particle s final momentum , as seen in the colorbar . for particles incident on rectangular barriers described by @xmath115 , @xmath254 , @xmath255 , @xmath256 and @xmath257 . ( a ) represents particles approaching from the left , and ( b ) represents particles approaching from the right . ( c ) sum of @xmath262 for particles approaching the barriers from both directions with @xmath263 . red indicates both particles scatter to the right , and blue indicates both particles scatter to the left . ( d ) fractional transport , @xmath64 . ] the lowermost blue region in fig . [ rbnospace](a ) represents particles approaching from the left that are initially reflected from the left barrier , and the lowermost yellow region in fig . [ rbnospace](b ) represents particles approaching from the right that are directly reflected by the right barrier . in both cases , if there were no other barrier , then the region above this lowermost boundary would be entirely of the opposite color , as all particles not initially reflected would be transmitted . it follows that all of the striping effects just above this boundary are due to the presence of the second barrier . just above this boundary , a particle has just enough energy to hop onto the first barrier it encounters . consequently , its momentum @xmath264 on the first barrier is small , it moves across the barrier slowly , and the barrier may oscillate many times while the particle is on it . in the limit that @xmath265 , an infinite number of oscillations occurs while the particle is on the barrier . hence , there is an infinite number of stripes converging from above upon the boundary . [ rbnospace](c ) sums @xmath262 for both particles which arrive at the barriers at the same time @xmath258 and @xmath61 , but which arrive from opposite directions . red ( online ) represents cases in which both particles scatter to the right of the barriers ( @xmath266 for both particles ) . blue ( online ) represents cases in which both particles scatter to the left of the barriers ( @xmath260 for both particles ) . if the particles scatter to opposite sides of the barrier , e.g. if both are reflected or transmitted , no color is plotted . the intensity of the color corresponds to the magnitude of the sum of @xmath262 for both particles . [ rbnospace](d ) shows fractional particle transport @xmath64 ( see eq . ) . this function is considered over the entire range of @xmath258 for each @xmath61 , i.e , @xmath64 accounts for all particles at a given @xmath61 . when @xmath64 is averaged over all @xmath61 , the symmetry theorem tells us that there is no net particle transport . however , there is transport ( in either direction ) within finite ranges of @xmath61 . fractional particle transport at a given @xmath61 can be understood by comparing figs . [ rbnospace](c ) and ( d ) . @xmath267 in the range @xmath268 , indicating net particle flow to the left of the barriers . examining fig . [ rbnospace](c ) , we see that only one colored lobe extends into this @xmath61 range . its color ( blue online ) indicates @xmath269 values for which both particles have @xmath260 , meaning that both particles scatter to the left of the barriers . since no red ( online ) lobes extend into this @xmath61 range , there are no @xmath269 values for which both particles scatter to the right . therefore , for any @xmath269 in this range , both particles can either scatter to the left of the barriers , or scatter to opposite sides , causing net particle transport to the left . , but with barriers four times as wide as those in fig . [ rbnospace ] ( @xmath270 and @xmath271 have been increased from @xmath272 to @xmath273 ) . increasing barrier width causes thinner ribbons of transmission and reflection for particles incident on the barriers from both sides . ] figure [ wider ] illustrates the effects of increasing the barrier widths . in this calculation , all parameters are the same as those in fig . [ rbnospace ] ( @xmath115 , @xmath257 , @xmath254 , and @xmath255 ) , except for @xmath270 and @xmath271 , which have been increased from @xmath272 to @xmath273 . one can see that the ribbons of transmission and reflection span a more narrow @xmath274 range at a given @xmath258 , and that the widths ( @xmath274 ) of the ribbons at a given @xmath258 do not decay as rapidly as in the previous case as @xmath275 . comparing figs . [ wider](c ) and ( d ) to figs . [ rbnospace](c ) and ( d ) , we see that in this case the net particle transport fluctuates more rapidly with @xmath61 . however , the magnitude of @xmath64 within these smaller @xmath274 regions can be just as large ( or larger ) as in the case of narrow barriers . the change in transmission and reflection ribbons for wide barriers can be understood by examining the condition for particles to transmit . let us examine particles which approach from the left with @xmath222 and arrive at the left barrier @xmath253 , when the height of that barrier is zero . all particles arriving at @xmath253 hop onto the left barrier and traverse it with momentum @xmath276 . the condition for them to transmit over the right barrier is @xmath277 when @xmath278 , particles transmit over the right barrier , and when @xmath279 , particles reflect from the right barrier . the zeroes of @xmath280 thus mark the boundaries between transmission and reflection ribbons . we illustrate this for the barrier parameters from the preceding two examples ( @xmath281 , @xmath255 ) . for these barrier parameters , eq . reduces to @xmath282 eq . shows that @xmath280 oscillates about @xmath283 , and the ratio @xmath284 governs its oscillation frequency . as @xmath275 , @xmath280 passes through zero an infinite number of times , resulting in an infinite number of transmission and reflection ribbons for any @xmath270 . higher @xmath270 values ( wider barriers ) cause @xmath280 to oscillate more rapidly as @xmath285 . the maximum amplitude of oscillation is @xmath286 ; thus , for @xmath287 , i.e. @xmath288 , all particles will transmit , no matter the width of the barriers . . the right column shows zooms of @xmath262 for the curves to the left . the top row has @xmath256 , the middle row has @xmath289 , and the bottom row has @xmath290 with all other barrier parameters equal . ] eq . is plotted for the selected barrier parameters in the left column of fig . [ zeroes ] while varying @xmath270 , the barrier width . the oscillatory curve is @xmath280 , and the two quadratic curves are @xmath291 , which bound @xmath280 . the threshold @xmath292 is the intersection between the left - most quadratic curve and the vertical line . the right column contains a zoom of @xmath262 about @xmath253 for particles incident from the left for the respective @xmath270 . the top row represents @xmath256 ( see fig . [ rbnospace ] ) , the middle row is @xmath289 ( see fig . [ wider ] ) , and the bottom row has @xmath290 . the infinite number of bands , and the reduction in their widths as the barriers get wider , are evident in these pictures . inserting space between the barriers leaves many of the features of the preceding section intact , but introduces a critical difference in particle trajectories . previously , a particle could reflect from a barrier no more than once . however , particles may now become trapped between the barriers for a long time , reflecting back and forth between them before finally arriving at the edge of a barrier with enough energy to transmit over it . these particle trajectories are thus very sensitive to initial conditions and the system is a model of chaotic scattering . numerical calculation of final momentum is performed in similar fashion as before . if a particle beginning on the left of the pump with positive momentum has enough energy to hop onto the left - hand barrier , we calculate its momentum @xmath264 and the time @xmath100 at which it reaches the end of this barrier using eqs . and , respectively . however , instead of either reflecting from the right barrier or transmitting over it , the particle instead falls off the first barrier into the region between barriers with momentum @xmath293 with @xmath294 . the particle reaches the second barrier at time , but with space between the barriers . the barriers are now centered at @xmath295 , and have a distance of @xmath296 between their inner edges . ] @xmath297 where @xmath298 is the distance between the inner edges of the barriers . if the height of the second barrier is greater than the particle s energy , i.e. , @xmath299 , the particle reflects from the second barrier , and we increment @xmath300 by @xmath301 . the index @xmath300 thus counts the number of trips between the barriers for each trajectory . each time a particle arrives at the edge of a barrier , we compare its kinetic energy @xmath302/2 with the height of that barrier ( for odd @xmath300 we compare to @xmath303 , and for even @xmath300 we compare to @xmath304 ) until it has enough energy to hop onto a barrier . once on top of a barrier , the particle s momentum is given by @xmath305 where @xmath306 is positive for odd @xmath300 , and negative for even @xmath300 . the particle then falls off the second barrier at time @xmath307 with final momentum @xmath308 where @xmath180 if @xmath309 , and @xmath310 if @xmath311 . the calculation is similar for particles approaching the barriers from the right . . all curves are as described in fig . [ rectrect ] ] figure [ rbwspace ] shows results for a pump with potentials given by eqs . and with @xmath312 , @xmath313 , @xmath115 , @xmath257 , @xmath254 , and @xmath255 . this pump is the same as the one from fig . [ wider ] , except the inner edges of the barriers are now separated by a distance @xmath296 . similar to the effect of making the barriers wider , inserting space between the barriers affects the ribbons of transmission and reflection for particles approaching the barriers from both sides . the width ( @xmath274 ) of the ribbons decays more quickly as @xmath314 for a given @xmath258 . consequently , the width @xmath274 for regions of large net particle transport is smaller . the magnitude of @xmath64 has decreased in this example ( although increasing the space between the barriers can also cause it to increase ) . predicting the effect of increasing barrier separation on the magnitude of fractional particle transport is not possible without detailed calculations . [ scspace ] shows @xmath135 for two particle packets which approach the barriers from opposite sides with the same initial energy ( @xmath3153.8 ) . the initial packets are described by eq . with @xmath316 and @xmath317 . their initial energy is large enough such that all particles transmit over both barriers . particles approaching from the left are scattered to a larger range of @xmath8 than those approaching from the right . this results in more peaks for @xmath180 in the semiclassical probability density @xmath140 shown in fig . [ scspace](b ) . [ scspace](c ) and ( d ) show expansions of fig . [ scspace](b ) ( note the different scales ) . this calculation has not been extended into the classically - forbidden regions . in fig . [ scspace](c ) , the classical probability density @xmath103 ( dashed curve ) and single - cycle semiclassical probability density ( oscillatory curve ) @xmath142 diverge at each of the four turning points of @xmath135 . interference among the four branches of @xmath135 within one cycle causes @xmath142 to be larger than @xmath318 at the location of the largest peak , @xmath319 . @xmath142 and @xmath318 are scaled ( multiplied by the same constant ) in order to be plotted with @xmath140 ( sharply - peaked curve ) , and the relative heights in the floquet peaks can be seen to align closely with the discrete values of @xmath142 at momenta corresponding the floquet energies . analysis of transmission and reflection ribbons is more difficult in the present case because of the possibility of multiple reflections between the barriers . however , we can gain insight by analyzing criteria for particles which directly transmit past both barriers with no reflection . for particles approaching from the left with @xmath222 and arriving at the left - hand barrier at @xmath253 ( i.e. when the height of the left barrier is zero ) , the condition for direct transmission past both barriers is @xmath320 with @xmath294 , where the subscript @xmath321 is the distance between the inner edges of the barriers ( see eq . ) . eq . reduces to @xmath322 for our selected barrier parameters . ( a ) about @xmath253 with color schemed changed . black represents reflection ; yellow represents transmission . right column : zooms of fig . [ rbwspace](a ) about @xmath253 . the zeroes of the thick ( blue ) oscillatory curve in the middle column mark the boundaries of transmission and reflection seen in the left column . when the thin ( red ) oscillatory curve is positive , particles in the right column transmit , but this function does not reveal all transmission ribbons in the right column . ] the left and right columns of fig . [ f2zero ] show zooms of figs . [ wider](a ) ( @xmath289 ) and [ rbwspace](a ) ( @xmath312 and @xmath313 ) , respectively , about @xmath253 . the color scheme has been changed to enhance visibility ; black ribbons represent reflection and the lighter ( yellow online ) ribbons represent transmission . the middle column of fig . [ f2zero ] shows @xmath323 for these two pumps . the thick ( blue online ) curve is @xmath324 ( @xmath325 ) , and is the same curve seen in fig . [ zeroes ] . its zeroes mark the boundaries of transmission and reflection in the left column . the thin curve ( red online ) is @xmath326 ( @xmath296 ) , and corresponds to the right column . when @xmath327 , particles arriving at the pump in the right column transmit , and the ribbon in the right column is the light color ( yellow online ) . when @xmath328 , the particle reflects from the right - hand barrier , and its ultimate fate is unspecified . it is evident that @xmath326 oscillates more rapidly than @xmath324 . consequently , regions which reflect when the barriers touch are split into multiple transmission and reflection ribbons as the barriers are moved apart . this is illustrated in the present case in regions where @xmath329 and @xmath327 . for each region in which @xmath330 is negative , there is a reflection ribbon in the left column . however , in each such region , @xmath326 oscillates through zero many times , and each positive segment of @xmath326 represents a transmission ribbon in the right column . increasing the barrier separation distance thus creates many transmission and reflection ribbons in regions where there is only pure reflection when the barriers touch . the minimum @xmath3 above which all particles arriving at @xmath253 transmit has also been greatly increased by moving the barriers apart . with no barrier separation , this @xmath331 , but increasing @xmath271 to @xmath332 increases this minimum to @xmath333 . in each case , there are an infinite number of ribbons as @xmath285 . this level of analysis predicts only the outcome of each particle s first arrival at the right - hand barrier . what happens after that is `` left as an exercise for the reader . '' the effects of increasing barrier width and separation can be summarized as follows . increasing the width causes more transmission and reflection ribbons below arbitrary @xmath61 , up to a maximum @xmath61 above which all particles will transmit for a given @xmath258 . the width of the ribbons ( in terms of @xmath274 ) decays more slowly as @xmath334 for wider barriers . ribbons produced by particles incident upon barriers with no separation are split into multiple ribbons by moving the barriers apart . increasing barrier separation can also allow particles of much higher @xmath61 to reflect for a given @xmath258 . increasing either of these parameters causes the widths ( @xmath274 ) of regions in which there is significant fractional particle transport to decrease , although its magnitude is not systematically changed . predictions on fractional transport are highly sensitive to the choice of initial conditions and parameters , and do not display any obvious pattern . therefore , general predictions beyond what we have mentioned can not be made without detailed calculations specific to a configuration and choice of parameters . while rectangular barriers provide a simplified model that addresses the essential pumping physics , gaussian barriers are more likely to be used in experimental implementations using laser - based optical dipole barriers for ultracold atoms . in this section , we examine a turnstile pump ( such as those in section [ symm ] ) with gaussian potentials described by eqs . and with @xmath335 , @xmath257 , @xmath254 , @xmath255 , @xmath336 , and @xmath167 . both barriers oscillate at the same frequency , but not in phase with one another . as in the previous section , particle trajectories for this type of pump are classically chaotic . [ redo](a ) and ( b ) show @xmath262 for particles incident upon the barriers from the left and right , respectively . unlike the previous cases with rectangular barriers , there is a minimum @xmath61 below which there is no particle transmission . as particles approach gaussian - shaped repulsive barriers , they lose momentum , resulting in a minimum initial energy required to transmit past the first barrier encountered . in this case , all particles with @xmath337 reflect from the first barrier . different types of structure can be seen in @xmath262 than for rectangular barriers , but qualitative features remain . the regions in which striping can be seen indicate particle trajectories which are temporarily trapped between the barriers before finally transmitting or reflecting . the lobe with significant striping seen in fig . [ redo](a ) is much wider than the narrow one seen in fig . [ redo](b ) in the range @xmath338 , indicating that particles approaching from the left in this energy range are much more likely to become temporarily trapped between the barriers than those approaching from the right with equal energy . however , above this range , all particles approaching from the left transmit , while some approaching from the right are trapped between the barriers until @xmath339 . the complete description of particle transport through the barrier region lies outside the scope of this paper , but a detailed topological analysis is given in @xcite . [ redo](c ) shows @xmath269 for which both particles scatter to the left ( blue online ) or right ( red online ) . whereas the previous cases with rectangular barriers have structure as @xmath334 , no structure is present in this region for gaussian barriers because of the nonzero minimum @xmath61 required to transmit past the first barrier . [ redo](d ) shows @xmath64 . the vertical region in this curve below @xmath337 corresponds to the region in which all particles directly reflect . above this range , fractional particle transport occurs in both directions until @xmath61 is large enough for all particles to transmit past both barriers . , but for gaussian barriers . ] for particles approaching a pump with two oscillating gaussian barriers from both sides . ( b ) final uniform semiclassical momentum probability @xmath140 ( plotted upward , blue ) , and quantum mechanical final momentum probability @xmath173 ( plotted downward , red ) , for the particles in ( a ) . ( c ) and ( d ) @xmath103 ( dashed curve ) , @xmath142 ( oscillatory curve , red online ) , @xmath340 ( green online ) and @xmath140 ( blue online ) for the particles in ( a ) . see the text for a discussion of these functions . ] figure [ gauss2osc ] shows classical , semiclassical , and quantum - mechanical comparisons for two packets of particles approaching the barriers from opposite directions with @xmath341 . the initial packets are decribed by eq . with @xmath316 and @xmath317 . figure [ gauss2osc](a ) shows classical initial position as a function of final momentum for particles approaching the barriers from both sides . all particles have enough energy to transmit over both barriers . particles incident upon the barriers from the right scatter to a larger range of @xmath342 than those approaching from the left . figure [ gauss2osc](b ) shows @xmath140 , the uniform semiclassical final momentum probability density ( plotted upward ) , and @xmath173 , the quantum mechanical final momentum probability ( plotted downward ) . the uniform semiclassical calculation has been repaired near turning points of @xmath343 , where the primitive form is divergent , and has been extended into classically - forbidden regions . @xmath173 has been mirrored about the amplitude axis for ease in comparing the two calculations . the horizontal lines ( red online ) in the upper half - plane are the heights of the peaks in @xmath173 , and are plotted to allow one to compare the calculations more easily . very good agreement between the two methods is evident . figures [ gauss2osc](c ) and ( d ) show the classical probability density @xmath344 ( dashed curve ) , and the primitive semiclassical single - cycle probability density @xmath142 ( thick oscillatory curve , red online ) . the two sharply - peaked functions are the primitive semiclassical probability density @xmath345 , given by summing eq . for all branches ( lighter peaked curve , green online ) , and the uniform semiclassical probability density @xmath140 ( darker peaked curve , blue online ) . the functions @xmath103 and @xmath142 are scaled ( multiplied by the same constant ) . the function @xmath345 ( green online ) takes on discrete values of the curve @xmath142 ( red online ) at momenta satisfying @xmath346 , showing that the single - cycle probability governs the relative heights of the floquet peaks ( the single - cycle probability shown is the primitive form ) . the classical dynamics underlying the quantum treatment are therefore necessary to fully understand the quantum mechanical result . while quantum theory tells us that the density will be peaked at momenta satisfying @xmath346 , it does not tell us the range of @xmath6 for which the peaks will be of appreciable height . the final momentum region in which particles are classically scattered governs the range of @xmath6 for which the quantum result yields large peaks . quantum theory also does not indicate why some momentum states are more highly populated than others , but semiclassical tools give an intuitive explanation for that . the double barrier turnstile pump might be viewed ( with some caution ) as a momentum - space interferometer . in this picture , each oscillating barrier acts as a multichannel beamsplitter which takes an incoming planewave and transforms it into a superposition of outgoing planewaves with different momenta ( with energies @xmath346 ) . in a pure transmission case ( such as fig . [ gauss2osc ] ) , the first barrier produces multiple planewave states , and then the second barrier mixes these and produces additional planewave states . in this way a turnstile pump may be viewed as a discrete multipath momentum space interferometer . however , this description can not be accurate if the barriers are not well - separated . the barriers must be sufficiently far apart that the configuration - space wave function in the region between them is approximately a superposition of plane waves , but not so far apart that packets associated with different floquet states have separated . in summary , we have defined and described ballistic atom pumps , showing that for finite ranges of initial particle energies , such systems can create net particle transport in either direction . the direction of particle pumping is highly sensitive to barrier parameters and to the initial energy of the particles . it is not possible to predict the direction or magnitude of particle pumping without detailed calculations . if tunneling can be neglected , diode pumps which only allow net transport in one direction for particles below a certain initial energy can be constructed . at sufficiently high incident particle energies , these diodes only allow net particle transport in the opposite direction . we have studied these pumps classically , semiclassically , and quantum mechanically . while classical theory gives a slowly - varying final momentum probability for scattered particles , quantum theory yields final momentum probabilities sharply - peaked at momenta satisfying @xmath346 . the range of @xmath6 for which there are appreciable peaks is governed by the underlying classically - allowed momentum range of scattered particles . semiclassical theory gives an intuitive explanation for the relative heights of the peaks , and agrees well with the quantum description . t. a. b. and m. k. i. contributed equally to this paper . k. k. d. acknowledges support of the nsf under grant no . k. a. m. acknowledges nsf support from grant no . j. b. d. and t. a. b. acknowledge nsf support via grant no . phy-1068344 .

non - adiabatic contribution of environmental degrees of freedom yields effective inertia of spin in effective spin dynamics . in this paper , we study several aspects of the inertia of spin in metallic ferromagnets . ( i ) a concrete expression of the spin inertia : , where is the spin polarization of conduction electrons and is the coupling constant . ( ii ) dynamical behavior of spin with inertia , discussed from viewpoints of a spinning top and of a particle on a sphere . ( iii ) behavior of spin waves and domain walls in the presence of inertia , and behavior of spin with inertia in the case of a time - dependent magnetic field .

1502.04107

different from other fundamental quantities such as mass and charge , spin is a dynamical quantity , and its dynamics have been widely studied and applied in science and technology . in particular , recent rapid growth of spintronics provides a stage where deeper understandings of spin dynamics directly lead to practical applications . the dynamics of spin are governed by the spin berry phase @xcite , and its equation of motion includes only the first - order time derivative of spin . this is natural because spin is an angular momentum and its equation of motion takes the familiar form : the time - derivative of the angular momentum ( i.e. spin ) is given by the torque acting on it . without any torque , the solution of the equation of motion of spin is only a static one . this is in contrast with , for example , the case of a massive point particle , which has inertia and can move at non - zero speed as its free motion . in this sense , spin does not have inertia . however , for systems where spin interacts with other environmental degrees of freedom , spin dynamics are affected by those environmental degrees of freedom , and the dynamical law of spin is changed to be an effective one . for example , in metallic ferromagnets , conduction electrons affect the dynamics of localized spins ( i.e. spins of atoms on lattice cites ) . a typical effect is spin - damping ( e.g. @xcite ) , where the energy and the angular momentum of spins are transferred to the environmental degrees of freedom and , as the result , spins relax to their ground state within a certain time scale . without this effect , spins undergo the larmor precession around the applied magnetic field forever . in this respect , the existence of the degrees of freedom other than spins changes the dynamical behavior of spins significantly . in the equation of motion of spins , this damping is represented by the gilbert damping term @xcite . this damping term includes only the first - order time derivative of spins . therefore , spin still does not have inertia even when we take into account the gilbert damping effect . the effects of the environments other than the gilbert damping can be studied systematically by the derivative expansion , where the effects are expanded in powers of the time ( and spatial ) derivative of spin . from that point of view , the gilbert damping term gives the leading order term in that expansion . in the higher orders , there appear terms which include the second - order time derivative , the third - order time derivative and so on , in the equation of motion of spin . these terms with higher order time derivatives are interesting in that , like the gilbert damping term , they change the dynamical law of spin itself , more than give additional torque on spin . in particular , from a comparison with the form of the newton s equation of motion of a massive point particle , the term with the second - order time derivative of spin plays the role of the inertia of spin . such spin inertia has been discussed in the literature ( e.g. ref.@xcite ) . in particular , recent progress in ultrafast magnetization @xcite motivated several works . in refs.@xcite , the inertia of spin was introduced phenomenologically and was shown to give additional nutation to the motion of spin . the time scale where the effect of the inertia is significant was discussed , based on the work of brown @xcite , to be sub - picosecond order . in ref.@xcite , the equivalence between the dynamics of spin with inertia and a spinning top was discussed . microscopic derivation of spin inertia was performed in refs.@xcite and @xcite . in ref.@xcite , an extended breathing fermi surface model was used , and the relation between the gilbert damping coefficient and the spin inertia was given in terms of physical quantities ( such as fermi - dirac occupation numbers ) of conduction electrons . the time scale for the nutational motion to be damped by the gilbert damping was estimated to be sub - picosecond order . in ref.@xcite , a general expression of the contribution of the conduction electrons to spin dynamics was discussed . the spin effective dynamics was shown to be non - local in general , which can be approximated as local dynamics by the derivative expansion of spin . the inertial term of spin arises in that derivative expansion and its general expression was given in term of the green s function of the conduction electrons . since the inertia of spin is conceptually interesting in its own and gives the first step toward the understanding of non - adiabatic contribution of environmental degrees of freedom to spin effective dynamics , further investigations are worthwhile . although the expressions of the inertia of spin were given as integral forms @xcite , an explicit expression of the inertia of spin in terms of parameters of a model has not been obtained so far . furthermore , the effect of the inertia of spin on the dynamics of spin has been discussed only for spatially homogeneous spin system under time - independent magnetic field . in this paper , we present detailed theoretical study of the effects induced by the spin inertia based on an @xmath4 model . in section [ sec : effective action ] , we derive a concrete expression of spin inertia in terms of the parameters in the @xmath4 model . in section [ sec : spin and spinning top ] , the basic behavior of spin with finite inertia is studied with the help of its two equivalents : a symmetric spinning top and a massive charged particle on a sphere subject to a monopole field . in section [ sec : phenomena ] , we study spatially inhomogeneous system , and discuss that spin waves and magnetic domain walls acquire an additional oscillation mode due to spin inertia . we also study the behavior of spin under large and time - dependent magnetic field , and find an unusual behavior of spin where the velocity of spin is parallel to the direction of the time - derivative of magnetic field . the inertia of spin arises naturally in its effective dynamics , which takes into account the effects of the environmental degrees of freedom @xcite . consider a system where a classical field of localized spin , @xmath5 ( with @xmath6 fixed ) , and a field @xmath7 representing the other environmental degrees of freedom are interacting with each other . for concreteness , we consider the case of metallic ferromagnets in this paper . conduction electrons are represented by annihilation and creation operators , @xmath8 and @xmath9 . the total action is given by @xmath10+\mathcal s_{\rm e}[c , \bar c , \bm n]$ ] where @xmath10 $ ] is the action of spin @xmath11 and @xmath12 $ ] is that of conduction electrons with their interaction with spin @xmath11 . when we are interested only in the dynamics of spin @xmath11 , it is convenient to integrate out @xmath8 and @xmath9 , and derive the effective action of spin @xmath11 . the contribution from electrons is given by path integration as @xmath13\right ) \equiv \int \mathcal d \bar c\mathcal d c ~ \exp\left(\frac{i}{\hbar}\mathcal s_{\rm e}[c , \bar c , \bm n]\right).\ ] ] the sum of this @xmath14 $ ] and the original spin action @xmath10 $ ] gives the total spin effective action . it is difficult to calculate @xmath14 $ ] exactly , so we should rely on a perturbative analysis . we here perform derivative expansion , where @xmath14 $ ] is expanded in powers of @xmath15 ( @xmath16 ) . when the system is isotropic , the general form is @xmath17 = \int \frac{d^3x}{a^3 } dt\left [ s_c\dot\phi(\cos\theta-1 ) - \frac{j_cs_c^2}{2}(\partial_i \bm n)^2 + \frac{m_s}{2}\bm { \dot n}^2 \right ] + \mathcal o((\partial_\mu \bm n)^3 ) , \label{delta seff in main text}\ ] ] where @xmath18 and @xmath19 . we have divided the lagrangian density entirely by lattice volume @xmath20 so that each coefficient represents a quantity per each lattice cite . the first term is the spin berry phase with @xmath2 the spin polarization of the conduction electrons and the second is the spin - spin exchange interaction induced by electrons with @xmath21 the coupling constant . in the final term , there arises the inertial term of spin with the inertia @xmath0 . this @xmath0 has the dimension of @xmath22 $ ] , the same as that of the moment of inertia . in ref.@xcite , a general expression of spin inertia @xmath0 is derived in term of the green s function of the conduction electrons . let us here calculate a concrete expression of spin inertia . as in ref.@xcite , a typical example of @xmath23 $ ] is the @xmath4 model @xcite , where conduction electrons interact with localized spins @xmath11 as @xmath24 = \int d^3x dt~ \bar c \left(i\hbar \partial_t + \frac{\hbar^2\partial_i^2}{2 m } + \epsilon_f + g_{\rm sd } \bm n\cdot \bm \sigma\right)c . \label{s - d model}\ ] ] here , @xmath25 is the mass of the conduction electrons , @xmath26 is the fermi energy , @xmath3 is the @xmath4 coupling constant , and @xmath27 is the pauli matrix vector . to obtain the derivative expansion of @xmath14 $ ] , we perform @xmath28 gauge transformation @xmath29 with an @xmath28 matrix @xmath30 acting on the spinor indices , so that the @xmath4 interaction becomes diagonal , @xmath31 . due to this unitary transformation , there appears a so - called spin gauge field @xmath32 in through @xmath33 . this @xmath34 contains the first - order derivative @xmath15 . therefore , expanding @xmath35/\hbar)$ ] in powers of @xmath34 , we can calculate @xmath14 $ ] perturbatively in powers of @xmath15 . the spin polarization @xmath2 and the inertia @xmath0 can be calculated as ( see appendix [ sec : calculation of spin effective action ] for details ) @xmath36 with @xmath37 . the inertia can be rewritten as @xmath38 . \label{expression of inertia 2}\ ] ] for @xmath39 , the function @xmath40 is a only slightly decreasing function from @xmath41 to @xmath42 . therefore , spin inertia does not depend much on the @xmath4 coupling constant @xmath3 and is proportional to @xmath43 , when @xmath44 and @xmath45 are regarded as independent parameters . adding this @xmath14 $ ] to the original spin action of the form @xmath46 = \int \frac{d^3x}{a^3 } dt\left [ s_l\dot\phi(\cos\theta-1 ) - \frac{j_ls_l^2}{2}(\partial_i \bm n)^2 \right ] , \label{spin original action}\ ] ] with @xmath47 the exchange coupling between spins , we obtain the total spin effective action as @xmath48 = \mathcal s_{\rm s}[\bm n]+\delta \mathcal s_{\rm eff}[\bm n]$ ] . including the zeeman coupling with external magnetic field we obtain @xmath49 = \int \frac{d^3x}{a^3 } dt\left[s\bm b \cdot \bm n + s\dot\phi(\cos\theta-1 ) - \frac{js^2}{2}(\partial_i \bm n)^2 + \frac{m_s}{2}\bm { \dot n}^2 \right ] + \mathcal o((\partial_\mu \bm n)^3 ) , \label{spin effective action}\ ] ] with @xmath50 the total spin amplitude per lattice cite and @xmath51 . we have set the gyromagnetic ratio as unity . as we will see below , spin with finite inertia has a typical precession mode with frequency @xmath52 . using @xmath0 or , and assuming @xmath53 , @xmath54 and @xmath55 , the energy scale of this frequency becomes @xmath56 , so that its period is @xmath57 therefore , as far as this simple estimation suggests , the existence of the inertia is significant for the dynamics of sub - picosecond scale . the equation of motion derived from this effective action is @xmath58 thus , the inertial term produces acceleration - dependent torque . we can rewrite this equation of motion , by taking vector product with @xmath11 , as @xmath59 the gilbert damping effect adds a term @xmath60 , with @xmath61 the dimensionless constant , to the right hand side of the equation of motion . therefore , the gilbert damping plays the same role as the familiar linear damping force for a point particle , and the time scale for this damping term to be significant is @xmath62 . on the other hand , the time scale for the inertial term to be effective is , as we will see below , @xmath63 . therefore , for @xmath64 , we can neglect the gilbert damping term as long as we are interested in the dynamics within the time scale @xmath65 . the equation of motion can be rewritten ( when @xmath66 ) in the conservation form of the angular momentum current @xmath67 , @xmath68 ( @xmath69 ) . note that the angular momentum @xmath70 ( per lattice cite ) , which is the noether charge corresponding to the invariance of the action under @xmath71 rotation in the internal spin space ( see appendix [ sec : derivation of the angular momentum ] for details ) , is no longer proportional to @xmath11 but includes the non - adiabatic contribution of the conduction electrons , @xmath72 . originally , the total angular momentum consists of that of the localized spin and that of the conduction electrons : @xmath73 with @xmath74 the amplitude of the localized spin . in the lowest order of the derivative expansion , i.e. in the adiabatic limit , the spin of the conduction electron aligns with that of the localized spin , so that @xmath75 with @xmath2 the spin polarization of the conduction electrons . beyond the adiabatic limit , the direction of the spin of the conduction electron is generally different from that of the localized spin . the derivative expansion incorporates this difference systematically , and the next order term in @xmath76 is given by @xmath77 . in this section , we describe classical dynamics of spin with inertia . for that purpose , it is helpful to use two equivalent pictures , which are summarized in fig.[fig : spin top particle ] . one is a symmetric spinning top , and the other is a massive charged particle on a sphere subject to a monopole magnetic field . we consider here only spatially homogeneous spin , @xmath97 , under time - independent magnetic field , for simplicity . the equivalence between the classical dynamics of spin and a spinning top has been recognized in the literature , e.g. refs.@xcite and @xcite . the content of this subsection is essentially a recapitulation of these facts , which we describe here in order for this paper to be self - contained . first , let us write down the lagrangian of spin with finite inertia and that of a spinning top . the lagrangian of spin is @xmath114 while the lagrangian of a spinning top in terms of the euler angles ( @xmath115 ) is @xmath116 where @xmath117 are the principle moments of inertia , @xmath118 is the mass of the top , @xmath119 is the gravitational acceleration constant , and @xmath120 is the distance between the center of mass and the fixed extremity of the top . here we have taken a symmetric spinning top and set two moments of inertia equal , @xmath121 . we take the positive directions of the external magnetic field @xmath105 and the gravity both in the positive @xmath122 direction . note that @xmath123 is a function of @xmath124 while @xmath125 is that of @xmath126 . let us see below that they have equivalent dynamics concerning @xmath124 . the equivalence can be directly seen at the level of their equations of motion . the equation of motion of spin is @xmath127 = 0 , \label{eom of quadratic spin}\end{aligned}\ ] ] while the equation of motion of the spinning top is @xmath128 = 0 , \notag \\ & \frac{d}{dt}\left [ i_3 ( \dot\psi+\dot\phi\cos\theta ) \right ] = 0 . \label{eom of spinning top}\end{aligned}\ ] ] from the last equation in eq . , the canonical momentum @xmath129 conjugate to @xmath130 , @xmath131 is conserved . substituting this @xmath129 for @xmath132 in the other two equations in , we obtain the same equations as with replacements @xmath133 thus , the classical behaviors of @xmath134 and @xmath135 are the same for spin and a spinning top . we can see the correspondence more explicitly through their hamiltonians : @xmath136 and @xmath137 where @xmath138 and @xmath139 are the canonical momenta of @xmath134 and @xmath135 , respectively . these two hamiltonians are completely the same under the replacements . [ the last term in does not contribute to the dynamics of @xmath134 and @xmath135 . ] the lagrangians and are related by the legendre transformation about @xmath130 : @xmath140 with replacements . in the right hand side , @xmath141 is substituted by @xmath142 ( or @xmath129 ) via . the situation is quite similar to that of the familiar centrifugal force problem . there , the original lagrangian is given as @xmath143 , and we can obtain @xmath135-reduced lagrangian by the legendre transformation about @xmath135 : @xmath144 . in exchange for reducing @xmath135 , there appears a fictitious potential @xmath145 . likewise , spin dynamics is the @xmath130-reduced dynamics of a spinning top , and the spin berry phase ( the second term in eq . ) appears as the fictitious potential arising from the @xmath130-reduction . when we perform the legendre transformation also about @xmath134 and @xmath135 on both sides in eq . , we are led to the same hamiltonians and . 0 0 we can regard the dynamics of the spinning top as those of a free particle on a three - sphere @xmath146 , whose metric is @xmath147 . [ to be precise , the @xmath146 is squashed in the @xmath130-direction in by the ratio @xmath148 , but this is not important for the discussion here . ] from such a viewpoint , the monopole magnetic field @xmath149 is just an effective potential arising due to dimensional reduction from @xmath146 to @xmath150 . a typical example of such effective potential is the familiar centrifugal potential : the particle gets the centrifugal effective potential after the azimuthal angle @xmath135 is reduced , @xmath151 with @xmath152 the angular momentum . at the case @xmath146 to @xmath150 , the effective potential is the coupling with the gauge field , contrary to the scalar centrifugal potential at the case @xmath153 to @xmath154 . the difference stems from the form of the metric in the dimensionally reduced direction . the monopole gauge field comes from the non - diagonal metric @xmath155 in @xmath156 , while the centrifugal potential comes from the prefactor @xmath157 in @xmath158 . the gauge transformation of the monopole field corresponds to unphysical @xmath124-dependent shift of @xmath130 , i.e. , @xmath159 with @xmath160 an arbitrary function of @xmath134 and @xmath135 . since the classical dynamics of spin with inertia and of a spinning top are equivalent , spin with inertia behaves in the same manner as a spinning top does ( fig.[fig : free precession and nutation ] ) . when a magnetic field is not applied , spin undergoes free precession : the spin precesses around the total angular momentum @xmath161 ( eq . ) , which is a constant of motion . when a magnetic field is applied , spin undergoes the larmor precession accompanied by the nutation : the total angular momentum @xmath70 precesses around the magnetic field ( the larmor precession ) , and the spin precesses around @xmath70 at each time ( the nutation ) @xcite . the free precession solution of the equation of motion ( with @xmath66 ) is @xmath162 therefore , the free precession frequency , or the nutation frequency , @xmath163 , is @xmath52 , assuming that @xmath164 is order of unity ( that is , the radius of the spin free precession or nutation is not very large ) . this gives the typical time scale for the inertial term , @xmath63 . finally , we mention that usual spin with zero inertia , @xmath165 , can be regarded as follows . it corresponds to the case of @xmath166 under replacements . this means that the other principle moment of inertia , @xmath167 , vanishes since , for an actual rigid body , any one of the principle moments of inertia is equal to or less than the sum of the other two , e.g. , @xmath168 @xcite . thus , setting @xmath169 leads to @xmath170 . therefore , there does not exist a spinning top corresponding to spin with zero inertia , in a non - relativistic framework . 0 the free motion of a symmetric spinning top is precessional motion of its symmetric axis . this precession is intrinsic to the system in that it occurs without any external force . when external force such as the gravitational field is applied , this precession is moved as a whole in direction perpendicular to both the applied force and the radial direction of the center of the precession . the precession is called nutation especially when such external force is applied . when the nutation radius is small , the frequency of this nutation is @xmath171 where @xmath129 is the component of the angular momentum in direction of the symmetric axis and @xmath172 is the moment of inertia about an arbitrary axis perpendicular to the symmetric axis . since the dynamics of a spin and a spinning top are equivalent , the behavior of spin with finite inertia is completely the same . spin has its intrinsic precession mode as the free motion . when or nutation mode whose frequency is about @xmath52 ( see the replacements ) . since the spin berry phase corresponds to a monopole gauge field @xcite , the classical dynamics of spin is also equivalent to that of a charged particle on a sphere subject to a monopole background . let us describe this equivalence and use it to understand the behavior of spin with inertia . a magnetic monopole yields a magnetic field in the radial direction with the strength inversely proportional to the square of the distance from the monopole . being put at the origin @xmath173 , the gauge field @xmath174 and the magnetic field @xmath175 of the monopole are @xmath176 where @xmath177 , @xmath178 is the unit vector in @xmath135-direction , and @xmath179 is the magnetic charge of the monopole . the lagrangian of a massive charged particle coupled to this monopole magnetic field and a constant electric field @xmath180 is @xmath181 where we take the electromagnetic scalar potential @xmath182 so that @xmath183 . this lagrangian is identical with that of spin with inertia when we assume that the particle is constrained on a unit sphere and identify the direction of the particle @xmath184 with the spin direction @xmath11 , the mass @xmath25 with the spin inertia @xmath0 , the magnetic charge @xmath179 with the spin amplitude @xmath142 , and the electric field @xmath180 on the particle with the magnetic field @xmath185 on spin ( we use calligraphic letters for electromagnetic fields on the particle ) . being restricted on the unit sphere , the dynamical degrees of freedom of the particle are the spherical angles @xmath134 and @xmath135 , and the equation of motion is , in vectorial form , @xmath186 which is identical with ( with @xmath187 ) . the last term proportional to @xmath184 in the right hand side of the second line is the constraint force , which keeps @xmath188 . 0 which makes the radial ( @xmath184-direction ) component of the equation of motion be zero and whose coefficient is given by a scalar product of @xmath184 with the other terms in the equation of motion , that is , @xmath189 the particle is on a unit sphere and is subject to the monopole magnetic field @xmath190 ( on the sphere @xmath188 ) and the electric field @xmath180 . the general motion of the particle is , in the absence of the electric field @xmath180 , the cyclotron motion , and , in the presence of the electric field @xmath180 , the @xmath191 drift motion , which yields drift motion of the particle in @xmath192 direction , rather than @xmath193 direction . ] . in view of the equivalence between the dynamics of spin with inertia and of the charged particle on a sphere , the free precession of spin described in the last subsection corresponds to the cyclotron motion of the particle with the frequency @xmath194 , and the larmor precession of spin accompanied by nutation corresponds to the @xmath191 drift motion . we summarize here the three pictures for the classical dynamics described by . [ cols="^,^,^",options="header " , ] @xmath195 + we have seen in this section that spin with finite inertia has an intrinsic precession mode ( i.e. free precession or nutation ) . the magnetic field causing this precession is the monopole magnetic field intrinsic to spin , i.e. , the spin berry curvature . the frequency of the intrinsic precession mode , @xmath196 , is infinite when the inertia of spin @xmath0 is zero , but goes down to finite value when @xmath0 becomes non - zero . as related works , we mention refs.@xcite where nutational motion of spin in josephson junctions or tunnel junctions between ferromagnets was discussed by examining the landau - lifshitz - gilbert equation with time - dependent gilbert damping coefficient . 0 @xmath197 now when we take the @xmath198 limit with @xmath199 fixed , then @xmath200 and @xmath201 . thus , at this limit , the free precession is shrinked to a point , and the initial condition of the velocity of spin becomes irrelevant , as expected . for general motion of spin under magnetic field , 0 note that this free precession of spin breaks parity ( chirality ) . for free precession of a spinning top , parity is broken depending on how we make the top spinning , clockwise or counterclockwise . the spin is always spinning counterclockwise in this sense , and the resulting free precession is also always counterclockwise about the axis of precession . the information about broken parity is formally expressed by the positivity of the monopole charge , which is a pseudoscalar under parity transformation ( this is wrong ! ! ! ! ) . 0 this is partly because the spin vector @xmath202 is a pseudovector while the position vector @xmath184 of a particle is a real vector . the motion is generally solvable ( see landau - lifshiz ) , but only the case where the radius of the nutation is very small is enough in this paper . the corresponding solution of eq . is @xmath203 as is the case of free precession , we have rewritten the solution in term of the nutation velocity @xmath199 . as we take the limit @xmath204 with @xmath199 fixed , the radius of nutation goes to zero and the usual larmor precession is recovered . we have so far assumed spatially homogeneous configuration of spin under time - independent magnetic field . let us include spatial variation of spin and time - dependence of applied magnetic field . the most typical behavior of spin under time - dependent magnetic field is the resonance phenomenon . in ref.@xcite , it is shown that spin with finite inertia @xmath0 has a resonance peak at the intrinsic frequency @xmath196 . we discuss other behaviors of spin , such as spin wave , domain wall motion . we also discuss an unusual behavior of spin under a large and time - dependent magnetic field . 0 to test the existence of the inertia of spin experimentally , the most direct way is to see the resonance peak at the intrinsic frequency @xmath196 . first , let us see the resonance for free precession @xcite . including the gilbert damping and assuming @xmath205 with @xmath206 , the eom is @xmath207 we apply ac magnetic field on @xmath208-@xmath209 plane - direction . however , since the intrinsic frequency @xmath210 is generally much larger than the larmor frequency @xmath211 , the dc field has no effect on the resonance at @xmath212 . therefore , what they observed is the resonance mode for the free precession . ] , @xmath213 . after enough time , the motion of spin converges to a stationary state where it precesses at frequency @xmath214 . the solution of this state is given by @xmath215 where @xmath216 the energy dissipation rate @xmath217 is proportional to @xmath218 , which has resonance peak at the frequency @xmath219 . next , we consider resonance mode for larmor precession accompanied by nutation . this time , we do not assume @xmath205 with @xmath206 . the eom is eq . with the gilbert damping included . we apply two ac magnetic fields on @xmath208-@xmath209 plane and one dc field in @xmath122-direction : @xmath220 . we set @xmath221 near @xmath211 and @xmath222 near @xmath210 . the motion of spin converges to larmor precession of frequency @xmath221 with accompanying nutation of frequency @xmath222 . the energy dissipation rate @xmath223 , which is proportional to @xmath218 , has resonance peaks at @xmath224 and at @xmath225 , respectively . fig.[fig : resonance ] shows the numerical result of @xmath226 . let us consider spin wave solutions of spin with inertia , and see that there exists a new gapful spin wave mode with unusual handedness . when we assume @xmath227 with @xmath228 , the equation of motion is means that we consider the problem in the tangent plane on @xmath229 , so the constraint force in vanishes . ] , from the action , @xmath230 where we have set @xmath66 . since the equation of motion is of second order in the time derivative , we have now two spin wave modes . when we substitute a plane wave @xmath231 to the equation of motion , with @xmath232 an arbitrary small constant , we obtain the dispersion relation ( fig.[fig : spin wave ] ) , @xmath233 one is the usual spin wave mode and the other is the new spin wave mode , the latter of which is essentially the free precession of spin mediated in space by the interaction @xmath234 with neighboring spins . we have expanded the dispersion relation in powers of @xmath235 and cut higher order terms because we are in the framework of the derivative expansion in deriving eq .. 0 the usual spin wave mode is clockwise seen from the positive @xmath122-direction , while the new spin wave mode is counterclockwise . in fact , to the leading order of the dispersion relation , the former obeys the equation of motion @xmath236 , while the latter @xmath237 , where @xmath238 is the unit vector in @xmath122-direction . therefore , their handednesses are opposite . the total angular momentum @xmath70 in also rotates in counterclockwise for the new spin wave mode . when the spin and the angular momentum align in the same direction , @xmath239 , the fluctuation of spin around the stable configuration is always clockwise . this can be understood as follows ( see e.g. ref.@xcite ) . let the stable configuration be @xmath229 . the poisson commutation relation of the angular momenta , @xmath240=j^0_z\cong s$ ] , leads to that of spin , @xmath241=1/s$ ] . therefore , the @xmath208 and @xmath209 components of spin are canonically conjugate to each other . moreover , the hamiltonian of spin expanded around the stable point is @xmath242 $ ] with @xmath243 . these two facts , i.e. the canonical commutation relation of @xmath244 and @xmath245 , and the form of the hamiltonian , indicate that the fluctuation of spin around @xmath229 has the same dynamical structure as that of a one - dimensional harmonic oscillator with coordinate @xmath208 and its canonical momentum @xmath246 governed by the commutation relation @xmath247=1 $ ] and the hamiltonian @xmath248 . the trajectory of the harmonic oscillator is always clockwise in the phase space . therefore , the fluctuation of spin is also always clockwise . however , when the inertia of spin is finite , the spin and the angular momentum generally do not align due to the additional contribution @xmath77 in @xmath70 , and moreover the hamiltonian of spin such as eq . does not consist only of the potential term but also of the kinetic term therefore , the fluctuation of spin with inertia is not restricted to clockwise motion . let us see that a domain wall also has the free precession mode , corresponding to that of each spin . when easy - axis anisotropy @xmath249 is included to the action , the system has a static domain wall solution , @xmath250 , \phi=\phi_0 $ ] ( we take @xmath122-axis as the wall normal ) where @xmath251 and @xmath252 are integration constants , and @xmath253 . promoting @xmath251 and @xmath252 to dynamical variables , substituting the domain wall solution into the action and integrating it over space , we obtain the lagrangian of @xmath251 and @xmath252 : @xmath254 = \frac{m_{\rm w}}{2}(\dot x^2 + \lambda^2\dot \phi_0 ^ 2 ) + \frac{sn_{\rm w}}{2\lambda}(\dot x\phi_0-x\dot\phi_0 ) \label{l of domain wall}\ ] ] with @xmath255 and @xmath256 where @xmath232 is the cross sectional area of the domain wall ( we discarded irrelevant constant terms in the lagrangian ) . we are going to show that this lagrangian is of the same form as that of a charged particle on a cylinder with a constant magnetic field @xmath257 perpendicularly penetrating the surface of the cylinder . take an orthogonal coordinate frame @xmath258 on a cylinder , with @xmath259 the periodic direction . ( the directions @xmath260 and the direction outward normal to the surface of the cylinder make an orthogonal right - handed frame . ) then , the gauge potential @xmath261 for the constant magnetic field @xmath262 in the outward normal direction of the surface can be taken as @xmath263 , and the lagrangian of the particle is @xmath264 therefore , when we identify @xmath265 with @xmath266 , the lagrangian is that of a charged particle under a perpendicular magnetic field of flux @xmath257 ( fig.[fig : particle on a cylinder ] ) . the magnetic field @xmath262 for the particle originates in the spin berry phase , and the mass @xmath267 of the particle comes from the inertia of spin in the action . as a result , the particle undergoes cyclotron motion ( in other words , the domain wall undergoes the intrinsic free precession mode ) , where @xmath251 and @xmath268 oscillate , @xmath269 with @xmath270 a constant , at frequency @xmath271 . this frequency corresponds to that of spin free precession discussed in section [ sec : spin and spinning top ] . when an external field is applied on this domain wall , the particle on a cylinder @xmath265 feels the corresponding force on it , and it moves in the direction perpendicular to the force , accompanied by the cyclotron motion ( nutation ) . that is , the particle undergoes @xmath272 drift motion , where @xmath180 is the force and @xmath273 is the magnetic field penetrating the cylinder @xmath265 perpendicularly . 0 the discussion above is well valid when the inertia @xmath0 is so large that the intrinsic frequency @xmath52 is less than the frequency of spin waves : @xmath274 . when @xmath210 exceeds this frequency scale , the dynamics of the domain wall can be no longer described only by the collective coordinates @xmath251 and @xmath252 , and spin wave excitations should also be included into the dynamics . [ mathematically , @xmath280 is the general homogeneous solution of , and @xmath297 is a special solution of which vanishes when the electric field is zero . and @xmath298 for large @xmath102 is small enough so that the expansion , @xmath299 , converges and the operator @xmath300 is well defined . ] ] the guiding center motion @xmath301 is no longer perpendicular to the applied field @xmath180 and is modified by the product of the cyclotron period @xmath302 and the time rate of change of the applied field . when the particle is massless , then @xmath303 and the time - derivative of the electric field has no effects on the guiding center motion . as we saw in section [ sec : spin and spinning top ] , the classical dynamics of spin with inertia is a spherical version of this planar problem , with the role of an electric field for the particle played by a magnetic field for spin , and therefore the essential features will be the same ( especially when the nutation radius is small and the curvature of the sphere can be neglected ) : ( i ) the spin nutation and the guiding center motion are decoupled ; ( ii ) the effect of the inertia on the guiding center motion is significant when the time rate of change of the applied field is comparable to the intrinsic frequency @xmath210 . therefore , the effect of the inertia is not only the superimposition of the nutational motion to the larmor precession . the inertia of spin gives an independent effect on the trajectory of the larmor precession itself , cooperatively with the time - dependence of the applied field . even when the nutation radius is small and there are no signs of nutation in experimental data about the time - profile of magnetization , it never indicates that the inertia has no effects on the dynamics , especially when the applied magnetic field has very rapid time - dependence , such as ultrafast magnetization . we illustrate this fact by numerical calculation ( fig.[fig : reaction to time - dependent magnetic field ] ) . we fix a magnetic field in the @xmath122-direction . first , we make the magnetic field in the negative @xmath122-direction , then change it into the positive @xmath122-direction . we can see that the spin moves parallelly ( i.e. in the positive @xmath122-direction ) to the time derivative of the magnetic field when the magnetic field is changing - direction . only the time - dependence of the magnetic field is essential . ] . this phenomenon is similar to usual spin flip by the gilbert damping , but is different in that we do not have to vary the direction of the magnetic field and moreover the time change rate of the magnetic field can be arbitrary ( as long as our derivative expansion is valid ) and is irrelevant to the damping coefficient . 0 for simplicity , let us fix the direction of the applied magnetic field @xmath304 in @xmath122-direction . when spin does not have inertia , spin undergoes the larmor precession only on @xmath208-@xmath209 plane . in contrast , when spin has inertia , the velocity of the larmor precession has also a non - zero component in @xmath122-direction , which is roughly @xmath305 , from eq .. therefore , from the initial time to the final time , the guiding center of spin nutation shifts in @xmath122-direction in amount of roughly @xmath306 , where @xmath307 and @xmath308 are the magnetic fields at the initial time and the final time , respectively . ( we here mean the velocity and the distance as that of @xmath11 , not of @xmath309 . ) when @xmath310 , or in other words @xmath311 is comparable to the intrinsic frequency ( nutation frequency ) @xmath210 , this shift changes the @xmath122-component of the spin significantly . to illustrate this fact , we numerically solve the equation of motion under an external magnetic field @xmath312 in @xmath122-direction with @xmath313 , @xmath314 and @xmath315 . we set @xmath316 and @xmath317 so that the period of the nutation is @xmath318 , as estimated in section [ sec : effective action ] . the initial time is @xmath319 and the final time is @xmath320 . the initial condition of spin , and is not essential for the shift of the spin in @xmath122-direction very much . ] is that @xmath321 , @xmath322 and @xmath323 , @xmath324 . fig.[fig : reaction to time - dependent magnetic field ] shows the numerical result of the trajectory of spin . the spin precesses in counterclockwise seen from the positive @xmath122-direction in the time domain i , where the magnetic field points in the negative @xmath122-direction . the spin stops precessing on @xmath208-@xmath209 plane and moves upward in @xmath122-direction in the time domain ii , where the direction of the magnetic field is changed from the negative to the positive @xmath122-direction . finally , the spin precesses in clockwise in the time domain iii , where the magnetic field points in the positive @xmath122-direction . [ the switching of the sign of the magnetic field from negative to positive is not essential for the shift of the spin in @xmath122-direction ; only the difference of the magnetic fields between at the initial and the final time , @xmath311 , is essential . ] 0 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * this unusual behavior of spin can be simply understood when we look at the angular momentum rather than spin itself . as in , the angular momentum of spin with finite inertia is no longer @xmath325 but has additional non - adiabatic contribution @xmath326 . when the larmor frequency @xmath327 is much smaller than the intrinsic frequency ( nutation frequency ) , @xmath328 , then the vector @xmath326 points inward to the center of the nutation circle , and the trajectory of the angular momentum of spin is overlapped with that of spin ( fig.[fig : contribution of orbital angular momentum](a ) ) . on the other hand , when @xmath329 , nutation does not form circles but rather ripples on the trajectory of spin . in this case , the vector @xmath326 points in almost the same direction at each time . as the result , the trajectory of the angular momentum deviates from that of spin ( fig.[fig : contribution of orbital angular momentum](b ) ) . fig.[fig : trajectories of spin and angular momentum ] is a replot of the trajectory of spin in fig.[fig : reaction to time - dependent magnetic field ] together with that of the angular momentum . the vector @xmath326 points in the positive @xmath122-direction in the time domain i ( described in fig.[fig : reaction to time - dependent magnetic field ] ) , and in the negative @xmath122-direction in the time domain iii , resulting in the conservation of the @xmath122-component of the angular momentum . this parallel shift of spin in the direction of the time derivative of the magnetic field can be understood also from a viewpoint of a spinning top , where the magnetic field on spin corresponds to the gravitational field on the top . the time - variation of the gravitational field can be realized fictitiously by putting the top in an elevator accelerated in the negative @xmath122-direction . there , the top feels , besides the real gravitational field in the negative @xmath122-direction , a fictitious inertial force in the positive @xmath122-direction due to the acceleration . this inertial force in the positive @xmath122-direction makes the top stand up in the positive @xmath122-direction , which explains the shift of spin in @xmath122-direction . thus , the inertia of spin yields an inertial force , literally . in this respect , the connection between the inertia of spin and the inertial effect on spin @xcite will also be interesting . 0 0 we examined the inertia of spin in metallic ferromagnets , which arises in the derivative expansion of spin effective action and represents non - adiabatic contribution from environmental degrees of freedom . we derived an explicit expression of spin inertia in an @xmath4 model . the equivalence between the dynamics of spin with inertia , spinning top and a charged particle on a sphere was explained . the behavior of spin with spatially inhomogeneous configuration or under time - dependent magnetic field is studied . the finite inertia has mainly two effects : the superimposition of the nutation on the usual larmor precession and the parallel shift of the larmor precession . as explained several times in this paper , most of unusual behavior of spin with finite inertia originates in the non - adiabatic component @xmath330 in the relation between spin and the angular momentum , @xmath331 . due to this difference between them , spin acquires a kind of separate freedom from the angular momentum , whose behavior is relatively restricted due to the conservation law . as the result , spin undergoes free precession or nutation discussed in section [ sec : spin and spinning top ] , and furthermore parallel shift along the direction of the time derivative of applied magnetic fields , discussed in section [ sec : phenomena ] . since one of the main purposes of spintronics is understanding and application of diverse cooperative phenomena between localized spins and their environmental degrees of freedom , researches beyond the adiabatic limit will be well - motivated for this spirit of spintronics ; in the adiabatic limit , the environmental degrees of freedom are rather obedient to the localized spins . in this paper , we estimated that in metallic ferromagnets such non - adiabaticity is effective for sub - picosecond dynamics , i.e. ultrafast magnetization . for broader applications , physical situations where non - adiabaticity is effective for sub - nanosecond dynamics are also desirable . apart from the context of effective dynamics ( i.e. the dynamics including the effects of environmental degrees of freedom ) , the inertial term , together with the spin berry phase term , can be present in the dynamics of the order parameters in systems whose ground states can be intermediate states between ferromagnet and antiferromagnet ( see e.g.@xcite ) . the discussion of this paper will be also applicable to such systems . t. k. would like to thank aron beekman , yan - ting chen , hideo kawaguchi , se kwon kim , nguyen thanh phuc , henri saarikoski , members in eiji saitoh group in tohoku university , and other people for helpful comments . in this appendix , we calculate the spin effective action by integrating out a conduction electron coupled to spin through the @xmath4 interaction , and derive the expression of spin inertia . we always make summation over repeated indices unless otherwise stated . the action of a conduction electron , represented by field operators @xmath8 and @xmath9 , coupled to localized spin @xmath11 is @xmath332=\int d^3x dt~\bar c \left(i\hbar \partial_t + \frac{\hbar^2\partial_i^2}{2 m } + \epsilon_f + g_{\rm sd}\bm n \cdot \bm \sigma \right)c . \label{action of conducting fermion}\ ] ] with @xmath3 the @xmath4 coupling constant . what we calculate in this appendix is the contribution to spin effective action from the conduction electron , @xmath14 $ ] , defined as @xmath13\right ) \equiv \int \mathcal d\bar c \mathcal d c~ \exp\left(\frac{i}{\hbar}\mathcal s_{\rm e}[c,\bar c , \bm n]\right ) . \label{def of delta seff}\ ] ] adding this @xmath14 $ ] to the original spin action @xmath46 = \int \frac{d^3x}{a^3}\left[s_l \dot \phi(\cos\theta-1 ) - j_ls_l^2(\partial_i \bm n)^2\right],\ ] ] where @xmath134 and @xmath135 are the polar angles of the unit vector @xmath11 , we obtain the total spin effective action @xmath48 = \mathcal s_{\rm s}[\bm n]+\delta \mathcal s_{\rm eff}[\bm n]$ ] . first , we diagonalize the @xmath4 coupling in . in order for that purpose , it is convenient and essential to employ the unitary - transformed electron field @xmath44 such that @xmath333 . here , the @xmath28 matrix @xmath30 is defined so that @xmath334 as such , @xmath30 has indefiniteness about the right @xmath335 gauge transformation , @xmath336 with @xmath337 an arbitrary function . [ due to the defining relation @xmath338 of @xmath44 , this unitary transformation corresponds to that of @xmath44 , i.e. @xmath339 . the original electron field @xmath8 remains invariant . ] one choice of @xmath30 is @xmath340 with @xmath341 . geometrically , @xmath30 rotates the electron spin in the positive @xmath122-direction , @xmath342 , to the direction of @xmath11 . for the choice @xmath340 above , @xmath343 with @xmath344 . the indefiniteness of @xmath30 , @xmath345 , corresponds to the phase rotation of @xmath346 , i.e. @xmath347 . the @xmath4 coupling becomes diagonal in the @xmath44-frame , @xmath348 but , in compensation for the @xmath28 gauge transformation by @xmath30 , there appears a spin gauge field @xmath349 in the derivative terms , @xmath350 . then , the action becomes @xcite @xmath351 + \int d^3x dt \left [ a_t^a ( -\hbar \bar a\sigma^a a ) + a_i^a \left(\frac{i\hbar^2}{2m}\bar a \sigma_a \overleftrightarrow{\partial}_i a\right ) + a_i^a a_i^a \left(-\frac{\hbar^2}{2m}\bar a a\right ) \right ] . \label{action of conducting fermion2}\end{aligned}\ ] ] here @xmath352 $ ] is @xmath353\equiv \int d^3x dt~\bar a \left(i\hbar \partial_t + \frac{\hbar^2\partial_i^2}{2 m } + \epsilon_f + g_{\rm sd } \sigma_3 \right)a , \label{s0}\ ] ] and @xmath354 is the spin gauge field ( make distinction between the electron field @xmath44 and the @xmath28 index@xmath355 ) , which is , in terms of the polar angles of @xmath11 , @xmath356 where @xmath357 . because @xmath34 is proportional to the first order derivative of @xmath11 , the derivative expansion of @xmath87 in powers of @xmath15 corresponds to that of @xmath34 : @xmath358 = & \frac{i}{\hbar}\int d^4x \left [ a_t^a ( -\hbar \langle \bar a\sigma^a a\rangle ) + a_i^a \frac{i\hbar^2}{2m}\langle \bar a \sigma_a \overleftrightarrow{\partial}_i a\rangle + a_i^a a_i^a \left(-\frac{\hbar^2}{2m}\langle \bar a a\rangle \right ) \right ] \notag \\ & + \frac{1}{2}\left(\frac{i}{\hbar}\right)^2 \int d^4x d^4x ' \big [ a_t^a(x ) a_t^b(x')\hbar^2 \langle \bar a \sigma^a a(x)\bar a \sigma^b a(x ' ) \rangle \notag \\ & + a_i^a(x)a_j^b(x')\left(\frac{i\hbar^2}{2m}\right)^2 \langle \bar a\sigma^a \overleftrightarrow{\partial_i } a(x ) \bar a\sigma^b \overleftrightarrow{\partial_j } a(x ' ) \rangle \notag \\ & + a_t^a(x)a_i^b(x')\left(-\frac{i\hbar^3}{m}\right ) \langle \bar a \sigma^a a(x)\bar a \sigma^b \overleftrightarrow{\partial_i}a(x')\rangle \big ] + \mathcal o(a_\mu^3 ) \label{expansion in powers of a}\end{aligned}\ ] ] where time - ordered quantum expectation values are taken about @xmath359 , @xmath360},\ ] ] and we take only the connected diagrams to evaluate the quantum expectation values . because the non - perturbed action @xmath359 is isotropic in space , @xmath361 and @xmath362 vanish . we define the notation for correlation functions as @xmath363 using the fourier components , we expand these non - local quantities . for example , @xmath364 when we substitute this into the expansion , the terms other than the first term in increase the order of the derivative on @xmath11 after partial integration , e.g. @xmath365 \delta^{(4)}(x - x')$ ] . therefore , we take only the first term in as long as we are interested in at most @xmath366-terms . after all , @xmath14 $ ] is , to @xmath367-order , @xmath17 = \int d^4x \left [ ( -\hbar \langle \bar a\sigma^a a\rangle)a_t^a + \left(-\frac{\hbar^2}{2m}\langle \bar a a\rangle \delta_{ab}\delta_{ij } - \frac{\hbar^3}{8m^2}{\rm re}\chi^{ab}_{ij}|_{q=0 } \right)a_i^a a_j^b + \left(\frac{\hbar}{2}{\rm re}\rho^{ab}|_{q=0}\right)a_t^a a_t^b \right ] \label{delta seff}\ ] ] ( the imaginary parts of @xmath368 and @xmath369 contribute to damping and are neglected here ) . we calculate each term in the following . secondly , the number density of the conducting electrons @xmath381 is given by @xmath382 . for non - diagonal elements , it is clear that @xmath411 regardless of the order of the @xmath391 and @xmath395 limits . finally , let us calculate @xmath412 . the calculation is almost the same as that of @xmath413 : just drop the factor @xmath414 from eq .. then , @xmath415 \notag \\ = & \frac{\hbar}{g_{\rm sd}}\frac{k_{f+}^3-k_{f-}^3}{6\pi^2}.\end{aligned}\ ] ] the order of taking the @xmath416 limits matters again for @xmath417 . this time , @xmath418 while @xmath419 . again , we determine the order of the limits by the isotropy about @xmath11 . when @xmath420 does not vanish , @xmath421 appears in @xmath87 , which breaks the isotropy . therefore , contrary to the case of @xmath404 , we employ @xmath422 as @xmath420 . finally , @xmath423 obviously vanishes . adding this @xmath14 $ ] to the action of localized spin , we obtain the total effective action @xmath48 $ ] , @xmath49 = \int \frac{d^3x}{a^3}dt \left[s\dot \phi(\cos\theta-1 ) - \frac{js^2}{2}(\partial_i \bm n)^2 + \frac{m_s}{2}\bm{\dot n}^2\right ] . \label{total spin effective action}\ ] ] with @xmath50 and @xmath431 . 0 the above results are obtained in the clean limit . let us include spin independent dilute impurity . the effect of the impurity is only to introduce the self energy into the green function , @xmath432 with @xmath428 the life time of the electron , because the vertex correction due to the impurity is negligible when @xmath433 . to see that , we take as an example the diagram with one impurity line over vertex , whose expression is @xmath434\ ] ] with @xmath199 the strength of the impurity . using @xmath435 = & ( \delta_{ab}-\delta_{a3}\delta_{b3})(g_1^+ g_2^+ g_3 ^ - g_4 ^ - + g_1 ^ - g_2 ^ - g_3^+ g_4^+ ) \notag \\ & ~+ \delta_{a3}\delta_{b3}(g_1^+ g_2^+ g_3^+ g_4^+ + g_1 ^ - g_2 ^ - g_3 ^ - g_4 ^ - ) + i \varepsilon_{ab3}(g_1 ^ - g_2 ^ - g_3^+ g_4^+ - g_1^+ g_2^+ g_3 ^ - g_4 ^ -),\end{aligned}\ ] ] we obtain @xmath436 thus , we see that one impurity line adds extra factor @xmath437 @xmath438 therefore , the vertex correction is of order @xmath439 . using the green function including the self energy , the inertia of spin becomes . 0 let us rewrite @xmath0 as @xmath440.\ ] ] for @xmath39 , the function @xmath40 is a decreasing function from @xmath41 to @xmath42 . therefore , assuming @xmath54 , numerically @xmath441 . in this appendix , we describe the derivation of the angular momentum current @xmath442 from the effective action @xmath443 , for the readers who are not familiar with such derivation . consider a @xmath71 rotation in internal space ( spin space ) of @xmath71 group and an element @xmath30 of @xmath28 group are related by @xmath444 with @xmath445 the pauli matrices . we are interested in the classical dynamics of a spin vector , that is , the behavior of @xmath446 , where @xmath447 is a quantum spin state . a @xmath28 rotation on @xmath448 causes a @xmath71 rotation on @xmath11 , i.e. @xmath449 . ] . under this rotation , a vector @xmath450 in internal space , such as spin @xmath11 and magnetic field @xmath105 , is rotated forms a scalar product with @xmath11 in the zeeman term , which means that @xmath105 is also a vector in internal space with the same transformation property as @xmath11 , within the framework presented here . ] . when the rotation is infinitesimally small , it is given by @xmath451 , where @xmath452 are constant infinitesimal transformation parameters . the effective action is invariant under this rotation . the noether current corresponding to this invariance , i.e. the angular momentum current , can be obtained by the following trick . make the transformation parameters @xmath452 be dependent on the position in the real space and time , @xmath453 . then , the variation of the action under this space - time dependent internal rotation does not vanish , but is given in the form , @xmath454 here , terms without any time- and spatial - derivatives on @xmath452 are absent , because the rotation would be a symmetry of the action if @xmath452 were constants . we have performed the partial integration in the last equality in eq .. when @xmath11 satisfies its equation of motion , then @xmath455 because such @xmath11 gives a saddle point of @xmath443 . since @xmath455 for arbitrarily position - dependent @xmath453 , it follows that @xmath456 when @xmath11 satisfies its equation of motion . thus , @xmath442 is the conserved current corresponding to the symmetry under internal rotations , i.e. the angular momentum current . [ in fact , since @xmath11 is a unit vector , an arbitrary variation of @xmath11 is given by a rotation in internal space . therefore , from the variation , the conservation law @xmath456 is the equation of motion itself . ] concretely , the variation of each term in the effective action is given as follows . for the spin berry phase term , @xmath457 \notag \\ = & \int \frac{d^3x}{a^3}dt~ \bm { \dot \theta}\cdot s\bm n , \label{variation of the spin berry phase}\end{aligned}\ ] ] where we have temporarily used @xmath458 which is defined over extended space - time @xmath459 with @xmath460 a dummy direction and with boundary conditions @xmath461 and @xmath462 the transformation parameters @xmath463 are also extended to @xmath464 similarly . in the last equality in eq . , we have performed the partial integration with respect to @xmath465 . for the inertial term and the exchange coupling term , @xmath466 the zeeman term is irrelevant for deriving the angular momentum because it does not involve the derivative on @xmath11 , which means that it does not yield terms proportional to @xmath467 or @xmath468 in eq . under the rotational transformation . in all , from eq . , the angular momentum current can be read as @xmath469 in this appendix , we derive higher order terms in spin effective action @xmath87 in the way discussed in the last part of section [ sec : effective action ] . we concentrate only on time - derivative terms , but derivation of higher order terms with spatial - derivatives is essentially the same . first , all possible terms in spin effective action to the fourth - order are @xmath470.\ ] ] the angular momentum @xmath78 can be derived from this action in the way described in appendix [ sec : derivation of the angular momentum ] , as @xmath471 this @xmath70 must satisfy @xmath472 . substitution of eq . makes , with @xmath473 , @xmath474 the unique solution in order that each coefficient vanishes is that @xmath475 0 y. b. bazaliy , b. a. jones , s .- c . zhang , modification of the landau - lifshitz equation in the presence of a spin - polarized current in colossal- and giant- magnetoresistive materials , " phys . rev . b 57 , r3213 ( 1998 ) . j. shibata , h. kohno , `` spin and charge transport induced by gauge fields in a ferromagnet , '' phys . b84 , 184408 ( 2011 ) . h. suhl , theory of the magnetic damping constant , " ieee trans . 34 , 1834 ( 1998 ) . g. tatara , h. kohno , j. shibata , `` microscopic approach to current - driven domain wall dynamics , '' phys . 468 , 213 - 301 ( 2008 ) . h. kawaguchi , g. tatara , coupling theory of emergent spin electromagnetic field and electromagnetic field , " j. phys . 83 , 074710 ( 2014 ) .

we present a study of globular clusters in the inner region of the giant elliptical galaxy ngc 4472 , based on the wfpc2 archive data . we have found about 1560 globular cluster candidates at the galactocentric radius arcmin , most of which are located within arcmin . diagram of these objects shows a dominant vertical structure which consists obviously of two components : blue globular clusters ( bgcs ) and red globular clusters ( rgcs ) . the luminosity function of the globular clusters is derived to have a peak at(max ) mag and from gaussian fitting . from this result the distance to ngc 4472 is estimated to be mpc , for the foreground reddening and the peak luminosity for the galactic globular clusters of(max ) mag . the peak luminosity for the rgcs is similar to that for the bgcs , which indicates that the rgcs may be several gyrs younger than the bgcs . the luminosity function of the globular clusters shows little systematic variation depending on the galactocentric radius at arcmin . however , the mean luminosity of the bright bgcs decreases by 0.2 mag with increasing galactocentric radius over the range of 9 arcmin , while that of the rgcs does not . the observed color distribution of these globular clusters is distinctively bimodal with peaks at and , which correspond to the metallicities of [ fe / h ] and dex , respectively . the mean observed color of all the globular clusters with mag is derived to be ( m.e . ) , corresponding to the metallicity of [ fe / h ] dex . these colors are exactly the same as those for the globular clusters in m87 . it is found that the relative number of the bgcs to the rgcs is increasing with the increasing galactocentric radius . there are more rgcs than the bgcs within arcmin , while there are fewer rgcs than the bgcs in the outer region of ngc 4472 . radial and azimuthal structures of the globular cluster system in ngc 4472 are also investigated . surface number density profiles of both the bgcs and rgcs get flat in the central region , and the core radii of the globular cluster systems are measured to be arcmin for the bgcs , arcmin for the rgcs , and arcmin for the total sample , which are much larger than the stellar core of the galaxy . in general the properties of the globular clusters in the inner region of ngc 4472 are consistent with those of the globular clusters in the outer region of ngc 4472 .

astro-ph0004116

ngc 4472 ( m49 ) is a giant elliptical galaxy located at 4@xmath22 south of the center of the virgo cluster . ngc 4472 is the brightest member of the virgo cluster , some 0.2 mag brighter than the cd galaxy m87 . ngc 4472 is an outstanding example of giant elliptical galaxies showing a bimodality in the color distribution of globular clusters ( gebhardt & kissler - patig 1999 , kundu 1999 ) . geisler , lee , & kim ( 1996 ) and lee , kim , & geisler ( 1998 ) have found from the deep wide ( @xmath23 arcmin@xmath24 ) field ccd imaging of ngc 4472 that the color distribution of the globular clusters in ngc 4472 is remarkably bimodal . the bimodal color distribution of the globular clusters in ngc 4472 has shown that there are two kinds of cluster populations in this galaxy : a metal - poor population with a mean metallicity of [ fe / h ] @xmath25 dex and a metal - rich population with a mean metallicity of [ fe / h ] @xmath26 dex . interestingly it is found that the metal - rich globular clusters show some properties in common with the galaxy halo stars in their spatial distribution and color profiles , while the metal - poor globular clusters do not show such behavior . also it is found that the metal - rich globular clusters are more centrally concentrated than the metal - poor globular clusters . this result indicates that there may exist some connection between the metal - rich globular clusters and halo stars in ngc 4472 in the formation and evolution processes . however , these studies based on the ground - based observations could not investigate the properties of the globular clusters close to the center of ngc 4472 , because of the severe crowding problem and strong background halo light in ground - based images of the inner region of the galaxy . hubble space telescope ( @xmath0 ) data are ideal for the study of the globular clusters close to the center of bright galaxies such as ngc 4472 . they are especially useful for the study of the structure of the globular cluster system in the central region of ngc 4472 for which little is known by now . in this paper we present a study of the globular clusters in the inner region of ngc 4472 using the @xmath0 archive data . during the preparation of this paper puzia et al . ( 1999 ) published a paper based on the analysis of the data which are the same as three sets among the four sets used in this study . however , there are several differences between the two studies , details of which will be described later . here is given a brief summary of the differences between the two studies . first , we use one more data set in addition to the data set used by puzia et al . , with which the size of the bright globular cluster sample is increased by 29% and the complete areal coverage of the bright globular cluster sample is increased significantly ( by a factor of three in the range of the galactocentric radius ) . secondly , we use different methods of data reduction and analysis from the puzia et al.s . as a result , our photometry goes about 1 mag deeper and is more complete than the puzia et al.s before incompleteness correction . third , @xmath4 and @xmath27 magnitudes in our photometry are on average about 0.2 mag fainter than the puzia et al.s for the common objects , while the colors are almost the same between the two studies . fourth , our results for the peak luminosity of the globular cluster luminosity functions are significantly different from the puzia et al.s , leading to an opposite conclusion on the age of the globular clusters . fifth , we present for the first time a study of the structure of the central region of the globular cluster systems which was not done by puzia et al . sixth , we use both the @xmath0 data of the inner region of ngc 4472 and the ground - based data of the outer region of ngc 4472 to investigate any systematic variation over a wide range , while puzia et al . worked with only the @xmath0 data of the inner region . this paper is composed as follows . sections 2 describes the data used in this study , and 3 explains the reduction of the data . section 4 presents the color - magnitude diagram of the measured objects in the ngc 4472 field . sections 5 and 6 show , respectively , the luminosity functions and color distribution of globular clusters . section 7 investigates the spatial structure of the globular cluster system in ngc 4472 . primary results are discussed in section 8 , and a summary of this study is given in the final section . we have used four sets of @xmath0 wide field and planetary camera 2 ( wfpc2 ) imaging data on ngc 4472 available in the @xmath0 archive for this study : two fields including the center of ngc 4472 , one field 3 arcmin north , and one field 3 arcmin south from the center of ngc 4472 . we name these @xmath0 fields : c1 field , c2 field , n field and s field . the images were obtained with the wide - band filters f555w and f814w . table 1 lists the observation log of these data and fig . 1 displays the positions and orientations of the @xmath0 fields in the digitized palomar sky survey image of ngc 4472 . areal coverage of the data used in this study extends out to @xmath28 arcmin , and is almost complete inside @xmath29 arcmin . c1 and c2 fields include the nucleus of ngc 4472 so that the background halo light is much brighter than the globular clusters in the images . therefore we need to subtract the galaxy halo light before doing photometry of the point sources in the images . we first construct smooth models of halo light using ellipse fitting and median smoothing , then subtract these models from the original images . the subtraction works very well and the resulting images show clearly the point sources close to the center of ngc 4472 . cosmic rays in the images were removed and individual images of the same filter were combined into a single image using crrej task in iraf / stsdas . crowding problem of the point sources seen in the @xmath0 images of ngc 4472 is negligible so that simple aperture photometry is good enough to derive reliable photometry of point sources . we have obtained the photometry of point sources in the images as follows . first , we find objects with 4 @xmath30 threshold from the images prepared by combining all f555w and f814w images , using find in iraf / daophot . secondly , we derive aperture magnitudes of these objects from the combined images of each filter . for this we use ccdcap designed for accurate aperture photometry of point sources in the @xmath0 images ( @xcite ) . we use the aperture radius of 2 pixels , the sky radius of 58 pixels , and the hardness parameter of 0.8 in ccdcap . thirdly , we derive the aperture corrections for the aperture radius of 0.5 arcsec using the bright point sources ( which are mostly globular clusters ) in each image , the results of which are listed in table 2 . finally we derive standard magnitudes and colors of these objects following the description given in holtzman et al . ( 1995a , b ) and kundu et al . we adopt the zero points of the magnitudes given in holtzman et al . ( 1995a , b ) . we have tried to measure the excess light of bright globular clusters compared with the stellar point spread function over the aperture region larger than the radius of 0.5 arcsec , as puzia et al . did . however , it is found that it is not possible to measure reliably the amount of the excess light , because of low signal - to - noise ratios for the outer region . the final photometry reaches @xmath31 mag , and the total number of objects in the photometry list is about 5670 . it is not easy to secure reliable photometry of the objects in the @xmath0 wfpc2 images as ours , especially due to difficulty in measuring the sky value and deriving the aperture correction . fortunately we can check the accuracy of the photometry using the objects located in the overlapped region among the @xmath0 fields . c2 field ( that was not used by puzia et al . 1999 ) is overlapped with c1 field and n field so that it is very useful for checking externally the accuracy of the photometry of c1 and n fields . table 3 and fig . 2 show the comparison of the photometry for the point sources common among the @xmath0 fields . in general the photometry of the objects show good agreement among the overlapped fields , and the mean differences of the photometry for the total sample are @xmath32 ( n = 52 ) and @xmath33 ( n = 59 ) . this result shows that the sky measurement and aperture correction in our photometry are reliable . there are included both point sources and extended sources in the list of the detected objects , and these are foreground galactic star , globular clusters in ngc 4472 , or background galaxies . here we mean galaxies by extended sources , and mean stars and globular clusters ( even though they are somewhat resolved ) by point sources . point sources are indeed mostly globular clusters rather than stars , as will be shown later . we have selected point sources from the list of the detected objects using the morphological classifier @xmath34 moment , an indicator for the degree of central concentration of light of an object ( @xcite ) . @xmath34 moment is defined as @xmath35^{1/2}$ ] . we have calculated the values of @xmath34 for @xmath36 pixels for the detected objects . the range of @xmath34 defined by the obvious bright point sources are derived to be @xmath37 , @xmath38 for the pc images , and @xmath39 , and @xmath40 for the wf images . the objects with @xmath34 in these range are classified as point sources . it turns out that most of the objects in the photometry list are extended sources . about 1560 objects among the 5670 detected objects were classified as point sources , and the rest as extended source . these point sources are considered as a final sample of the point sources for the following analysis . photometry of the bright point sources with @xmath41 mag are listed in table 4 . most of these objects are located at @xmath42 arcmin from the center of ngc 4472 , and some of them are located out to @xmath43 arcmin ( in the n field and s field ) . we have compared our photometry with the puzia et al.(1999 ) s for the globular clusters common between the two studies , as displayed in fig . 3 shows that our @xmath4 and @xmath27 magnitudes are on average fainter than the puzia et al.s , but that the colors agree very well between the two . the mean differences between this study and puzia et al . are @xmath44 mag for 157 objects with @xmath45 mag , @xmath46 mag for 176 objects with @xmath47 mag , and @xmath48 . it appears that there are two causes for the photometric differences between the two studies which used the same original @xmath0 data . first , part of the difference ( 0.1 mag ) is due to the fact that puzia et al . made a mistake in standard calibration . if the holtzman zero points that are given for the 0.5 radius arcsec aperture are used , the additional aperture correction of 0.1 mag is not needed to derive the total magnitudes ( see holtzman et al . 1995a , b ) . however , puzia et al . applied this correction , making systematically their magnitudes 0.1 mag too bright . secondly , part of the difference ( 0.040.05 mag ) may be due to different methods used in the two studies . we used ccdcap for the aperture photometry of the point sources in the images where the galaxy light is subtracted , while puzia et al . used sextractor ( bertin & arnouts 1996 ) for the original images . it is not unexpected that 0.040.05 mag difference exists between the different methods of photometry . it is also noted that the scatter in the difference in fig . 3 is larger than the scatter in the difference among the objects in the overlapped images in our photometry shown in fig . 2 , indicating that the mean errors in our photometry are smaller than the puzia et al.s . puzia et al . derived also additional aperture corrections for the 0.5 arcsec radius aperture to the infinite radius aperture for the globular clusters , using the modeled light profiles : @xmath49 mag , @xmath50 mag , and @xmath51 . since the modeled light profiles given by puzia et al . do not match well the bright globular cluster profiles at the outer radius ( as shown in their fig . 1 ) , it is difficult to estimate how reliable the above corrections are . above comparison of the photometry is based on the puzia et al.s photometry before these corrections are applied . if these corrections are included , the magnitude differences between the two studies become larger : @xmath52 mag , @xmath53 mag , and @xmath54 . we have tested completeness of our photometry using artificial cluster experiment . we have chosen as a test field a field covered by the pc chip for the c1 field . this pc field includes the nucleus of ngc 4472 in the center of the image so that incompleteness for this field is the most severe among the fields used in this study . we have derived empirical point spread function of the clusters using several isolated brightest clusters in the images and have created artificial clusters with different magnitudes by scaling down the images of the selected ones . we have created 1000 artificial clusters in 10 images for each filter . then we have repeated the same reduction procedure to estimate how many objects are recovered from the artificial images . the completeness factors and mean photometric errors obtained in this way are displayed in fig . 4 shows that our photometry is complete for @xmath55 mag , 90% complete at @xmath56 mag , and 50% complete at @xmath57 mag ( at @xmath58 mag ) for the region at @xmath59 arcsec . in the innermost region at @xmath60 arcsec , completeness of our photometry is much lower than the outer region : the completeness is 100% for @xmath61 mag , and 50% at @xmath62 mag . the photometry of the other fields must be more complete than this pc field , because the contribution due to varying galaxy light in the other field is much weaker than this pc field . from this test it is concluded that our photometry is almost complete ( @xmath63 ) for @xmath64 mag except for the innermost region at @xmath60 arcsec . we created a final sample of bright globular clusters with @xmath15 mag used for the analysis of the color of globular clusters , and this sample is considered to be complete . the number of the bright globular clusters with @xmath15 mag in the region at @xmath60 arcsec is only 11 so that it is negligible compared with the total sample of bright globular clusters used in this study . figs . 5(a ) and 5(b ) display the @xmath4@xmath65 diagrams of the measured point sources ( about 1560 objects ) and extended objects ( about 4110 objects ) , respectively . in general , basic features seen in fig . 5 are similar to those seen in the color - magnitude diagrams of ngc 4472 obtained from the ground - based observations by geisler et al . ( 1996 ) , but the @xmath0 data go much deeper than the ground - based data . 5(a ) shows a dominant vertical feature at the color range of @xmath66 , extending up to @xmath67 mag . the objects seen in this feature are mostly globular clusters in ngc 4472 . contamination due to foreground stars is considered to be negligible for this feature . we consider the point sources with @xmath68 as globular cluster candidates from now on . a small number of point sources bluer or redder than this range are probably not the members of ngc 4472 . the vertical structure in the color - magnitude diagram appears obviously to consist of two components : one relatively blue component ( @xmath69 ) and the other relatively red component ( @xmath70 ) . we call them , the blue globular clusters ( bgcs ) and red globular clusters ( rgcs ) , respectively . we create a sample of the bright globular clusters with @xmath15 mag for the detailed analysis of the globular cluster system , considering the photometric errors and the degree of non - member contamination . the boxes in fig . 5 represent the boundary of these bright globular clusters . the numbers of the blue and red globular clusters with @xmath15 mag are 263 and 346 , respectively . the relative abundance of the bgcs and rgcs derived for the inner region of ngc 4472 in this study is opposite to that based on the outer region of ngc 4472 where there are more bgcs than the rgcs ( @xcite ) . these bright globular clusters are a final sample used for the analysis of the colors of the globular clusters later . 6 displays the positions of the bgcs , rgcs and faint globular cluster candidates with @xmath71 and @xmath72 mag . the property of these two components will be discussed in detail later . on the other hand , most of the extended objects seen in fig . 5(b ) are fainter than @xmath73 mag . these are mostly faint background galaxies , as are often seen in the deep @xmath0 images such as the hubble deep field ( @xcite ) . the striking difference between the color - magnitude diagram of the point sources and that of the extended sources shows that the image classification used in this study works very well for the @xmath0 data . a sharp drop in the number of the extended objects seen at @xmath74 mag in fig . 5(b ) shows that there are few globular clusters candidates among the extended objects , while a large color scatter in the faint magnitudes seen in fig . 5(a ) indicates that there are some contamination due to galaxies in the faint point sources . we have cross - identified about 190 bright point sources with @xmath75 mag common between this study and geisler et al . 4 displays the relation between @xmath65 and @xmath76 colors of these objects . using the colors and magnitudes of these objects , we have derived transformation relations between @xmath77 system and @xmath78 system : @xmath79 for @xmath80 with @xmath81 , @xmath82 for @xmath83 with @xmath84 , and @xmath85 with @xmath86 . fig . 7 shows the color transformation relation obtained in this study in comparison with the relation derived from the photometry of the standard stars by geisler ( 1996 ) : @xmath87 and @xmath88 . the @xmath65 colors derived from the relation obtained in this study is systematically redder than those derived from geisler s relation ( by @xmath89 mag for the color of @xmath90 corresponding to the mean color of the globular clusters ) . we have used the relations derived here for combining the @xmath0 @xmath77 data in this study with the ground - based @xmath91 data ( @xcite;@xcite ) for further analysis . in this study we adopt the foreground reddening @xmath92 given by schlegel , finkbeiner , & davis ( 1998 ) , while we adopted a zero foreground reddening given by burstein & heiles ( 1982 ) in our previous studies ( @xcite ; @xcite ) . for the extinction to reddening ratio @xmath93 , corresponding values are @xmath94 , @xmath95 , and @xmath8 . we have derived the luminosity functions of the globular clusters ( gclf ) , which are listed in table 5 and displayed in fig . 8 . in fig . 8 the luminosity functions of the total sample look almost gaussian , with a peak at @xmath96 mag . we have fit a gaussian function to the data for the range @xmath97 mag , for which our photometry is almost complete . we derive the peak luminosities : @xmath4(max ) = @xmath98 mag and the width , @xmath99 , and @xmath27(max ) = @xmath100 mag and the width , @xmath101 . assuming the peak luminosity of the galactic globular clusters of @xmath9(max ) @xmath10 mag ( @xcite;@xcite ) and the foreground extinction of @xmath94 mag , we derive a value for the distance to ngc 4472 , @xmath102 and @xmath103 mpc . we have also compared the luminosity functions in this study with the puzia et al.s in fig . 9 . for this comparison we used the same fields as used by puzia et al . , c1 , n and s fields and plotted puzia et al.s luminosity functions before incompleteness correction . also we adjusted the puzia et al.s magnitudes by + 0.14 mag to match our photometry . 9 shows that the luminosity functions in the two studies agree well for @xmath104 mag , but that the luminosity function in this study is somewhat higher than puzia et al.s for @xmath105 mag , especially for the bgcs . @xmath27-band luminosity functions also show similar results , with some deviations for @xmath106 mag . this indicates that our photometry is more complete than puzia et al.s before incompleteness correction . we could not compare the luminosity functions after incompleteness correction , because the information given in puzia et al . is not enough for comparison . we have investigated whether there is any systematic difference in the peak value of the luminosity functions of the bgcs and rgcs . the @xmath4-band luminosity functions of the bgcs and rgcs are also listed in table 5 and displayed in fig . 8 shows that the shapes of the luminosity functions of the two populations are similar and that the peaks are at similar luminosity . from gaussian fitting to the data , we derive the values for the peak luminosity : @xmath4(max ) @xmath107 mag ( @xmath108 ) for the bgcs , @xmath4(max ) @xmath109 mag ( @xmath110 ) for the rgcs , @xmath27(max ) @xmath111 mag ( @xmath112 ) for the bgcs , and @xmath27(max ) @xmath113 mag ( @xmath114 ) for the rgcs . therefore the peak luminosities of the bgcs and rgcs are similar within the errors . this result is consistent with the results based on the data of the globular clusters in the outer region of ngc 4472 ( @xcite ) . the derived values of the sigma for the two populations are exactly the same for the @xmath4-band , and are slightly different for the @xmath27-band . if we constrain the sigma as the mean of the two , 1.28 , for the @xmath27-band , we get @xmath27(max ) @xmath115 mag and @xmath116 mag , respectively , for the bgcs and rgcs . this shows that the difference between the bgcs and rgcs is changed little by constraining the sigma . the color chosen as the boundary between the bgcs and rgcs , combined with observational errors in color , may affect somewhat the luminosity function that is determined for each group . however , above results are expected to be changed , if any , little , because there is a clear minimum separating between the bgcs and rgcs in the color distribution , and because the total numbers of the clusters in each population are not much different . we performed a kolmogorov - smirnov test for the samples derived using the color boundary which is 0.05 mag different from the chosen value . this value of 0.05 mag is five times larger than the mean error of the chosen value . the kolmogorov - smirnov test shows that the two populations derived using the color boundary which is 0.05 mag different from the chosen value are the same as the original samples with a probability of 80% . therefore this effect is found to affect little above results . on the other hand , puzia et al . suggested that the peak luminosities of the bgcs are 0.40.5 mag brighter than those of the rgcs : @xmath4(max ) @xmath117 mag and @xmath27(max ) @xmath118 for the bgcs , and @xmath4(max ) @xmath119 mag and @xmath27(max ) @xmath120 for the rgcs . if we adjust puzia et al s magnitudes to match our photometry as shown before , these values will be changed to @xmath4(max ) @xmath121 mag and @xmath27(max ) @xmath122 for the bgcs , and @xmath4(max ) @xmath123 mag and @xmath27(max ) @xmath124 for the rgcs . therefore the peak luminosities for the bgcs in this study are similar to the puzia et al.s , but the peak luminosities for the rgcs in this study as much as 0.8 mag brighter than the puzia et al.s . the causes for this large difference for the rgcs between the two studies are not clear . however , there are a few reasons that our results are considered to be more reliable than the puzia et al.s . first , the peak luminosity differences between the bgcs and rgcs given by puzia et al . are significantly different between the c1 field and ( n+s ) fields : @xmath125 mag and @xmath126 mag for the c1 field , and @xmath127 mag and @xmath128 mag for the ( n+s ) fields . this indicates that there may be some problems in the data even after the incompleteness correction of their sample . secondly , our sample of globular clusters is about 30% larger than puzia et al.s ( because we included the c2 field which was not used by puzia et al . ) . thirdly , our photometry is more complete than puzia et al.s . puzia et al . used further these differences in magnitudes between the bgcs and rgcs to investigate any age differences between the two components . from the comparison of the peak luminosities of the bgcs and rgcs with the simple stellar population models ( worthey 1994 ; bruzual and charlot 1996 ; maraston 1998 ; kurth et al 1999 ) puzia et al . concluded that 0.4 0.5 mag differences between the two components indicate that the bgcs and rgcs are coeval within the errors of @xmath129 gyrs . however , our results indicate that the rgcs are several gyrs younger than the bgc , if we use the same models used by puzia et al . ( their figures 9 , 10 , 11 and 12 ) . our result on the age difference is similar to the results for the globular clusters in the inner region of m87 ( kundu et al . 1999 ) . kundu et al . ( 1999 ) found from the analysis of the wfpc images of the inner region of m87 that the bgcs are 0.20.3 mag brighter than the rgcs at @xmath4-band , and that the rgcs are 0.06 mag brighter than the bgcs at @xmath27-band , and suggested that the rgcs may be 36 gyrs younger than the bgcs . globular clusters in the inner region of giant galaxies are ideal targets to study the dynamical effects on the globular cluster system due to external processes . primary external processes driving dynamical evolution of globular clusters are tidal shocks and dynamical friction . according to the theoretical prediction , the most massive clusters will be preferentially depleted by dynamical friction , while less massive clusters with lower density will be more efficiently destroyed by tidal shocks . these external dynamical effects are expected to be strongly enhanced in the inner region of the galaxies ( @xcite;@xcite;murali & weinberg 1997a , b , c ) . as results of these effects , it is expected that the globular cluster luminosity function for the inner region of a galaxy will get flatter than that for the outer region , and the peak luminosity for the inner region will get brighter than that for the outer region . however , several observational studies found little evidence supporting these theoretical predictions in the case of the globular clusters of m87 ( @xcite;@xcite ) . ngc 4472 is another good galaxy to study this aspect . we have investigated whether the globular cluster luminosity function of ngc 4472 shows any systematic variation on the galactocentric distance , using three methods : a ) the slope of the bright globular cluster luminosity function , b ) the peak luminosity of the globular cluster luminosity function , and c ) the mean luminosity of the bright globular clusters . fig . 10 displays the globular cluster luminosity function for five radial bins . we have fit the data for the bright part in the range of @xmath130 mag using a logarithmic line to check any systematic variation in the slope , the results of which are shown by the dashed line . there is seen little systematic variation depending on the galactocentric radius in the logarithmic slope of the bright part of the luminosity function . next we display the peak luminosity of the @xmath4 and @xmath27 bands globular cluster luminosity function for each radial bin in fig . 11 shows little systematic variation of the peak luminosity depending on the galactocentric radius within @xmath131 arcmin . the peak luminosity at the outermost position of @xmath132 arcmin seems to be 0.3 mag fainter than the mean value shown by the dashed line in fig . . this may be either an intrinsic property of the globular clusters or may be caused by the fact that the degree of field contamination due to background galaxies is increasing in the outer region ( see the asymmetric globular cluster luminosity function for @xmath133 in fig . 10 ) . this point needs to be checked with similar @xmath0 observations for larger areas in ngc 4472 . however , the following result indicates that the first possibility is low . we have calculated the mean luminosity of the bright globular clusters with @xmath134 mag . the magnitude range is chosen that the resulting mean luminosity is not affected at all by incompleteness in our photometry . 12 displays the mean luminosity of the globular clusters versus galactocentric radius based on the @xmath0 data , and table 6 lists the linear fits to the data . 12 shows that there is little , if any , change in the mean luminosity of the bright globular clusters for the region at @xmath1 arcmin from the center of ngc 4472 ( the determined values of the slope are smaller than the errors ) . this result is similar to that of the globular clusters in the inner region of m87 ( @xcite;@xcite ) . then we have combined these @xmath0 data with those for the outer region of ngc 4472 given by geisler et al . ( 1996 ) to investigate any gradient of the mean luminosity over a large range of radius , the results of which are illustrated in fig 13 shows that the mean luminosity of the bgcs is clearly decreasing by about 0.2 mag over the range of @xmath135 arcmin ( the slope = @xmath136 for @xmath4 and @xmath27 ) , while the mean luminosity of the rgcs is almost constant over the same range of radius . this result of the difference between the bgcs and rgcs is not affected by the errors involved with combining the @xmath0 data and the ground - based data , because it is based on the globular clusters with the same magnitude range . therefore it is concluded that there is little systematic variation of the globular cluster luminosity function within the range of @xmath137 arcmin , and that the bgcs on average become fainter by 0.2 mag with the increasing galactocentric radius over the range of 9 arcmin , while the rgcs do not . this result shows that the external dynamical effects are much weaker than the theoretical predictions for the globular cluster systems in ngc 4472 . the cause for the 0.2 mag variation of the mean luminosity seen only for the bright bgcs is unclear . if it is due to dynamical effects , it may indicate that the bgcs are older than the rgcs . we have derived a color distribution of the bright globular clusters with @xmath15 mag in ngc 4472 , which is listed in table 7 and displayed in fig . 14 . 14 shows that the color distribution of the globular clusters is clearly bimodal . from double - gaussian fitting to the data , we have derived the peak colors @xmath138 with @xmath139 , and @xmath140 with @xmath141 . these results agree very well with those given for @xmath142 mag by puzia et al . ( 1999 ) : @xmath143 and @xmath144 , respectively . using the relation between @xmath65 and metallicity of galactic globular clusters given by kundu & whitmore ( 1998 ) , [ fe / h ] @xmath145 , and adopting the reddening of @xmath8 , we derive the peak metallicities [ fe / h ] @xmath13 and 0.23 dex for the bgcs and rgcs , respectively , from the peak colors . ( if we use the slightly different relation used by harris et al . ( 1998 ) , [ fe / h ] @xmath146 , we derive [ fe / h ] @xmath147 and + 0.00 dex for the bgcs and rgcs , respectively . ) the mean color of the bright globular clusters with @xmath15 mag is derived to be @xmath16 ( m.e . ) , corresponding to the metallicity of [ fe / h ] @xmath17 dex . these results for the peak metallicity are in good agreement with the results for the globular clusters in the outer region of ngc 4472 ( @xcite ; @xcite ) . however , the relative abundance of the bgcs and rgcs is opposite between the inner region ( this study ) and the outer region ( @xcite ) . in the inner region , the rgcs are more numerous than the bgcs , while it is opposite in the outer region . 15 displays a radial variation of the color distribution of the bright globular clusters . it shows that the relative number of the bgcs to that of the rgcs is increasing as the galactocentric distance is increasing . in fig . 16 we display a radial variation of the number of the bgcs to that of the rgcs (= n(bgc)/n(rgc ) ) . we also plot the data of the globular clusters with @xmath148 mag in the outer region of ngc 4472 derived from geisler et al . 16 shows that the radial variation of the ratio is consistent between the inner region and the outer region . the ratio n(bgc)/n(rgc ) increases continuously out to the limit of the data , @xmath149 arcmin . we have investigated the spatial structure of the globular cluster system in ngc 4472 using the sample of the bright globular clusters with @xmath15 mag . without the @xmath0 data it is very difficult to study the spatial structure of the globular cluster systems in the central region of ngc 4472 . 17 displays the spatial positions of these objects in the central @xmath150 arcmin@xmath151 region of ngc 4472 . it is seen immediately in fig . 17 that the rgcs are more centrally concentrated than the bgcs . radial and azimuthal structures of the globular cluster system are studied in the following . we have derived radial profiles of the surface number density of the globular clusters with @xmath15 mag , which are listed in table 8 and are displayed in fig . 18 shows ( a ) that the surface density increase with decreasing radius , ( b ) that the surface density profiles of both the rgcs and bgcs get flat in the central region , and ( c ) that the surface density profile of the rgcs is steeper than that of the bgcs . the surface density profiles within @xmath152 arcmin are approximately fit by king models(@xcite ) : core radius @xmath21 arcmin (= 6.0 kpc ) , concentration parameter @xmath153 ( where @xmath154 is a tidal radius ) , and peak surface density @xmath155 clusters arcmin@xmath156 for the total sample , @xmath19 arcmin (= 8.7 kpc ) , @xmath157 , and @xmath158 clusters arcmin@xmath156 for the bgcs , and @xmath20 arcmin (= 5.5 kpc ) , @xmath157 , and @xmath159 clusters arcmin@xmath156 for the rgcs . the core radius of the globular cluster system derived here is much larger than that of the stellar halo , @xmath160 arcsec (= 280 pc ) given by kim , lee & geisler ( 2000 ) . in fig . 19 we display both the surface density profiles for the globular clusters in the inner region derived in this study and those for the globular clusters with @xmath161 mag in the outer region given by lee et al . the lower limit for the outer globular cluster sample , @xmath162 mag , corresponds to @xmath163 mag for the mean color of the globular clusters . we plot the surface density profile of the inner globular clusters with @xmath61 mag in fig . note that the surface density profiles of both data agree approximately for the overlapped region at @xmath164 arcmin . the surface density profiles of the combined data of the outer region at @xmath165 arcsec are well fit by de vaucouleurs law : @xmath166 for the total sample , @xmath167 for the bgcs , and @xmath168 for the rgcs , where @xmath169 is given in units of arcmin . we have derived the radial profiles of the mean and median color of the globular clusters , which are listed in table 9 and displayed in fig . we also displayed the radial profile of the color of the halo of ngc 4472 in fig . 20 to compare the globular clusters and halo of ngc 4472 . we have transformed the @xmath76 color of the halo of ngc 4472 ( kim , lee & geisler 2000 ) into the @xmath65 color using the relation derived in this study . 20 shows ( a ) that the mean color of the total sample of globular clusters decreases with increasing radius for @xmath170 arcmin , ( b ) that the mean colors of the bgcs and rgcs change little with increasing radius , and ( c ) that the profile of the rgcs is very similar to that of the halo . these results are consistent with those for the outer region given in lee et al . ( 1998 ) . fig . 21 shows the radial profiles of the metallicity of the globular clusters derived from the colors based on this study ( filled circles ) and geisler et al . ( 1996 ) ( open circles ) . the mean metallicity is increasing with decreasing galactocentric radius . however , the metallicity becomes almost constant within @xmath171 arcmin , where the surface number density of the globular clusters is almost constant as shown in fig . interestingly harris et al . ( 1998 ) also pointed out a similar trend from the ground - based data for m87 : the mean metallicity of the globular clusters in m87 is almost constant for @xmath172 arcmin . however , kundu et al . ( 1999 ) mentioned that their observations of the inner region of m87 based on the @xmath0 data do not provide any compelling supporting evidence for the claim by harris et al . ( 1998 ) . the metallicity profile for the combined sample of the inner region and outer region ( @xmath173 arcsec ) is fit by [ fe / h]@xmath174 with @xmath175 , where @xmath169 is given in terms of arcmin . this result is very similar to that derived for the outer region , shown by the dashed line with a slope of 0.41 in fig . 21 ( @xcite ) . we have investigated the azimuthal distribution of the globular clusters . 22 displays azimuthal variations of the globular cluster number density for @xmath42 arcmin where the areal coverage of the @xmath0 data is complete . it shows that the total sample has obvious peaks at a position angle of 160 and 340 deg and that the rgcs have a peak at 160 deg , while the bgcs show an almost uniform distribution . the presence of only one peak instead of two for the rgcs is due to the asymmetric distribution of the globular clusters extended along the south - east direction , as shown in fig . in this study we have investigated several aspects of the globular clusters in the inner region of ngc 4472 : color - magnitude diagram , luminosity function , color distribution , and spatial structure of the globular cluster system . @xmath0 data enable us to study the globular clusters located close to the nucleus of ngc 4472 , which was difficult in the case of ground - based data . we have compared these results to those of the globular clusters in the outer region of ngc 4472 ( @xcite;@xcite ) , finding that the properties of the globular clusters in the inner region and the outer region of ngc 4472 are in general similar . the only exception is that there are more rgcs than the bgcs in the inner region , while it is opposite in the outer region . however , this is also expected from the relation of n(bgc)/n(rgc ) and galactocentric distance derived from the data of the outer region . implications of the results based on the globular clusters in the outer region of ngc 4472 for the origin of the globular clusters were discussed in detail in lee et al . ( 1998 ) , which remain still valid with the new results for the globular clusters in the inner region of ngc 4472 . in summary , the observed properties are consistent with many of the predictions of both the model of episodic in situ formation plus tidal stripping of globular clusters given by forbes , brodie & grillmair ( 1997 ) , and the gaseous merger model given by ashman & zepf ( 1992 ) , but each of the models also has some problems . ngc 4472 and m87 ( ngc 4486 ) are the two most dominant galaxies in the virgo cluster . ngc 4472 is the brightest member of the virgo cluster , but not a cd galaxy , while m87 is a cd galaxy at the center of the virgo cluster and is only slightly fainter than ngc 4472 . m87 has been a target of much more studies than ngc 4472 . recently kundu et al . ( 1999 ) presented a study of 1057 globular clusters in the inner region ( @xmath176 arcmin ) based on the @xmath0 wfpc2 data of a field centered on the nucleus of m87 . the areal coverage of their data is similar to that of c1 and c2 fields in our study . kundu et al . adopted foreground extinctions of @xmath177 and @xmath178 in their study of m87 , which are the same adopted for ngc 4472 in this study . we have compared the properties of the globular clusters in the inner region of these two galaxies based on the similar data ( this study and @xcite ) . the luminosity function of the m87 globular clusters has a peak at @xmath179(max ) @xmath180 mag with @xmath181 , which are slightly larger than that of ngc 4472 , @xmath179(max ) @xmath182 mag with @xmath6 . thus ngc 4472 is considered to be slightly closer than m87 . the luminosity function of the m87 globular clusters does not show any significant spatial variation over the radius of 1.5 arcmin , similarly to the case of ngc 4472 . the color distribution of the globular clusters in m87 is also bimodal . the peak colors are @xmath183 and @xmath184 and the mean color of the total sample is @xmath185 . these values are essentially the same as those of the globular clusters in ngc 4472 , @xmath183 , 1.20 and 1.08 . thus the characteristic colors of the globular clusters in ngc 4472 and m87 are almost the same . the surface number density profiles of the globular clusters are flat in the central regions of both m87 and ngc 4472 . the core radii of the globular clusters in the inner region of m87 and the stellar halo are , respectively , 56 arcsec and 6.8 arcsec ( kundu et al . 1999 ) . therefore the core of the globular cluster system in m87 is somewhat smaller than that of ngc 4472 , 78 arcsec , while the core of the stellar halo in m87 is twice larger than that of ngc 4472 , 3.6 arcsec . the relative number of the bgcs to that of the rgcs in m87 is also increasing with increasing galactocentric radius , as in the case of ngc 4472 . the absence of the spatial variation of the globular cluster luminosity function and the core radius of the globular cluster systems larger than the stellar core of the galaxies are considered to support the suggestion that the large core of the globular cluster systems is a relic of the cluster formation process rather than a result due to destruction through tidal shocking and dynamical friction ( grillmair , pritchet , & van den bergh 1986 , harris et al . 1998 ) . however , the surface central number density of the globular clusters is about six times larger in m87 ( 460 clusters arcmin@xmath156 ) than in ngc 4472 ( 81 clusters arcmin@xmath156 ) . this indicates that the conditions of formation process of globular clusters might have been different between m87 and ngc 4472 . in summary , we conclude that the properties of the globular clusters in the inner regions of ngc 4472 and m87 are very similar in general . the only significant difference is that the surface central number density of the globular clusters ( and the specific frequency of the globular clusters correspondingly ) is much higher in m87 than in ngc 4472 . we have presented @xmath77 photometry of the globular cluster candidates in the inner ( @xmath1 arcmin ) region of ngc 4472 , derived from the @xmath0 wfpc2 archive data for the four fields of ngc 4472 . we have found about 1560 globular cluster candidates from the colors and image classifiers . primary results and conclusions obtained in this study are summarized as follows . 1 . the color - magnitude diagram of the measured point sources shows a dominant vertical structure at the color range of @xmath71 , most of which are globular clusters in ngc 4472 . this structure consists obviously of two populations : the blue globular clusters and the red globular clusters . the luminosity function of the globular clusters is well fit by a gaussian function with a peak at @xmath186 mag and @xmath99 . from this the distance to ngc 4472 is derived to be @xmath7 mpc . the globular cluster luminosity function shows little systematic variation depending on the galactocentric radius . there is little difference in the peak luminosity between the bgcs and rgcs . this indicates that the rgcs may be several gyrs younger than the bgcs , contrary to the conclusion given by puzia et al . the color distribution of the 609 bright globular clusters with @xmath15 mag is distinctively bimodal with peaks at @xmath187 and @xmath188 . the metallicities corresponding to these peaks are [ fe / h ] @xmath13 and @xmath14 dex . the mean color of the bright globular clusters with @xmath15 mag is derived to be @xmath16(m.e . ) , corresponding to the metallicity of [ fe / h ] @xmath17 dex . 4 . the ratio of the number of the bgcs to that of the rgcs is increasing with increasing galactocentric radius from the center out to about 9 arcmin . thus the rgcs are more centrally concentrated than the bgcs . the surface number density profile of the globular clusters in the inner region is approximately fit by a king model with core radius @xmath21 arcmin , while that of the globular clusters in the outer region is better fit by de vaucouleurs law . the core radius of the globular cluster system obtained in this study is much larger than that of the stellar halo . the mean colors of the bgcs and rgcs change little with increasing radius , and the mean color of the rgcs is very similar to that of the stellar halo of ngc 4472 . the radial gradient of the metallicity of the entire sample of the globular clusters with @xmath15 mag is approximately fit by [ fe / h ] @xmath189 @xmath190 with @xmath175 , where @xmath169 is given in the unit of arcmin . 8 . in general the property of the globular clusters in the inner region of ngc 4472 is consistent with that of the globular clusters in the outer region of ngc 4472 , except that there are more rgcs than the bgcs in the inner region . the authors are grateful to sang chul kim for careful reading of the manuscript of this paper and to thomas puzia for providing the photometry data of the globular clusters of ngc 4472 used in puzia et al . ( 1999 ) and useful information . anonymous referee is thanked for useful comments . this research is supported in part by the kosef / kistep international collaboration research program ( 1 - 99 - 009 ) . ashman , k.m . , & zepf , s. e. 1992 , apj , 384 , 50 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 bruzual , a. g. , & charlot , s. 1996 , electronically available . see leitherer , c. , et al . 1996 , pasp , 108 , 996 murali , c. , & weinberg , m. d. 1997a , mnras , 288 , 749 murali , c. , & weinberg , m. d. 1997b , mnras , 288 , 767 murali , c. , & weinberg , m. d. 1997c , mnras , 291 , 717 ostriker , j. p. , & gnedin , o. y. 1997 , apj , 487 , 667 puzia , t. h. , kissler - patig , m. , brodie , j. p. , & huchra , j. p. 1999 , aj , 118 , 2734 cccccc c1 field & @xmath191 s & @xmath191 s & 1995 feb . 4,5 & go.5236 westphal c2 field & @xmath192 s & @xmath192 s & 1995 apr . 18 & go.5236 westphal n field & 900 s , 1300 s & 900 s , 1400 s & 1996 may 3 & go.5920 brodie s field & 900 s , 1300 s & 900 s , 1400 s & 1996 may 2 & go.5920 brodie ccc c1 field & & pc & @xmath193 & @xmath194 wf2 & @xmath195 & @xmath196 wf3 & @xmath197 & @xmath198 wf4 & @xmath199 & @xmath200 c2 field & & pc & @xmath201 & @xmath202 wf2 & @xmath203 & @xmath204 wf3 & @xmath205 & @xmath206 wf4 & @xmath207 & @xmath208 n field & & pc & @xmath209 & @xmath210 wf2 & @xmath211 & @xmath212 wf3 & @xmath213 & @xmath214 wf4 & @xmath215 & @xmath216 s field & & pc & @xmath217 & @xmath218 wf2 & @xmath219 & @xmath220 wf3 & @xmath221 & @xmath222 wf4 & @xmath223 & @xmath224 ccccc c1 field(pc ) minus c2 field(pc ) & @xmath225 & 10 & @xmath226 & 10 c1 field(wf2 ) minus c2 field(wf4 ) & @xmath227 & 28 & @xmath228 & 32 c1 field(wf4 ) minus s field(wf4 ) & @xmath229 & 5 & @xmath230 & 5c2 field(wf3 ) minus n field(wf2 ) & @xmath231 & 10 & @xmath232 & 11total objects & @xmath233 & 52 & @xmath234 & 59 cccccc | cccccc 1061 & -91.3 & -30.6 & 21.89 & 1.32 & 96.2 & 90 & 8.3 & -109.6 & 21.58 & 0.89 & 109.9 617 & -86.4 & 14.6 & 21.70 & 1.25 & 87.7 & 1246 & 12.0 & -44.8 & 21.65 & 1.20 & 46.4 911 & -78.6 & -12.8 & 21.26 & 1.21 & 79.6 & 39 & 14.6 & -138.1 & 21.69 & 1.33 & 138.9 741 & -73.8 & -1.8 & 20.45 & 1.24 & 73.8 & 23 & 19.4 & -122.4 & 21.07 & 0.85 & 124.0 236 & -69.3 & 21.0 & 21.87 & 1.16 & 72.4 & 179 & 19.6 & -218.8 & 21.43 & 1.05 & 219.7 1062 & -68.0 & -44.6 & 21.83 & 1.18 & 81.3 & 371 & 20.6 & -25.1 & 21.45 & 1.26 & 32.5 398 & -67.0 & 12.8 & 21.82 & 1.07 & 68.2 & 514 & 25.6 & 65.0 & 19.82 & 1.01 & 69.9 417 & -64.8 & 10.8 & 21.22 & 0.99 & 65.6 & 238 & 25.7 & 130.9 & 21.64 & 1.15 & 133.4 833 & -64.5 & -14.7 & 20.90 & 1.31 & 66.2 & 584 & 26.0 & -19.5 & 21.62 & 0.93 & 32.5 1129 & -64.1 & -63.7 & 21.57 & 0.93 & 90.4 & 295 & 28.9 & 110.6 & 20.35 & 1.12 & 114.3 705 & -60.2 & -7.2 & 21.85 & 0.99 & 60.6 & 542 & 29.2 & 211.9 & 21.49 & 1.29 & 213.9 972 & -55.9 & -32.7 & 21.49 & 0.97 & 64.8 & 263 & 31.1 & -193.2 & 21.54 & 1.25 & 195.7 233 & -54.1 & 12.0 & 21.30 & 1.27 & 55.4 & 723 & 32.7 & -26.2 & 21.25 & 1.04 & 41.9 925 & -51.7 & -30.5 & 20.40 & 1.12 & 60.0 & 243 & 33.7 & 28.1 & 20.78 & 0.96 & 43.8 772 & -50.0 & -18.1 & 21.47 & 1.11 & 53.1 & 541 & 34.3 & 75.8 & 21.54 & 1.00 & 83.2 749 & -47.6 & 202.2 & 21.62 & 1.17 & 207.7 & 512 & 36.2 & 63.6 & 21.77 & 1.08 & 73.2 1153 & -46.3 & 68.3 & 21.64 & 1.15 & 82.5 & 388 & 38.3 & 40.4 & 20.93 & 1.27 & 55.6 1084 & -42.4 & 37.5 & 21.77 & 0.98 & 56.6 & 305 & 38.7 & -177.3 & 21.64 & 0.92 & 181.5 256 & -41.2 & -36.6 & 21.69 & 1.30 & 55.1 & 417 & 40.7 & 95.1 & 21.60 & 1.09 & 103.5 314 & -39.3 & -0.1 & 20.03 & 1.19 & 39.3 & 327 & 42.8 & -124.8 & 20.74 & 1.25 & 131.9 808 & -38.8 & -27.8 & 21.76 & 0.92 & 47.8 & 824 & 43.8 & -4.8 & 21.21 & 1.12 & 44.1 648 & -38.8 & -15.6 & 21.51 & 0.95 & 41.8 & 379 & 44.4 & 39.4 & 21.26 & 1.32 & 59.3 480 & -38.0 & -128.3 & 21.07 & 0.98 & 133.8 & 503 & 45.5 & 61.5 & 21.43 & 0.95 & 76.5 1104 & -36.2 & 54.0 & 21.43 & 1.28 & 65.0 & 549 & 48.4 & 79.5 & 21.76 & 0.97 & 93.1 617 & -35.0 & 188.5 & 19.39 & 1.05 & 191.7 & 407 & 50.3 & -179.5 & 21.85 & 0.96 & 186.4 450 & -34.4 & -8.7 & 21.78 & 1.20 & 35.4 & 432 & 51.0 & 140.6 & 20.74 & 0.95 & 149.6 347 & -33.8 & -4.7 & 20.92 & 0.93 & 34.1 & 99 & 52.1 & 20.0 & 21.40 & 0.99 & 55.8 402 & -26.6 & -106.2 & 21.68 & 1.25 & 109.5 & 368 & 55.2 & 38.1 & 21.16 & 1.03 & 67.1 467 & -19.5 & -20.3 & 20.31 & 1.13 & 28.2 & 259 & 61.4 & 159.9 & 21.88 & 1.21 & 171.3 322 & -18.7 & 8.1 & 21.35 & 1.15 & 20.4 & 872 & 61.8 & -22.7 & 21.52 & 1.14 & 65.8 1198 & -18.4 & -86.6 & 21.40 & 1.18 & 88.5 & 556 & 64.5 & 136.9 & 21.56 & 0.95 & 151.4 846 & -16.5 & -42.6 & 21.44 & 1.22 & 45.7 & 881 & 65.9 & 6.4 & 20.63 & 1.27 & 66.2 317 & -16.2 & -103.2 & 21.54 & 0.97 & 104.4 & 270 & 68.0 & 29.7 & 21.58 & 1.16 & 74.2 179 & -15.6 & 9.5 & 20.25 & 1.19 & 18.2 & 142 & 69.2 & 21.8 & 20.59 & 0.97 & 72.6 441 & -13.1 & -170.3 & 21.52 & 1.14 & 170.8 & 554 & 69.9 & 81.4 & 21.70 & 0.99 & 107.3 766 & -12.5 & -30.1 & 21.87 & 1.08 & 32.6 & 427 & 71.0 & 43.9 & 20.48 & 1.11 & 83.5 905 & -8.0 & 55.3 & 21.23 & 1.09 & 55.9 & 312 & 74.2 & 164.1 & 20.71 & 1.05 & 180.1 683 & -7.6 & -16.5 & 21.34 & 1.31 & 18.1 & 901 & 74.4 & -43.8 & 21.78 & 0.91 & 86.3 60 & -7.5 & -7.9 & 21.13 & 1.26 & 10.9 & 902 & 75.9 & 0.3 & 21.08 & 1.36 & 75.9 344 & -6.8 & 190.0 & 21.89 & 1.01 & 190.1 & 568 & 76.1 & 89.9 & 21.40 & 0.94 & 117.8 1289 & -5.6 & -87.7 & 21.85 & 1.37 & 87.9 & 419 & 76.5 & 43.0 & 21.68 & 1.28 & 87.7 1257 & -3.9 & -74.8 & 21.88 & 1.25 & 74.9 & 555 & 79.5 & 206.6 & 21.86 & 1.12 & 221.3 413 & 0.2 & 22.9 & 21.86 & 1.26 & 22.9 & 389 & 79.8 & 39.9 & 21.68 & 1.04 & 89.2 1199 & 0.7 & -54.3 & 20.99 & 1.22 & 54.3 & 507 & 82.6 & -110.7 & 21.88 & 1.22 & 138.1 339 & 3.1 & 31.6 & 21.46 & 1.30 & 31.8 & 911 & 83.9 & -11.7 & 21.83 & 0.98 & 84.7 1366 & 4.3 & -103.6 & 19.50 & 1.08 & 103.6 & 284 & 85.4 & -137.2 & 21.82 & 1.32 & 161.6 292 & 4.6 & 32.3 & 21.82 & 1.27 & 32.6 & 499 & 85.5 & 191.5 & 21.48 & 1.02 & 209.8 79 & 6.9 & -12.9 & 21.46 & 1.16 & 14.6 & 523 & 86.5 & -109.2 & 20.23 & 1.02 & 139.3 1182 & 7.4 & -39.4 & 21.43 & 0.92 & 40.1 & 265 & 88.7 & 155.9 & 20.79 & 1.08 & 179.3 ccccc|cccc 19.3 & 1 & 1 & 0 & & 23.7 & 54 & 28 & 26 19.5 & 1 & 0 & 1 & & 23.9 & 52 & 27 & 25 19.7 & 0 & 0 & 0 & & 24.1 & 57 & 28 & 29 19.9 & 1 & 1 & 0 & & 24.3 & 51 & 19 & 32 20.1 & 1 & 0 & 1 & & 24.5 & 25 & 12 & 13 20.3 & 5 & 1 & 4 & & 24.7 & 42 & 19 & 23 20.5 & 3 & 1 & 2 & & 24.9 & 32 & 18 & 14 20.7 & 6 & 3 & 3 & & 25.1 & 41 & 13 & 28 20.9 & 4 & 1 & 3 & & 25.3 & 25 & 8 & 1721.1 & 5 & 3 & 2 & & 25.5 & 24 & 10 & 14 21.3 & 10 & 3 & 7 & & 25.7 & 19 & 8 & 1121.5 & 24 & 12 & 12 & & 25.9 & 14 & 7 & 7 21.7 & 20 & 8 & 12 & & 26.1 & 20 & 10 & 10 21.9 & 35 & 13 & 22 & & 26.3 & 18 & 8 & 1022.1 & 33 & 17 & 16 & & 26.5 & 19 & 12 & 722.3 & 34 & 16 & 18 & & 26.7 & 16 & 7 & 922.5 & 51 & 27 & 24 & & 26.9 & 19 & 8 & 11 22.7 & 37 & 13 & 24 & & 27.1 & 23 & 9 & 14 22.9 & 59 & 24 & 35 & & 27.3 & 10 & 4 & 6 23.1 & 56 & 23 & 33 & & 27.5 & 8 & 6&2 23.3 & 65 & 27 & 38 & & 27.7 & 8 & 5 & 3 23.5 & 72 & 25 & 47 & & 27.9 & 1 & 0 & 1 cccccccc total & @xmath4 & @xmath235 & @xmath236 & 0.091 & @xmath237 & @xmath238 & 0.052 & @xmath27 & @xmath239 & @xmath240 & 0.098 & @xmath241 & @xmath242 & 0.058 bgc & @xmath4 & @xmath243 & @xmath244 & 0.083 & @xmath245 & @xmath246 & 0.052 & @xmath27 & @xmath247 & @xmath248 & 0.094 & @xmath245 & @xmath249 & 0.054 rgc & @xmath4 & @xmath250 & @xmath251 & 0.063 & @xmath252 & @xmath253 & 0.040 & @xmath27 & @xmath254 & @xmath255 & 0.065 & @xmath256 & @xmath257 & 0.043 ccc|cc 0.625 & 1 & & 1.125 & 68 0.675 & 5 & & 1.175 & 58 0.725 & 4 & & 1.225 & 68 0.775 & 4 & & 1.275 & 65 0.825 & 11 & & 1.325 & 53 0.875 & 24 & & 1.375 & 15 0.925 & 65 & & 1.425 & 8 0.975 & 74 & & 1.475 & 1 1.025 & 59 & & 1.525 & 1 1.075 & 53 & & 1.575 & 0 0.29 & @xmath258 & 0.33 & @xmath259 & 0.32 & @xmath260 0.48 & @xmath261 & 0.54 & @xmath262 & 0.55 & @xmath263 0.86 & @xmath264 & 0.96 & @xmath265 & 0.82 & @xmath266 1.18 & @xmath267 & 1.31 & @xmath268 & 1.13 & @xmath269 1.55 & @xmath270 & 1.72 & @xmath271 2.38 & 1.46 & @xmath272 2.17 & @xmath273 & 2.36 & @xmath274 & 2.04 & @xmath275 3.26 & @xmath276 & 3.33 & @xmath277 & 3.16 & @xmath278 ccc | ccc | ccc 0.413 & @xmath279 & 1.145 & 0.454 & @xmath280 & 0.977 & 0.385 & @xmath281 & 1.253 0.732 & @xmath282 & 1.155 & 0.786 & @xmath283 & 0.976 & 0.701 & @xmath284 & 1.232 0.991 & @xmath285 & 1.139 & 1.048 & @xmath286 & 0.985 & 0.955 & @xmath287 & 1.230 1.227 & @xmath288 & 1.121 & 1.304 & @xmath289 & 0.967 & 1.176 & @xmath290 & 1.240 1.479 & @xmath291 & 1.125 & 1.604 & @xmath292 & 0.951 & 1.415 & @xmath293 & 1.208 1.852 & @xmath294 & 1.121 & 2.043 & @xmath295 & 0.954 & 1.762 & @xmath296 & 1.235 2.477 & @xmath297 & 1.068 & 2.678 & @xmath298 & 0.948 & 2.355 & @xmath299 & 1.250 3.381 & @xmath300 & 1.064 & 3.462 & @xmath301 & 0.990 & 3.286 & @xmath302 & 1.196

in this chapter the physical aspects of the global optimization of the geometry of atomic clusters are elucidated . in particular , i examine the structural principles that determine the nature of the lowest - energy structure , the physical reasons why some clusters are especially difficult to optimize and how the basin - hopping transformation of the potential energy surface enables these difficult clusters to be optimized . atomic clusters , basin - hopping , multiple funnels

cond-mat0007338

global optimization ( go ) is essentially a mathematical task . namely , for the class of go problems i will be particularly considering here , it is to find the absolute minimum ( or maximum ) of a cost function , @xmath0 , where @xmath1 belongs to a subset @xmath2 of euclidean n - space , @xmath3 @xcite , i.e.@xmath4 although the applications of global optimization span a wide range of fields from the economics of business in the travelling salesman problem to biophysics in the lowest - energy structure of a protein this does not take away from the essentially mathematical nature of the optimization problem . so why do i wish to discuss the _ physical _ aspects of global optimization ? to begin with we should realize that even for go problems that do not correspond to a physical system , physical properties can be associated with the system by thinking of the cost function as a potential energy function , @xmath5 . this allows the thermodynamics of the system to be defined . when the system is at equilibrium at a temperature @xmath6 each point @xmath1 in configuration space will be sampled with a probability proportional to its boltzmann weight , @xmath7 , where @xmath8 is the boltzmann constant . furthermore , for systems with continuous coordinates the forces , @xmath9 , associated with each coordinate can be obtained from the gradient of the cost function , i.e. @xmath10 . once masses are associated with each coordinate , the dynamics are then defined through newton s equations of motion . if one wishes the system s dynamics can then be simulated by integrating these equations of motion , as in the molecular dynamics method @xcite . even when the coordinates can only take discrete values , monte carlo ( mc ) simulations can still provide a pseudo - dynamics with the number of steps taking the role of time . of course , this connection to physics is most transparent , and most natural , when the system being optimized is a physical system , which has a real ( and potentially observable ) thermodynamics and dynamics . furthermore , in those cases where the cost function does truly correspond to the potential energy of the system , there is another physical dimension to the problem how is the structure of the global minimum determined by the physical interactions between the atoms and molecules that make up @xmath11 ? given that we have established that physical properties can be associated with any system being optimized , what relevance does this physics have to the task of global optimization ? firstly , many go algorithm have drawn their inspiration from physics . most famously , simulated annealing is analogous to the slow cooling of a melt to allow the formation of a near perfect crystal , the idea being that if equilibrium is maintained in the simulation as the system is cooled , then at zero temperature it must end up in the global minimum @xcite . there are many other physically - motivated go approaches . the extension of statistical thermodynamics to systems with non - extensive thermodynamics through the use of tsallis statistics @xcite has led to a generalized simulated annealing @xcite which is no longer tied to the boltzmann distribution and is often more efficient than standard simulated annealing . genetic algorithms imitate the biophysical evolution of the genome @xcite . and i could go on . however , this is not the link between physics and global optimization that is my focus here . rather , i wish to show how the ease or difficulty of global optimization is often intimately linked to the physics of the system . the insights obtained from understanding the physical basis for the success or failure of an algorithm not only provide an understanding of the limitations of the method and a basis for assessing the likelihood of success in future applications , but also aid the development of new algorithms by identifying the main physical challenges that need to be overcome to enable greater efficiency and suggesting the type of physical behaviour that would need to be incorporated into an improved algorithm . i will attempt to achieve this aim by concentrating on one class of problems , namely the global minimization of the potential energy of an atomic cluster . furthermore , i will mainly concentrate on model systems where the cost function is computationally cheap to evaluate , enabling the physical properties of these systems to be comprehensively examined and understood . as outlined by hartke elsewhere in this book @xcite , this class of problems is of great general interest to the chemical physics community , because the identification of a cluster s structure is often a prerequisite for understanding its other physical and chemical properties . in this chapter i start at the ` end ' , first showing the structures of the putative global minima for a number of cluster systems in order that the reader can understand some of the physical principles that determine the structure and how these relate to the interatomic interactions . furthermore , the structure provides a basis for understanding a cluster s thermodynamic and dynamic properties , especially when , as in some of our examples , the competition between different structural types plays an important role . i then consider some of the go algorithms that are most successful for these systems focussing on those that use the basin - hopping transformation of @xmath11 @xcite and on how the performance of these algorithms depend on the system and the cluster size . i then look at the physical properties of some of the clusters , relating these back to the ease or difficulty of global optimization . i firstly examine the topography of the multi - dimensional surface defined by @xmath11 ( the so - called potential energy surface ( pes ) or energy landscape ) , then the thermodynamics and dynamics . finally , i show why basin - hopping is able to locate the global minimum in those clusters where the pes has a multiple - funnel topography , and make some suggestions as to how further gains in efficiency might be secured . in this section i mainly concentrate on the structures of model clusters , where the interactions have simple idealized forms that are isotropic , thus favouring compact geometries . the models have been chosen so that they span a wide range of structural behaviour that is likely to be relevant to rare gas , metal and molecular clusters bound by dispersion forces , but not to clusters with directional covalent bonding or molecular clusters with directional intermolecular forces , as with the hydrogen bonding in water clusters . three examples of the structures clusters can adopt : ( a ) a 38-atom truncated octahedron , ( b ) a 55-atom mackay icosahedron , and ( c ) a 75-atom marks decahedron . these clusters have the optimal shape for the three main types of regular packing seen in clusters : close - packed , icosahedral and decahedral , respectively . the mackay icosahedron is a common structure and is observed for rare gas @xcite and many metal @xcite clusters . the truncated octahedron has been recently observed for nickel @xcite and gold @xcite clusters , and the marks decahedron for gold clusters @xcite.,width=340 ] -0.85 cm -0.3 cm most of the clusters we consider have only pair interactions , i.e. @xmath12 where @xmath13 is the pair potential and @xmath14 is the distance between atoms @xmath15 and @xmath16 . in this case we can partition the energy into three terms @xcite : @xmath17 where @xmath18 is the pair well depth , @xmath19 is the number of nearest neighbours , @xmath20 and @xmath21 is the equilibrium pair distance . two atoms are defined as nearest neighbours if @xmath22 . @xmath23 should lie between the first and second coordianation shells and a typical value would be @xmath24 , although the exact value is somewhat arbitrary . the first term in equation [ eq : esplit ] is the ideal pair energy if all @xmath19 nearest - neighbour pairs lie exactly at the equilibrium pair distance , the strain energy is the energetic penalty for the deviation of nearest - neighbour distances from the equilibrium pair distance and @xmath25 is the contribution to the energy from non - nearest neighbours . @xmath25 is usually smaller than the other two terms and is relatively independent of the detailed structure . therefore , the global minimum usually represents the best balance between maximizing @xmath19 and minimizing @xmath26 . for an atom in the interior of a cluster this is usually achieved through the atom having a coordination number of twelve . this can be achieved as in close - packing , but another possibility is an icosahedral coordination shell . @xmath19 is further increased through the cluster having a compact spherical shape , and through the surface mainly consisting of faces with a high co - ordination number . for example , an atom on a face - centred - cubic ( fcc ) @xmath27 face has nine nearest neighbours , whereas an atom on a @xmath28 face has eight nearest neighbours . planes corresponds to the close - packed planes , where the atoms in the planes form a grid of equilateral triangles . the atoms in the @xmath28 planes form a square grid . ] examples of the strain involved in packing tetrahedra . ( a ) five regular tetrahedra around a common edge produce a gap of @xmath29 . ( b ) twenty regular tetrahedra about a common vertex produce gaps equivalent to a solid angle of 1.54 steradians.,width=264 ] -1.0 cm -0.3 cm the three main types of cluster structure found for systems with isotropic interactions , namely icosahedral , decahedral and close - packed structures , are depicted in figure [ fig : structure_main ] . these examples have the optimal shape for each structural type , and all have been identified experimentally . one of the unusual properties of clusters is that they can exhibit non - crystallographic symmetries , because there is no requirement for translational periodicity . decahedral clusters have a single five - fold axis and are based on a pentagonal bipyramid that can be thought of as five strained fcc tetrahedra sharing a common edge . the symmetry axis corresponds to this common edge . the marks decahedron @xcite , which represents the optimal shape for this structural type , can be formed from a pentagonal bipyramid by exposing @xmath28 faces at the equatorial edges then introducing reentrant @xmath27 faces . mackay icosahedra @xcite have six five - fold axes of symmetry and can be thought of as twenty strained fcc tetrahedra sharing a common edge . the fcc cluster represented in figure [ fig : structure_main]a is simply a fragment of the bulk fcc lattice . icosahedral structures generally have the largest @xmath19 because of their spherical shape and @xmath27 faces , and close - packed clusters the smallest @xmath19 because of their higher proportion of @xmath28 faces . by contrast , close - packed clusters can be unstrained , whereas , as figure [ fig : gaps ] illustrates , decahedra and icosahedra are increasingly strained . the strain energy is proportional to the volume of the cluster , but differences in @xmath19 are due to surface effects . therefore , icosahedra are likely to be found at small sizes , but at sufficiently large size , the cluster must take on the bulk structure . at intermediate sizes decahedra can be most stable . the sizes at which the crossovers between structural types occur is system dependent . the structures illustrated in figure [ fig : structure_main ] involve exactly the right number of atoms to form a cluster of the optimal shape . for example , complete mackay icosahedra can be formed at @xmath30 at intermediate sizes clusters have an incomplete surface layer . marks decahedra with square @xmath28 faces occur at @xmath31 . other complete marks decahedra that are less spherical can be found in between these sizes , e.g. at @xmath32=101 and 146 . fcc truncated octahedra with regular hexagonal @xmath27 faces can be found at @xmath33 , and other less spherical truncated octahedra can be found , for example , at @xmath32=79 , 116 , 140 @xcite . furthermore , because the energy of a twin plane is often small , close - packed structures with other forms can also be particularly stable . four examples are given in figure [ fig : cp ] for @xmath34 . the 26-atom structure has a hexagonal close - packed ( hcp ) structure ; the 50-atom structure consists of two fragments of the 38-atom structure joined at a twin plane ; the 59-atom structure consists of a 31-atom fcc truncated tetrahedron with each face covered by a 7-atom hexagonal overlayer that occupies the hcp surface sites ; and the 79-atom structure , which is similar to the 50-atom structure , is formed by the introduction of a twin plane into the 79-atom truncated octahedron @xcite . four examples of stable close - packed structures for @xmath35 . the sizes are as labelled . the 59-atoms structure has @xmath36 point group symmetry , and the rest @xmath37.,width=415 ] -0.8 cm -0.3 cm recently , a new structural type called a leary tetrahedron @xcite has been discovered . an example with 98 atoms is illustrated in figure [ fig : leary_td ] . at the centre of this structure is an fcc tetrahedron . to each of the faces of this tetrahedron , further fcc tetrahedra ( minus an apical atom ) are added , to form a stellated tetrahedron . finally , the edges of the original tetrahedron are covered by 7-atom hexagonal overlayers . the coordination along the edges of the central tetrahedron is the same as along the symmetry axis of the decahedron and so the strain energy of this structure is intermediate between icosahedra and decahedra . it is not yet clear how general this class of structures is . the 98-atom example is the global minimum for a model potential @xcite and mass spectroscopic studies of clusters of c@xmath38 molecules suggest that ( c@xmath38)@xmath39 has this structure @xcite . however , it may be that the stability of this structural class is restricted to @xmath32=98 , because this size results in a particularly spherical shape , and that equivalent structures at larger sizes ( e.g. @xmath32=159 , 195 ) are never competitive . in this section i focus on clusters bound by the lennard - jones ( lj ) potential @xcite : @xmath40,\ ] ] where @xmath18 is the pair well depth and @xmath41 is the equilibrium pair separation . the potential is illustrated in figure [ fig : potentials]a , and provides a reasonable description of the interatomic interactions of rare gases , such as argon . lj clusters have become probably the most common test system for go algorithms for configurational problems . the number of papers with applications to this model system is now very large , but unfortunately many are distinctly unimpressive , only reporting results for small sizes or failing for relatively simple cases . i do not attempt to review this literature , but instead refer the interested reader elsewhere @xcite . front and back views of the 98-atom leary tetrahedron.,width=226 ] -0.9 cm -0.3 cm at small sizes the lj potential is able to accommodate the strain associated with icosahedral packing relatively easily @xcite . indeed , only for @xmath42 are the majority of global minimum expected to be decahedral and the crossover to fcc clusters has been estimated to occur at @xmath43 @xcite . this preference for icosahedral packing is also evident from figure [ fig : lj_evn ] where i compare the energies of icosahedral , decahedral and close - packed clusters . sizes where complete mackay icosahedra are possible ( @xmath32=13 , 55 ) stand out as particularly stable . the icosahedra are least stable when the overlayer is roughly half - filled . therefore , when especially stable non - icosahedral clusters coincide with these sizes there is a possibility that the global minimum will be non - icosahedral . there are eight such cases for @xmath44 . at @xmath32=38 the global minimum is the fcc truncated octahedron @xcite ; at @xmath32=7577 @xcite and 102104 @xcite the global minima are marks decahedra ; and at @xmath32=98 the global minimum is a leary tetrahedron @xcite . at these sizes the lines for the decahedral or close - packed structures in figure [ fig : lj_evn ] dip just below the line for the icosahedra . for @xmath45 there are a further eight non - icosahedral global minima @xcite , all of which are decahedral and which divide into two sets that are based on the complete marks decahedra possible at n=192 and 238 . ( a ) a comparison of the lennard - jones ( lj ) potential with the dzugutov potential ( dz1 ) and a modified version of it ( dz2 ) . ( b ) the morse potential for several values of the range parameter , @xmath46 . , width=415 ] -0.5 cm -0.3 cm in this section i focus on clusters bound by the the morse potential @xcite : @xmath47 where @xmath18 is the pair well depth and @xmath21 is the equilibrium pair separation . in reduced units there is a single adjustable parameter , @xmath46 , which determines the range of the interparticle forces . figure [ fig : potentials]b shows that decreasing @xmath46 increases the range of the attractive part of the potential and softens the repulsive wall , thus widening the potential well . values of @xmath46 appropriate to a range of materials have been catalogued elsewhere @xcite . the lj potential has the same curvature at the bottom of the well as the morse potential when @xmath46=6 . girifalco has obtained an intermolecular potential for c@xmath38 molecules @xcite that is isotropic and short - ranged relative to the equilibrium pair separation with an effective value of @xmath46=13.62 @xcite . the alkali metals have longer - ranged interactions , for example @xmath46=3.15 has been suggested for sodium @xcite . the global minima for this system have been found as a function of @xmath46 for all sizes up to @xmath32=80 @xcite.=30 and @xmath46=14 the energy of the global minimum is -106.835790 . ] equation ( [ eq : esplit ] ) enables us to understand the effect of @xmath46 on cluster structure . as @xmath46 increases and the potential well narrows , the energetic penalty for distances deviating from the equilibrium pair separation increases . thus , @xmath26 increases for strained structures , and so icosahedral and decahedral structures become disfavoured as @xmath46 increases . this is illustrated in figure [ fig : mphased ] , which shows how the structure of the global minimum depends on @xmath32 and @xmath46 . the global minimum generally change from icosahedral to decahedral to close - packed as @xmath46 is increased . it can be seen that the value of @xmath46 appropriate for the lj potential lies roughly in the middle of the icosahedral region of figure [ fig : mphased ] . comparison of the energies of icosahedral ( i ) , decahedral ( d ) and close - packed ( c ) lj@xmath48 clusters . the energy zero is @xmath49 , a function fitted to the energies of the first four mackay icosahedra at @xmath32=13 , 55 , 147 and 309 : , width=340 ] -0.5 cm -0.3 cm alternatively , the effect of @xmath46 can be thought of in terms of its effect on the crossover sizes at which a particular structural type becomes dominant . as @xmath46 increases , the less strained structures become dominant at smaller sizes . these effects can also be found in real materials . for example , sodium clusters have been shown to exhibit icosahedral structures up to at least 22000 atoms @xcite , whereas the thermodynamically stable structure of clusters of c@xmath38 molecules have recently been shown to be non - icosahedral for @xmath50 @xcite . as well as these trends , figure [ fig : mphased ] , of course , also reflects the specifics of the structures that are possible at each size , so the boundaries between structural types are not smooth lines but show a lot of detailed structure . for example , the range of @xmath46 values for which icosahedral structures are most stable is a local maximum at @xmath32=55 because of the complete mackay icosahedra possible at this size . at sizes where close - packed structures have a greater or equal number of nearest neighbours to the best decahedral structure , the global minimum changes directly from icosahedral to close - packed . a new class of structures appears in the bottom right - hand corner of figure [ fig : mphased ] . they are polytetrahedral clusters with disclination lines running through them . a polytetrahedral structure can be decomposed into tetrahedra without any interstices . the 13-atom icosahedron is an example , and one of the possible ways of adding atoms to the surface of the icosahedron the so - called anti - mackay overlayer continues the polytetrahedral packing . this overlayer does not lead to the next mackay icosahedron but instead to the 45-atom rhombic tricontahedron ( figure [ fig : ptet]a ) , which can be thought of as an icosahedron of interpenetrating icosahedra . if one imagines adding regular tetrahedra to the form in figure [ fig : gaps]b one soon realizes that the 45-atom structure must be extremely strained , and for this reason it is the global minimum only at low @xmath46 , where this strain can be accommodated . for @xmath51 similar polytetrahedral clusters can be formed but based not on the 13-atom icosahedron but on polyhedra with a higher coordination number . the two examples in figure [ fig : ptet]a have a 14- and 16-coordinate central atom . these structures can be described in terms of disclination lines , where the lines pass along those nearest - neighbour contacts that are the common edge for six tetrahedra @xcite . these types of structures might be thought to be fairly esoteric , but they form the basis for the crystalline frank - kasper phases @xcite where atoms of different size create a preference for coordination numbers higher than 12 , and so they might be good candidate structures for certain mixed metal clusters . furthermore , they have recently been observed in a model of clusters of heavy metal atoms @xcite , and recent experimental diffraction and electron microscopy data for small cobalt clusters can best be modelled by a disclinated polytetrahedral structure that is a fragment of a frank - kasper phase @xcite . however , at the size corresponding to this experiment ( @xmath52 ) the long - ranged morse clusters have disordered polytetrahedral global minima . zero temperature ` phase diagram ' showing the variation of the lowest - energy structure with @xmath32 and @xmath46 . the data points are the values of @xmath46 at which the global minimum changes . the lines joining the data points divide the phase diagram into regions where the global minima have similar structures . the solid lines denote the boundaries between the four main structural types icosahedral , decahedral , close - packed and those associated with low @xmath46 ( l)and the dashed lines are internal boundaries within a structural type , e.g. between icosahedra with different overlayers , or between decahedra with different length decahedral axes.,width=434 ] -0.6 cm -0.3 cm in contrast to the potentials that we have so far examined , the dzugutov potential @xcite has a maximum that penalizes distances near to @xmath53 times the equilibrium pair distance ( figure [ fig : potentials]a ) , the distance across the diagonal of the octahedra in close - packed structures . this maximum loosely resembles the first of the friedel oscillations @xcite often found in effective metal potentials . the potential was originally designed to suppress crystallization in bulk simulations so that the properties of supercooled liquids and glasses could more easily be studied in a one - component system @xcite . however , under certain conditions it was found that a dodecagonal quasicrystal could be formed on freezing @xcite . polytetrahedral structures that are the global minima for the ( a ) long - ranged morse potential and ( b ) dzugutov potential . sizes as labelled . the point group symmetries of the structures in ( a ) are @xmath54 , @xmath55 and @xmath36 and in ( b ) are @xmath56 , @xmath57 , @xmath58 and @xmath59 . the disclination networks associated with the 53- and 61-atom structures are illustrated next to the clusters . , width=377 ] -0.5 cm -0.3 cm for clusters the potential will penalize close - packed , decahedral and mackay icosahedral structures ( the latter two because octahedra are found within the fcc tetrahedral units from which the structures are made ) and will favour polytetrahedral clusters . therefore , one might think that this potential would provide a good model for small cobalt clusters . however , as can be seen from figure [ fig : potentials]a the potential is narrower than the lj potential , and matching the second derivative at the equilibrium pair separation to that of the morse potential gives an effective value of @xmath46 of 7.52 . therefore , the potential can not accommodate the strain in compact polytetrahedral clusters . instead , the global minima are non - compact polytetrahedral structures , such as the needles , disc and torus illustrated in figure [ fig : ptet]b @xcite . these structures are made up of face - sharing or interpenetrating 13-atom icosahedral units . in terms of equation ( [ eq : esplit ] ) they represent the best balance between maximizing @xmath19 whilst minimizing both @xmath26 and @xmath25 , where the latter now corresponds to the total energetic penalty for distances close to @xmath60 . in order to generate a model that exhibits ordered compact polytetrahedral clusters a modified dzugutov potential was constructed with an effective value of @xmath46 of 5.16 , allowing it to accommodate more strain ( figure [ fig : potentials]a ) . preliminary results indicate that the global minima do have the desired structural type , and so this system should be useful in generating realistic candidate structures to compare with the cobalt experiments @xcite . one of the remarkable features of these simple model potentials is that the structures they exhibit do provide good candidates for the structures of real clusters . indeed , frequently the structures first identified for these model systems are subsequently identified in experiments . for example , the special stability of the structures exhibited by the fcc ni@xmath61 @xcite and au@xmath61 @xcite , the decahedral au@xmath62 @xcite and the tetrahedral ( c@xmath38)@xmath39 @xcite were first identified through calculations on lj and morse clusters @xcite . furthermore , as experiments can rarely identify a cluster s structure directly , but often have to rely on comparison with properties calculated using candidate structures , it is extremely useful to have databases of plausible structures available . this is the philosophy behind internet repositories such at the cambridge cluster database ( http://brian.ch.cam.ac.uk/ccd.html ) , which contains the global minima for all the potentials described here , and the birmingham cluster website ( http://www.tc.bham.ac.uk/bcweb/ ) . in comparisons between experiment and theory the role of temperature and kinetics should be remembered . the global minimum is only rigorously the equilibrium structure at zero temperature . at higher temperatures other structures may become more stable due to entropic effects @xcite as we will see in section [ sect : therm ] . furthermore , it is not always clear whether equilibrium has been achieved under the experimental conditions , especially for clusters formed at low temperature @xcite . the type of go algorithms in which i am interested are those that find global minima , not those that are also able to _ prove _ that the best structure found is in fact truly global . unsurprisingly , the latter is a much more demanding task . for example , for lj clusters good putative global minima have been found up to @xmath32=309 , but only up to @xmath32=7 have these structures been proven to be global @xcite . of course , the problem with settling for obtaining putative global minima is that it is difficult to know when to give up looking for a lower - energy solution . for example , to my surprise , at least , a new putative global minimum was recently found for lj@xmath39 @xcite , even though powerful go algorithms had previously been applied to this cluster @xcite . the failure of these previous attempts to locate the global minimum was not because the algorithms are unable to locate the leary tetrahedron , but simply because the computations had been terminated too soon . i also wish to concentrate on go algorithms that are unbiased , i.e. those that do not artificially bias the system towards those structures that physical insight would suggest are low in energy , for example , by seeding the algorithm with fragments of a certain structural type @xcite or by searching on a lattice for a specified structural type @xcite . of course , such biased algorithms are usually more efficient . for example , most of the lj global minima were first found using such methods @xcite . however , they can not cope with ` surprising ' structures that fall outside the expected categories , and lack transferability to other systems , because they have sacrificed generality for greater efficiency in the specific problem instances . furthermore , they require a sufficient prior understanding of the structure . this may be possible for the model potentials we consider here , but it is a much more difficult task with the complex interactions that are often necessary to realistically describe a system . virtually all the global optimization algorithms that are most successful at locating the global minima of clusters have a common feature . namely , they make extensive use of local minimization . all go algorithms require elements of both local and global search . the algorithm has both to be able to explore all regions of configuration space ( overcoming any energy barriers that might hinder this ) whilst also sufficiently sampling the low - energy configurations within each region . performing local minimizations from configurations generated by a global search is one way of combining these two elements . simulated annealing provides a perhaps more traditional way of achieving this goal . in simulated annealing , by varying a parameter , the temperature , the nature of the search is changed from global ( at high temperature ) to local ( as @xmath63 ) . however , this approach has a number of weaknesses . there is effectively only one local minimization , so if the configuration does not become confined to the basin of attraction of the global minimum as the temperature is reduced the algorithm will fail , even if the system had passed through that basin of attraction at higher temperature . this condition for success is unnecessarily restrictive and leads to inefficiency . there is a further element to the most successful algorithms , namely that the energies of the local minima , not of the configurations prior to minimization , are the basis for comparing and selecting structures . this approach was first used in 1987 by li and scheraga in the application of their ` monte carlo plus minimization ' to polypeptides @xcite . however , despite this approach being independently adopted a number of times subsequently @xcite , it was only in 1997 that it was realized that in this approach one is effectively searching a transformed pes , @xmath64 , where the energy associated with each point in configuration space is that of the minimum obtained by a local minimization from that point @xcite , i.e. @xmath65 where ` min ' signifies that an energy minimization is carried out starting from @xmath1 . unlike many pes transformations proposed in the name of global optimization , this ` basin - hopping ' transformation is guaranteed to preserve the identity of the global minimum . the transformation maps the pes onto a set of interpenetrating staircases with plateaus corresponding to the basins of attraction of each minimum ( i.e. the set of configurations which lead to a given minimum after optimization ) . a schematic view of the staircase topography that results from this transformation is given in figure [ fig : trans ] . a schematic diagram illustrating the effects of the basin - hopping potential energy transformation for a one - dimensional example . the solid line is the potential energy of the original surface and the dashed line is the transformed energy , @xmath66.,width=264 ] -0.5 cm -0.3 cm the potential advantages of using the basin - hopping transformation become clear when we contrast the inter - minimum dynamics on the original and transformed pess . in molecular dynamics simulations on the original pes much time is wasted as the system oscillates back and forth within the well surrounding a minimum , waiting for the kinetic energy to become sufficiently localized along the direction of a transition state valley to enable the system to pass into an adjacent minimum . a similar dynamical ( albeit as a function of steps rather than time ) picture holds for mc when only local moves are used . the biased random walk is confined to the well around a minimum , frequently being reflected back off the walls of this well , until by chance the system happens to wander over a transition valley into a new minimum . however , a completely different picture is appropriate to the dynamics in simulations ( using mc or discontinuous molecular dynamics @xcite ) on the transformed pes . the transformation removes vibrational motion ( the hessian has no positive eigenvalues ) and transitions out of a basin are possible anywhere along the boundary of the basin . therefore , steps in any direction can lead directly to a new minimum . furthermore , downhill transitions are now barrierless . however , as figure [ fig : trans ] illustrates , significant barriers between low - energy minima can remain if they are separated by high - energy intervening minima . consequently , on @xmath64 the system can hop directly between basins ; hence the name of this transformation . furthermore , much larger mc steps can be taken on @xmath64 ; such steps would virtually always be rejected on the original pes because atoms would become too close and an extremely high energy would result . after the transformation atoms can even pass through each other . the method of searching @xmath64 is of secondary importance compared to the use of the transformation itself . indeed , the performance is fairly similar for the two main methods used , genetic algorithms and the constant temperature mc used in basin - hopping . here , we mainly concentrate on the basin - hopping approach and refer readers to hartke s chapter for more detail on the genetic algorithm methodology @xcite . in the basin - hopping or monte carlo plus minimization method , standard metropolis mc is used , i.e. moves are generated by randomly perturbing the coordinates , and are always accepted if @xmath67 decreases and are accepted with a probability @xmath68 if @xmath67 increases . using constant temperature is sufficient , since there is no great advantage to using an annealing schedule because the aim is not to trap the system in the global minimum , but just to visit it at some point in the simulation . one of the advantages of this method is its simplicity there are few parameters to adjust . it is usually satisfactory to dynamically adjust the step size to produce a 50% acceptance ratio . an appropriate temperature also needs to be chosen , but fortunately the temperature window for which the method is effective is usually large , and can be quickly found after some experimentation . furthermore , there is a well - defined thermodynamics associated with the method @xcite that makes understanding the physics behind the approach easier , as we shall see in section [ sect : go_sol ] . typically , a series of basin - hopping runs of a specified length will be performed starting from a random geometry . this is advantageous over a single longer run because it can provide a loose gauge of success . if all the runs return the same lowest - energy structure one would imagine that the true global minimum had been found . it can also often prove useful to perform runs starting from the best structures at sizes one above and below , with the lowest - energy atom removed or an atom added , respectively . the local minimization method that we have found to be most efficient for clusters is a limited memory bfgs algorithm @xcite . the basin - hopping approach is also found to be more efficient when the configuration is reset to the configuration of the local minimum after each accepted step @xcite ; this avoids problems with evaporation of atoms from the cluster since the basin - hopping transformation also reduces the barriers to dissociation . in addition to the usual steps , it is advantageous to have occasional angular steps for low - energy surface atoms . these have a similar aim to the directed mutations introduced by hartke into his genetic algorithm @xcite ; they both enable the best arrangement of the surface atoms to be found more rapidly . for biopolymers other system - specific step types have been introduced to increase efficiency @xcite . one of the main differences between the basin - hopping and genetic algorithms is that only local moves are used in basin - hopping , whereas genetic algorithms employ co - operative crossover moves in which a new structure is formed from fragments of two ` parent ' clusters . however , recently non - local moves have been introduced into a variant of basin - hopping by rotating or reflecting a fragment of the structure @xcite . mean number of steps to reach the lj@xmath48 ( up to @xmath32=74 ) global minimum from a random starting point with the basin - hopping approach . the average is over a hundred runs at @xmath69 . reproduced from ref . @xcite . , width=321 ] -0.5 cm -0.3 cm two examples of the performance of basin - hopping algorithms for lj clusters as a function of size are given in figures [ fig : bh ] and [ fig : leary_msbh ] . the basin - hopping transformation must lead to a considerable speed - up ( in terms of steps ) if the method is to be cost - effective , because the transformation is computationally expensive and requires many evaluations of the energy and the forces at each step . clearly , one would not want to use the many millions of steps and cycles that are typically used in molecular dynamics and mc simulations on the original pes . however , the results in figure [ fig : bh ] show that the number of steps required to find the global minimum is remarkably few , only of the order of hundreds or thousands of steps . this is even more remarkable when the number of minima on the pes is considered . in line with theoretical expectations @xcite the number of minima for small lj clusters increases exponentially with size @xcite . extrapolating this trend provides , for example , an estimate of @xmath70 minima for lj@xmath71 . therefore , a levinthal - type paradox @xcitemark / levinthal / levinthal.html . ] can be formulated for locating the global minimum of a cluster : the number of minima of an atomic cluster quickly becomes so large that beyond a fairly small size , if these minima were _ randomly _ searched , even at an extraordinarily fast rate , it would take an unfeasibly long time to locate the global minimum so how is it possible to find the global minimum ? yet the basin - hopping algorithm using optimal parameters finds the lj@xmath71 global minimum from a random starting configuration on average within 150 steps @xcite . the fallacy in levinthal s paradox has long been known to be the assumption of random searching @xcite however it does emphasize the extremely non - random nature of basin - hopping for lj@xmath71the runs are extremely biased towards the global minimum . observed probability of hitting the global minimum in a monotonic sequence basin - hopping run starting from a random starting point , averaged over 1000 runs . reproduced from ref . , width=321 ] -0.5 cm -0.3 cm figure [ fig : bh ] shows some interesting variations in the ease of global optimization with cluster size . the 38-atom global minimum particularly stands out as being difficult to locate , suggesting that competition between different structural types makes global optimization more difficult . indeed for lj@xmath62 , the next largest cluster with a non - icosahedral global minimum , the number of steps required is so large that it was not possible to obtain good enough statistics to be included in figure [ fig : bh ] . subsequent calculations on a super - computer by leary suggest that the mean first passage time is of the order of @xmath72 steps @xcite . the other six non - icosahedral global minima for @xmath73 are of roughly similar difficulty to locate . it is also noteworthy that the global minimum for lj@xmath74 is relatively difficult to find . in this case , there is some structural competition between the two types of icosahedral overlayer it is the first size at which the overlayer that leads to the next mackay icosahedron is lowest in energy . an alternative perspective on these effects can be obtained from figure [ fig : leary_msbh ] , which shows the probability that a ` monotonic sequence ' basin - hopping run ends at the global minimum . in this variation of the basin - hopping algorithm @xcite only downhill steps are accepted ( i.e. @xmath6=0 ) and the run is stopped after there is no further improvement for a certain number of steps . for those sizes with non - icosahedral global minima there is a much smaller probability of the run ending in the global minimum , and so again global optimization is more difficult . in these cases the majority of runs end at low - energy icosahedral structures . those examples at larger sizes are again more than an order of magnitude more difficult than lj@xmath61 , but interestingly the marks decahedra at @xmath32=102104 are somewhat more easy to locate than those at @xmath32=7577 . these results enable us to comment on the use of lj clusters as a test system for go methods . they show that the icosahedral global minima are relatively easy to locate , and in these cases optimization only starts to become more difficult as @xmath32 approaches 100 ( figure [ fig : leary_msbh ] ) . as for the non - icosahedral global minima , the number of unbiased go methods that have found the lj@xmath61 global minimum is now quite large @xcite but those that can find the lj@xmath62 global minimum is still small @xcite . therefore , a good test for a go method is to attempt to find all the global minima up to @xmath32=110 . any go method ` worth its salt ' for clusters should be able find all the icosahedral global minima and the truncated octahedron at @xmath32=38 . success for the other non - icosahedral global minima would indicate that the method has particular promise . however , far too many go algorithms have only been tested on cluster sizes where global optimization is relatively trivial . the weakness of lj clusters as a test system is that they have a relatively uniform structural behaviour . morse clusters could provide a much more varied test system , as figure [ fig : mphased ] illustrates . a suitable test would be to aim to find all the global minima at @xmath46=3 , 6 , 10 and 14 up to @xmath32=80 , as putative global minima have been tabulated for this size and parameter range @xcite . one would generally expect the difficulty of global optimization to increase with @xmath46 because the number of minima increases @xcite and the energy landscape becomes more rough @xcite . the system also provides many examples of structural competition , particularly for the short - ranged potentials where decahedral and close - packed clusters can have similar energies . a number of studies have begun to use morse clusters as a test system @xcite . the aim of this section is to provide a physical perspective that can help us understand why the global optimization of a system is easy or difficult , for example , to explain the size - dependence of figures [ fig : bh ] and [ fig : leary_msbh ] . as mentioned in section 3 , an equivalent of levinthal s paradox , which was originally formulated to capture the difficulty of a protein folding to its native state , can be applied to a cluster locating its global minimum . the flaw in that paradox is its assumption that conformations will be sampled randomly , i.e. all configurations are equally likely , because we know that in an equilibrium physical sampling of the conformation space , say in the canonical ensemble , each point will be sampled with a probability proportional to the boltzmann weight , @xmath7 . thus , the boltzmann factor favours low - energy conformations . therefore , we can begin to see the vital role played by the potential energy surface . this role extends beyond purely thermodynamic considerations to the dynamics : does the topography and connectivity of the pes naturally lead the system towards or away from the global minimum ? the levinthal assumption of random sampling is equivalent to assuming that the energy landscape has the topography of a perfectly flat putting green with no thermodynamical or dynamical biases towards the global minimum at the bottom of the ` hole ' . similarly , the np - hard character of the global optimization of atomic clusters @xcite , which results in part from the exponential increase in the search space with size , considers a general case where no assumptions about the topography of the pes can be made . in the protein folding community , after the fallacy in the levinthal paradox was recognized , attention focussed on the more important question of how does the topography of the pes differ for those polypeptides that are able to find their native states from those that can not @xcite . here , i address similar questions for the global optimization of clusters . disconnectivity graphs for ( a ) lj@xmath75 , ( b ) lj@xmath74 , ( c ) lj@xmath61 , ( d ) lj@xmath71 , ( e ) lj@xmath62 and ( f ) lj@xmath76 . in ( a ) all the minima are represented . in the other parts only the branches leading to the ( b ) 200 , ( c ) 150 , ( d ) 900 , ( e ) 250 and ( f ) 200 lowest - energy minima are shown . the numbers adjacent to the nodes indicate the number of minima the nodes represent . pictures of the global minimum , and sometimes the second lowest - energy minimum , are adjacent to the corresponding branch . the units of the energy axes are @xmath18 . , title="fig:",width=453 ] -0.5 cm -0.5 cm one of the topographical features of the energy landscape that the protein folding community has found to be common is , what has been termed , a ` funnel ' @xcite . by this they mean a region of configuration space that can be described in terms of a set of downhill pathways that converge on a single low - energy structure or a set of closely - related low - energy structures . as its name suggest a protein pes with a single funnel converging on the native state will be a good folder because the topography helps in guiding the protein towards that native state . if these ideas are to be useful , one needs a way of depicting the physically - relevant aspects of the topography of a complicated @xmath77-dimensional energy landscape . one technique that has proven to be helpful in characterizing the pess of proteins @xcite and clusters @xcite is the disconnectivity graph . this graph provides a representation of the connectivity of the multi - dimensional energy landscape and by depicting the effective barriers between minima it is especially useful in the interpretation of dynamics . to construct a disconnectivity graph , at a series of energy levels the minima on the pes are divided into sets which are connected by paths that never exceed that energy level . in the graph each set is represented by a node at the appropriate energy and lines connect a node to the sets at higher and lower energy which contain the minima corresponding to the original node . a line always ends at the energy of the minimum it represents . disconnectivity graphs can be understood by analogy to the effects of the water level in a geographical landscape . the number of nodes in a graph at a given energy is equivalent to the number of distinct seas and lakes for a given water level . inspection of actual disconnectivity graphs , e.g. figure [ fig : ljtree ] , also helps to clarify these ideas . at sufficiently high energy , all minima are mutually accessible and so there is only one node , but as the energy decreases sets of minima become disconnected from each other and the graph splits , until at sufficiently low energy , again there is only one node left , that of the global minimum . the pattern of the graph can reveal particularly interesting information about the pes topography . for a pes with a single funnel there is a single dominant stem with the other minima branching directly off it as the energy is decreased . by contrast for ` multiple - funnel ' pess the graph is expected to split at high energy into two or more major stems . in figure [ fig : ljtree ] disconnectivity graphs for a selection of lj clusters are presented , in particular some of those clusters that figures [ fig : bh ] and [ fig : leary_msbh ] indicated are more difficult to optimize . in the graph for lj@xmath75 all the minima in our near - exhaustive sample are represented . the graph shows the form for an ideal single - funnel pes , because the icosahedral global minimum is particularly low in energy , and dominates the energy landscape . the pes has a remarkable connectivity : 911 distinct transition states are connected to the global minimum and all minima are within three rearrangements of the global minimum . the disconnectivity graph for lj@xmath71 , another ` magic number ' lj cluster , also has a single - funnel . unlike for lj@xmath75 , we are unable to represent all the minima that we found on the graph , so instead we concentrate on the lower - energy minima in our sample . indeed , the two bands of minima in the graph represent mackay icosahedra with one or two defects . the graph only reveals the bottom of a funnel which extends up into the liquid - like minima @xcite . the lowest - energy path from the global minimum to the second lowest - energy minimum for ( a ) lj@xmath61 , ( b ) lj@xmath62 and ( c ) lj@xmath76 . in each case the zero of energy corresponds to the energy of the global minimum . , width=302 ] -0.5 cm -0.3 cm in contrast to these two clusters , the bottom of the lj@xmath74 pes is much flatter , and there are significant barriers between the low - energy minima , in particular between the two lowest - energy minima , which are icosahedral structures , but with different types of surface overlayer . these effects of structural competition are found in more extreme form in the graphs of those clusters with non - icosahedral global minima . the graphs of lj@xmath61 , lj@xmath62 and lj@xmath76 split at high energy into stems associated with icosahedral and fcc or decahedral structures , and so these energy landscapes have two major funnels . this splitting is most dramatic for lj@xmath62 and lj@xmath76 , where the barrier between the two funnels is much greater than between any of the other sets of minima . the barriers between the two lowest - energy minima are @xmath78 and @xmath79 for lj@xmath61 , @xmath80 and @xmath81 for lj@xmath62 and @xmath82 and @xmath83 for lj@xmath76 . the corresponding lowest - barrier paths are represented in figure [ fig : lj_epath ] . these pathways pass over many transition states13 , 65 and 30 for lj@xmath61 , lj@xmath62 and lj@xmath76 , respectively . there are many other pathways connecting the funnels , but they are either longer or involve higher effective barriers . for lj@xmath61 the pathway passes through disordered liquid - like minima . however , for the two larger clusters all the minima along the pathways are ordered and the main structural changes are achieved by rearrangements that involve cooperative twists around the five - fold axis of the decahedron the conservation of this axis throughout the structural transformation has also been observed in simulations of the decahedral to icosahedral transition in gold clusters @xcite . for lj@xmath62 the marks decahedron is oblate whilst the low - energy icosahedral minima are prolate , so the pathway involves a greater amount of reorganization of the surface layer either side of these cooperative transitions than for lj@xmath76 . for the laterr cluster the decahedral and icosahedral structures have fairly similar shapes ; hence the shorter pathway for lj@xmath76 ( figure [ fig : lj_epath ] ) , even though it is larger . equilibrium thermodynamic properties of ( a ) lj@xmath61 and ( b ) lj@xmath71 in the canonical ensemble . both the probability of the cluster being in the labelled regions of configuration space and the heat capacity , @xmath84 are depicted . the label m+@xmath85d stands for a mackay icosahedron with @xmath85 surface defects . , title="fig:",width=453 ] -0.3 cm -0.3 cm as expected from the general dominance of icosahedral structures in this size range , there are many more low - energy icosahedral minima than low - energy decahedral or fcc minima for these three clusters . there are many low - energy arrangements of the incomplete surface layer of the icosahedral structures , whereas the decahedral or fcc structures have especially compact structures and so any alteration in structure leads to a significant increase in energy . the number of minima in our samples that lie within the respective funnels is marked on figure [ fig : ljtree ] and is indicative of the greater width of the icosahedral funnels . from characterizing the energy landscapes of these clusters , the physical origins of some of the differences in difficulty for go algorithms should be becoming apparent . the single funnels of lj@xmath75 and lj@xmath71 make the global minimum particularly accessible , and the system is strongly directed towards the global minimum on relaxation down the pes . for lj@xmath74 , once the system has reached a low - energy structure , the flatness and the barriers at the bottom of the pes mean that further optimization is relatively slow compared to lj@xmath75 and lj@xmath71 . for the clusters with non - icosahedral global minima , the icosahedral funnel is much more accessible because of its greater width . furthermore , after entering the icosahedral funnel , subsequent escape into the fcc or decahedral funnel is likely to be very slow because of the large barriers that need to be overcome . the icosahedral funnel acts as a kinetic trap hindering global optimization . these effects will come out even more clearly in the next two sections as we look at the thermodynamics and dynamics associated with these clusters . from the disconnectivity graphs one would expect trapping to be a significantly greater hindrance to global optimization for lj@xmath62 and lj@xmath76 than for lj@xmath61 , and the longer interfunnel pathway for lj@xmath62 provides a possible explanation of why lj@xmath76 seems to be somewhat less difficult to optimize than lj@xmath62 . the typical thermodynamic properties of a cluster are illustrated in figure [ fig : lj_therm]b for lj@xmath71 . the heat capacity peak is associated with a melting transition that is the finite - size analogue of a first - order phase transition @xcite and up to melting the structure is based on the global minimum , perhaps with some surface defects . the thermodynamics of the clusters with non - icosahedral structures are significantly different . now , as well as the melting transition there is a further transition associated with a transition from the global minimum to the icosahedral structures that gives rise to a second lower - temperature peak in the heat capacity ( figure [ fig : lj_therm]a and [ fig : ljss ] ) . for lj@xmath61 the transition occurs fairly close to melting , and [ fig : ljss ] have been calculated using a method where the thermodynamic properties of the individual minima are summed @xcite . however , recent simulations using parallel tempering indicate that for lj@xmath61 the two transitions are slightly closer than indicated by figure [ fig : lj_therm ] and so the fcc to icosahedral transition results only in a shoulder in the heat capacity curve @xcite . ] however as the size increases the transition temperature generally decreases @xcite . canonical heat capacity peaks associated with the structural transitions from the global minimum to icosahedral structures for the lj clusters with non - icosahedral global minima . the sizes are as labelled . , width=302 ] -0.5 cm -0.3 cm these solid - solid transitions have a number of implications for global optimization . firstly , on cooling from the melt it is thermodynamically more favourable for the cluster to enter the icosahedral funnel than that associated with the global minimum . secondly , the transitions , particularly those for the larger clusters , lie below the ` glass transition ' temperature where the cluster is effectively trapped in the current local minimum . this presents a nightmare scenario for simulated annealing . on cooling the cluster would first enter the icosahedral funnel , where it would then become trapped , even when the global minimum becomes thermodynamically more stable , because of the large free energy barriers ( relative to @xmath86 ) for escape from this funnel @xcite . in the protein folding literature , good folders have been shown to have a large value of the ratio of folding temperature to the glass transition temperature , @xmath87 , because this ensures the kinetic accessibility of the native state of the protein at temperatures where it is thermodynamically most favoured @xcite . by contrast , these clusters have effective @xmath87 values less than one and are archetypal ` bad folders ' . it is impractical to examine the interfunnel dynamics by standard molecular dynamics simulations because of the extremely long time scales involved . however , it is possible to calculate the rate of interfunnel passage by applying a master equation approach to the large samples of minima and transition states used to construct the disconnectivity graphs @xcite . the rate constants for lj@xmath61 have been computed and show that the interfunnel dynamics obeys an arrhenius law ( i.e. the rate is proportional to @xmath88 , where @xmath89 is the activation energy ) well with the activation energies corresponding to the barriers associated with the lowest - energy pathway between the two funnels @xcite . for lj@xmath61 this gives a value of 43s@xmath90 for the interfunnel rate constant at the centre of the fcc to icosahedral transition ( using parameters appropriate for ar ) . as exected , this is beyond the time scales accessible to molecular dynamics simulations . the equivalent rate constants for the larger clusters are much slower because of the larger activation energies and the lower transition temperatures . relaxation of lj@xmath61 from high - energy minima showing the fast and slow contributions to the final probability of the fcc funnel . the time is in units of @xmath91 . , width=264 ] -0.8 cm -0.3 cm figure [ fig : lj38dyn ] illustrates for lj@xmath61 the dynamics of relaxation from high - energy states . the initial relaxation is relatively rapid and the majority of the population enters the icosahedral funnel ( the approximations in this calculation actually lead to an overestimation of the probability of initially entering the fcc funnel ) . these processes are separated by a couple of decades in time from the subsequent equilibration between the two funnels . the combined effects of the thermodynamics and dynamics can be illustrated by some simulated annealing results . for lj@xmath71 the probability of reaching the global minimum in annealing simulations of @xmath92 and @xmath72 mc cycles is 29% and 94% . the equivalent values for lj@xmath61 are 0% and 2% , and for lj@xmath62 the annealing simulations were never able to locate the global minimum . the previous sections illustrate the difficulty of finding the global minimum of the lj clusters with non - icosahedral global minima if the natural thermodynamics and dynamics of the system are followed . however , optimization approaches do not have to be restricted to this behaviour . for example , as we mentioned in section 3 the basin - hopping transformation accelerates the dynamics , allowing hops directly between basins . however , the transformation only reduces the interfunnel energy barriers by @xmath93 for lj@xmath61 , @xmath94 for lj@xmath62 and @xmath95 for lj@xmath76 . therefore , multiple - funnels are still potentially problematic . equilibrium thermodynamic properties of lj@xmath61 on the transformed pes in the canonical ensemble . ( a ) the probability of the cluster being in the fcc , icosahedral and liquid - like regions of configuration space . ( b ) the configurational heat capacity , @xmath84 . , title="fig:",width=453 ] -0.3 cm -0.3 cm figure [ fig : lj38_gmin_therm ] shows that the thermodynamic properties of lj@xmath61 are dramatically changed by the transformation . the transitions have been significantly broadened . there is now only a single heat capacity peak and a broad temperature range where all states are populated . in particular , the global minimum now has a significant probability of occupation at temperatures where the free energy barriers between the funnels can be surmounted . this effect is illustrated by the basin - hopping simulations in figure [ fig : gmin_dynamics ] . at low temperature the system is localized in one of the funnels with transitions between the two funnels only occurring rarely . however , at higher temperatures transitions occur much more frequently @xcite . as we noted earlier the rate of interfunnel passage is proportional to @xmath88 . although the transformation only reduces @xmath89 by a small amount , the temperature for which the occupation probability for the global minimum still has a significant value is now over ten times larger . hence , the increased interfunnel rates . the transformation has a second kinetic effect : it increases the width of the funnel of the global minimum . on the original lj@xmath61 pes only 2% of the long annealing runs entered the icosahedral funnel , whereas 13% of leary s downhill basin - hopping runs ended at the global minimum ( figure [ fig : leary_msbh ] ) . on relaxation down the energy landscape the system is much more likely to enter the fcc funnel on the transformed pes , thus making global optimization easier . these changes to the thermodynamics and dynamics also reduce the difficulty of global optimization for the larger non - icosahedral clusters , making it possible , if still very difficult , to reach the global minimum . the results of leary indicate that for these clusters the increased accessibility of the funnel of the global minimum is the more important effect of the transformation . he found that it is more efficient to restart a run when it gets stuck in a funnel ( and hope that it enters the funnel of the global minimum next time ) than to wait for the cluster to escape from that funnel @xcite . we can understand why the pes transformation so dramatically changes the thermodynamics of the system , by examining @xmath96 , the occupation probability of a minimum @xmath15 . for the untransformed pes within the harmonic approximation @xmath97 , where @xmath98 is the potential energy of minimum @xmath15 and @xmath99 is the geometric mean vibrational frequency . for the transformed pes @xmath100 , where @xmath101 is the hyperarea of the basin of attraction of minimum @xmath15 . the differences between these expressions , the vibrational frequency and hyperarea terms , have opposite effects on the thermodynamics . the higher - energy minima are generally less rigid and so the vibrational term entropically stabilizes the icosahedra and , even more so , the liquid - like state , pushing the transitions down to lower temperature and sharpening them . by contrast the hyperarea of the minima decreases with increasing potential energy , thus stabilizing the lower - energy states and broadening the thermodynamics . @xmath67 as a function of the number of steps in basin - hopping runs for lj@xmath61 at ( a ) @xmath6=0.4@xmath102 and ( b ) @xmath6=0.9@xmath102 . , title="fig:",width=377 ] -0.3 cm -0.3 cm in this chapter i hope to have shown how physical insight can play an important role in understanding the behaviour of global optimization algorithms and hope that these insights can help provide a firmer physical ( rather than just empirical or intuitive ) foundation for the design of new improved algorithms . i also hope that it will encourage some to think about the physical aspects of other types of go problems . in particular i have highlighted some examples where a multiple - funnel energy landscape strongly hinders global optimization . such challenging cases are likely to be a common feature for any cluster system where there is not a single strongly dominant structural type . more generally , multiple funnels probably represent the most difficult problem for this class of go problems , in which the lowest - energy configuration of a system is sought . for example , for this reason , the main criterion in the design of polypeptides that fold well is the avoidance of multiple funnels by optimizing the energy gap between the native state and competing low - energy structures @xcite . i have also shown why the basin - hopping transformation of the energy landscape makes global optimization for these multiple - funnel cases easier . the method s success results from a broadening of the thermodynamics , so that the occupation probability of the global minimum is significant at temperatures where the interfunnel free energy barriers can be surmounted . this idea should act as a design principle in the development of any new go algorithms that hope to overcome the challenge of multiple funnels . although basin - hopping is successful for the lj multiple - funnel examples , further improvements in efficiency are required before one can hope to succeed for similar cases at larger size or for potential energy functions that are significantly more computationally intensive to evaluate . a number of avenues by which this may be achieved suggest themselves . firstly , the basin - hopping algorithm searches the transformed potential energy surface using simple mc . however , there are a whole raft of methods that have been developed in order to speed up the rate of rare events in simulations on the untransformed pes , which could potentially be applied to the transformed pes . these include parallel tempering @xcite , jump - walking @xcite , and the use of non - boltzmann ensembles , such as tsallis statistics @xcite . secondly , the basin - hopping transformation can be combined with other pes transformations , of which there have been many suggestions @xcite , to get a double transformation ( and hopefully a greater simplification ) of the pes @xcite . the potential problem with this type of approach is that as well as smoothing the pes , most transformations also change the relative stabilities of different structures . this can work in one s favour if the transformation stabilizes the global minimum . for example , a recent transformation proposed by locatelli and schoen , which favours compact spherical clusters , stabilizes the non - icosahedral lj global minima @xcite ; for the 38-atom cluster the pes , when sufficiently deformed , has a single - funnel topography with the truncated octahedron at its bottom @xcite . but just as often a transformation will destabilize the global minimum this is why the performance of many pes - transformation go methods is erratic , perhaps solving some ` hard ' instances while failing for ` easier ' examples . however , the proposed approach could be run alongside standard basin - hopping runs , and so would perhaps succeed in instances where the standard algorithm struggles . thirdly , hartke recently proposed a modification to the genetic algorithm approach , in which a diversity of structures is maintained in the population @xcite . this leads to significant increases in efficiency for the lj clusters with multiple - funnels , because it prevents the whole population being confined ( and trapped ) within the icosahedral funnel . such an approach could also potentially increase the efficiency of other algorithms . for example , a diversity of structures could be maintained between a set of parallel basin - hopping runs . the author is the sir alan wilson research fellow at emmanuel college , cambridge . i would like to acknowledge the role played by david wales and mark miller in this research program and to thank bob leary for providing the data for figure [ fig : leary_msbh ] . f. dassenoy , m .- j . casanove , p. lecante , m. verelst , e. snoeck , a. mosset , t. ould ely , c. amiens and b. chaudret , _ experimental evidence of structural evolution in ultrafine cobalt particles stabilized in different polymers from a polytetrahedral arrangement to the hexagonal structure _ , j. chem . phys . * 112 * , 8137 ( 2000 ) . c. levinthal , _ how to fold graciously _ , in _ mssbauer spectroscopy in biological systems , proceedings of a meeting held at allerton house , monticello , illinois _ , edited by j. t. p. debrunner and e. munck , pp . 2224 ( university of illinois press , illinois , 1969 ) . j. d. bryngelson , j. n. onuchic , n. d. socci and p. g. wolynes , _ funnels , pathways , and the energy landscape of protein folding : a synthesis _ , proteins : structure , function and genetics * 21 * , 167 ( 1995 ) . j. pillardy , a. liwo and h. a. scheraga , _ an efficient deformation - based global optimization method ( self - consistent basin - to - deformed basin mapping ) . application to lennard - jones atomic clusters _ , j. phys . a * 103 * , 9370 ( 1999 ) . k. doye and d. j. wales , _ the effect of the range of the potential on the structure and stability of simple liquids : from clusters to bulk , from sodium to c@xmath38 _ , j. phys . b * 29 * , 4859 ( 1996 ) . s. schelstrate , w. schepens and h. verschelde , _ energy minimization by smoothing techniques : a survey _ , in _ molecular dynamics : from classical to quantum mechanics _ , edited by p. b. balbuena and j. m. seminario , pp . 129185 ( elsevier , amsterdam , 1999 ) .

we analyze the spectral energy distributions ( seds ) of lyman break galaxies ( lbgs ) at selected using the _ hubble space telescope _ ( _ hst _ ) wide field camera 3 ( wfc3 ) uvis channel filters . these _ hst_/wfc3 observations cover about 50 arcmin in the goods - south field as a part of the wfc3 early release science program . these lbgs at are selected using dropout selection criteria similar to high redshift lbgs . the deep multi - band photometry in this field is used to identify best - fit sed models , from which we infer the following results : ( 1 ) the photometric redshift estimate of these dropout selected lbgs is accurate to within few percent ; ( 2 ) the uv spectral slope is redder than at high redshift ( ) , where lbgs are less dusty ; ( 3 ) on average , lbgs at are massive , dustier and more highly star - forming , compared to lbgs at higher redshifts with similar luminosities ( ) , though their median values are similar within 1 uncertainties . this could imply that identical dropout selection technique , at all redshifts , find physically similar galaxies ; and ( 4 ) the stellar masses of these lbgs are directly proportional to their uv luminosities with a logarithmic slope of.46 , and star - formation rates are proportional to their stellar masses with a logarithmic slope of.90 . these relations hold true within luminosities probed in this study for lbgs from to . the star - forming galaxies selected using other color - based techniques show similar correlations at , but to avoid any selection biases , and for direct comparison with lbgs at , a true lyman break selection at is essential . the future _ hst _ uv surveys , both wider and deeper , covering a large luminosity range are important to better understand lbg properties , and their evolution .

1206.6116

the high redshift frontier has moved to @xmath11 as a result of the high resolution near - infrared ( nir ) images from the _ hubble space telescope _ ( _ hst _ ) wide field camera 3 ( wfc3 ) , and the lyman break ` dropout ' technique . the lyman break technique was first applied to select lyman break galaxies ( lbgs ) at @xmath12 @xcite , and since then it has been extensively used to select and study lbg candidates at redshifts @xmath12@xmath13 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? this dropout technique has generated large samples of faint star - forming galaxy candidates at @xmath12@xmath13 . however , at highest redshifts ( @xmath4 ) , it is very difficult to understand the details of their stellar populations using current space and ground - based telescopes . their faint magnitudes make it extremely difficult to do spectroscopic studies , and limited high resolution rest - frame optical photometry make it challenging to investigate their spectral energy distributions ( seds ) . these limitations make it imperative to identify and study lbgs at lower redshifts ( @xmath14 ) . the primary reason for the lack of dropout selected lbgs at @xmath0@xmath1 is that we need highly sensitive space - based cameras to observe the mid- to near - ultraviolet ( uv ) wavelengths required to identify lyman break at @xmath0@xmath1 . the peak epoch of global star - formation rate at @xmath0@xmath1 is now accessible using the dropout technique with the wfc3 uvis channel . * hereafter h10 ) and @xcite have used the _ wfc3 with its superior sensitivity to photometrically identify lower redshift ( @xmath0@xmath1 ) lbgs . understanding the lbgs at @xmath14 is vital for two main reasons . first , we need to study the star - formation properties of these lbgs , because they are at redshifts corresponding to the peak epoch of the global star - formation rate ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? second , they are likely lower redshift counterparts of the high redshift lbgs because of their identical dropout selection and similar physical properties whose understanding will help shed light on the process of reionization in the early universe ( e.g. , * ? ? ? * ; * ? ? ? there are primarily three techniques to select star - forming galaxies at @xmath10 : ( 1 ) _ sbzk _ ( using the @xmath15 , @xmath16 , @xmath17 bands , * ? ? ? * ; * ? ? ? * ) , ( 2 ) bx / bm ( using the @xmath18 , @xmath19 , @xmath20 bands , * ? ? ? * ; * ? ? ? * ) , and ( 3 ) lbg ( using the bands which bracket the redshifted lyman limit , h10 ; * ? ? ? all these approaches select star - forming galaxies , and yield insight into the star - forming properties of these galaxies , but they have differing selection biases , and so these samples do nt completely overlap ( see * ? ? ? * ; * ? ? ? * for details ) . therefore , it is essential to apply identical selection criteria at all redshifts to properly compare galaxy samples and accurately trace their evolution . the lbg selection is widely used to select high redshift ( @xmath4 ) galaxies , and to do equal comparison with these galaxies , here we investigate physical properties of lbgs at @xmath14 . h10 used uv observations of the wfc3 science oversight committee early release science extragalactic program ( pid : 11359 , pi : oconnell ; hereafter `` ers '' ) , which covers approximately 50 arcmin@xmath2 in the northern - most part of the great observatories origins deep survey ( goods ; * ? ? ? * ) south field , to identify lbgs at @xmath0@xmath1 . the high sensitivity of the wfc3 uvis channel data @xcite , along with existing deep optical data obtained with the advanced camera for surveys ( acs ) as part of the goods program are ideal to apply dropout technique in observed uv filters to select lbg candidates at @xmath0@xmath1 . in this paper , we use this h10 sample of lbgs to investigate their physical properties by fitting stellar synthesis models to their observed seds . this paper is organized as follows : in [ data ] , we summarize the wfc3 ers observations , and discuss our lbg sample at @xmath0@xmath1 as well as the comparison sample of lbgs at @xmath21@xmath9 . in [ seds ] , we fit observed seds of lbgs at @xmath0@xmath1 and @xmath21@xmath9to stellar population synthesis models , and discuss the best - fit parameters ( redshift , uv spectral slope , stellar mass , stellar age , and star - formation rates ) obtained from these sed fits . in [ results ] , we discuss correlations between best - fit physical parameters and their implications on our understanding of lbgs . in [ conclusion ] , we conclude with a summary of our results . in the remaining sections of this paper we refer to the _ hst_/wfc3 f225w , f275w , f336w , f098 m , f125w , f160w , filters as @xmath22 , @xmath23 , @xmath24 , @xmath25 , @xmath26 , @xmath27 , to the _ hst_/acs f435w , f606w , f775w , f850lp filters as @xmath28 , @xmath29 , @xmath30 , @xmath31 , and to the _ spitzer_/irac 3.6 @xmath32 m , 4.5 @xmath32 m filters as [ 3.6 ] , [ 4.5 ] , respectively , for convenience . we assume a _ wilkinson microwave anisotropy probe _ ( wmap ) cosmology with @xmath33=0.274 , @xmath34=0.726 and @xmath35=70.5 km s@xmath36 mpc@xmath36 , in accord with the 5 year wmap estimates of @xcite . this corresponds to a look - back time of 10.4 gyr at @xmath10 . magnitudes are given in the ab@xmath37 system @xcite . the wfc3 ers observations @xcite were done in both the uvis ( with a fov of 7.30 arcmin@xmath2 ) and the ir ( with a fov of 4.65 arcmin@xmath2 ) channels . here , we briefly summarize the uv imaging observations . the wfc3 ers uv observations were carried out in three broad - band filters @xmath22 , @xmath23 and @xmath24 . the @xmath22 and @xmath23 filters were observed for 2 orbits ( @xmath75688s ) per pointing , while the @xmath24filter was observed for 1 orbit ( @xmath72778 s ) per pointing , for a total of 40 orbits over the full ers field ( 8 pointings ) . we used the existing goods v2.0 reduction of the acs images in four optical bands ( @xmath28 , @xmath29 , @xmath30 , @xmath31 ) , which were re - binned to a pixel size of 0.09 . to match the ers ir ( @xmath25 , @xmath26 , @xmath27 ) and re - pixellated acs optical images , the uv mosaics have a pixel scale of 0.090 pix@xmath36 and cover @xmath750 arcmin@xmath2 area of the goods - south field . details of these observations and reduction process are described in @xcite . the combination of the three wfc3 uv filters and the four acs optical filters provide an excellent ability to select lbgs at @xmath0@xmath1(h10 ; * ? ? ? * ) , using the dropout technique to detect the lyman - break at rest - frame 912 @xcite . h10 used dropout color selection technique in the ers uv field to identify three sets of uv - dropouts @xmath22-dropouts , @xmath23-dropouts and @xmath24-dropouts which are lbg candidates at @xmath38 , @xmath39 and @xmath40 , respectively . they found 66 @xmath22- , 151 @xmath23- and 256 @xmath24-dropouts to a magnitude limit of [email protected] mag , respectively . in this paper , we start with the h10 lbg sample . our goal is to investigate seds of reliable lbg candidates with at least 10-band hst coverage ( augmented by additional data as described in [ seds ] ) from the h10 sample , so we apply the following filtering criteria . first criterion is the availability of the wfc3 ir ( @xmath25 , @xmath26 , @xmath27 ) data . h10 used the wfc3 uvis and acs data to select lbgs at @xmath0@xmath1 . the wfc3 uvis channel has a larger final ers mosaic than the wfc3 ir channel , so we exclude lbg candidates that do nt have wfc3 ir data from our sed analysis . this criterion reduces the h10 sample size by about 10% . secondly , galaxies with poor sed fits ( see [ seds ] ) as measured by their larger @xmath42 were excluded from the sample . the galaxies which fail the sed fit are usually fainter , does not have all nir photometric data or have highly uncertain nir photometry , has poor best - fit sed ( indicated by high @xmath42 ) and could have a primary lower - redshift ( @xmath43 ) solution . this criterion removes additional @xmath710% of galaxies from the h10 sample . this fraction of catastrophic @xmath42 outliers is consistent with the outlier fraction in the photometric redshift distribution of the dropouts in the h10 sample . the final sample of lbgs for the sed analysis after applying above mentioned criteria is 47 @xmath22- , 126 @xmath23- and 213 @xmath24-dropouts . to compare sed properties of lbgs at @xmath0@xmath1 , we select @xmath28- and @xmath29-dropouts in the wfc3 ers field . these dropouts lbg candidates at @xmath44 and @xmath45 , respectively were selected following the @xcite selection criteria . the goal of this paper is to compare the hst / wfc3 selected uv - dropout galaxies at @xmath0@xmath1 with similar galaxies at higher redshifts ( @xmath46@xmath9 ) . the limited area and depth of the ers data puts brightness limitations on our uv - dropout selection , which was confined to comparatively brighter part of the rest - uv luminosity function ( around the knee and brighter as shown in h10 ) . therefore , we have selected @xmath28-dropouts and @xmath29-dropouts in the ers field whose luminosities are similar to the uv - dropouts . the luminosity range is @xmath5 ( based on @xmath47 corresponding to m=21 mag ) . applying similar filtering criteria as lbgs at @xmath0@xmath1 , we have a comparison sample of 155 @xmath28-dropouts and 27 @xmath29-dropouts in the ers field . based on @xcite x - ray catalog , there are four active galactic nuclei ( agn ) in the h10 lbg sample ( @xmath14 ) , three agn in the @xmath28-dropout sample , and none in the @xmath29-dropout sample . these small number of x - ray agn does not affect our results or conclusions . all subsequent analysis in this paper is done identically on these 5 samples ( @xmath22- , @xmath23- , @xmath24- , @xmath28- , and @xmath29-dropouts ) for proper comparison . to show general evolutionary trends , we combine three dropout samples from h10 as a uv - dropout sample ( @xmath0@xmath1 ) , and two high redshift samples as a @xmath28-,@xmath29-dropout sample ( @xmath21@xmath9 ) . the ` le phare ` software package @xcite was used to measure the photometric redshifts , and to fit the broadband seds of lbgs . the primary goal of sed fitting is to find the best - fitting synthetic stellar population model to the observed photometry . from this best - fit model , we can estimate the redshift , stellar age , stellar mass , star - formation rate ( sfr ) , dust extinction , and other physical properties of each galaxy . we use the 2007 version of the ( * ? ? ? * hereafter cb07 ) models , which has improved prescription of thermally pulsating agb stars . we generated a set of stellar population models assuming a salpeter initial mass function , and varying the redshift ( @xmath486.0 , @xmath49z=0.1 , though a parabolic interpolation is used to refine the photometric redshift solution within @xmath49z intervals ) , metallicity ( 0.2 , 0.4 and 1 @xmath50 ) , age ( 1 myr @xmath51 @xmath52 ) , dust extinction ( 0 @xmath53 e(b - v ) @xmath53 0.7 mag , using a modified @xcite attenuation law ) , and @xmath54-folding timescale ( @xmath55=0.1,0.3,1,2,3,5,10,15,30 gyr ) for a star - formation history ( sfh)@xmath56exp(-t/@xmath55 ) . the ` le phare ` code assumes the @xcite prescription to estimate inter - galactic medium ( igm ) opacity . the model that gives the lowest @xmath42 is chosen as the best - fit sed . the contribution of major emission lines in different filters can be included in the models using the ` le phare ` code . neglecting emission lines during the sed fitting process can overestimate the best - fit stellar ages and masses by as much as 0.3 dex ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the ` le phare ` code accounts for the contribution of emission lines with a simple recipe based on the @xcite relations between the sfr and uv luminosity , h@xmath57 and [ oii ] lines . the code includes the ly@xmath57 , h@xmath57 , h@xmath3 , [ oii ] , oiii[4959 ] and oiii[5007 ] lines with different line ratios with respect to [ oii ] line , as described in @xcite . the observed photometry is available in up to 13 filters : three _ hst_/wfc3 uvis , four _ hst_/acs , three _ ir , one vlt ks , and two _ spitzer_/irac [ 3.6 ] , [ 4.5 ] bands . we perform matched aperture photometry in 10 _ hst _ bands as discussed in h10 , while we use vlt and _ spitzer _ photometry from the publicly available goods - music catalog @xcite . the photometry in music catalog has accurate psf - matching of space and ground - based images of different resolution and depth . figure [ fig : seds ] shows example best - fit seds for lbgs at @xmath0@xmath1 , and the comparison sample of lbgs at @xmath21@xmath9 . one of the free parameters during the sed fitting process is the redshift . to assess the accuracy of our sed - based photometric redshifts ( @xmath58 ) at @xmath0@xmath1 , we compare them with the spectroscopic redshifts ( @xmath59 ) from various vlt / magellan campaigns in the goods - s field ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we find that only a small number ( @xmath6030% ) of h10 dropout sample has spectroscopic redshifts , most likely due to the lack of strong features in observed 45009000 range at @xmath61 , where most ground - based spectrographs on large telescopes are optimized . we matched 91 spectroscopic redshifts for the uv - dropout sample ( @xmath0@xmath1 ) selected with the criteria discussed in figure [ fig : redshifts ] shows the comparison between the sed based photometric redshifts and the publicly available spectroscopic redshifts . the catastrophic outliers shown by concentric circles in figure [ fig : redshifts ] have quality flags that indicate the spectroscopic redshift is unreliable ( in most catalogs c or worst ) . so it is likely that these spectroscopic redshifts are not correct and hence , redshift comparison for these objects is not credible . the histogram in figure [ fig : redshifts ] shows the distribution of photometric redshift uncertainties @xmath62=(@xmath63/1+@xmath59 ) . based on this distribution , we estimate @xmath6(@xmath64)@xmath650.05 , and @xmath66=0.03 . the fraction of catastrophic outliers ( @xmath673@xmath6 ) is @xmath77% , excluding objects with unreliable spectroscopic redshifts . our photometric redshift uncertainties are consistent with @xcite , who used the deepest and the most comprehensive photometric data in the goods - s field . @xcite selected fairly bright lbgs at @xmath10 using the _ galex _ data , and found similar photometric redshift uncertainties and outlier fraction for their dropout sample . the photometric redshift uncertainty in the implied redshift is also consistent with the dropout selection method applied to select these galaxies . the dropout selection technique uses the location of a spectral break within a photometric bandpass ( filter ) , and therefore , the redshift uncertainty depends on the width of the bandpass , and could be as high as @xmath70.5 in @xmath16 . the distribution of photometric redshift uncertainties ( @xmath64 ) is asymmetric , even after excluding objects with unreliable spectroscopic redshifts . there are more galaxies in the distribution with spectroscopic redshifts lower than their photometric redshifts i.e. , ( @xmath68 ) . detail investigation of each ground - based spectra ( if available ) is needed to figure out what is causing spectroscopic redshift to be lower than photometric redshift . such an investigation is beyond the scope of this paper , but we should point out that such asymmetric distribution is also observed for _ galex_selected lbgs ( e.g. , * ? ? ? * ) , and is totally consistent within the estimated photometric redshift uncertainties . the uv spectral slope @xmath3 is determined from a power - law fit to the uv continuum spectrum @xcite , @xmath69 , where @xmath70 is the flux density per unit wavelength ( ergs s@xmath36 @xmath71 @xmath36 ) . we use the best - fit seds of dropout selected lbgs to estimate their uv spectral slope @xmath3 by fitting a straight line between rest - frame 1300 and 1900 in their model spectrum . this wavelength range covers 7 out of 10 spectral fitting windows identified by @xcite to estimate the uv spectral slope . this wavelength range is also ideal for comparing @xmath3 values at higher redshifts , because those are usually measured between rest - frame 1600 and 2000 . figure [ fig : beta_fit ] shows the slope - fitting method applied to the best - fit seds to estimate @xmath3 , where the solid line is the best - fit uv spectral slope , the dashed line is the best - fit sed , and the black filled circles are observed magnitudes . by selection ( [ data ] ) , we only consider galaxies with good sed fits so the choice of model should not affect the @xmath3 estimate . though uncertainties in observed photometry could affect the best - fit sed parameters , and hence , the @xmath3 estimate . in figure [ fig : beta_fit ] , we also quote typical intrinsic uncertainty in @xmath3 for galaxies at different redshifts . we estimate @xmath3 for each galaxy , and then fit a gaussian to the @xmath3 distribution to find median ( and sigma ) value in each redshift bin . table [ tab : beta ] shows median @xmath3 values and their corresponding uncertainties for the uv - dropout , and the @xmath28-,@xmath29-dropout samples . the evolution in the uv spectral slope @xmath3 as a function of redshift may indicate change in stellar populations of galaxies over cosmic time . we compare our @xmath3 values with the higher redshift measurements from the literature ( e.g. , * ? ? ? * ; * ? ? ? figure [ fig : beta_z ] shows the uv spectral slope @xmath3 as a function of redshift . blue filled squares are median @xmath3 values measured between rest - frame 1300 and 1900 for our dropout samples . to test how @xmath3 measurements are affected by the selection of rest - frame uv wavelength range , we also measured @xmath3 between rest - frame 1300 and 3400 , which are shown by blue open squares in figure [ fig : beta_z ] . both @xmath3 values agree within 1@xmath6 uncertainties . the red filled diamonds are measurements from @xcite and purple filled circles are from @xcite . the @xmath3 values from @xcite are for the galaxies around @xmath72 , which is consistent with our sample , while @xcite measurements are based on all galaxies extending to those fainter than @xmath72 in their respective redshift bins . the uncertainties on the median values of @xmath3 are the standard error of the mean in the case of @xcite and our measurements , while @xcite uncertainties represent 1@xmath6 scatter . for comparison , our estimated 1@xmath6 scatter in median @xmath3 values are listed in table [ tab : beta ] . figure [ fig : beta_z ] shows that the median values of @xmath3 decreases as redshift increases ( @xmath731.6 at @xmath38 to @xmath732.4 at @xmath74 ) , which could imply variations in one or more physical properties of lbgs as a function of redshift . figure [ fig : beta_all ] shows evolution in @xmath3 as a function of best - fit sed parameters ( stellar mass , stellar age , dust content , sfr ) and redshift . the lowest redshift bin ( @xmath38 ) is shown by the smallest circles , and the highest redshift bin ( @xmath21@xmath9 ) is shown by the largest circles . the largest change ( factor of @xmath72 or 0.3 dex ) is seen in the dust content e(b v ) of galaxies as @xmath3 changes from 1.6 ( at @xmath38 ) to 1.9 ( at @xmath75@xmath9 ) , while other three parameters vary much less than a factor of 2 . this could imply that change in the dust content of lbgs has largest effect on the uv spectral slope @xmath3 , and any variation in @xmath3 as a function of redshift could most likely be due to changing dust content of galaxies . therefore , based on our @xmath3 estimates , as shown in figure [ fig : beta_z ] , we could say that lbgs at lower redshift ( @xmath38 ) have more dust than lbgs at higher redshift ( @xmath76 ) . this trend of @xmath3 is consistent with previous studies , which argued that galaxies at @xmath77 tend to be bluer than those at @xmath12 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . those @xmath3 measurements , on uniformly selected lbgs , were limited to lbgs at @xmath78 , and our results below @xmath12 extends this observed trend to @xmath8 . the evolution in the uv spectral slope @xmath3 could also be due to changing star - formation history , initial mass function ( imf ) , and/or metallicity . these effects are believed to be much smaller than the effects from changing dust content of the galaxy . many authors have investigated various stellar population models to estimate these effects . @xcite explored sensitivity of the uv - continuum slope @xmath3 to changes in the mean metallicity , age , or dust extinction by choosing one fiducial model as a benchmark , and then changing various model parameters to assess changes in @xmath3 . they conclude that a factor of 2 ( or 0.3 dex ) changes in metallicity , age or dust content result in 0.07 , 0.15 , 0.35 changes in the uv spectral slope @xmath3 , respectively . this implies that changes in the dust content have much larger effect on the uv - continuum slope than similarly - sized changes in the age , metallicity , or the stellar imf . similar studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have come to the same conclusion that though @xmath3 could be affected by various stellar population properties , the change in dust content of galaxies is the predominant effect which causes @xmath3 to change . we should also emphasize that it is very challenging to completely understand these various effects based on observations only , rigorous modeling and/or simulations are required to fully assess the contributions of these various effects on the uv spectral slope @xmath3 . we compare observed sed with a suite of model templates from cb07 to find the best - fit model through @xmath42 minimization . all sed parameters are fit simultaneously . the best - fit model allows us to estimate photometric redshift ( as shown in [ photz ] ) , and physical properties of stellar populations such as stellar age , stellar mass , dust extinction e(b v ) and sfrs for each galaxy . the estimated uncertainties ( @xmath70.30.4 dex ) in stellar ages , masses and sfrs are estimated by marginalizing the uncertainties in observed photometry and redshift . one of the main limitations of sed fitting is the need to assume a sfh , which can not be reliably constrained for a galaxy from limited photometric data points . we have assumed an exponentially declining sfh . different sfhs ( e.g. , rising , constant , declining ) introduce systematic uncertainties in the stellar mass determinations , mostly at redshift greater than @xmath12 ( e.g. , * ? ? ? * ; * ? ? ? these uncertainties are typically @xmath790.3 dex ( e.g. , * ? ? ? * ) , and are within our estimated uncertainties . stellar ages are highly sensitive to the assumed sfh . any prior star - forming episode can be overshadowed by newly born stars from the most recent star - formation , totally neglecting possible existence of older population in a given galaxy . therefore , based on assumed histories , it is possible to get an older or a younger age for the same galaxy . hence , interpreting the stellar ages derived from sed fitting can be tricky , and at the very least , the uncertainties on the stellar ages could be much larger than estimated uncertainties ( @xmath70.30.4 dex ) . because of these issues , in subsequent analysis , we will not elaborate on stellar population ages and focus on other physical properties . figure [ fig : sed_hist ] shows the distributions of stellar age , stellar mass , sfrs , and e(b v ) for lbgs at @xmath0@xmath1 ( black ) , and the comparison sample of lbgs at @xmath21@xmath9 ( red ) . the median values are shown by dashed vertical lines and 1@xmath6 uncertainties in these distributions for lbgs at @xmath0@xmath1 are shown by an error bar at the top of the black histogram . a two - sided k - s test in each panel indicates a probability _ less _ than 0.006 that the distributions ( red and black histograms ) are drawn from the _ same _ parent distribution . figure [ fig : sed_hist ] shows a general trend that on average higher redshift lbgs have low sfrs , less dust , and are less massive than their lower redshift counterparts , though median values of two distributions ( red and black ) are similar within 1@xmath6 uncertainties . this result is in good agreement with previous studies comparing lbgs at @xmath12 and @xmath76 ( e.g. , * ? ? ? the average e(b v ) at @xmath80 is consistent with studies based on star - forming galaxies selected using bx / bm color technique ( e.g. , * ? ? ? * ; * ? ? ? the distribution of e(b v ) completely agrees with the uv spectral slope evolution as discussed in [ beta ] , implying that the lbgs at @xmath0@xmath1 are more dusty ( redder ) compared to lbgs at @xmath21@xmath9 . the rest - frame uv light traces recent or instantaneous sfr , while rest - frame optical and nir data help us to estimate stellar masses of galaxies . if the galaxy stellar mass and uv luminosity are related then we can directly use rest - frame uv light to estimate stellar mass without needing rest - frame optical / nir data . figure [ fig : mass_abs ] shows stellar mass of lbgs at @xmath8@xmath9 as a function of their uv absolute magnitude . these quantities are based on best - fit seds , and their typical uncertainties are shown in the lower - left corner . the dotted lines are best - fit line obtained by keeping the logarithmic slope fixed at 0.46 , which was estimated by @xcite for star - forming galaxies at @xmath10 . the dot - dash lines show the scatter from the best - fit line , which is @xmath70.3 dex for lbgs at @xmath0@xmath1 and about 0.2 dex for lbgs at @xmath21@xmath9 . we also tested the validity of this relation by fitting the slope of the line rather than fixing it . we find that the fitted slope is in the range of [email protected] for our lbg samples , which is consistent with 0.46 within the estimated 1@xmath6 scatter in this relation . therefore , we find that a proportionality relation between these two parameters with a logarithmic slope of 0.46 provides a good fit to the data . the stellar masses of the brighter lbgs with uv luminosities near the @xmath82 value of lbgs at @xmath12 from @xcite are about a factor of 2 lower than 10@xmath83 m@xmath84 estimated by @xcite . this discrepancy , though within our estimated uncertainties , could be due to the fact that we include emission lines in our sed fitting which could affect stellar masses by as much as a factor of @xmath72 . the stellar mass uv luminosity relation is fairly tight with a small scatter ( @xmath600.3 dex ) , which is consistent with other studies at similar redshifts ( e.g. , * ? ? ? * ) , and it points to a nearly constant mass - to - light ratio ( log(m / l)@xmath410.5 ) for lbgs between @xmath8 and @xmath9 . the tightness / lower scatter of the stellar mass uv luminosity relation in figure [ fig : mass_abs ] could be in part due to the fact that both these quantities are output parameters from the best - fit seds , and therefore , it is possible that these parameters are not totally independent . we have addressed this issue and discussed its implications in [ discuss ] . a similar correlation between stellar mass and absolute magnitude has been reported for lbgs at @xmath76@xmath85 by @xcite . figure [ fig : mass_abs ] shows that lbgs at @xmath8@xmath9 follow similar linear correlation between stellar mass and uv absolute magnitude ( within uncertainties ) for @xmath86 between 19 and 22.5 mag . it is important to note that @xcite does not find any correlation between the stellar mass and uv absolute magnitude for star - forming galaxies at @xmath10 with stellar masses @xmath8710@xmath83 m@xmath84 . this could be due to different color - selection technique ( bx / bm ) used by the @xcite to select star - forming galaxies , whose physical properties could differ from the dropout selected lbgs at these masses ( e.g. , * ? ? ? * ; * ? ? ? it is also possible that their sample which consists of spectroscopically confirmed bright galaxies with stellar masses greater than or equal to 10@xmath83 m@xmath84 has more massive galaxies than our sample and it is uncertain how massive galaxies would follow this correlation . the ers observations are too limited in area and depth to cover a larger luminosity range , so we can not predict how this relation will evolve for luminous ( @xmath8822.5 mag ) or dwarf ( @xmath8919 mag ) galaxies at these redshifts . the correlation between the current sfr and stellar mass in star - forming galaxies , also known as ` main sequence of star - formation ' ( ms ) , has been observed at @xmath90 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? these studies have shown that the ms relation seems to be not evolving strongly with redshift , but the zeropoint does : that is high redshift ( @xmath10 ) star - forming galaxies are forming stars at a higher rate than similar mass local galaxies . in figure [ fig : mass_sfr ] , we investigate this relation and star - formation histories for lbgs at @xmath8@xmath9 . these quantities are based on best - fit seds , and their typical uncertainties are shown in the lower - right corner . the dotted lines are best - fit line obtained by keeping the logarithmic slope fixed at 0.90 , estimated for star - forming galaxies at @xmath91 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the dot - dash lines show the scatter from the best - fit line , which is @xmath70.6 dex for lbgs at @xmath0@xmath1 and about 0.4 dex for lbgs at @xmath21@xmath9 . we also obtained the best - fit logarithmic slope for this relation , and found the fitted slope in the range of [email protected] for our lbg samples , which is consistent with 0.90 within the estimated 1@xmath6 scatter in this relation . we find that a proportionality with a logarithmic slope of 0.90 provides a good fit to the data with few outliers at stellar mass greater than 10@xmath83 m@xmath84 . @xcite have shown that tight relation exists between sfr and stellar mass for galaxies at @xmath21 using the cosmological hydrodynamic simulations , which is also consistent with the observations @xcite . @xcite also point out that the scatter in the figure [ fig : mass_sfr ] could be a measure of sfr ` burstiness ' as a function of stellar mass . this means that the linear relation ( with a logarithmic slope of @xmath70.90 ) indicate an average sfr for a given stellar mass , but galaxies can also experience bursts of up to two times the average sfr value at the same stellar mass as shown by the scatter . the scatter in the sfr versus stellar mass relation for lbgs at @xmath0@xmath1 is slightly larger than @xmath70.3 dex observed at @xmath10 by @xcite possibly because of few galaxies forming a sharp edge towards high sfr values , as seen in the relation for @xmath22- and @xmath23-dropouts ( upper panel in figure [ fig : mass_sfr ] ) . these galaxies have low stellar ages ( less than 10 myr ) , which could be highly uncertain as discussed in [ age_mass_sfr ] . it is also possible that this edge could be an artifact due to lower limits on the model parameters @xmath55 and _ t _ ( e.g. , * ? ? ? we also note that @xcite argue that the tightness in the sfr - stellar mass relation depends on the assumed sfh . the scatter in this relation is much less for a constant sfh , while it is much larger for other sfhs . therefore , it is also likely that the larger scatter we see in figure [ fig : mass_sfr ] could be due to different sfhs . figure [ fig : mass_sfr ] shows that , though our data has little more scatter compared to the ms relation at @xmath90 , the majority of our galaxies fall on to this relation characterized by a logarithmic slope of 0.90 . a similar correlation is observed at @xmath77@xmath13 by @xcite , and supported by cosmological hydrodynamic simulations of @xcite . our observations confirm this ms relation for star - forming galaxies from @xmath8 to @xmath9 , implying that on average their star - formation histories are similar . in previous sections , we have shown that lbgs at @xmath0@xmath1 on average are massive , dustier , and have higher star - formation rates than lbgs at @xmath21@xmath9 with similar luminosities , though it should also be noted that they are not very different within estimated 1@xmath6 uncertainties . as pointed out by @xcite , the number densities of galaxies at fixed luminosity could change substantially over this redshift range , which could lead to potential biases when comparing galaxies at different redshifts . however , the general trends we observe in stellar masses , sfrs , and dust extinction are supported by other independent means . the characteristics uv luminosity ( @xmath82 ) is increasing as a function of redshift from @xmath92 to @xmath93 ( e.g. , h10 ) , which implies increase in sfrs with time , while @xcite have shown that stellar masses for @xmath94 lbgs grow from @xmath74 to @xmath93 . the uv spectral slope @xmath3 shows evolution as a function of redshift ( figure [ fig : beta_z ] ) , which could indicate lower dust content at higher redshifts . the higher dust content in lbgs at lower redshift is also in accordance with the studies at @xmath0 ( e.g. , * ? ? ? * ; * ? ? ? * ) , while the @xcite supports the lower dust content in lbgs at @xmath76 . therefore , the ensemble properties of lbgs in our sample are in general agreement with the expected results . the stellar mass uv luminosity relation ( figure [ fig : mass_abs ] ) and the sfr stellar mass relation ( figure [ fig : mass_sfr ] ) are based on measurements from best - fit seds , therefore , it is possible that these quantities are not totally independent , which might affect their observed correlations . to investigate this , we show distributions of mass - to - light ( m / l ; mass/@xmath95 ) ratios and specific sfrs ( ssfr ; sfr / mass ) in figure [ fig : ssfr ] . the black ( red ) histograms show distribution for lbgs at @xmath0@xmath1 ( @xmath21@xmath9 ) , and the median values are shown by dashed vertical lines . the median values of m / l ratio and ssfr for lbgs at @xmath21@xmath9 are slightly lower than that at @xmath0@xmath1 , but are still consistent within the 1@xmath6 uncertainties as shown by the error bar on the top of the black histogram . a two - sided k - s test in each panel indicates a probability _ less _ than 0.05 that the distributions ( red and black histograms ) are drawn from the _ same _ parent distribution . the constancy of the m / l ratio and ssfr between @xmath8 and @xmath9 agrees very well with the constant slope we find in figure [ fig : mass_abs ] and figure [ fig : mass_sfr ] for our sample of lbgs , though with a slightly larger scatter . stellar masses of lbgs at @xmath0@xmath1 are generally well correlated with uv absolute magnitude and current sfr , as expected for star - forming galaxies at similar redshifts ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? these correlations implies very similar mass assembly and sfh for these galaxies , but the exact nature of sfhs is still not clearly understood . @xcite showed that the cosmologically averaged sfrs of star - forming galaxies at @xmath96 at constant co - moving number density increase smoothly from @xmath97 to @xmath1 , and the stellar mass growth in these galaxies is consistent with this derived sfh . the scenario of rising sfh ( see also * ? ? ? * ) is also supported by recent results from the cosmological hydrodynamic simulations ( e.g. , * ? ? ? * ) . the models with rising sfhs conflicts with the assumptions that the sfr in distant galaxies is either constant or decreasing exponentially with time ( e.g. , * ? * ; * ? ? ? * ; * ? ? ? though , we remind the reader that the models with rising sfhs advocated by @xcite and others correspond to a cosmologically averaged sfhs for typical galaxies , and not individual galaxies , because they could involve random events that changes their instantaneous sfr . @xcite and @xcite also argue that rising sfhs are most beneficial to higher redshift ( @xmath78 ) galaxies . we find that for our assumed sed model parameters , the lbgs between redshift @xmath8 and @xmath9 on average have similar sfhs , though the precise nature of sfhs at all redshift is still under debate , and could also affect the sfr stellar mass correlation . our analysis demonstrates that the dropout selected galaxies at @xmath0@xmath1 within luminosities probed here show similar correlations between physical parameters ( sfr , stellar mass , uv luminosity ) as other star - forming galaxies selected using different color criteria ( e.g. , @xmath98 , bx / bm ) at @xmath10 . this is consistent with the @xcite conclusion that majority ( @xmath78090% ) of the dropout selected galaxies overlap with other color selected star - forming galaxies with stellar masses less than 10@xmath83 m@xmath84 . the stellar mass range for our current sample is between @xmath710@xmath99 and @xmath710@xmath83 m@xmath84 , with sample completeness around 10@xmath100 m@xmath84 . significant differences between the dropout selected sample and other color selected samples of star - forming galaxies at @xmath10 exists for massive galaxies ( @xmath8710@xmath83 m@xmath84 ; @xcite ) . therefore , it is vital to use uniform selection technique at all redshifts to avoid any selection biases . the lyman break dropout technique is the most convenient and widely used method to select galaxies at @xmath78 , and we have shown that lbgs at @xmath0@xmath1 selected using this dropout technique have similar physical properties ( within uncertainties ) as lbgs at @xmath21@xmath9 with similar luminosities . hence , lbg selection at @xmath101 is important to understand properties of lbgs and properly investigate their evolution as a function of redshift . the validity of lbg properties over wide luminosity and mass range can be investigated in detail with the upcoming and future wfc3 uv surveys such as candels @xcite and the wfc3 uv udf ( * ? ? ? * pi : h. teplitz ) . in this paper , we investigated stellar populations of lbgs at @xmath0@xmath1 selected using _ hst_/wfc3 uvis filters in the goods - s field . we used deep multi - wavelength observations from the _ hst _ , vlt , and _ spitzer _ to compare observed seds with the spectral synthesis models to infer physical properties ( stellar masses , stellar ages , sfrs , and dust extinction ) of these lbgs . we also compared these lbgs with their higher redshift ( @xmath21@xmath9 ) counterparts with similar luminosities ( @xmath5 ) . our results can be summarized as follows : @xmath102 the uv continuum slope @xmath3 for lbgs at @xmath0@xmath1 is redder ( @xmath731.6 at @xmath38 ) compared to their higher redshift counterparts ( @xmath732.4 at @xmath74 ) , implying higher dust content in these lbgs . @xmath102 on average , lbgs at @xmath0@xmath1 are massive , dustier and more highly star - forming compared to lbgs at @xmath21@xmath9 , though their median values are very similar within estimated 1@xmath6 uncertainties . this similarity emphasizes the importance of identical lyman break selection technique at all redshifts , which selects physically similar galaxies . @xmath102 the stellar mass absolute uv magnitude relation for lbgs between @xmath8 and @xmath9 show linear correlation with a logarithmic slope of @xmath70.46 , while the sfr stellar mass relation show similar correlation with a logarithmic slope of @xmath70.90 . to properly compare and interpret such relations at higher ( @xmath4 ) redshift , and to avoid any selection biases due to different selection techniques , a true lyman break selection is required at @xmath10 . @xmath102 we need larger _ hst _ uv surveys to cover full range in luminosity / mass and better understand lbg properties , and their evolution . both deeper and wider uv surveys are needed . the wider one to probe the high mass end , while the deeper one will probe the sub-@xmath47 population . a large number of _ hst _ orbits have been used for dropout selected galaxies at @xmath4 , and the lower redshift regime needs to be explored in a comparable manner . we thank the referee for helpful comments and suggestions that significantly improved this paper . this paper is based on early release science observations made by the wfc3 scientific oversight committee . we are grateful to the director of the space telescope science institute for awarding director s discretionary time for this program . support for program # 11359 was provided by nasa through a grant hst - go-11359.08-a 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 . this research was ( partially ) supported by a grant from the american astronomical society . adelberger , k. l. , steidel , c. c. , shapley , a. e. , et al . 2004 , apj , 607 , 226 arnouts , s. , cristiani , s. , moscardini , l. , et al . 1999 , mnras , 310 , 540 atek , h. , siana , b. , scarlata , c. , et al . 2011 , apj , 743 , 121 balestra , i. , mainieri , v. , popesso , p. , et al . 2010 , a&a , 512 , 12 basu - zych , a. r. , hornschemeier , a. e. , hoversten , e. a. , lehmer , b. , & gronwall , c. 2011 , apj , 739 , 98 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 bouwens , r. j. , illingworth , g. d. , blakeslee , j. p. , & franx , m. 2006 , apj , 653 , 53 bouwens , r. j. , illingworth , g. d. , franx , m. , & ford , h. 2007 , apj , 670 , 928 bouwens , r. j. , illingworth , g. d. , franx , m. , et al . 2009 , apj , 705 , 936 bouwens , r. j. , illingworth , g. d. , oesch , p. a. , et al . 2010 , apj , 709 , l133 bouwens , r. j. , illingworth , g. d. , oesch , p. a. , et al . 2012 , apj , 754 , 83 bruzual , g. , & charlot , s. 2003 , mnras , 344 , 1000 burgarella , d. , le floh , e. , takeuchi , t. t. , et al . 2007 , mnras , 380 , 986 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , apj , 429 , 582 calzetti , d. , armus , l. , bohlin , r. c. , et al . 2000 , apj , 533 , 682 cooper , m. c. , yan , r. , dickinson , m. , et al . 2012 , mnras , 425 , 2116 daddi , e. , cimatti , a. , renzini , a. , et al . 2004 , apj , 617 , 746 daddi , e. , dickinson , m. , morrison , g. , et al . 2007 , apj , 670 , 156 dahlen , t. , mobasher , b. , dickinson , m. , et al . 2010 , apj , 724 , 425 elbaz , d. , daddi , e. , le borgne , d. , et al . 2007 , a&a , 468 , 33 erb , d. k. , shapley , a. e. , pettini , m. , et al . 2006 , apj , 644 , 813 finkelstein , s. l. , papovich , c. , giavalisco , m. , et al . 2010 , apj , 719 , 1250 finkelstein , s. l. , hill , g. j. , gebhardt , k. , et al . 2011 , apj , 729 , 140 finkelstein , s. l. , papovich , c. , salmon , b. , et al . 2012 , apj , 756 , 164 finlator , k. , dav , r. , papovich , c. , & hernquist , l. 2006 , apj , 639 , 672 finlator , k. , dav , r. & oppenheimer , b. d. 2007 , mnras , 376 , 1861 finlator , k. , oppenheimer , b. d. , & dav , r. 2011 , mnras , 410 , 1703 giavalisco , m. , ferguson , h. c. , koekemoer , a. m. , et al . 2004 , apj , 600 , l93 grazian , a.,fontana , a. , de santis , c. , et al . 2006 , a&a , 449 , 951 grogin , n.a . , kocevski , d. d. , faber , s. m. , et al . 2011 , apjs , 197 , 35 guhathakurta , p. , tyson , j. a. , & majewski , s. r. 1990 , apj , 357 , l9 haberzettl , l. , williger , g. , lehnert , m. d. , nesvadba , n. , & davies , l. 2012 , apj , 745 , 96 hainline , k. n. , shapley , a. e. , greene , j. e. , et al . 2012 , apj , 760 , 74 hathi , n. p. , malhotra , s. , & rhoads , j. 2008a , apj , 673 , 686 hathi , n. p. , jansen , r. a. , windhorst , r. a. , et al . 2008b , aj , 135 , 156 hathi , n. p. , ryan , r. e. , cohen , s. h. , et al . 2010 , apj , 720 , 1708 ( h10 ) ilbert , o. , arnouts , s. , mccracken , h. j. , et al . 2006 , a&a , 457 , 841 ilbert , o. , capak , p. , salvato , m. , et al . 2009 , apj , 690 , 1236 kennicutt , r. c. , jr . 1998 , ara&a , 36 , 189 koekemoer , a. m. , faber , s. m. , ferguson , h. c. , et al . 2011 , apjs , 197 , 36 komatsu , e. , dunkley , j. , nolta , m. r. , et al . 2009 , apjs , 180 , 330 labb , i. , gonzlez , v. , bouwens , r. j. , et al . 2010 , apj , 708 , l26 lee , s .- k . , ferguson , h. c. , somerville , r. s. , wiklind , t. , & giavalisco , m. 2010 , apj , 725 , 1644 leitherer , c. , schaerer , d. , goldader , j. d. , et al . 1999 , apjs , 123 , 3 ly , c. , malkan , m. a. , treu , t. , et al . 2009 , apj , 697 , 1410 ly , c. , malkan , m. a. , hayashi , m. , et al . 2011 , apj , 735 , 91 madau , p. 1995 , apj , 441 , 18 mclure , r. j. , dunlop , j. s. , de ravel , l. , et al . 2011 , mnras , 418 , 2074 noeske , k. g. , weiner , b. j. , faber , s. m. , et al . 2007 , apj , 660 , l43 oesch , p. a. , bouwens , r. j. , carollo , c. m. , et al . 2010 , apj , 725 , l150 oke , j. b. , & gunn , j. e. 1983 , apj , 266 , 713 papovich , c. , dickinson , m. , & ferguson , h. c. 2001 , apj , 559 , 620 papovich , c. , dickinson , m. , giavalisco , m. , conselice , c. , & ferguson , h. c. 2005 , apj , 631 , 101 papovich , c. , finkelstein , s. l. , ferguson , h. c. , lotz , j. m. , & giavalisco , m. 2011 , mnras , 412 , 1123 rafelski , m. , teplitz , h. , grogin , n. , et al . 2012 , aas 220 , 421.04 ravikumar , c. d. , puech , m. , flores , h. , et al . 2007 , a&a , 465 , 1099 reddy , n. a. , et al . 2008 , apjs , 175 , 48 reddy , n. a. , & steidel , c. c. 2009 , apj , 692 , 778 santini , p. , fontana , a. , grazian , a. , et al . 2009 , a&a , 504 , 751 sawicki , m. 2012 , mnras , 421 , 2187 schaerer , d. , & de barros , s. 2009 , a&a , 502 , 423 shapley , a. e. , steidel , c. c. , erb , d. k. , et al . 2005 , apj , 626 , 698 stanway , e. r. , mcmahon , r. g. , & bunker , a. j. 2005 , mnras , 359 , 1184 stark , d. p. , ellis , r. s. , bunker , a. , et al . 2009 , apj , 697 , 1493 stark , d. p. , ellis , r. s. , chiu , k. , ouchi , m. , & bunker , a. 2010 , mnras , 408 , 1628 steidel , c. c. , giavalisco , m. , pettini , m. , dickinson , m. , & adelberger , k. l. 1996 , apj , 462 , l17 steidel , c. c. , adelberger , k. l. , giavalisco , m. , dickinson , m. , & pettini , m. 1999 , apj , 519 , 1 steidel , c. c. , shapley , a. e. , pettini , m. , et al . 2004 , apj , 604 , 534 vanzella , e. , cristiani , s. , dickinson , m. , et al . 2008 , a&a , 478 , 83 verma , a. , lehnert , m. d. , frster schreiber , n. m. , bremer , m. n. , & douglas , l. 2007 , mnras , 377 , 1024 wilkins , s. m. , bunker , a. j. , stanway , e. , lorenzoni , s. , & caruana , j. 2011 , mnras , 417 , 717 windhorst , r. a. , cohen , s. h. , hathi , n. p. , et al . 2011 , apjs , 193 , 27 wuyts , s. , labb , i. , frster schreiber , n. m. , et al . 2008 , apj , 682 , 985 xue , y. q. , luo , b. , brandt , w. n. , et al . 2011 , apjs , 195 , 10 yan , h. , windhorst , r. a. , hathi , n. p. , et al . 2010 , ra&a , 10 , 867 example best - fit seds of lbgs at @xmath0@xmath1 . we have used the ` le phare ` code @xcite to compute photometric redshifts and to perform sed fitting . grey concentric circles are observed magnitudes in three _ hst_/wfc3 uvis , four _ hst_/acs , three _ ir , one vlt ks , and two _ spitzer_/irac [ 3.6 ] , [ 4.5 ] bands . [ bottom panel ] same as the top panel but for the comparison sample of lbgs at @xmath21@xmath9 . stellar masses are in solar units and stellar ages are in years . ] the concentric circles indicate unreliable spectroscopic redshifts based on their quality flags . the dotted line shows the 1-to-1 relation between two redshifts . the histogram shows distribution of ( @xmath104/1+@xmath59 ) , and the 1@xmath6 uncertainties in this distribution is @xmath75% . the dot - dash lines show 3@xmath6 limits . ] to best - fit seds for two lbgs from the @xmath0@xmath1 sample . the dotted black line is the best - fit sed , while solid black line shows the estimated uv continuum slope @xmath3 . the black filled circles are observed magnitudes . we test two different rest - frame uv wavelength range ( 13001900 and 13003400 ) to address the robustness of our estimated uv continuum slope @xmath3 , and find similar median slopes within 1@xmath6 scatter . ] as a function of redshift . blue filled squares indicate the median @xmath3 values when we fit rest - frame 1300 to 1900 wavelengths , and blue open squares indicate @xmath3s when we fit 1300 to 3400 . the higher redshift ( @xmath105 ) measurements from @xcite and @xcite are also shown for comparison . ours and the @xcite uncertainties are the standard error of the mean , while uncertainties from bouwens et al . are 1@xmath6 scatter . for comparison , our 1@xmath6 uncertainties / scatter are listed in table [ tab : beta ] . ] versus best - fit sed parameters ( e(b - v ) , stellar mass , stellar age , star - formation rate ) . the size of the circle increases with redshift i.e. , the smallest circle correspond to the lowest redshift bin ( @xmath38 ) , and the largest circle correspond to @xmath21@xmath9 bin . the changes in the uv - continuum slope @xmath3 from @xmath38 to @xmath9 are most likely due to the change ( factor of @xmath72 or 0.3 dex ) in the dust content e(b v ) of the galaxies , as other parameters show smaller variation as a function of redshift . ] @xmath1 ( black ) and the comparison sample at @xmath21@xmath9 ( red ) . all four parameters are estimated from the best - fit seds . median values are shown by vertical dashed lines and 1@xmath6 uncertainties are shown by an error bar at the top of the black histogram . though the higher redshift lbgs have , on average , lower values of most of these physical parameters than lbgs at @xmath0@xmath1 , the median values are similar within estimated uncertainties . stellar ages do not show any clear evolution with redshift mainly because of higher uncertainties in their measurements ( see [ age_mass_sfr ] ) . the histograms show normalized numbers for both samples ( total number of galaxies in each sample is shown in table [ tab : beta ] ) , and a two - sided k - s test in each panel indicates a probability @xmath106 that the two distributions are drawn from the same parent distribution . ] - , @xmath23- , @xmath24-dropout samples . bottom right panel shows the same relation for the comparison sample of lbgs at @xmath21@xmath9 ( black circles for @xmath21 , and black squares for @xmath107 ) . the average uncertainties are shown in the bottom - left corner . the dotted black line is the best - fit line with a logarithmic slope of 0.46 . the dot - dash line shows the 1@xmath6 scatter ( @xmath70.3 dex ) from the best - fit linear relation . ] - , @xmath23- , @xmath24-dropout samples . bottom right panel shows the same relation for the comparison sample of lbgs at @xmath21@xmath9 ( black circles for @xmath21 , and black squares for @xmath76 ) . the average uncertainties are shown in the bottom - right corner . the dotted black line is the best - fit line with a logarithmic slope of 0.90 . the dot - dash line shows the 1@xmath6 scatter ( @xmath70.6 dex , @xmath70.4 dex for @xmath21@xmath9 ) from the best - fit linear relation . in the upper panel , few galaxies form a sharp edge towards high sfr values , which could be an artifact due to lower limits on the model parameters @xmath55 and _ t _ ( e.g. , * ? ? ? ] @xmath1 ( black ) and for the comparison sample at @xmath21@xmath9 ( red ) . median values are shown by vertical dashed lines and 1@xmath6 uncertainties are shown by an error bar at the top of the black histogram . though the higher redshift lbgs have , on average , lower values of these parameters than lbgs at @xmath0@xmath1 , the median values are similar within estimated uncertainties . the histograms show normalized numbers for both samples ( total number of galaxies in each sample is shown in table [ tab : beta ] ) , and a two - sided k - s test in each panel indicates a probability @xmath108 that the two distributions are drawn from the same parent distribution . ] 1.6 ( @xmath810.2 ) & 47 & 1.59 ( 1.67 ) & 0.29 ( 0.27 ) & 0.04 ( 0.04 ) + 2.2 ( @xmath810.3 ) & 126 & 1.71 ( 1.81 ) & 0.34 ( 0.37 ) & 0.03 ( 0.03 ) + 2.6 ( @xmath810.4 ) & 213 & 1.71 ( 1.78 ) & 0.47 ( 0.50 ) & 0.03 ( 0.03 ) + 3.8@xmath109 ( @xmath810.5 ) & 182 & 1.88 ( 1.93 ) & 0.45 ( 0.41 ) & 0.03 ( 0.03 ) +

we perform a systematic study of several models that have been proposed for the purpose of understanding the motion of driven interfaces in disordered media . we identify two distinct universality classes : ( i ) one of these , referred to as directed percolation depinning ( dpd ) , can be described by a langevin equation similar to the kardar - parisi - zhang equation , but with quenched disorder . ( ii ) the other , referred to as quenched edwards - wilkinson ( qew ) , can be described by a langevin equation similar to the edwards - wilkinson equation but with quenched disorder . we find that for the dpd universality class the coefficient of the nonlinear term diverges at the depinning transition , while for the qew universality class either or as the depinning transition is approached . the identification of the two universality classes allows us to better understand many of the results previously obtained experimentally and numerically . however , we find that some results can not be understood in terms of the exponents obtained for the two universality classes _ at _ the depinning transition . in order to understand these remaining disagreements , we investigate the scaling properties of models in each of the two universality classes _ above _ the depinning transition . for the dpd universality class , we find for the roughness exponent for the pinned phase , and for the moving phase . for the growth exponent , we find for the pinned phase , and for the moving phase . furthermore , we find an anomalous scaling of the prefactor of the width on the driving force . a new exponent , characterizing the scaling of this prefactor , is shown to relate the values of the roughness exponents on both sides of the depinning transition . for the qew universality class , we find that and are roughly the same for both the pinned and moving phases . moreover , we again find a dependence of the prefactor of the width on the driving force . for this universality class , the exponent is found to relate the different values of the local and global roughness exponent at the depinning transition . these results provide us with a more consistent understanding of the scaling properties of the two universality classes , both at and above the depinning transition . we compare our results with all the relevant experiments . 2

cond-mat9506091

recently the growth of rough interfaces has witnessed an explosion of theoretical , numerical , and experimental studies , fueled by the broad interdisciplinary aspects of the subject @xcite . applications can be so diverse as imbibition in porous media , fluid fluid displacement , fire front motion , and the motion of flux lines in superconductors @xcite . in general , a @xmath13-dimensional self - affine interface , described by a single - valued function @xmath14 , evolves in a @xmath15-dimensional medium . usually , some form of disorder @xmath16 affects the motion of the interface leading to its roughening . two main classes of disorder have been discussed in the literature . the first , called thermal or `` annealed , '' depends only on time . the second , referred to as `` quenched , '' is frozen in the medium . early studies focused on _ time - dependent _ disorder as being responsible for the roughening @xcite . here , we consider in detail the effect of _ quenched _ disorder on the growth . the presence of quenched disorder introduces an interesting analogy between the motion of driven interfaces in disordered media and the theory of critical phenomena . the continual motion of the interface requires the application of a driving force @xmath17 . there exists a critical force @xmath18 , such that for @xmath19 , the interface will become pinned by the disorder after some finite time . for @xmath20 , the interface moves indefinitely with an average velocity @xmath21 . this suggests that the motion of driven rough interfaces in disordered media can be studied as a phase transition which we shall call the _ depinning transition_. the velocity of the interface @xmath22 plays the role of the _ order parameter _ , since as @xmath23 , @xmath22 vanishes as @xmath24 we call @xmath25 the _ velocity _ exponent , and @xmath26 the _ reduced force _ ( fig . [ phases ] ) . for @xmath23 , large but finite regions of the interface are pinned by the disorder . at the transition , the characteristic length @xmath27 of these pinned regions diverges as @xmath28 where @xmath29 is the _ correlation length _ exponent . the added phenomenological richness introduced by the quenched disorder leads to an increased difficulty . in fact , the problem of identifying the universality classes for interface roughening in the presence of quenched disorder and determining the corresponding sets of critical exponents has so far remain unsolved . experimental and numerical measurements of some of the critical exponents vary considerably , and the question of the existence of universality classes bas been raised @xcite . the problem is made more complex by the disagreement of most experimentally measured values with analytical calculations @xcite . in this paper , we address the question of why the same exponents vary in value for different systems . we study several numerical models introduced to study interface roughening in the presence of quenched disorder , and identify two universality classes . we show that one of those universality classes can be identified with the case studied analytically . the exponents for the other universality class can not be determined from renormalization groups calculations . however , mappings to directed percolation ( dp ) and isotropic percolation allow us to estimate most scaling exponents . the identification of the universality classes and the calculation of the respective scaling exponents _ at _ the depinning transition enable us to understand the results for most of the models proposed and for some of the experiments . we shall see that several experimental results particularly for the case of the moving interface in fluid - fluid displacement experiments still do not fit the framework provided by the universality classes we discuss . for this reason , we study models in each of the universality classes _ above _ the depinning transition i.e. , in the _ moving _ phase . the results obtained allow us to re - interpret the experimental measurements . furthermore , we find that for one of the universality classes , some of the critical exponents change values at the depinning transition . the paper is organized as follows : in sec . ii we introduce the problem , by describing the set of relevant scaling properties and associated exponents , for driven interfaces moving in disordered media . in sec . iii we discuss the experimental and numerical results motivated our study . in sec . iv we consider several numerical models , and identify two distinct universality classes . in sec . v we try to link the results obtained for the two universality classes with the experimental results . to achieve this goal , we study models , representative of each of the universality classes , above the depinning transition . finally , in sec . vi , we summarize our results and discuss some of the new questions arising from this work . an interface moving in a disordered medium becomes rough due to the action of the disorder . this roughening process can be quantified by studying the _ global _ interface width @xmath30 ^{1/2 } \right \rangle , \label{wid}\ ] ] where @xmath31 is the system size , the overbar denotes a spatial average , and the angular brackets denote an average over realizations of the disorder . the study of discrete models @xcite and continuum growth equations @xcite leads to the observation that during the initial period of the growth , i.e. for @xmath32 , the global width @xmath33 grows with time as @xmath34 where @xmath35 is the _ growth _ exponent . for times much larger than @xmath36 the width saturates to a constant value . it was observed that the saturation width of the interface @xmath37 scales with @xmath31 as @xmath38 where @xmath39 is the _ roughness _ exponent . the dependence of @xmath36 on @xmath31 allows the combination of ( [ bet ] ) and ( [ wsat ] ) into a single scaling law @xcite @xmath40 where @xmath41 [ scl1 ] here @xmath42 is the _ dynamical _ exponent , and @xmath43 is a universal scaling function that grows as @xmath44 when @xmath45 , and approaches a constant when @xmath46 . an alternative way of determining the scaling exponents is to study the _ local _ width @xmath47 in a box of length @xmath48 . the scaling law ( [ scl1 ] ) , and the fact that the interface is self - affine ( and with @xmath49 ) , allow us to conclude @xmath50 where @xmath51 or @xmath52 [ scl2 ] here @xmath43 is a universal scaling function that decreases as @xmath53 when @xmath46 and approaches a constant when @xmath45 . not all of the exponents introduced so far are independent ; e.g. , we can calculate the velocity exponent from the other scaling exponents , as we show by the following argument . near the depinning transition , except for a few growing regions , most of the interface is pinned by some pinning path . the growth occurs by the propagation of these growing regions through the system . the characteristic time required for this propagation must be of the order of @xmath36 , since this is the time for correlations to propagate across the entire system , as defined by ( [ scl1]b ) . during this process the interface advances from one blocking path to the next , the distance advanced is typically of the order of @xmath37 . since , close to the transition , @xmath54 , we can use ( [ corl ] ) , ( [ wsat ] ) , and ( [ scl1 ] ) to obtain @xmath55 upon comparison with ( [ vel ] ) , we conclude that @xmath56 this exponent `` scaling law '' can also be derived in a different way @xcite , by considering the interplay between the range of action of the quenched disorder and the length scales for which the disorder can be considered to be time - dependent because of the motion of the interface . the quenched disorder is a function of @xmath57 and @xmath58 , but we can consider the change of variables to a reference frame moving with the velocity of the interface @xmath59 now , we can consider the conditions for which @xmath60 . this will happen for the length scale characterizing the action of the quenched disorder , @xmath27 . thus we have @xmath61 leading to ( [ thet ] ) . the most commonly measured exponent in experiments and simulations is the roughness exponent @xmath39 . in the last few years , several experiments in which quenched disorder plays a dominant role have been performed . _ ( i ) _ stokes _ et al . _ performed fluid - fluid displacement experiments , and observed that for some conditions a self - affine interface would form , while for other , the growth would generate a percolation - like cluster @xcite . _ ( ii ) _ rubio _ et al . _ measured @xmath39 for fluid - fluid displacement experiments @xcite , consistently obtaining @xmath62 regardless of the velocity of the interface @xmath22 . moreover , they observed that the total width of the interface decreased with @xmath22 @xcite . _ ( iii ) _ horvth _ et al . _ performed similar fluid - fluid displacement experiment , obtaining @xmath63 @xcite . _ ( iv ) _ buldyrev _ et al . _ studied the imbibition of coffee in paper towels observing @xmath64 for the pinned interface @xcite . similar experiments were performed by buldyrev _ et al . _ for @xmath65 dimensions leading to @xmath66 @xcite . _ ( v ) _ he _ et al . _ conducted fluid - fluid displacements experiments for different velocities of the interface @xcite . they reported values of @xmath39 varying from @xmath67 for high velocities , to @xmath68 for low velocities . furthermore , they also observed that the total width of the interface decrease with velocity @xcite . _ ( vi ) _ zhang _ et al . _ performed flameless paper burning experiments @xcite . they reported @xmath69 for the moving interface . thus the experiments reveal values of the roughness exponent as low as @xmath67 and as high as @xmath68 , leading many to question whether one can divide all systems into a small number of universality classes ( fig . [ f : summary ] ) . several discrete models with quenched disorder have been proposed to explain the experimental results . _ ( i ) _ early on , cieplak and robbins noticed the importance of quenched disorder in many surface phenomena @xcite . they studied two models , motivated by fluid - fluid displacement experiments . the first , called the _ fluid invasion model _ , is a quite realistic model and calculations reveal @xmath70 in @xmath71 dimensions @xcite . the second , called the _ random field ising model _ ( rfim ) @xcite , does not have a self - affine phase in @xmath71 dimensions , so no roughness exponent can be calculated . on the other hand , for @xmath65 dimensions , a self - affine regime exists , and a roughness exponent of @xmath72 was calculated @xcite . _ ( ii ) _ kessler _ et al . _ integrated the edwards - wilkinson ( ew ) equation with quenched disorder , with the purpose of explaining the results for the fluid - fluid displacement experiments @xcite . they reported values of @xmath39 that vary with the driving force , i.e. with the velocity of the interface . for low velocities they found @xmath73 , and for large velocities , @xmath74 @xcite . furthermore , they found a decrease of the total width with the velocity of the interface . similar results were obtained earlier by koplik and levine @xcite . _ ( iii ) _ buldyrev _ et al . _ introduced a discrete model to explain their imbibition experiments @xcite . they found that in @xmath71 dimensions the scaling behavior of the pinned interface can be obtained exactly by mapping the interface onto dp . in higher dimensions they found that the interface can be mapped onto a _ directed surface _ ( ds ) @xcite . in @xmath71 dimensions , dp and ds are equivalent , so this model is referred to as _ directed percolation depinning _ ( dpd ) model @xcite . _ ( iv ) _ a similar model was introduced independently by tang and leschhorn @xcite . for the dpd class of models , it was observed that @xmath64 for the pinned interface @xcite . _ ( v ) _ parisi also introduced a model for interface roughening ; however , no measurement was made of the roughness exponent @xcite . several authors studied parisi s model and found it to possess problems ( see appendix b ) . for this reason , amaral introduced an alternate version of the model , which led to exponents identical to the ones obtained for the dpd class of models @xcite . _ ( vi ) _ dong _ et al . _ also integrated numerically the qew equation , and reported @xmath75 from the study of the height - height correlation function @xcite . _ ( vii ) _ leschhorn introduced a model , corresponding essentially to a discretization of the qew equation , and found @xmath76 from measurements of the global width . the disagreement with ref . @xcite was later explained by leschhorn and tang , by showing that the height - height correlation function is limited to a scaling with a roughness exponent smaller or equal to @xmath77 @xcite . _ ( viii ) _ roux and hansen studied what amounts to a self - organized version of leschhorn s model finding that @xmath78 for the local width and @xmath79 for the global width @xcite . using a similar model , galluccio and zhang found identical results @xcite . _ ( x ) _ csahk _ et al . _ studied the effect of multiplicative quenched disorder in the scaling properties of the interface @xcite . they found @xmath80 and @xmath81 , in close agreement with the experimental results of horvth _ et al . _ @xcite . on the continuum front , some attempts were made to explain the previous results . the first theoretical studies to introduce quenched disorder were performed independently by nattermann _ _ @xcite and narayan and fisher @xcite . analyzing the kardar - parisi - zhang ( kpz ) equation @xcite with quenched disorder , they argued that the dependence of the noise on @xmath82 introduces an infinite hierarchy of nonlinearities . they also noticed that in the annealed disorder case the kpz nonlinearity has a kinematic origin . thus @xmath83 , and @xmath0 should vanish at the depinning transition . for these reasons , they argue that at the depinning transition , it is sufficient to consider the equation @xcite @xmath84 where @xmath85 represents the quenched disorder . this equation was studied by means of the functional renormalization group @xcite , yielding to first order @xmath86 @xmath87 and @xmath88 [ theory ] where @xmath89 . hence @xmath90 in @xmath71 dimensions , in disagreement with many of the experimental and numerical results . reference @xcite also argued that all but ( [ theory]c ) are exact for all orders in @xmath91 . in view of the disagreement between the theoretical predictions of the ew equation with quenched disorder and the results presented in the previous subsections , csahk _ et al . _ numerically integrated the kpz equation with quenched disorder @xcite @xmath92 the numerical integration of ( [ qkpz ] ) for the _ moving _ phase @xcite yielded exponents in agreement with the calculations for the models in the dpd universality class . the fact that for ( [ qew ] ) we can calculate the exponents , and they do not agree with the numerical results obtained for ( [ qkpz ] ) , suggests that the nonlinear term @xmath93 may play a crucial role in determining the scaling properties of the interface . unfortunately no analytical results are available for eq . ( [ qkpz ] ) . in this section , we investigate the presence of the nonlinear term in several of the models discussed in the previous section @xcite . if we consider an equation of the kpz - type , and consider its average over the spatial coordinates , we obtain @xmath94 now , if we impose an average _ tilt _ to the interface , @xmath95 , and make the transformation @xmath96 we obtain @xmath97 so , it follows that the average velocity of the interface changes with the tilt as @xcite @xmath98 thus by varying the tilt @xmath99 , we can study the presence of nonlinear terms in the growth equation and calculate the effective coefficient @xmath0 of a given model @xcite . for discrete models the tilt can be easily implemented introducing _ helicoidal _ boundary conditions . in @xmath71 dimensions , we simply impose the condition , @xmath100 and @xmath101 . in @xmath102 dimensions , we use the same boundary conditions , so that the tilt occurs only in one direction . we begin by applying the method described above to the model introduced by tang and leschhorn @xcite . in @xmath71 dimensions , the interface becomes pinned by a dp cluster @xcite , and the critical dynamics is controlled by a divergent correlation length parallel to the interface @xmath103 , with @xmath104 . this model , referred to as `` dpd-1 '' , excludes overhangs , and gives rise to a self - affine interface at the depinning transition , with a roughness exponent @xmath64 . in the dpd-1 model , in @xmath71 dimensions , we start from a horizontal interface at the bottom edge of a lattice of size @xmath31 . at every site of the lattice we define a random , uncorrelated quenched variable , the noise @xmath16 , with magnitude in the range @xmath105 $ ] . during the time evolution of the interface , we choose one of the @xmath31 columns at random @xcite . if the difference in height to the lowest neighbor is larger than @xmath106 , this lowest neighboring column grows by one unit . otherwise , the chosen column grows one unit provided the noise on the site above the interface is smaller than the driving force @xmath17 . the unit time is defined to be @xmath31 growth attempts . we measure the velocity of the interface for different values of the reduced force @xmath107 and different values of the tilt @xmath99 . the results for @xmath71 dimensions are shown in fig . [ dpd1](a ) . for a fixed force @xmath107 , we find that the interface velocity depends on the tilt @xmath99 , indicating the existence of nonlinear terms . near the depinning transition ( @xmath108 ) , the velocity curves become `` steeper '' and , from ( [ vel1 ] ) , we infer that @xmath0 is coupled to the external force and that it increases as @xmath108 . to measure @xmath0 , we first attempt to fit a parabola to the tilt - dependent velocities in the vicinity of zero tilt . the calculations indicate that as we approach the depinning transition , @xmath0 diverges as @xcite @xmath109 however , in the vicinity of @xmath18 , the velocity curves lose their parabolic shape for large tilts ( see figs . [ dpd1](a ) and [ cross ] ) , indicating a change to a different universality class @xcite . we can understand the breakdown of ( [ vel1 ] ) for large @xmath99 using scaling arguments . substituting eqs . ( [ vel ] ) and ( [ phi ] ) into ( [ vel1 ] ) , we find @xcite @xmath110 equation ( [ velo ] ) indicates that the velocity curves corresponding to two different forces @xmath111 and @xmath112 , with @xmath113 , will intersect at a tilt @xmath114 ( see fig . [ cross ] ) . for tilts greater than @xmath114 , @xmath115 , a clearly unphysical prediction since @xmath113 , and since the average velocity , for the same tilt @xmath99 , should be larger for the larger force . thus the velocity can not follow a parabola for arbitrarily large @xmath99 , and a crossover to a different behavior than that of eq . ( [ velo ] ) must occur for values of the tilt larger than @xmath114 . letting @xmath116 , we find from ( [ velo ] ) that the crossing point of the two corresponding parabolas scales as @xmath117 equations ( [ velo ] ) and ( [ range ] ) motivate the scaling form for the velocities @xcite @xmath118 where @xmath119const for @xmath120 , and @xmath121 for @xmath122 @xcite . figure [ dpd1 ] shows the data collapse we obtain using ( [ scal ] ) , with the exponents @xmath123 for @xmath71 dimensions . below , we will show that an entire class of models produces similar results , which will allow us to group them into a single universality class . the scaling behavior ( [ scal ] ) is not limited to the dpd-1 model in @xmath71 dimensions . for @xmath65 dimensions and for the models introduced in refs . @xcite , we find a very similar behavior ( table [ tab1 ] ) . the `` dpd-2 '' model was introduced by buldyrev _ @xcite ( see also amaral _ et al . _ let us define the model for @xmath71 dimensions . in a square lattice of edge @xmath31 , assign an uncorrelated random number , the disorder @xmath124 , with magnitude uniformly distributed in the interval @xmath105 $ ] , to each cell @xmath125 . we compare the random pinning forces @xmath124 in the lattice with the driving force @xmath17 , where @xmath126 . if the pinning force at a certain cell @xmath124 is larger than the driving force , the cell is labeled `` blocked '' , otherwise it is labeled `` unblocked '' . thus a cell is blocked with a probability @xmath127 . since the model was developed to study imbibition , we will refer to the growing , invading , region as `` wet '' , and to the invaded region as `` dry '' . at time @xmath128 , we wet all cells in the bottom row of the lattice . then , we select a column at random @xcite , and wet all dry _ unblocked cells _ in that column that are nearest - neighbors to a wet cell . to obtain a single - valued interface , we impose the auxiliary rule that all dry _ blocked cells _ below a wet cell become wet as well @xcite . we refer to this rule as _ erosion of overhangs_. the time unit is defined as @xmath31 growth attempts . figure [ dpd2 ] shows the dependence of the velocity on the tilt for the `` dpd-2 '' model and the respective data collapse according to ( [ scal ] ) , for @xmath71 dimensions . the values of the exponents in the data collapse are @xcite @xmath129 finally , we simulate the model proposed by amaral @xcite , which we will refer to as `` dpd-3 '' . we simulate it in @xmath71 dimensions , and rescale the tilt dependent velocities according to ( [ scal ] ) using the exponents ( fig . [ dpd3 ] ) @xmath130 three questions are raised by these results : _ ( i ) _ is the new exponent @xmath131 an independent exponent or can it be expressed in terms of the other known exponents ? _ ( ii ) _ what are the scaling properties of the tilted interface ? _ ( iii ) _ what are the reasons for the divergence of @xmath0 at the depinning transition ? questions _ ( i ) _ and _ ( ii ) _ have recently been answered by tang , kardar and dhar ( tkd ) @xcite . they noticed that for the dpd class of models there is a characteristic slope @xmath132 where @xmath133 and @xmath134 are , respectively , the perpendicular and parallel correlation lengths of a dp cluster ( see appendix a ) . for values of the average tilt smaller than @xmath135 , the interface still belongs to the dpd universality class . however , for larger values of the tilt a qualitative change should occur . but we showed above that such a change occurs for @xmath136 , so we conclude that @xmath137 , from which follows @xcite @xmath138 combining ( [ e : t+p ] ) with ( [ vel ] ) , we find the scaling law @xmath139 tkd also showed that for @xmath136 a _ new _ universality class is obtained @xcite , and they were able to relate its scaling exponents to the exponents of the kpz equation with annealed disorder @xcite @xmath140 to address question _ ( iii ) _ , let us consider the effect of the nonlinear term . its role is to make the interface grow wherever it has a nonzero slope , so it corresponds to what we can refer to as lateral growth . when an interface is moving , we can consider two types of growth , forward growth and lateral growth . the meaning of @xmath141 is then the irrelevance of forward growth compared with lateral growth when we approach the depinning transition . these ideas were studied quantitatively by makse @xcite , who found that , near the depinning transition , the probability of occurrence of lateral growth vanishes much slower than the probability of occurrence of forward growth . furthermore , makse argued that the application of constraints to the local slopes of the interfaces , as is the case for the dpd class of models , appears to be an _ essential _ ingredient for @xmath0 to diverge @xcite . we mentioned in sec . iii that several authors numerically integrated eq . ( [ qew ] ) , and others developed models corresponding to the discretization of this equation . thus , we expect results in agreement with the theoretical calculations . furthermore , we do not expect , for these models , to detect any dependence of the velocity on the tilt of the interface . in fact , we have simulated the model introduced by leschhorn @xcite , which we will refer to as `` qew-1 '' , and find that for any reduced force , @xmath1 ( fig . [ qew1 ] ) . a hamiltonian model , which we will refer to as `` qew-2 '' , was introduced by makse @xcite . in @xmath71 dimensions , the hamiltonian is defined as @xmath142 . \label{hamilt}\ ] ] here , the first term represents the elastic energy that tends to smooth the interface , and @xmath143 is an uncorrelated random number which mimics a random potential due to the disorder of the medium . in the simulation , a column @xmath125 is chosen and its height is updated to @xmath144 if the change in ( [ hamilt ] ) is negative . thus , only motions that decrease the total energy of the system are accepted . backward motions are neglected since these are rare events . equation ( [ qew ] ) can be easily derived from this hamiltonian simply by considering that @xmath145 we study the dependence of the velocity with the tilt for this model and find that @xmath1 for any force , just as for the qew-1 model . an interesting result , presented in sec . iii , is the reasonable agreement between the exponents measured for the fim and the rfim @xcite with the predictions of ( [ qew ] ) . since these models are not simple discretizations of eq . ( [ qew ] ) , it is interesting to investigate the presence of any nonlinear term . for this reason , we study the rfim in both @xmath71 and @xmath65 dimensions . this model allows for overhangs , and for certain values of its parameters can be mapped to the isotropic percolation problem @xcite . in the rfim , spins on a square lattice interact through the hamiltonian @xmath146s_i , \label{rfim}\ ] ] where @xmath147 , @xmath17 now denotes the external magnetic field , and @xmath148 is the time - independent local random field ( i.e. , quenched noise ) whose values are uniformly distributed in the interval @xmath149 $ ] . the strength of the quenched disorder is characterized by the parameter @xmath150 . at time zero , all spins are `` down'' except those in the first row , which are initially `` up '' . the interface consists of all down spins that are nearest - neighbors to an up spin . during the time evolution of the system , we flip any down spin that belongs to the interface and is `` unstable , '' i.e. , whenever the flip will lower the total energy of the system . the control parameter of the depinning transition is the external magnetic field @xmath17 ; the unit time corresponds to the parallel flipping of all unstable spins @xcite . for @xmath71 dimensions , there are two morphologically - different regimes , depending on the strength @xmath150 of the disorder ( i.e. , of the random fields ) . for @xmath151 , the interface is self - similar ( ss ) , while for @xmath152 it is faceted or flat ( fa ) . for @xmath65 dimensions , there is again a fa regime ( @xmath153 ) , a ss regime ( @xmath154 ) , and also a self - affine ( sa ) regime in between ( @xmath155 ) @xcite . the sa regime , which exists only for the case of @xmath65 dimensions , is the only regime of the rfim for which either eqs . ( [ qew ] ) or ( [ qkpz ] ) could apply . in the ss regime , overhangs in the interface are larger enough for the approximation of small slopes not to be valid . on the other hand , in the fa regime , lattice effects dominate the growth @xcite . for the sa and ss regimes we find that ( [ scal ] ) is still valid ; however , we find a negative @xmath131 @xmath156 this behavior can be understood , for the ss regime , by considering that near the depinning transition , the morphology of the interface corresponds to the hull of an isotropic percolation cluster , which has no well - defined orientation @xcite . thus a change in the boundary conditions will not affect the growth process , and we can not expect any divergence of a possible nonlinear term when the magnetic field approaches its critical value . on the other hand , for large fields the effect of the quenched disorder diminishes , and we can observe an average interface orientation . for such values of field , we expect the presence of nonlinear terms , of _ kinematic _ origin , to be relevant . although for the sa regime the behavior of @xmath0 is similar ( fig . [ rfim2 ] ) , we do not have a simple argument that would explain the observed behavior . these results lead us to conclude that in the sa regime the rfim belongs to the universality class of eq . ( [ qew ] ) . this conclusion is further supported by the agreement between the numerically determined exponents , @xmath157 and @xmath158 for @xmath65 dimensions , and the theoretical predictions of ( [ qew ] ) . the results of table [ tab1 ] show , for @xmath71 dimensions , a separation into two groups in the values of the critical exponents for the six models studied @xcite . this separation reflects the existence of two distinct universality classes , described by the two continuum growth equations , ( [ qew ] ) and ( [ qkpz ] ) . for the qew-1 , qew-2 models , and for the rfim in the sa regime , we find that either @xmath1 or @xmath159 at the depinning transition ( fig . [ f : lambda ] ) . thus , the scaling behavior of these models should be correctly described by ( [ qew ] ) . for the dpd-1 , dpd-2 , and dpd-3 models we observe a divergent @xmath0 , indicating that nonlinearities are relevant near the depinning transition . thus to properly describe the scaling properties of these models it is necessary to study ( [ qkpz ] ) , since ( [ qew ] ) does not include the nonlinear term @xmath93 . further evidence of the existence of the two universality classes is given by the values of the roughness exponents . the models for which @xmath0 diverges at the depinning transition , have @xmath64 , in agreement with the mapping to dp @xcite . on the other hand , models in the universality class of eq . ( [ qew ] ) , gave roughness exponents typically larger , and in better agreement with the theoretical prediction , @xmath73 . it is worthwhile to discuss the implications of the previous results on other models and theoretical approaches studied in the literature . recently , considerable attention has been focused on a self - organized version of the dpd model , the so - called self - organized depinning ( sod ) model @xcite . in this model , the constant driving force of the dpd model is replaced by an algorithm that selects the site on the interface for which the disorder is weakest i.e , it selects the point in the interface offering the least resistance . thus , the local update rule of the dpd model is replaced by a global search process , similar to the one used in invasion percolation . this update rule assures that the model will evolve to criticality and that it will remain in the critical state . this implies that the sod model is always at the depinning transition . thus the properties of the sod model are described by the dpd model at criticality , which implies , since the sod model is always at criticality , that the sod model can not be described by a continuum equation of the kpz - type , since this would require an infinite coefficient of the nonlinear term . however , this argument _ does not _ rules out that the sod model can be described by a continuum equation that upon a mapping to the quenched kpz equation will generate an effective coefficient @xmath0 that would diverge @xcite . in the previous section we identified the universality classes for interface roughening in the presence of quenched disorder . however , the experimental situation still leaves us with several unsolved puzzles , such as the change in the roughness exponent or the dependence of the global width on the driving force for the fluid - fluid displacement experiments . in trying to shed some light over the remaining problems , we observe that most of the numerical and the analytical studies focus on the `` pinned phase '' ( @xmath160 ) while nearly all experimental results are for the `` moving phase '' ( @xmath20 ) . a logical next step is therefore to study the moving phase for models representative of each of the two universality classes . let us start by considering the dpd universality class . in fig . [ sdpd](a ) , we show the local width for the pinned and the moving phases . we find @xmath161 for the pinned phase , and @xmath162 for the moving phase ( table [ tab_exp ] ) . so far , no explanation for this change in roughness exponent at the depinning transition is available . the study of the scaling of the local width for the moving phase reveals a _ novel _ dependence of the prefactor of the width on the external force . we also find that the exponents characterizing this anomalous dependence change values for the two different regimes . in the first @xmath163 where @xmath27 is the correlation length , and @xmath164 is a new exponent that characterizes the dependence of the prefactor of the width on the driving force . in the second regime , the effect of the quenched disorder becomes irrelevant compared to the annealed disorder @xcite , and we obtain @xmath165 where @xmath166 is a new exponent , and @xmath167 is the roughness exponent corresponding to annealed disorder . depending on the absence or presence of nonlinear terms , we recover the results of either the ew or the kpz equations with annealed disorder . this crossover was observed both for experiments and simulations of discrete models @xcite . due to the dependence of the prefactor of the width on the driving force , and since @xmath103 , we propose the scaling _ ansatz _ @xmath168 upon comparison with ( [ mov1 ] ) and ( [ mov2 ] ) , we find that the scaling function @xmath169 satisfies @xmath170 const , and @xmath171 . using ( [ scal_w ] ) we also obtain @xmath172 in fig . [ sdpd](b ) we show the data collapse of the widths for different forces and length scales . the deviations from scaling for large values of @xmath173 are due to finite - size effects ( see fig . [ sdpd](d ) for discussion ) . while previous studies considered @xmath174 as an effective exponent , and concluded that no self - affine scaling exists in the moving phase for @xmath175 @xcite , we find a consistent value for @xmath174 regardless of how closely we approach the depinning transition . the self - affinity of the interface is supported by the good data collapse obtained with ( [ scal_w ] ) , as shown in fig . [ sdpd ] . to derive a second relation for the new exponent @xmath164 , let us consider the approach to the depinning transition for a system of size @xmath31 . for a finite system , the transition does not occur for @xmath176 , as it would for an infinite system , but rather for an effective critical force @xmath177 such that @xmath178 , implying @xmath179 . using this result and ( [ mov1 ] ) , we obtain @xmath180 for the pinned phase we have @xmath181 at the depinning transition , ( [ mov ] ) and ( [ pin ] ) should scale in the same way . if we replace @xmath182 by @xmath31 we obtain @xcite @xmath183 replacing the known values of @xmath29 , @xmath11 , @xmath174 , and @xmath167 into ( [ chi1 ] ) and ( [ chi3 ] ) , we find @xmath184 to be compared with the measured values @xmath185 and @xmath186 ( table [ tab_exp ] ) . although the agreement with the measured values is not perfect , the error bars do not rule out the validity of ( [ chi3 ] ) . for the dpd universality class , we find that @xmath35 changes value at the depinning transition ( see table [ tab_exp ] ) . we find @xmath187 for the pinned phase and @xmath188 for the moving phase . we obtain for both sides of the transition @xmath189 , implying that @xmath190 remains unchanged at the transition @xcite . the exponent @xmath190 characterizes the time scale @xmath36 for the propagation of correlations in the interface ( cf . ( [ scl1 ] ) ) . this time scale is not expected to depend on the external force @xcite . these results are in good agreement with a numerical integration of eq . ( [ qkpz ] ) @xcite . for the time evolution of the width , we again find an anomalous dependence of the width on the reduced for the two different scaling regimes . in regime i , @xmath191 where @xmath192 is the characteristic time for the propagation of correlations over a distance @xmath27 , and @xmath193 is a new exponent that characterizes the dependence of the prefactor of the width on the driving force . in regime ii , the effect of the pinned disorder becomes irrelevant compared to the annealed disorder @xcite , and we obtain @xmath194 where @xmath195 is a new exponent , and @xmath196 is the growth exponent corresponding to annealed disorder . since @xmath197 , we propose the scaling _ ansatz _ @xmath198 upon comparison with ( [ tmov1 ] ) and ( [ tmov2 ] ) , we find that the scaling function @xmath169 satisfies @xmath170 const , and @xmath199 . from ( [ scal_wt ] ) we also obtain @xmath200 in fig . [ tdpd ] we display the rescaling of our simulation results for the dependence of the global width on time and driving force according to ( [ scal_wt ] ) . to determine a second relation for the new exponent @xmath193 , let us again consider the approach to the depinning transition for a system of size @xmath31 . as discussed above , the transition occurs for an effective critical force such that @xmath201 , implying @xmath202 . using this result and ( [ tmov1 ] ) , we obtain @xmath203 for the pinned phase , we have @xmath204 at the transition , ( [ tmov ] ) and ( [ tpin ] ) must scale in a similar way . so , if we replace @xmath205 by @xmath36 , we obtain @xmath206 replacing the known values of @xmath29 , @xmath190 , @xmath207 , @xmath208 , and @xmath196 into ( [ kappa1 ] ) and ( [ kappa3 ] ) , we find @xmath209 in good agreement with the simulation results , @xmath210 and @xmath211 ( table [ tab_exp ] ) . for the study of the qew universality class , we focus on the qew-2 model . we observe a dependence of the local width on the system size @xmath31 , that can be described by the scaling function @xcite @xmath212 where @xmath213 is a scaling function that for @xmath45 scales as @xmath214 , and @xmath215 is the global roughness exponent , which is defined by the scaling of the global width with the system size : @xmath216 . equation ( [ self ] ) is valid only when @xmath217 , and it implies that @xmath218 for @xmath219 . hence , the local slopes are unbounded , i.e. , they diverge with the system size . this anomalous characteristic of the qew models is displayed in fig . [ f : anom](a ) , where we plot interfaces generated for the qew-2 model for different values of the system size . we calculate the global width as a function of @xmath31 and find at the depinning transition that @xmath220 . we also determine the scaling of the local width @xmath47 in a window of size @xmath182 , for different values of the reduced force . in fig . [ sqew](a ) , we show the local width for the pinned and the moving phases . analysis of consecutive slopes yield roughness exponents @xmath221 for the pinned phase , and @xmath222 for the moving phase ( table [ tab_exp ] ) . these results are in agreement with values commonly found for the qew class of models when the roughness exponent is calculated from the local width i.e. , @xmath223 . the slightly smaller value of our estimates is is likely due to finite size effects . as for the dpd universality class , the study of the local width in the moving phase reveals an anomalous dependence on the reduced force , which leads to eqs . ( [ scal_w ] ) and ( [ chi1 ] ) being also valid for the qew universality class . in fig . [ sqew](b ) , we show the data collapse of the local widths obtained using ( [ scal_w ] ) . to determine a second relation for the new exponent @xmath164 , we again consider the approach to the depinning transition of a system of size @xmath31 . as before , we obtain ( [ mov ] ) for the moving phase . however , for the pinned phase , the scaling behavior ( [ self ] ) , leads to @xmath224 at the transition , ( [ mov ] ) and ( [ pin0 ] ) must be identical . hence , we find @xmath225 and @xmath226 [ chi2 ] replacing the measured values of @xmath29 , @xmath215 , @xmath11 , and @xmath167 into ( [ chi1 ] ) and ( [ chi2 ] ) , we find @xmath227 in good agreement with the values obtained in our simulations , @xmath228 and @xmath229 ( table [ tab_exp ] ) . we study the scaling of @xmath33 with @xmath205 , and find @xmath230 and @xmath231 . these results imply that @xmath190 remains unchanged at the transition . in fig . [ tqew ] , we show the scaling of the global width as a function of time for different values of the reduced force , and its data collapse using ( [ scal_wt ] ) . for the qew universality class , we find by replacing the measured values of @xmath29 , @xmath190 , @xmath207 , and @xmath196 into ( [ kappa1 ] ) and ( [ kappa3 ] ) that @xmath232 in good agreement with the values obtained in our simulations , @xmath233 and @xmath234 ( table [ tab_exp ] ) . the results obtained in the previous subsections allow us to discuss the experimental and numerical results reported in the literature . a problem with the interpretation of experimental and numerical results for the roughness exponent , has been the wide range of values for @xmath39 : @xmath235 , measured in the moving phase . we note that , in this regime , the crossover to the annealed disorder regime leads to `` effective '' exponents that change with the velocity ( or the driving force ) . for this reason , we suggest that the scaling function ( [ scal_w ] ) might be useful in the determination of the exponents from the study of the local width @xmath47 . as shown in table [ tab_exp ] , the exponent @xmath164 has different signs for the two universality classes , leading to sharply distinct scaling behaviors for the prefactor of the width with the driving force . since in many experiments it is possible to monitor the velocity of the interface , and therefore the driving force , the study of this prefactor may also lead to an easier identification of the universality class to which the experiments belong . in light of this discussion , the interpretation of the results of ref . @xcite is clear . the numerical integration of the ew equation with quenched disorder performed in ref . @xcite must belong to the qew universality class , and the reported exponents are `` effective '' exponents whose values were affected by the crossover to the annealed regime . the determination of the universality class of the _ fluid invasion model _ ( fim ) of ref . @xcite is not trivial . however , we note that @xmath29 is identical for the fim and for the hamiltonian model . on the other hand , the value of @xmath39 measured in ref . @xcite is somewhat smaller than the value obtained for the qew model using the local width . regarding the fluid - fluid displacement experiments of refs . @xcite , we find that they share several scaling properties with the hamiltonian model . we first notice that the range of roughness exponents measured in the experiments of refs . @xcite is consistent with the values that could be obtained with the hamiltonian model . moreover , rubio _ et al . _ @xcite found that the amplitude of the width decreases with the increase of the velocity . this property is entirely consistent with our findings for the qew universality class : if we replace @xmath236 in ( [ mov1 ] ) , then we obtain that the width decreases with @xmath22 as @xmath237 . the imbibition experiments of refs . @xcite and the paper burning experiments of ref . @xcite are believed to belong to the dpd universality class . this conclusion is supported by the values of the exponents measured in both experiments . the reason for the different universality classes for the two groups of experiments , fluid - fluid displacement on one hand and imbibition and paper burning on the other , is not totally clear . however , a possible explanation might be the importance of lateral growth in the second case compared with the first . when we analyze the imbibition and paper burning experiments , it becomes clear that the interface can be described by a single - valued interface , because only the highest position of the interface in each column is relevant for the dynamics . this effectively corresponds to eroding any overhangs and , as shown by makse @xcite , leads to a diverging @xmath0 at the transition . on the other hand , in the fluid - fluid displacement experiments , we must at all times consider the full interface , because of the trapped fluid that might exist below the interface . since it will require some effort to displace this fluid or to separate it from the rest of the displaced fluid , we can not physically justify a rule such as the erosion of overhangs . this implies that @xmath0 vanishes at the depinning transition , and the experiments must belong to the qew universality class . to summarize , we perform a systematic study of several models proposed to understand the motion of driven interfaces in disordered media . we are able to identify _ two _ distinct universality classes . for one of these universality classes , qew , we observe that it can be described by a langevin equation similar to the ew equation but with quenched disorder . for the qew universality class nearly all exponents for the pinned phase can be obtained by renormalization group calculations . for the other universality class , dpd , we find that it can be described by a langevin equation similar to the kpz equation , but with quenched disorder . furthermore , we find that the coefficient of the nonlinear term @xmath0 diverges at the depinning transition . because of this divergence , no perturbative analytical calculations have been possible so far . however , a mapping of the static properties of the interface , at the depinning transition , to dp yields all static exponents of the problem . a mapping of the dynamics to isotropic percolation yields the dynamical exponent . the identification of these two universality classes leads to some understanding of many of the results obtained previously for models and experiments . however , many other results could not be understood in terms of the exponents obtained for the two universality classes _ at _ the depinning transition . for this reason , we also investigated the scaling properties of the models in the two universality classes _ above _ the depinning transition . we find that for the dpd universality class , @xmath39 and @xmath35 change their values at the depinning transition . furthermore , we find a dependence of the prefactor of the width on the driving force . the new exponent @xmath164 , characterizing the scaling of this prefactor , can be used to relate the values of the roughness exponents on both sides of the transition . for the qew universality class , we find that @xmath39 and @xmath35 remain unchanged at the depinning transition . as for the dpd universality class , we also find a dependence of the prefactor of the width on the driving force . the exponent @xmath164 is , in this case , found to relate the different values of the local , @xmath11 , and global , @xmath215 , roughness exponents at the depinning transition . these results provide us with a more consistent understanding of the scaling properties of the two universality classes both at and above the depinning transition . this knowledge enables us to interpret most of the experimental results obtained previously . however , several questions , that demand an answer , are still unsolved . one of them is the reason for the change in values of the roughness exponent at the depinning transition for the dpd universality class . another is the understanding of the different values of @xmath215 and @xmath11 for the qew universality class . we thank s. v. buldyrev , r. cuerno , s. t. harrington , s. havlin , v. k. horvth , p. ch . ivanov , k. b. lauritsen , h. leschhorn , i. procaccia , p. rey , r. sadr , s. tomassone , t. vicsek , and p .- z . wong for valuable contributions and discussions . l.a.n.a . acknowledges support from junta nacional de investigao cientfica e tecnolgica . the center for polymer studies is supported by the national science foundation . near @xmath238 , the size of dp clusters is characterized by a longitudinal ( parallel ) correlation length @xmath134 and a transverse ( perpendicular ) correlation length @xmath133 that diverge as @xcite @xmath239 the parallel and perpendicular correlation length exponents for dp clusters have been calculated @xcite , with the results @xmath240 the mapping of the _ pinned interface _ to dp enables us to estimate the static exponents of this problem from the characteristic exponents of dp clusters . the characteristic length @xmath27 of the pinned regions must be of the order of @xmath134 , so we can identify the exponent @xmath29 to be @xmath241 the global width @xmath37 of the pinned interface should scale as @xmath133 , since its advance is blocked by a dp path . on the other hand , @xmath134 must be larger than the system size @xmath31 for the interface to become pinned , from which follows @xcite @xmath242 comparing with ( [ wsat ] ) , we conclude that the roughness exponent is given in terms of the correlation exponents for dp , @xmath243 substituting ( [ exp0 ] ) into ( [ alp ] ) , we predict @xmath244 for any fractal set with a fractal dimension smaller than the dimension of the embedding space , the distance between to points of the set @xmath182 is not given by the euclidean distance @xmath245 between those points but by the minimum path distance ( or chemical distance ) @xmath246 . numerical studies show that @xmath246 scales as @xcite @xmath247 where @xmath248 is called the _ minimum path _ exponent . recently , havlin _ et al . _ showed that the dynamics of the dpd model in @xmath102 dimensions could be mapped to the minimum path of isotropic percolation in @xmath13 dimensions @xcite . they argued that close to the depinning transition only a few cells on the interface are active , i.e. , unblocked . these active cells perform a global search of unblocked cells and propagate correlations in the interface . since the cells through which the active cells can move are confined to unblocked or eroded blocked cells in a region of thickness @xmath249 , we can identify this region with an isotropic percolation cluster embedded in a space of dimension @xmath13 . thus , the path through which active cells move is the minimum path . so , the time @xmath205 for the correlations to propagate an euclidean distance @xmath245 , is proportional to @xmath246 , which scales as @xmath250 for isotropic percolation . from this follows that we can connect the dynamical exponent @xmath190 to the static exponent @xmath248 of _ isotropic _ percolation @xcite @xmath251 in this appendix , we discuss the model introduced by parisi to study interface roughening in the presence of quenched disorder @xcite . we show that the original growth rule leads to an unphysical behavior . we then proceed to describe the model proposed by amaral that solves the problems with the original model @xcite . let us start by describing the model originally introduced by parisi . in a square lattice of edge @xmath31 , with periodic boundary conditions in that direction , we assign to each cell an uncorrelated random number , uniformly distributed in the interval @xmath252 $ ] . we start , at @xmath253 , by invading the bottom row in the lattice , and then , we calculate the driving force for each column @xmath254 , \label{force}\ ] ] where @xmath255 is the controlling parameter of the model . if this driving force @xmath256 is larger than the noise in the cell above the interface , @xmath257 , then we update the height of column @xmath125 to @xmath258 . \label{grow1}\ ] ] the time evolution of this model leads to the development of huge jumps in the height of neighboring columns that propagate along the system . for small enough values of @xmath255 , only one or two such growing regions will be propagating in the system . the reason for such `` unnatural '' behavior of the model is the growth rule ( [ grow1 ] ) . the fact that the advancing column grows to an higher value than the highest neighbor is a kind of `` bootstrap '' that is difficult to justify . in fact , it does not seem reasonable that any growth mechanism would pull a lower column to an height larger than that of its neighbors . for this reason , amaral proposed a new growth rule to replace ( [ grow1 ] ) @xmath259 in this way , the invading region only advances one cell at a time ; thus , avoiding the formation of big jumps between neighboring columns @xcite . the study of this model , which we will refer to as `` dpd-3 '' , leads to the result that a critical value of the controlling parameter exists , @xmath260 . below this threshold the interface stops moving after some finite time , and above it it moves indefinitely . to determine some of the scaling exponents , we applied the gradient method , introduced by sapoval _ @xcite , and already used in the study of the dpd-2 model @xcite and in several other studies @xcite . figure [ grad ] displays our results for the case of a linear gradient , @xmath261 . the scaling of the interface changes with the concentration rate and is characterized by an exponent @xmath262 according to @xmath263 where @xmath107 is an universal scaling function that satisfies @xmath264 const , and @xmath265 . figure [ grad](a ) shows the width @xmath47 as a function of the size @xmath182 for different values of the concentration gradient @xmath266 . figure [ grad](b ) shows the data collapse of these results according to ( [ scaling ] ) . the best collapse was obtained with the exponents @xmath267 since we know that @xcite @xmath268 we obtain @xmath269 in agreement with the known results for dp @xcite . t. vicsek , _ fractal growth phenomena _ , 2nd ed . , part iv ( world scientific , singapore , 1992 ) ; _ dynamics of fractal surfaces _ , edited by f. family and t. vicsek ( world scientific , singapore , 1991 ) ; j. kertsz and t. vicsek , in _ fractals in science _ , edited by a. bunde and s. havlin ( springer - verlag , heidelberg , 1994 ) . m. a. rubio , c. a. edwards , a. dougherty , and j. p. gollub , phys . lett . * 63 * , 1685 ( 1989 ) ; v. k. horvth , f. family , and t. vicsek , _ ibid . _ * 65 * , 1388 ( 1990 ) , m. a. rubio , c. a. edwards , a. dougherty , and j. p. gollub , _ ibid . _ * 65 * , 1389 ( 1990 ) . s. havlin , a .- barabsi , s. v. buldyrev , c. k. peng , m. schwartz , h. e. stanley , and t. vicsek , in _ growth patterns in physical sciences and biology _ [ proc . 1991 nato advanced research workshop , granada ] , ed . j. m. garcia - ruiz , e. louis , p. meakin , and l. m. sander ( plenum press , new york , 1993 ) . s. v. buldyrev , a .- barabsi , s. havlin , j. kertsz , h. e. stanley and h. s. xenias , physica a * 191 * , 220 ( 1992 ) ; a .- barabsi , s. v. buldyrev , s. havlin , g. huber , h. e. stanley , and t. vicsek , in _ surface disordering : growth , roughening and phase transitions _ , ed . r. jullien , j. kertsz , p. meakin , and d. e. wolf ( nova science , new york , 1992 ) . e. medina , t. hwa , m. kardar , and y .- c . zhang , phys . a * 39 * , 3053 ( 1989 ) ; c. k. peng , s. havlin , m. schwartz , and h. e. stanley , phys . a * 44 * , 2239 ( 1991 ) ; j. g. amar , p .- m . lam , and f. family , phys . a * 43 * , 4548 ( 1991 ) . zhang , j. physique * 51 * , 2129 ( 1990 ) ; s. v. buldyrev , s. havlin , j. kertsz , h. e. stanley , and t. vicsek , phys . a * 43 * , 7113 ( 1991 ) ; s. havlin , s. v. buldyrev , h. e. stanley , and g. h. weiss , j. phys . a * 24 * , l925 ( 1991 ) . c. s. nolle , b. koiller , n. martys , and m. o. robbins , phys . lett . * 71 * , 2074 ( 1993 ) ; b. koiller , m. o. robbins , h. ji , and c. s. nolle , in _ new trends in magnetic materials and their applications _ , edited by j. l. moran - lopez and j. m. sanchez ( plenum , new york , 1993 ) ; m. o. robbins , m. cieplak , h. ji , b. koiller , and n. martys , in _ growth patterns in physical sciences and biology _ , edited by j. m. garcia - ruiz , e. louis , p. meakin , and l. m. sander ( plenum , new york , 1993 ) . we introduce this time - dependent randomness in the model to mimic the experimental fact that for @xmath20 , a transition to the kpz behavior with annealed disorder is observed for @xmath270 . in fact , eqs . ( 3.1 ) and ( 3.3 ) should also include annealed disorder terms . however , this annealed disorder is much weaker than the quenched disorder . for this reason , for @xmath175 the presence of time dependent disorder is irrelevant . a different situation occurs for @xmath270 . in this case the effect of quenched disorder is no longer relevant , and even a week annealed disorder becomes important . the justification of this particular rule lies not only on the computational simplifications that it affords but also on the experimental finding that overhangs do not play an important role in the scaling properties of the interface for imbibition . as was shown in refs . @xcite , the dpd-1 and dpd-2 models belong to the same universality class , with exponents @xmath64 and @xmath271 @xcite . the deviation of the numerical values in the exponents for the dpd-2 model can be explained by the slow convergence of this model to the asymptotic scaling regime . the large error bars for all models arise from the fact that simulations were done for relatively small systems . despite the differences in the values of the scaling exponents , based on the numerical results combined with the exact results of refs . @xcite , we conclude that the models , dpd-1 , dpd-2 , and dpd-3 , share the same set of scaling exponents , thus belonging to the same universality class . as was noted by koiller _ @xcite , different methods of defining time in the rfim lead to different values of the exponents . here we applied one of the methods suggested in @xcite , in which the unit time corresponds to flip all unstable interface spins . the value of the exponent given in @xcite was obtained assuming that the time it takes a spin to flip depends on the energy gain , for the system , in the flipping . this mapping to dp is valid only for one - dimensional interfaces , for higher dimensions the properties of these models are related to _ directed surfaces _ , of which dp is a particular case , for a discussion see @xcite . our results show that for the fa regime , the rfim behaves in a similar fashion to the models in the dpd universality class , in that the coefficient of the nonlinear term diverges at the depinning transition . however , we stress that this regime has little physical motivation , being its existence due to lattice effects . since we want to calculate the scaling exponents we only need to consider the leading order terms . hence , although eqs . ( [ mov ] ) and ( [ pin ] ) are only valid for @xmath219 , we can , to leading order , compare the two expressions when @xmath272 . reference @xcite measured @xmath273 . however , the large intrinsic width @xcite of their variant of the dpd model , leads the authors of ref . @xcite to add a constant to their data , which probably altered the scaling of @xmath33 for @xmath274 . llcccc & & + & & @xmath25 & @xmath131 & @xmath25 & @xmath131 + dpd-1 & & @xmath275 & @xmath275 & @xmath276 & @xmath277 + dpd-2 & & @xmath278 & @xmath279 & & + dpd-3 & & @xmath280 & @xmath281 & & + qew-1 & & @xmath282 & & & + qew-2 & & @xmath283 & & & + & fa & @xmath284 & @xmath285 & & + rfim & sa & & & @xmath286 & @xmath287 + & ss & @xmath288 & @xmath289 & & + . critical exponents for the two universality classes studied in this paper . the subscript `` @xmath290 '' refers to the _ pinned _ phase , the subscript `` @xmath291 '' denotes the quenched disorder regime in the _ moving _ phase , and the subscript `` @xmath292 '' refers to the _ annealed disorder _ regime . the exponents @xmath190 and @xmath293 were calculated from scaling relations , while the remaining exponents were calculated directly in the simulations . [ cols="<,^,^ " , ]

this study investigates a multiplicative integer system by using a method that was developed for studying pattern generation problems . the entropy and the minkowski dimensions of general multiplicative systems can thus be computed . a multi - dimensional decoupled system is investigated in three main steps . ( i ) identify the admissible lattices of the system ; ( ii ) compute the density of copies of admissible lattices of the same length , and ( iii ) compute the number of admissible patterns on the admissible lattices . a coupled system can be decoupled by removing the multiplicative relation set and then performing procedures similar to those applied to a decoupled system . the admissible lattices are chosen to be the maximum graphs of different degrees which are mutually independent . the entropy can be obtained after the remaining error term is shown to approach zero as the degree of the admissible lattice tends to infinity . the first author would like to thank the national science council , r.o.c . ( contract no . nsc 100 - 2115-m-259 - 009-my2 ) and the national center for theoretical sciences for partially supporting this research . ] the second author would like to thank the national science council , r.o.c . and the st yau center for partially supporting this research . ] the third author would like to thank the national science council , r.o.c . ( contract no . nsc 98 - 2115-m-009 ) and the st yau center for partially supporting this research . ]

1207.7154

multiplicative integer systems have been intensively studied in recent years ; see @xcite and the references therein . one of the main related issues is to compute minkowski ( box ) dimension and hausdorff dimension of such systems and to compare them . these two dimensions are equal in an additive shift . however , for most known examples of multiplicative integer systems , they are different . since the computations of these two dimensions are difficult , effective methods for computing these dimensions for general multiplicative systems must be developed . this investigation is motivated directly by the work of kenyon _ et al . _ @xcite , who utilized a variational method to obtain the results on @xmath0 and also pointed out that the method fails for the system @xmath1 this work provides an approach to general multiplicative systems , including ( [ eqn:1.1 ] ) and ( [ eqn:1.0 - 2 ] ) . this study emphasizes the computation of entropy @xmath2 of the multiplicative system @xmath3 . let @xmath4 be the number of the symbols of system @xmath3 . the minkowski dimension @xmath5 is given by @xmath6 where @xmath7 @xmath8 is the set of @xmath9-blocks in @xmath3 and @xmath10 is the number of @xmath8 . definition ( [ eqn:1.0 - 4 ] ) is standard for symbolic dynamical systems @xcite , and it specifies the growth rate of @xmath10 . however , @xmath3 is not invariant under the shift map @xmath11 . bowen @xcite used @xmath12-spanning and @xmath12-separation sets to define the topological entropy @xmath13 for an arbitrary set @xmath14 in a topological dynamical system @xmath15 . if @xmath16 is a shift space with the shift map @xmath17 , then the topological entropy @xmath18 equals the entropy @xmath19 @xcite . hence , @xmath20 . recently , feng and huang @xcite defined several topological entropies of subsets @xmath14 in a topological dynamical system @xmath15 . in particular , they defined the upper capacity topological entropy @xmath21 and packing topological entropy @xmath22 . the variational principle for those entropies are proved therein . for any @xmath23 , @xmath24 @xcite . if @xmath14 is @xmath25-invariant and compact , then they are coincident . the multi - dimensional shifts of finite type have been studied intensively ; see @xcite and the references therein . among them , the authors studied pattern generation problems on multi - dimensional shifts of finite type and developed some efficient means of studying the generation of admissible patterns , and then computing the topological entropy ; see @xcite . this study shows that these methods can be used to study the multi - dimensional decoupled systems and one - dimensional coupled systems of multiplicative integers , including @xmath26 . to illustrate our method , ( [ eqn:1.1 ] ) is investigated first . the topological entropy @xmath27 has been shown to be @xmath28 where @xmath29 is a fibonacci number with @xmath30 , @xmath31 and @xmath32 for all @xmath33 @xcite . the derivation of ( [ eqn:1.2 ] ) is as follows . denote by the multiplicative relation set @xmath34 of the integers of 2-power , i.e. , @xmath35 denote by the complementary index set @xmath36 of @xmath37 that contains all positive odd integers : @xmath38 the set of all natural number @xmath39 can now be rearranged into @xmath40 where @xmath41 . clearly , @xmath42 if @xmath43 and @xmath44 . more precisely , the right - hand side of ( [ eqn:1.5 ] ) can be expressed as an @xmath45 table , table 1.1 . @xmath46{m2.eps } \end{array}\ ] ] @xmath47 on integer lattice @xmath48 , for @xmath49 , a @xmath50-lattice @xmath51 can be represented by @xmath50-cells as fig . 1.1 ( a ) for drawing numbers or @xmath50-vertices as fig . 1.1 ( b ) for drawing graph latter . @xmath52{zk.eps } & & & \psfrag{c}{$z_{k}$ } \includegraphics[scale=0.7]{z_k_2.eps } \\ \text{figure 1.1 ( a ) . } & & & \text{figure 1.1 ( b ) . } \end{array}\ ] ] let @xmath53 and @xmath54 be the numbered lattices of the first @xmath50 elements in @xmath34 and in @xmath55 on the @xmath56 , respectively . @xmath57{mk.eps } \end{array}\ ] ] @xmath58 let @xmath59 be the set of natural numbers that are less than or equal to @xmath9 . for each @xmath60 and @xmath61 , let @xmath62 where @xmath63 is the largest integer that is less than or equal to @xmath64 . then , from table 1.1 , it is clear that @xmath65 for example , for @xmath66 , from table 1.1 , @xmath67 in terms of blank lattices , the numbers in @xmath68 lie on @xmath69 the result of general @xmath70 follows from the following proposition , which can be easily proven by mathematical induction . [ proposition:1.1 ] for integers @xmath71 and @xmath60 , @xmath72 in particular , @xmath73 therefore , ( [ eqn:1.11 ] ) states that the numbers in @xmath70 are spread out on blank lattices with one copy of @xmath74 and @xmath75 copies of @xmath56 , @xmath76 . in particular , setting @xmath66 in ( [ eqn:1.11 ] ) yields ( [ eqn:1.9 ] ) . now , consider again system @xmath37 and the target formula ( [ eqn:1.2 ] ) . for any @xmath60 , let @xmath77 be the set of all admissible @xmath78-sequences in @xmath37 : @xmath79 our purpose is to compute @xmath80 , which is the number of elements in ( [ eqn:1.12 ] ) . the entropy @xmath27 follows from @xmath81 the constraint @xmath82 in ( [ eqn:1.12 ] ) is the admissible condition of golden - mean shift on @xmath48 , and it states that symbol 1 is not allowed to follow symbol 1 immediately . then , the forbidden set on @xmath83 is @xmath84{sigma2_4.eps } \right\}$ ] . the transition matrix is @xmath85.\ ] ] le @xmath86 be the set of all admissible patterns on @xmath51 with respect to ( [ eqn:1.13 ] ) ; then latexmath:[\[\label{eqn:1.14 } which is the @xmath50-th fibonacci number . since the constraint ( [ eqn:1.12 - 1 ] ) applies to each @xmath55 independently for @xmath88 , @xmath89 hence ( [ eqn:1.2 ] ) follows easily from ( [ eqn:1.16 ] ) . by the similar argument , ( [ eqn:1.10 ] ) of proposition [ proposition:1.1 ] also recovers the following results @xcite . [ theorem:1.2 ] for any @xmath71 , denote the multiplicative integers system @xmath90 then @xmath91 consideration of the above reveals the following three main parts of our study of @xmath37 . 1 . identify the numbered lattice @xmath53 and the admissible blank lattice @xmath56 from the given system ; see figs . 1.1 and 1.2 compute the numbers of copies of independent admissible lattices of the same length ; see formulae ( [ eqn:1.10 ] ) and ( [ eqn:1.11 ] ) . 3 . determine the set of all admissible patterns @xmath86 , which can be generated on @xmath56 , and compute the number of @xmath92 . notably , step ( iii ) in the study of @xmath37 is the classical one - dimensional pattern generation problem ; see @xcite . based on the above observations , the rest of this paper will consider the following two classes of systems . 1 . multi - dimensional decoupled systems like + @xmath93 2 . one - dimensional coupled systems like + @xmath94 + i.e. , @xmath95 where @xmath96 is a proper additive shift of finite type . first , consider multi - dimensional decoupled systems . let @xmath97 be natural numbers , @xmath98 , such that @xmath99 and @xmath100 are relatively prime for all @xmath101 , i.e. , @xmath102 for all @xmath103 , where @xmath104 is the greatest common divisor of natural numbers @xmath105 and @xmath106 . denote by @xmath107 and @xmath108 let @xmath109 with @xmath110 if @xmath111 . then , ( [ eqn:2.37 ] ) defines a sequence of @xmath112-dimensional numbered lattices @xmath53 of @xmath50 cells . the blank lattices @xmath113 are defined analogously ; see ( [ eqn:2.2 ] ) and ( [ eqn:2.39 - 10 ] ) . the complementary index set @xmath114 of @xmath115 is defined by @xmath116 hence , @xmath117 the following theorem for multi - dimensional decoupled system is proven in theorem [ theorem:2.10 ] . [ theorem:1.10 ] let @xmath118 satisfy ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) . then the entropy of @xmath119 is given by @xmath120 where @xmath121\right)}{\gamma_{1}\gamma_{2}\cdots \gamma_{d}}.\ ] ] finally , consider one - dimensional coupled system : @xmath122 where @xmath96 is a shift of finite type generated by transition matrix @xmath123 . denote by @xmath124 the degree @xmath50 blank lattice of the admissible numbered lattice @xmath125 ; see fig . 3.2 for @xmath126 and figs . 3.7 and 3.8 for @xmath127 . the following theorem are proven by theorems 3.7 and [ theorem:3.7 ] . [ theorem:1.11 ] for any @xmath71 and @xmath33 , @xmath128 and @xmath129 where @xmath130 is the set of all admissible patterns on @xmath124 . after we completed our study of ( i ) and ( ii ) , we became aware of the work of peres _ _ @xcite on ( [ eqn:1.19 ] ) . these authors obtained the same results as ours for multi - dimensional system ( i ) . our methods for studying ( i ) differ from theirs by using the results from an investigation of pattern generation problems , and involving the three specified steps ( i ) , ( ii ) and ( iii ) . moreover , a modification of these procedures enables us to study the one - dimensional coupled system ( ii ) . for the multi - dimensional coupled system like @xmath131 which is much more delicate . the dimension of @xmath96 is one and @xmath26 is two . difference of the dimensions induces an intrinsic difficulty to decouple the system effectively . the problem of ( [ eqn:1.21 - 1 ] ) is still not solved by using our method which works well in studying one - dimensional coupled system ( [ eqn:1.21 ] ) . see section 4 for further discussion . the rest of this paper is arranged as follows . section 2 studies multi - dimensional systems . section 3 studies one - dimensional coupled systems . section 4 studies multi - dimensional coupled systems . this section concerns multi - dimensional systems . for simplicity , @xmath26 is considered first . recall that @xmath132 firstly , the admissible numbered and blank lattices , determined by the constraint @xmath133 must be identified in @xmath26 . let @xmath134 be the set of the numbers that are multiples of 2-power and 3-power : @xmath135 @xmath134 can be expressed as follows . @xmath136{m23_1.eps } & & & \psfrag{1}{$q_{1}$ } \psfrag{2}{$q_{2}$ } \psfrag{3}{$q_{4}$ } \psfrag{4}{$q_{6}$ } \psfrag{5}{$q_{9}$ } \psfrag{7}{$q_{3}$ } \psfrag{8}{$q_{5}$ } \psfrag{9}{$q_{8}$ } \psfrag{a}{$q_{11}$ } \psfrag{b}{$q_{15}$ } \psfrag{d}{$q_{7}$ } \psfrag{e}{$q_{10}$ } \psfrag{f}{$q_{14}$ } \psfrag{g}{$q_{18}$ } \psfrag{h}{$q_{23}$ } \psfrag{k}{$q_{12}$ } \psfrag{l}{$q_{16}$ } \psfrag{m}{$q_{21}$ } \psfrag{n}{$q_{26}$ } \psfrag{o}{$q_{32}$ } \psfrag{q}{$q_{19}$ } \psfrag{r}{$q_{24}$ } \psfrag{s}{$q_{30}$ } \psfrag{t}{$q_{36}$ } \psfrag{u}{$q_{43}$ } \psfrag{v}{$q_{27}$ } \psfrag{w}{$q_{33}$ } \psfrag{x}{$q_{40}$ } \psfrag{y}{$q_{47}$ } \psfrag{z}{$q_{55}$ } \psfrag{a}{$q_{13}$ } \psfrag{b}{$q_{20}$ } \psfrag{c}{$q_{29}$ } \psfrag{d}{$q_{39}$ } \psfrag{e}{$q_{51}$ } \psfrag{f}{$q_{64}$ } \includegraphics[scale=0.8]{m23_2.eps } \end{array}\ ] ] @xmath137 @xmath138 since @xmath134 inherits the natural ordering of natural numbers , the @xmath50-th number in @xmath134 can be denoted by @xmath139 , @xmath140 . it seems that no known formula relates @xmath141 to @xmath142 explicitly @xcite . from table 2.1 , for any @xmath49 , the l - shaped @xmath50-cell numbered lattice @xmath53 that contains @xmath143 can be defined , and @xmath50-cell blank lattice @xmath113 can also be defined by deleting the numbers of @xmath53 . indeed , for @xmath49 , @xmath144 fig . 2.1 presents the first eight numbered lattices @xmath53 . notably , the number @xmath139 in the bold square is the largest number in @xmath53 . @xmath145{m_1.eps } & \hspace{0.3cm}\psfrag{a}{$m_{2}$ } \includegraphics[scale=0.8]{m_2.eps } & \hspace{0.3cm}\psfrag{a}{$m_{3}$ } \includegraphics[scale=0.8]{m_3.eps } & \hspace{0.3 cm } \psfrag{a}{$m_{4}$ } \includegraphics[scale=0.8]{m_4.eps}\\ & & & \\ \psfrag{a}{$m_{5}$ } \includegraphics[scale=0.8]{m_5.eps } & \hspace{0.3cm}\psfrag{a}{$m_{6}$ } \includegraphics[scale=0.8]{m_6.eps } & \hspace{0.3 cm } \psfrag{a}{$m_{7}$ } \includegraphics[scale=0.8]{m_7.eps } & \hspace{0.3 cm } \psfrag{a}{$m_{8}$ } \includegraphics[scale=0.8]{m_8.eps}\\ \end{array}\ ] ] @xmath146 in contrast to @xmath37 , all lattices @xmath53 and @xmath113 are now two - dimensional . therefore , the system @xmath26 can be regarded as a two - dimensional system ; see @xcite . now , the complementary index set @xmath147 of @xmath26 is the set of all natural numbers that can not be divided by two and three : @xmath148 the set @xmath39 of natural numbers can be rearranged into the first octant of three - dimensional space as @xmath149 and for @xmath150 with @xmath44 , @xmath151 see table 2.2 . @xmath152{im23.eps } \end{array}\ ] ] @xmath153 therefore , for any @xmath154 , there exists a unique @xmath155 such that @xmath156 . next , proceed to step ( ii ) : compute the numbers of copies of @xmath53 for a given @xmath157 , defined in ( [ eqn:1.6 ] ) . let @xmath158 for some @xmath159 . denote by @xmath160 define @xmath161 then @xmath162 for any @xmath163 . denote by the set @xmath164 for @xmath165 . clearly , if @xmath166 , then @xmath167 therefore , a parallel result of ( [ eqn:1.7 ] ) and ( [ eqn:1.10 ] ) is as follows . @xmath168 with @xmath169\cap \mathcal{i}_{2,3}.\ ] ] the following example illustrates ( [ eqn:2.10 ] ) and ( [ eqn:2.12 ] ) . [ example:2.4 ] for @xmath170 , it is easy to verify @xmath171 the others are empty . therefore , @xmath172 for @xmath165 , denote by @xmath173 the number of copies of @xmath53 in @xmath174 . formula ( [ eqn:2.12 ] ) gives the density of copies of @xmath53 in @xmath175 $ ] , which is crucial in computing the entropy . indeed , following proposition is offered . [ proposition:2.5 ] on @xmath26 , for any @xmath49 , @xmath176 where @xmath177\right\}}{2\cdot3}=\frac{1}{3}.\ ] ] for any fixed @xmath49 , from ( [ eqn:2.12 ] ) and ( [ eqn:2.14 ] ) , ( [ eqn:2.15 ] ) follows . clearly , ( [ eqn:2.16 ] ) follows from ( [ eqn:2.4 ] ) . furthermore , for @xmath60 , denote by @xmath178 now , the following proposition can be verified . [ proposition:2.6 ] for @xmath49 , @xmath179 where @xmath180 is given by ( [ eqn:2.16 ] ) . it is easy to see that if @xmath181 , then @xmath182 as ( [ eqn:2.12 ] ) , we have @xmath183\cap \mathcal{i}_{2,3}.\ ] ] therefore , ( [ eqn:2.18 ] ) follows immediately . after the density of @xmath53 ( [ eqn:2.18 ] ) is obtained , step ( ii ) is completed . now , the final step ( iii ) is to compute the admissible patterns on @xmath113 for all @xmath49 . previously , two - dimensional pattern generation problems on l - shaped lattices has been studied by lin and yang in @xcite . the basic lattice @xmath184 is defined by @xmath185 i.e. , @xmath186{l23.eps}\ ] ] the l - shaped lattice with origin @xmath187 as the corner vertex . the constraint ( 2.2 ) implies that the forbidden local pattern on @xmath184 is @xmath188{f23.eps } \end{array } \right\}.\ ] ] therefore , the basic set of admissible patterns is @xmath189{b1.eps } , & \includegraphics[scale=0.7]{b2.eps } , & \includegraphics[scale=0.7]{b3.eps } , & \includegraphics[scale=0.7]{b4.eps } , & \includegraphics[scale=0.7]{b5.eps } , & \includegraphics[scale=0.7]{b6.eps } , & \includegraphics[scale=0.7]{b7.eps } \end{array } \right\};\ ] ] see @xcite . for @xmath190 , the set of all admissible patterns on @xmath113 that are determined by @xmath191 is defined by @xmath192 clearly , @xmath193{sigma1_1.eps } , & \includegraphics[scale=0.7]{sigma1_2.eps } \end{array } \right\}$ ] and @xmath194{sigma2_1.eps } , & \includegraphics[scale=0.7]{sigma2_2.eps } , & \includegraphics[scale=0.7]{sigma2_3.eps } , & \includegraphics[scale=0.7]{sigma2_4.eps } \end{array } \right\}$ ] , and no constraint applies on @xmath195 and @xmath196 . recall that @xmath197 and @xmath198 after the above procedures have been completed , the following result concerning the entropy @xmath199 is obtained . [ theorem:2.7 ] the entropy of @xmath26 is given by @xmath200 where @xmath92 is the number of all admissible patterns determined by @xmath191 on @xmath113 . for any @xmath60 , let @xmath201 be the set of all admissible @xmath78-sequences in @xmath26 . from the condition ( [ eqn:2.2 ] ) , it is easy to see that for any two @xmath202 , the admissible patterns on @xmath203 and the admissible patterns on @xmath204 are mutually independent . then , we have that for any @xmath60 , @xmath205 therefore , from propositions 2.3 , @xmath206 the proof is complete . [ remark:2.7 - 1 ] denote by latexmath:[\[\label{eqn:2.22 } the number of patterns in @xmath86 . since no exact formula relates @xmath141 to @xmath142 for @xmath134 in table 2.1 , unlike for a fibonacci number , no recursive formula exists for @xmath208 ; see @xcite . this fact creates the most difficulty in computing entropy for a multi - dimensional system ; see@xcite . however , for relatively small @xmath50 , @xmath208 can be computed using the transition matrices developed in @xcite . table 2.3 presents few cases for @xmath209 , @xmath210 . @xmath211 @xmath212 define the ratio of @xmath92 by @xmath213 for @xmath33 . from table 2.1 and 2.1 , it is easy to verify that @xmath214 for some @xmath60 . on the other hand , it can be shown that there exists @xmath215 such that @xmath216 therefore , @xmath217 can not have a limit as @xmath50 tends to @xmath218 , unlike the fibonacci sequence which has the limit @xmath219 . a further study of @xmath217 and @xmath208 is needed . in the following , an approximation of ( [ eqn:2.23 ] ) is given . for @xmath220 , let @xmath221 clearly , from theorem 2.4 , @xmath222 is a lower bound of @xmath199 , and @xmath222 increasingly approaches @xmath199 as @xmath78 tends to infinity . furthermore , let @xmath223 where @xmath224 . hence , @xmath225 table 2.4 presents cases for @xmath78 with @xmath226 and @xmath210 . @xmath227 @xmath228 moreover , @xmath229 and @xmath230 . now , the theorem just established can be easily extended to general multiplicative systems . for simplicity , the result is stated and necessary modifications to its proof are only sketched . first , consider @xmath119 that satisfies ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) . @xmath115 is defined in ( [ eqn:2.37 ] ) . for @xmath49 , the blank @xmath50-cell lattice @xmath113 is defined by @xmath231 after determining the @xmath50-cell lattice @xmath113 of @xmath119 , the basic lattice @xmath232 of @xmath119 is defined by @xmath233 the @xmath112-dimensional l - shaped lattice with the origin @xmath234 as the corner vertex ; see ( [ eqn:2.19 - 2 ] ) for @xmath26 . from the constraint @xmath235 we have the forbidden set @xmath236 then , the basic set of admissible patterns is defined by @xmath237 which induces a @xmath112-dimensional shift of finite type @xmath238 on the @xmath112-dimensional lattice space @xmath239 . denote by @xmath240 the set of all admissible patterns determined by @xmath241 on @xmath113 and @xmath242 by a similar argument , the entropy of @xmath119 can be obtained as follows . [ theorem:2.10 ] let @xmath118 satisfy ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) . then the entropy of @xmath119 is given by @xmath243 where @xmath244\right)}{\gamma_{1}\gamma_{2}\cdots \gamma_{d}}.\ ] ] by ( [ eqn:2.37 ] ) and ( [ eqn:2.38 ] ) , let @xmath245 as the proof of proposition 2.3 , it can be verified that @xmath246\bigcap \mathcal{i}_{\gamma}\ ] ] and @xmath247 from the constraint ( [ eqn:2.40 - 99 ] ) , for any @xmath248 , the admissible patterns on @xmath249 and the admissible patterns on @xmath250 are mutually independent . and , the admissible patterns on @xmath251 are completely determined by @xmath241 . then , it can be verified that @xmath252 therefore , the result follows immediately . the following three - dimensional system illustrates the methods and results . [ example:2.11 ] for @xmath253 , consider @xmath254 then , @xmath255{m235.eps } \\ \mathbb{m}_{2,3,5 } \end{array}\ ] ] @xmath256 the first five numbered lattices are listed as follows . @xmath257{m235_1.eps } & \hspace{0.3cm}\psfrag{a}{$m_{2}$ } \includegraphics[scale=0.4]{m235_2.eps } & \hspace{0.3cm}\psfrag{a}{$m_{3}$ } \includegraphics[scale=0.4]{m235_3.eps } & \hspace{0.3 cm } \psfrag{a}{$m_{4}$ } \includegraphics[scale=0.4]{m235_4.eps } & \psfrag{a}{$m_{5}$ } \includegraphics[scale=0.4]{m235_5.eps } \end{array}\ ] ] @xmath258 the basic lattice is @xmath259{l235.eps}\ ] ] clearly , @xmath260 therefore , it can be verified that @xmath261 where @xmath262 and the forbidden set of @xmath263 is @xmath264{f235.eps}\end{array}\right\}$ ] . the @xmath78-th order approximation of ( [ eqn:2.41 - 1 ] ) is as follows . for @xmath265 , let @xmath266 in table 2.6 , some cases for @xmath267 in @xmath268 are listed . @xmath269 @xmath270 moreover , @xmath271 . the previous idea also applies to system @xmath119 that does not satisfy the conditions ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) , where @xmath272 . let @xmath273 be the least common multiple of @xmath274 . denote by @xmath275 the set of prime factors of @xmath273 with @xmath276 , @xmath277 . clearly , @xmath278 satisfies both ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) . then , @xmath119 can be studied by using @xmath279 and @xmath280 . denote by @xmath281 with @xmath282 if @xmath111 . the complementary index set @xmath280 of @xmath283 is defined by @xmath284 for @xmath49 , let the @xmath50-cell lattice @xmath285 of @xmath119 be @xmath286 now , the constraint @xmath287 can be expressed in terms of @xmath278 . indeed , define the basic lattice @xmath232 of @xmath119 by @xmath288 then , the forbidden set @xmath289 is given by @xmath290 therefore , the basic set of admissible patterns can be written @xmath291 notably , @xmath241 induces a @xmath292-dimensional shift of finite type @xmath238 . let @xmath240 be the set of all admissible patterns that can be determined by @xmath241 on @xmath285 , @xmath49 . in the following , theorem 2.6 is generalized for @xmath119 without the conditions ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) . [ theorem:2.10 - 1 ] let @xmath118 . then , the entropy of @xmath119 is given by @xmath293 where @xmath294\right)}{p_{1}p_{2}\cdots p_{q}},\ ] ] @xmath278 , @xmath283 and @xmath280 are given by ( 2.41 ) , ( 2.42 ) and ( 2.43 ) , respectively . first , from the construction of ( [ eqn:2.35 - 10 ] ) , ( [ eqn:2.35 - 11 ] ) and ( [ eqn:2.35 - 12 ] ) , it is clear that @xmath295 and for @xmath296 with @xmath44 , @xmath297 it is easy to see that@xmath298 for @xmath299 . moreover , if @xmath300 for some @xmath301 , then @xmath302 for all @xmath299 . hence , from the constraint ( [ eqn:2.40 ] ) , the admissible patterns on @xmath303 and the admissible patterns on @xmath304 are mutually independent for @xmath305 . as in the proof of theorem 2.6 , we can define @xmath306 , @xmath307 , @xmath308 and @xmath309 . it can be proven that @xmath310\bigcap \mathcal{i}_{\gamma^{*}}\ ] ] and @xmath311 next , the constraint ( [ eqn:2.40 ] ) and the construction of @xmath241 imply that the admissible patterns on @xmath312 , @xmath313 , are completely determined by @xmath241 . hence , @xmath314 where @xmath240 the set of all admissible patterns determined by @xmath241 on @xmath315 . therefore , ( [ eqn:2.41 ] ) follows . the proof is complete . the following example illustrates theorem 2.8 . [ example:2.7 - 1 ] consider @xmath316 . it is easy to see that @xmath317 . from ( 2.46 ) , the basic lattice @xmath318{l28.eps}$ ] , here the third cell is deleted . the forbidden set @xmath319 is @xmath320{f28_0.eps}\right\}$ ] and @xmath321 . define the associated transition matrix @xmath322{a28.eps}\ ] ] then , @xmath323 where @xmath324 and @xmath325 for @xmath326 and @xmath327 . in the remaining of this section , the constraint ( [ eqn:2.40 ] ) is further relaxed . therefore , we can study more general case than @xmath119 . for simplicity , only @xmath328 that satisfies conditions ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) is studied . for any @xmath329 , consider a multiplicative system is of @xmath4-symbols , @xmath330 . for any @xmath331 , let the constraint set @xmath332 be a subset of @xmath333 . denote by @xmath334 the multiplicative integer system with constraint set @xmath332 : @xmath335 then , the basic set @xmath336 of admissible patterns on @xmath232 is given by @xmath337 the following theorem can be proven as theorem 2.6 . [ theorem:2.11 ] let @xmath118 satisfy ( [ eqn:1.20 - 1 ] ) and ( [ eqn:1.20 - 2 ] ) and @xmath338 . the entropy of @xmath334 is given by @xmath339 where @xmath340 is the set of @xmath341dimensional admissible local patterns that can be generated by @xmath342 on @xmath113 . this proof is similar to the proof of theorem 2.6 . the only difference between @xmath119 and @xmath334 is their constraints . by ( [ eqn:2.36 ] ) , it is easy to see that the basic set @xmath343 can completely determine the patterns on @xmath251 for @xmath344 . therefore , the result follows . the following example illustrates theorem 2.10 . [ example:2.11 - 1 ] let @xmath345 and @xmath346 . then @xmath347 the basic set of admissible local patterns is now given by @xmath348{f2c_1.eps } , & \includegraphics[scale=0.5]{f2c_2.eps } , & \includegraphics[scale=0.5]{f2c_3.eps } , & \includegraphics[scale=0.5]{f2c_4.eps } , & \includegraphics[scale=0.5]{f2c_5.eps } , & \includegraphics[scale=0.5]{f2c_6.eps } , & \includegraphics[scale=0.5]{f2c_7.eps } \end{array } \right\}.\ ] ] the associated transition matrix is @xmath349{a2nc.eps}\ ] ] therefore , as in theorem 1.2 , @xmath350 where @xmath351 , @xmath352 for all @xmath33 . this section investigates the one - dimensional coupled system which is an intersection of the multiplicative integer system @xmath353 with an additive proper shift of finite type @xmath96 , i.e. , @xmath354 a simple system is considered first ; the findings are then extended to general systems . consider @xmath355 to incorporate the effect of @xmath96 , table 1.1 is replaced by the following figure . @xmath356{x2a.eps}\ ] ] @xmath357 as in table 1.1 , the horizontal lines in fig . 3.1 connect the integers in @xmath55 for each @xmath88 , the effect comes from @xmath37 . on the other hand , the bold zigzag line in fig . 3.1 connects all natural integers comes from @xmath96 . therefore , for any @xmath44 in @xmath36 , @xmath55 and @xmath358 are no longer mutually independent . in fact , they are all coupled through the relation set @xmath34 . therefore , ( [ eqn:3.2 ] ) is regarded as a coupled system . before the system @xmath359 is decoupled , the following definition is needed . [ definition:3.0 ] two sets of integers of @xmath360 and @xmath361 are mutually independent in @xmath359 if @xmath362 and any numbers @xmath9 in @xmath360 and @xmath363 in @xmath361 are not consecutive and also not consecutive in @xmath364-power , i.e. , if @xmath365 for some @xmath78 then @xmath366 and @xmath367 . then , the following lemma can be obtained . [ lemma:3.0 - 1 ] suppose @xmath360 and @xmath361 are mutually independent in @xmath359 . then @xmath368 is the set of all admissible patterns on lattice @xmath360 , and @xmath369 and @xmath370 are defined analogously . since @xmath360 and @xmath361 are decoupled in @xmath359 , the patterns in @xmath371 are independent of the patterns in @xmath369 . therefore , the result holds . the strategy for studying ( [ eqn:3.2 ] ) is to decouple the whole system into disjoint pieces which are located in some proper subset @xmath372 of @xmath359 . from the reduced system @xmath372 , a sequence @xmath373 of independent decoupled subsystems are chosen . then , the entropy of the decoupled independent system @xmath374 can be computed easily . an appropriate choice of @xmath374 is demonstrated to enable the recovery of the entropy of @xmath359 , i.e. , @xmath375 as in decoupled system @xmath37 , the admissible numbered lattice @xmath53 in @xmath372 is firstly picked up . indeed , in fig . 3.2 , some @xmath125 are drawn for @xmath376 . @xmath377{m2a1.eps } & \hspace{1.0 cm } \psfrag{a}{{\scriptsize(b ) $ m_{1}(l)$ } } \psfrag{l}{$l$ } \includegraphics[scale=0.5]{m2a1_l.eps } & \hspace{1.0 cm } \psfrag{a}{{\scriptsize ( c ) $ m_{2}(3)$ } } \includegraphics[scale=0.5]{m2a2_3.eps } & \hspace{1.0 cm } \psfrag{a}{{\scriptsize ( d ) $ m_{2}(l)$ } } \psfrag{b}{{\scriptsize$l$ } } \psfrag{c}{{\scriptsize$2l$ } } \psfrag{d}{{\scriptsize$2l-1 $ } } \psfrag{e}{\scriptsize{$2l+1 $ } } \includegraphics[scale=0.5]{m2a2_l.eps } \end{array}\ ] ] @xmath378{m2a3_3.eps } & \hspace{1.0 cm } \psfrag{a}{{\tiny ( f ) $ m_{3}(l)$ } } \psfrag{b}{{\tiny$l$ } } \psfrag{c}{{\tiny$2l$ } } \psfrag{d}{{\tiny$4l$ } } \psfrag{e}{{\tiny$2l-1 $ } } \psfrag{f}{{\tiny$4l-2 $ } } \psfrag{g}{{\tiny$2l+1 $ } } \psfrag{h}{{\tiny$4l+2 $ } } \psfrag{j}{{\tiny$4l-3 $ } } \psfrag{k}{{\tiny$4l-1 $ } } \psfrag{l}{{\tiny$4l+1 $ } } \psfrag{m}{{\tiny$4l+3 $ } } \includegraphics[scale=0.3]{m2a3_l.eps } \end{array}\ ] ] @xmath379{m2a4_3.eps}\ ] ] @xmath380 the choice of @xmath53 is recursive and robotic . the basic idea is that any number can produce the next generation through @xmath381 or @xmath96 . more precisely , for each number @xmath78 , if @xmath382 , then @xmath78 can produce the next generation @xmath383 . if @xmath384 with @xmath385 , @xmath386 , then @xmath78 can produce @xmath387 through @xmath381 and @xmath388 through @xmath96 . in summary , a complete production cycle is as follows . if @xmath382 , then @xmath78 produces @xmath389 and then @xmath390 . if @xmath385 , @xmath386 , then @xmath78 produces @xmath391 and then @xmath392 . for example , @xmath393 has one cell , and number @xmath394 is regarded as the number of first generation . @xmath395 is constructed from @xmath393 by producing number @xmath396 from @xmath394 through @xmath397 . immediately , @xmath396 creates numbers @xmath398 and @xmath399 as the descendanta through @xmath96 . @xmath395 is of degree 2 since there are two numbers @xmath400 on the horizontal line . the construction of @xmath401 from @xmath402 is performed similarly : the number @xmath396 yields number 12 in @xmath397 . on the same time , numbers @xmath398 and @xmath399 yield the numbers @xmath403 and 14 in @xmath404 and @xmath405 , respectively . next , numbers 10 , 14 and 12 yield their descendants 9 , 11 , 13 , 15 and 11 , 13 in @xmath36 through @xmath96 , as presented in fig . 3.2 ( e ) . now , the three numbers @xmath394 , @xmath396 and @xmath406 are in the lowest horizontal direction , and @xmath407 is therefore of degree 3 . on @xmath408 , the maximal number of numbers of cells in the horizontal direction is @xmath50 , and @xmath408 is of degree @xmath50 . now , @xmath125 can be defined formally as follows . [ definition:3.1 ] for each @xmath409 , define @xmath410 . for @xmath49 , define @xmath411 then , define @xmath412 and for @xmath33 , @xmath413 see fig . 3.2 for @xmath414 , @xmath376 . notably , @xmath415 and @xmath416 are mutually independent when @xmath417 and @xmath418 . after the lattices @xmath53 and @xmath113 are identified , in a given range @xmath70 , ( ii ) is then to be carried out , i.e. , the number of disjoint copies of @xmath419 with @xmath420 is computed . for example , in fig . 3.3 , @xmath421\cap\mathbb{n}$ ] can be decoupled by @xmath395 , @xmath422 , @xmath423 , @xmath424 , @xmath425 and the numbers in @xmath426 are not used . there are one copy @xmath395 in @xmath427 and four copies @xmath422 , @xmath423 , @xmath424 , @xmath425 in @xmath428 ; see fig . @xmath356{x2a-0.eps}\ ] ] @xmath429 the results for disjoint copies in @xmath430 can be proven as follows . [ lemma:3.1 ] given @xmath49 and @xmath431 , define @xmath432 then , within @xmath430 , there are @xmath433 mutually independent copies of @xmath125 with @xmath420 . we begin with @xmath434 and @xmath435 . it is easy to show that in @xmath436 , each odd integer @xmath437 can produce a @xmath438 that lies in @xmath436 and they are all disjoint ; see fig . . therefore , there are totally @xmath439 copies of @xmath402 . similarly , as in fig . 3.5 , between @xmath440 , each @xmath441 produces a @xmath442 that lies in @xmath440 . they are all disjoint too . the total number of copies of @xmath442 in @xmath440 is @xmath443 . @xmath444{2_n.eps } & \hspace{1.3cm}\psfrag{a}{{\scriptsize$2^{n-2}$ } } \psfrag{b}{{\scriptsize$2^{n-1}$ } } \psfrag{c}{{\scriptsize$2^{n}$ } } \psfrag{e}{{\scriptsize$2^{n-3}$ } } \psfrag{f}{{\scriptsize$2^{n-4}$ } } \psfrag{g}{{\scriptsize$2^{n-5}$ } } \psfrag{h}{{\scriptsize$2^{n-6}$ } } \psfrag{d}{{\scriptsize$\mathcal{i}_{2}$ } } \includegraphics[scale=0.8]{2_n_1.eps}\\ & \\ \text{figure 3.4 . } & \hspace{1.3cm}\text{figure 3.5 . } \end{array}\ ] ] by using the similar argument , it can be verified that for any @xmath33 and @xmath445 , within @xmath446 , the number of disjoint copies of @xmath53 is @xmath447 therefore , @xmath448 the proof is complete . finally , for step ( iii ) , denote by @xmath86 the admissible patterns on @xmath113 . then , lemma 3.4 yields the lower bound of the entropy @xmath449 . [ lemma:3.2 ] for any @xmath49 , @xmath450 within @xmath451 $ ] , we put symbol 0 on the cells that are not used in deriving ( [ eqn:3.6 ] ) . then , @xmath452 now , it is easy to see @xmath453 therefore , @xmath454 the proof is complete . now , a good upper estimate of the entropy @xmath449 remains to be found . from the derivation of lemma 3.4 , we need to estimate the contribution to entropy of the vertices that are not used in deriving ( [ eqn:3.6 ] ) . the following lemma is obtained . [ lemma:3.3 ] for each @xmath49 , the number of the vertices of @xmath113 satisfies @xmath455 within @xmath451 $ ] , the number of the vertices that are not used in deriving ( [ eqn:3.6 ] ) is @xmath456 moreover , @xmath457 where @xmath92 is the number of all admissible patterns on @xmath113 . ( [ eqn:3.8 ] ) is easily proved by induction . then , for @xmath445 , within @xmath458 $ ] , by ( [ eqn:3.6 - 10 ] ) and ( [ eqn:3.8 ] ) , it can be verified that there are @xmath459 \\ & \\ = & k \left(2^{n - k(l+1)-1}\right ) \end{array}\ ] ] vertices that are not used in deriving ( [ eqn:3.6 ] ) . since @xmath460=\left(\underset{l=0}{\overset{m^{*}}{\bigcup}}\left(2^{n - k(l+1)},2^{n - kl}\right ] \right)\bigcup \left [ 1,2^{n - k(m^{*}+1)}\right].\ ] ] it is easy to see that there are @xmath461 vertices in @xmath451 $ ] that are not used in deriving ( [ eqn:3.6 ] ) . since two symbols 0 and 1 may appear on vertices in ( [ eqn:3.10 ] ) , the contribution to the entropy of these unused vertices is at most @xmath462/2^{n } \\ & \\= & \frac{k}{2(2^{k}-1)}\log 2 . \end{array}\ ] ] the upper estimate ( [ eqn:3.11 ] ) of entropy follows . lemmas 3.5 and 3.6 yield the following result . [ theorem:3.4 ] the entropy @xmath449 is given by @xmath463 furthermore , @xmath464 lemmas 3.5 and 3.6 imply ( [ eqn:3.13 ] ) , and then ( [ eqn:3.12 ] ) follows from ( [ eqn:3.13 ] ) immediately . [ example:3.4 - 1 ] consider the one - dimensional couple shifts @xmath465 where @xmath466\ ] ] and @xmath467 is the golden mean shift . table 3.1 presents a numerical approximation of ( [ eqn:3.13 ] ) . for @xmath60 , let @xmath468 @xmath469 @xmath470 [ remark:3.5 ] whether or not @xmath449 can be expressed in explicit form , as in ( [ eqn:1.2 ] ) for @xmath27 is not clear . now , the result for @xmath359 is ready to extend to @xmath471 for any integer @xmath472 . from the study of @xmath359 , the main steps for @xmath473 are : 1 . identify the lattice @xmath53 ( and @xmath113 ) , which is the maximal connected graph of degree @xmath50 . all @xmath125 , @xmath474 , are disjoint . 2 . compute the unused vertices in + @xmath475 3 . compute the number @xmath86 of admissible patterns on @xmath113 . step ( i)@xmath476 gives lower bound of @xmath477 , and ( ii)@xmath476 gives upper bound of @xmath477 . then , @xmath477 follows if the error term in ( ii)@xmath476 approaches zero as @xmath78 tends to infinity . to minimize the error in ( ii)@xmath476 , the amount of unused lattices should be as small as possible . therefore , the choice of @xmath53 or the graph @xmath113 in ( i)@xmath476 should be as large as possible as far as they are decoupled . @xmath478 can be used to illustrate the procedures . from @xmath479 the following fig . 3.6 is drawn , which is corresponding to fig . 3.1 for @xmath359 . @xmath480{x3a.eps}\ ] ] @xmath481 take @xmath482 as being on one vertex . in fig . 3.7 , a maximal connected graph @xmath483 with two horizontal vertices in @xmath484 can be identified as follows . notably , @xmath485 and @xmath486 are mutually independent . @xmath487{m_2_4.eps } & & & & \psfrag{a}{{\footnotesize $ m_{3;2}(7)$ } } \includegraphics[scale=0.5]{m_2_7.eps } \end{array}\ ] ] @xmath488 furthermore , @xmath489 can be constructed from @xmath485 as follows . @xmath490{m_3_4.eps}\ ] ] @xmath491 similar to definition 3.3 for @xmath126 , for any @xmath472 , @xmath492 can be defined as follows . [ definition:3.6 - 10 ] for any @xmath472 and @xmath493 , let @xmath494 , and for @xmath49 , define @xmath495 then , let @xmath496 and for @xmath33 , define @xmath497 for any @xmath472 and @xmath49 , denote @xmath124 be the degree @xmath50 blank lattice . the following lemma gives the number of the vertices of @xmath124 , an extension of lemma 3.6 . [ lemma:3.6 ] for any @xmath472 and @xmath33 , the number @xmath498 of the vertices of @xmath124 is @xmath499 ) is proven for the case @xmath127 . the other cases can be treated analogously . for @xmath127 , let @xmath500 the blank lattice @xmath501 can be obtained from @xmath485 in fig . 3.7 , and @xmath502 now , @xmath503 is obtained from @xmath489 in fig . 3.8 and can be grouped as follows . @xmath504{2618.eps}\ ] ] @xmath505 according to fig . 3.9 , it is easy to see that @xmath506 by induction , it can be proven @xmath507 ) follows for @xmath127 . for @xmath508 , @xmath509 is as in fig . @xmath510{m_4_2.eps}\ ] ] @xmath511 similarly , for any @xmath512 , @xmath513 is given in fig . @xmath514{m_q_2.eps}\ ] ] @xmath515 @xmath516 @xmath517 define @xmath518 careful mathematical induction proves ( [ eqn:3.17 ] ) . the details are omitted here . now , we can extend theorem 3.7 to all @xmath472 , as follows . [ theorem:3.7 ] for any @xmath472 and @xmath33 , @xmath519 and @xmath520 where @xmath130 is the set of all admissible patterns on @xmath124 . given @xmath33 and @xmath60 , @xmath521 is defined in ( [ eqn:3.5 - 50 ] ) . as the proof of lemma 3.4 , it can be verified that for @xmath445 , within @xmath522 , the number of mutually independent copies of @xmath523 is @xmath524 . hence , within @xmath525 , there are @xmath526 disjoint copies of @xmath492 with @xmath474 . as the proof of lemma 3.6 , by ( [ eqn:3.17 ] ) , the number of the remaining vertices that are not used in @xmath527 $ ] is @xmath528 therefore , the results follow immediately . like theorem 2.10 , theorem 3.12 can be generalized to any number of symbols , any constraints @xmath332 and any additive shifts of finite type @xmath96 . indeed , let @xmath529 where @xmath338 is a constraint set and @xmath123 is an @xmath530 @xmath531 matrix . then , the following theorem can easily be obtained . the details of the proof are omitted . [ theorem:3.7 - 10 ] for any @xmath71 , @xmath338 , @xmath329 , @xmath331 and @xmath33 , @xmath532 and @xmath533 where @xmath534 is the set of all admissible patterns on @xmath124 , the constraint of the vertices on the bold lines in @xmath124 is given by @xmath123 and the constraint of the vertices on thin lines in @xmath124 is given by @xmath4 and @xmath332 . this section discusses the multi - dimensional coupled systems and points out the difficulties when applying the method that works well for one - dimensional coupled systems . 1 . for @xmath49 , choose a suitable admissible numbered lattice @xmath125 as basic elements for @xmath473 ; see fig . 3.2 for @xmath359 . 2 . for each @xmath49 , split the natural numbers @xmath39 into two parts : + @xmath537 + where @xmath538 is used to select the mutually independent admissible numbered lattice @xmath125 for approximating the entropy , and @xmath539 is the set of the cells that is removed from @xmath39 to achieve independence of @xmath125 in @xmath538 . good splittings require + @xmath540\right|}{\left|u_{k}\cap [ 1,n]\right|}=0;\ ] ] + see lemmas 3.4 and 3.6 and theorems 3.7 and 3.12 . 3 . finally , compute the number of members of the set + @xmath541 + let + @xmath542 \right|,\ ] ] + and + @xmath543 + then , the number @xmath544 is an approximation to the entropy of @xmath473 , where @xmath86 is the set of all admissible patterns on @xmath53 . for @xmath359 , see lemma 3.4 . where @xmath26 is given by ( [ eqn:2.1 ] ) and @xmath96 is a shift of finite type . @xmath53 is the l - shaped @xmath50-cell numbered lattices given in table 2.1 and fig . @xmath546 is given in ( [ eqn:2.4 ] ) , and then @xmath547 given in ( [ eqn:2.5 ] ) is obtained . indeed , if @xmath574 or @xmath575 , then @xmath576 is consecutive to @xmath577 by choosing @xmath578 or @xmath579 in ( 1 ) ( i ) and ( ii ) , respectively . if @xmath580 , then ( 4.9 ) and lemma 4.1 implies @xmath581 , @xmath582 , @xmath583 , @xmath584 , @xmath585 and @xmath586 can be consecutive to @xmath587 which appears in ( 2 ) , ( 3 ) and ( 4 ) of lemma 4.1 . similar results hold for @xmath588 by using ( 4.10 ) and the results in ( 2 ) , ( 3 ) and ( 4 ) of lemma 4.1 . the consecutive diagram of @xmath592 for @xmath593\cap \mathcal{i}_{2,3}$ ] is drawn in fig . 4.2 ( i ) . furthermore , the projection of consecutive diagram of @xmath592 in @xmath147 for @xmath594\cap\mathcal{i}_{2,3}$ ] is drawn in fig . 4.2 ( ii ) . whether these arms can reach some @xmath599 is of interest , as a determination would reveal which elements in @xmath600 belong to @xmath601 with @xmath602 . the following proposition provides the answers . from proposition 4.3 , there are 12 cases for each @xmath604 and @xmath606 ; see the following table 4.1 . notably , in tables 4.1 , ( i ) and ( ii ) are symmetric in @xmath608 of @xmath609 and @xmath610 of @xmath611 . @xmath616{4u3d.eps } \end{array } & & \begin{array}{c } \psfrag{a}{{\footnotesize $ 77 $ } } \psfrag{b}{{\footnotesize$37 $ } } \psfrag{c}{{\footnotesize$29\cdot 2 $ } } \psfrag{d}{{\footnotesize$25 \cdot 3 $ } } \psfrag{e}{{\footnotesize $ 13\cdot 3 $ } } \psfrag{f}{{\footnotesize$5\cdot 4 $ } } \includegraphics[scale=0.8]{4u2d.eps } \end{array } \\ & & \\ \text{(a ) 4 upward arms and 3 downward arms } & & \text{(b ) 4 upward arms and 2 downward arms } \\ & & \\ \begin{array}{c } \psfrag{a}{{\footnotesize $ 101 $ } } \psfrag{b}{{\footnotesize$49 $ } } \psfrag{c}{{\footnotesize$37\cdot 2 $ } } \psfrag{d}{{\footnotesize$19 \cdot 4 $ } } \psfrag{e}{{\footnotesize $ 17\cdot 3 $ } } \psfrag{f}{{\footnotesize$13\cdot 2 $ } } \includegraphics[scale=0.8]{3u3d.eps } \end{array } & & \begin{array}{c } \psfrag{a}{{\footnotesize $ 19 $ } } \psfrag{b}{{\footnotesize$11 $ } } \psfrag{c}{{\scriptsize$7\cdot 2,7\cdot 3 $ } } \psfrag{d}{{\footnotesize$1 \cdot 4 $ } } \includegraphics[scale=0.8]{4u1d.eps } \end{array } \\ & & \\ \text{(c ) 3 upward arms and 3 downward arms } & & \text{(d ) 4 upward arms and 1 downward arm } \\ & & \\ \begin{array}{c } \psfrag{a}{{\footnotesize $ 29 $ } } \psfrag{b}{{\footnotesize$13 $ } } \psfrag{c}{{\footnotesize$11\cdot 2 $ } } \psfrag{d}{{\scriptsize$5 \cdot 4,5\cdot3 $ } } \includegraphics[scale=0.8]{3u2d.eps } \end{array } & & \\ & & \\ \text{(e ) 3 upward arms and 2 downward arms } & & \\ \end{array}\ ] ] now , consider the splitting procedure ( ii ) again . for the one - dimensional coupled system @xmath359 , the two - dimensional graph in fig . 3.1 looks like a bamboo - blind . the horizontal lines @xmath618 are the bamboo sticks and the zigzag lines are the strings that tie the sticks together . for each @xmath49 , taking @xmath619 , the blind falls apart into infinitely - many pieces . these disjoint pieces are mutually independent and are used to estimate the lower bound of the entropy . in the two - dimensional coupled system @xmath536 , for each @xmath620 , the numbers in @xmath621 are taken as the vertices ; the horizontal edges are segments in @xmath54 , and the vertical edges are segments that connect the consecutive numbers in @xmath622 ; see fig . 4.2 ( i ) for @xmath623 and @xmath624 . figure 4.2 ( ii ) is a projection of the graph in @xmath147 for @xmath623 and @xmath625 . from propositions 4.2 and 4.3 , for each @xmath626 , @xmath622 is a fully connected graph . this paper does not find any means to split @xmath622 into two parts : from propositions 4.2 and 4.3 , for @xmath626 , the connections among the vertices in @xmath622 are quite complicated . then , the topology of @xmath622 is far away from @xmath629 , a standard three - dimensional lattice . previous results concerning pattern generation problems can not apply successfully @xcite . a better understanding of @xmath622 is required before dealing with @xmath536 . s. j. lightwood , _ morphisms from non - periodic @xmath632 subshifts ii : constructing homomorphisms to square - filling mixing shifts of finite type _ , ergodic theory and dynamical systems * 24 * , ( 2004 ) 12271260 . n. g. markley and m. e. paul , _ maximal measures and entropy for @xmath633 subshift of finite type _ , classical mechanics and dynamical systems , lecture notes in pure and appl . math . * 70 * ( medford , mass . ) , ( 1979 ) 135157 .

we investigate the kinematics of a combined sample of 74 globular clusters around ngc 1399 . their high velocity dispersion , increasing with radius , supports their association with the gravitational potential of the galaxy cluster rather than with that of ngc 1399 itself . we find no evidence for rotation in the full sample , although some indication for rotation in the outer regions . the data do not allow us to detect differences between the kinematics of the blue and red sub populations of globular clusters . a comparison between the globular cluster systems of ngc 1399 and those of ngc 1404 and ngc 1380 indicates that the globular clusters in all three galaxies are likely to have formed via similar mechanisms and at similar epochs . the only property which distinguishes the ngc 1399 globular cluster system from these others is that it is ten times more abundant . we summarize the evidence for associating these excess globulars with the galaxy cluster rather than with ngc 1399 itself , and suggest that the over abundance can be explained by tidal stripping , at an early epoch , of neighboring galaxies and subsequent accumulation of globulars in the gravitational potential of the galaxy cluster .

astro-ph9811373

extragalactic globular clusters have in recent years established themselves as potential tracers of the formation and evolution of galaxies ( see ashman & zepf 1998 for a recent review ) . the number of detailed photometric studies has rapidly increased , and these studies reveal a number of interesting connections between globular cluster systems and their host galaxies . with the recent commissioning of 10m class telescopes , measuring absorption line indices of a large number of individual globular clusters has become feasible ( kissler - patig et al . 1998 ; cohen , blakeslee , & ryzhov 1998 ) . colors , magnitudes , total numbers , spatial distributions , radial density profiles , ages , and metallicities , can now be obtained and used to constrain the formation history of globular clusters and their host galaxies . another essential source of information for discriminating between different formation scenarios is the kinematics of globular cluster systems , as determined from the measured radial velocities of individual clusters . for example , the studies of m87 and ngc 1399 found a significantly higher velocity dispersion for the globular clusters than for the stars ( huchra and brodie 1987 ; mould et al . 1990 ; brodie and huchra 1991 ; grillmair et al . 1994 ; cohen & ryzhov 1997 ; minniti et al . 1998 ; kissler - patig et al . mould et al . ( 1990 ) demonstrated that , in m87 , this was consistent with the the surface density distribution of globular clusters being more extended than the surface density distribution of stars . grillmair et al . ( 1994 ) suggested that the globular clusters around ngc 1399 were reacting to the gravitational potential of the fornax cluster as a whole rather than just that of the host galaxy . in ngc 5128 , harris et al . ( 1988 ) and hui et al . ( 1995 ) reported rotation in the globular cluster system , though only for the metal rich clusters . this is contrary to the findings in ngc 4472 ( sharples et al . 1998 ) and m87 ( kissler - patig & gebhardt 1998 ) in which the metal poor globular clusters seem to dominate the rotation . unfortunately , few models exist to compare with the ( still sparse ) data . predictions for the kinematic signature in globular cluster systems after spiral spiral mergers have been presented by hernquist & bolte ( 1992 ) . they studied the kinematics of globular clusters already present in the progenitors and found that in the merger product these clusters are expected to be on radial orbits in the inner regions and to show some systemic rotation far out . the kinematics of globular clusters that might have formed during a merger was not addressed in their study ; it would depend on the kinematics of the in - falling gas from which they formed . other simulations ( e.g. muzzio 1987 ) studied the accretion and stripping of globular clusters in galaxy clusters , but no clear predictions for the kinematics of accreted globular clusters were formulated . the primary goal of the present paper is to combine all existing kinematic data on the globular cluster system of ngc 1399 to better constrain its origin . a secondary aim is to use the photometric properties of globular clusters in the brightest fornax galaxies to further constrain formation scenarios . the sample of radial velocities for globular clusters around ngc 1399 is compiled and presented in sect . 2 . these are used to investigate the kinematic properties of the full sample , as well as sub samples selected on the basis of radius and color in sect . 3 . in section 4 we compare the kinematics of the globular clusters with those of the stars , cluster galaxies and x - ray gas , and derive the mass - to - light ratio in the outer environs of the galaxy . in sect . 5 we compare the properties of the globular clusters in ngc 1399 with those of the globular cluster systems of the next two brightest early type galaxies in fornax , ngc 1380 and ngc 1404 . we then discuss the implications for different formation scenarios . a summary and our conclusions are given in sect . our sample is based on the compilations of grillmair ( 1992 ) , grillmair et al . ( 1994 ) , minniti et al . ( 1998 ) , and kissler - patig et al.(1998 ) . briefly , the data from grillmair et al . were obtained at the anglo australian telescope with the low dispersion survey spectrograph and the image photon counting system in the wavelength range 38004800 with @xmath0 resolution . minniti et al . obtained their data with the new technology telescope at the european southern observatory , using the eso multi mode instrument in the wavelength range 60009000 with a resolution of 7.5 . kissler - patig et al . observed with the low resolution imaging spectrograph on the keck 1 telescope ; their spectra covered a wavelength range from about 4000 to 6100 with 5.6 resolution . we refer the reader to the original papers for a more detailed description of the observations , the data reduction , and the velocity measurements . hereafter , we will refer to the respective samples as the aat sample ( grillmair 1992 , grillmair et al . 1994 ) , the ntt sample ( minniti et al . 1998 ) , and the keck sample ( kissler - patig et al . 1998 ) . the combined sample of 74 globular clusters is listed in table 1 . for each globular cluster we give the i d number ( taken from the papers with the prefix aat / ntt / keck added respectively ) , the equatorial coordinates ( b1950 ) , the heliocentric radial velocity ( with the weighted mean when multiple measurements were available ) , as well as the available photometric information . for 52 globular clusters v@xmath1i colors accurate to 0.035 mag where obtained from the work of kissler - patig et al . ( 1997a ) . the b@xmath2 magnitude and b@xmath3r color were taken from grillmair ( 1992 ) . we included in our combined sample all objects with radial velocities greater than 500 km s@xmath4 and less than 2500 km s@xmath4 ( i.e. within @xmath5 of the mean ) . a list of positions , velocities and colors of the 74 globular clusters in electronic form is available from the first author . figure 1 shows the positions with respect to ngc 1399 of all globular clusters observed ; the symbols are proportional to the difference between the globular cluster velocity and the systemic velocity of ngc 1399 . the two pairs of larger circles ( continuous and dashed lines ) show 1 and 5 r@xmath6 for the stellar light distribution of ngc 1399 ( centered ) and ngc 1404 ( in the se ) , respectively . in table 2 we list the adopted properties of the three brightest early type galaxies in the center of fornax : ngc 1399 , ngc 1404 , and ngc 1380 . figure 2 shows a histogram of the globular cluster velocities . as already noted by minniti et al . ( 1998 ) , the velocity distribution exhibits two peaks roughly centered on the systemic velocity of ngc 1399 , although a single gaussian can not be rejected at better than the 95% confidence level according to a @xmath7 or a kolmogorov smirnov test . the mean velocity of the combined sample of globular clusters is @xmath8 km s@xmath4 , similar to the velocity derived for the stellar component ( see table 2 ) . no rotation is detected in the combined sample . the maximum rotation along any axis is found to be @xmath9 km s@xmath4 . a possible cause for the double peak in the velocity distribution may be contamination of the sample by globulars belonging to ngc 1404 or ngc 1380 . we discuss this possibility further below . in figure 3 we plot the cluster velocities against projected radius from the center of ngc 1399 . the mean velocity does not change significantly with radius ( see also table 3 ) . however , we note that within 4 the velocities seem to cluster around the mean velocity of the stellar component of ngc 1399 , whereas beyond 5 the globular clusters seem to have either significantly higher or lower velocities than the mean . this is better illustrated in fig . 4 , where we have plotted histograms of the velocities in concentric annuli . using the _ rostat _ package ( beers et al . 1990 , bird & beers 1993 ) , we tested the samples for normality and unimodality . while the globular clusters in the inner ring are consistent at the 98% confidence level with a normal distribution , the statistics are inconclusive in the middle ring and inconsistent with normality in the outer ring at the 90% confidence level . however , the number statistics are rather small , and the distribution in the outermost annulus is only consistent with multi modality at the 80% confidence level . it seems reasonable to question whether beyond 6 one is starting to sample globular clusters belonging to ngc 1404 , which lies only 9 in projection from ngc 1399 . the answer comes from fig . 1 : there is no obvious sign of any concentration of high velocity clusters in the direction of ngc 1404 . the second velocity peak is therefore probably not due to contamination of the sample by globular clusters belonging to ngc 1404 . the recently derived density profile for the globular clusters around ngc 1404 ( forbes et al . 1998 ) indeed shows that contamination of our sample by ngc 1404 clusters is likely to be small ; half way between ngc 1399 and ngc 1404 , the surface density of globular cluster belonging to ngc 1399 is @xmath10 times higher than that of ngc 1404 . even at 5 r@xmath11 ( @xmath12 2.5 ) from ngc 1404 the contribution of the two galaxies is still equal . this makes it unlikely for ngc 1404 clusters to strongly influence our sample . however , we can not exclude the possibility that some of the globular clusters in our sample may have been stripped from ngc 1404 or ngc 1380 at some point in the past ( see sect . 5.3 ) . we have searched for evidence of rotation in various annular sub samples . no significant rotation is found in the inner regions . however , in the outer regions ( @xmath135 ) we find a maximum rotation amplitude of @xmath14 km s@xmath4 along a position angle @xmath15 degrees ( east of north ) , roughly along the isophotal major axis of ngc 1399 ( @xmath16 degrees , extrapolating from the measurements of goudfrooij et al . figure 5 shows the velocities of globular clusters with projected radii greater than 5 versus their position angles . the velocity profile predicted by the rotation amplitude computed above is shown superimposed . we ran monte carlo simulations ( 1000 realizations , computing random data points with the same velocity dispersion , velocity errors and position angles as spanned by our sample ) to estimate the significance of this result : the hypothesis of no rotation for this sample is ruled out at the 95% confidence level . ostrov et al . ( 1993 ) , kissler - patig et al . ( 1997a ) and forbes et al . ( 1998 ) found the color distribution of ngc 1399 globular clusters to exhibit two distinct peaks . we split the sample into red and blue globular clusters ( at v@xmath1i = 1.05 , following kissler - patig et al . the small number statistics do not allow us to see any difference in the kinematics of the blue and red samples . we only note that our sample is dominated by red clusters inside 5 ( mean v@xmath1i = @xmath17 , where the error is the standard error of the mean ) , while blue and red clusters are about equal in number beyond 5 from the center of ngc 1399 ( mean v@xmath1i = @xmath18 ) . in summary , rotation of the ngc 1399 globular cluster system as a whole can not account for the possibly double - peaked nature of the velocity histogram . however , we can not rule out relatively high angular momentum orbits among a subset of the sample clusters . such orbits would be a natural consequence of tidal interactions or mergers of cluster galaxies passing through the core of the fornax cluster ( cf.below ) . the mean velocities and velocity dispersions of globular clusters reported in previous studies ( the aat , ntt , and keck samples ) are given in table 3 , together with the values for our combined sample and the various sub samples discussed in previous sections . the mean velocity and velocity dispersions were computed using a maximum likelihood dispersion estimator ( see pryor & meylan 1993 ) . the dispersions of the different samples around a fixed mean ( taken to be the mean of the full sample , i.e. , 1429 km s@xmath4 ) are also given . we find that the velocity dispersion measured for the combined sample agrees with those derived individually for the aat , ntt , and keck samples . more interesting is the fact that the velocity dispersion increases with radius . this behavior is clearly shown by a locally - weighted , scatter - plot smoothing fit to the velocity deviation squared ( lowess , see cleveland & mcgill 1984 , as implemented by gebhardt et al . 1994 ) and shown as a solid line in fig . 6 , on top of the data for the globular clusters . also shown in fig . 6 are the measured velocity dispersions of the stars , the planetary nebulae , and the fornax cluster galaxies . the difference between the line of sight velocity dispersions of the globular clusters and the stars could be explained by very different density profiles of the two components , or by different degrees of anisotropy . wagner et al . ( 1991 ) , bridges et al . ( 1991 ) , kissler - patig et al . ( 1997a ) and forbes et al . ( 1998 ) found little or no difference in the surface density profiles of the stars and the globular clusters , accounting for @xmath1910% of the difference in the velocity dispersions . a difference in the degree of anisotropy is not excluded and assuming the stellar velocity dispersion to be isotropic , the higher velocity dispersion of the globular clusters would require that the tangential velocity dispersion of the globular cluster system is at least 30% larger than its radial component ( following mould et al . however , given that the stellar and globular cluster data are sampling different radial regimes , we could equally well postulate that the stars and the globular clusters are both isotropic systems . in this case , the differences in the measured velocity dispersions of the stars and the globular clusters would arise from a strong increase in the potential field strength with radius . as we show below , even assuming the extreme case that all globular clusters are on circular orbits , we find a mass luminosity ratio several times larger than that which has been derived for the stellar component within 1.5 of ngc 1399 . the spectroscopic measurements of the integrated stellar light almost reach the range in radius where the innermost globular clusters are measured . together with the data for the planetary nebulae , there is a consistent indication of a marked up turn in the velocity dispersion between 1 and 3 from ngc 1399 . this is seen independently in each of the stellar , planetary nebula , and globular cluster data . regardless of whether the orbit shapes are changing or the fornax cluster potential is becoming dominant , this suggests that all luminous components are similarly affected . this is not a trivial result ; whereas we find only little evidence for rotation in the globular cluster system , arnaboldi et al . ( 1994 ) claim that the system of planetary nebulae exhibits a rotation amplitude of @xmath20 km s@xmath4 with a pa @xmath21 , i.e. counter rotating with respect to the globular clusters . if real , this might constitute an important clue concerning the manner in which the cd envelope of ngc 1399 was built up . following grillmair et al . ( 1994 ) , we determine whether or not the increase in the measured velocity dispersion at larger radii must be attributed to an increase in @xmath22 by assuming the extreme case that all of our observed globular clusters are on perfectly circular orbits ( which for a given potential field strength will produce the largest observed line - of - sight velocity dispersion ) . assuming a spherical distribution of mass in hydrostatic equilibrium , and that the surface density of globular clusters and the stellar light both follow the surface brightness parameterization of killeen & bicknell ( 1988 ) , we use equations ( 2 ) and ( 3 ) of grillmair et al . ( 1994 ) to solve in a least - squares sense for a radially invariant @xmath22 ratio . sampling the velocity dispersion profile over the same distribution of projected radii as in our combined sample , we find that our observations are best fit using @xmath23 m@xmath24/l@xmath24 , where the quoted error reflects only the uncertainty in velocity dispersion computed for the full sample in table 3 . in a way similar to minniti et al . ( 1998 ) we use the projected mass estimator ( heisler et al . 1985 ) , under the hypothesis of a spherical distribution of matter and isotropy of the velocity dispersion . for the 74 globular clusters , we obtain @xmath25 . this mass estimate , along with @xmath26 ( grillmair et al . 1994 , scaled to our adopted distance ) , gives @xmath27 . the error in this quantity is dominated by the small sample size and the different underlying theoretical assumptions . e.g. , adopting radial instead of isotropic orbits for the globular clusters increases the mass estimate by a factor of two . other mass estimators ( heisler et al . 1985 ) applied to the same sample give the following results : the virial mass @xmath28 implies @xmath29 , the median mass @xmath30 implies @xmath31 , and the average mass @xmath32 implies @xmath33 . the virial , projected , and average masses share the same sensitivity to interlopers , i.e. , globular clusters from the intracluster medium or from the nearby galaxies . eliminating from the sample even a single such object can decrease the @xmath22 values by as much as 30% ( minniti et al . 1998 ) . these @xmath22 values , obtained from the globular cluster system of ngc 1399 , are similar to the corresponding values obtained with the same mass estimators by huchra & brodie ( 1987 ) from the globular cluster system of m87 . contrary to the velocity dispersion determinations , which suffer only from the observational errors on the radial velocities , the mass - to - luminosity estimates accumulate the errors associated with many various observed parameters , such as the total luminosity within a given radius , the distance modulus and the radial velocity of the host galaxy , not to mention the different theoretical assumptions . nevertheless , all our @xmath22 determinations ( 50 200 ) are larger than that of a typical old stellar population ( @xmath34 ) , supporting the existence of a substantial amount of dark matter . we also confirm that @xmath22 is not constant with radius . these values are significantly larger than the value @xmath35 determined by bicknell et al . ( 1989 ) , from the stellar velocity dispersion measurements at radii @xmath1986 . there is therefore strong evidence for a change in the character of the gravitational potential at about 2 3 from the center of ngc 1399 . we conclude , as suggested by grillmair et al . ( 1994 ) , that most of the globular clusters in our sample are orbiting in a gravitational potential which is more closely associated with the galaxy _ cluster _ rather than with ngc 1399 itself . as the velocity dispersion profile measured for planetary nebulae at large radii is very similar to that of the globular clusters , we must conclude that the stars in the outer cd envelope are similarly associated with the potential of the fornax cluster as a whole ( see also arnaboldi et al . 1996 , mendez et al . . this would be consistent with the recent detection of red giant branch stars in the intergalactic region of the virgo cluster ( ferguson , tanvir , & von hippel 1998 ) . recent radial velocity measurements by cohen & ryzhov ( 1997 ) of globular clusters in m87 show similar velocity dispersions for stars and globular clusters within 100 of the center of m87 , but a steadily rising velocity dispersion from there outward . these authors are similarly driven to conclude that the mass - luminosity ratio increases substantially with radius . this similarity may not be surprising as m87 and ngc 1399 both occupy the center of their respective cluster potentials . an examination of the velocity distribution of 68 fornax galaxies studied by ferguson ( 1989 ) using the statistical tests described in sect . 3 reveals multiple peaks of marginal significance similar to those apparent in figure 2 . this motivated us to look for sub structure within the fornax cluster . we applied a dressler & shectman ( 1988 ) test on the entire sample of galaxies ; with the exception of a group of galaxies surrounding ngc 1316 ( fornax a , located at a projected distance of @xmath12 1.2 mpc sw of ngc 1399 ) , we found no evidence for sub structure . the fornax cluster appears to be very homogeneous . this is in good accord with the general view that fornax is a very compact , relatively relaxed cluster . leaving the group around ngc 1316 out of the velocity distribution considerably diminishes the statistical significance of the double peak ( to @xmath36% ) . the mean velocity of the 57 remaining galaxies in fornax is @xmath37 km s@xmath4 , consistent with the systemic velocity of ngc 1399 . this confirms that ngc 1399 sits at the center of the cluster potential well . the velocity dispersion of the galaxy cluster rises from @xmath38 km s@xmath4 in the outer parts ( several degrees from the cluster center ) to @xmath39 km s@xmath4 within 24 of ngc 1399 ( den hartog & katgert 1997 ) . there is very good agreement ( figure 6 ) between the velocity dispersions of the galaxies and the globular clusters . jones et al . ( 1997 ) find that the hot x - ray gas around ngc 1399 has a temperature of @xmath40 kev ( derived within an annulus extending from 2 to 18 ) . this is energetically equivalent to a velocity dispersion of @xmath41 km s@xmath4 . the temperature profile , converted into a velocity dispersion profile , is shown as the dashed line in fig . once again the agreement with the globular cluster measurement is excellent . ikebe et al . ( 1996 ) showed that the x - ray gas profile could be modeled as the sum of two components , perhaps responding to the potentials of ngc 1399 and the fornax cluster , respectively . the potential of the galaxy evidently falls off steeply and only clearly dominates over the cluster potential in the inner 2 to 3 . beyond 5 or 6 , the potential of the cluster dominates . in order to better understand the origin of the ngc 1399 globular cluster system , with its high velocity dispersion and specific frequency ( number of globular clusters per unit luminosity , harris & van den bergh 1981 ) , we pose the question whether the properties of these globular clusters are peculiar in any way . we briefly compare the globular cluster system of ngc 1399 with those of the next brightest early type galaxies in fornax , ngc 1380 and ngc 1404 . we then discuss our findings within the framework of different scenarios that could explain the globular cluster over abundance . note that ngc 1399 is traditionally associated with high@xmath42 galaxies , although ostrov et al . ( 1998 ) recently suggested that it has only a moderate over abundance . the new value is a consequence of taking proper account of the light in the extended cd halo . the @xmath42 value for ngc 1399 , galaxy and cd envelope is still about a factor of two higher than for other galaxies in fornax . in particular , the @xmath42 of the cd envelope reaches values a factor of 3 higher than the mean of the brightest surrounding ellipticals , and when the globular clusters are associated with the `` galaxy '' component alone , the @xmath42 reaches values of a factor 5 to 6 higher than the fornax mean . in the following we will distinguish between ngc 1399 including the cd envelope , the `` galaxy '' component alone ( to which we associate the light of a de vaucouleurs fit to the central region ) , and the cd envelope . @xmath43 and @xmath42 for the different components are listed in table 2 . the globular cluster system of ngc 1399 has been extensively studied photometrically ( hanes & harris 1986 , geisler & forte 1990 , wagner , richtler & hopp 1991 , bridges , hanes & harris 1991 , ostrov , geisler & forte 1993 , kissler - patig et al . 1997a , forbes et al . 1998 , ostrov et al . 1998 , grillmair et al . 1999 ) . the most striking result of these investigations is the high number of globular clusters surrounding the galaxy , and the over abundance in the number of globular clusters per unit light when compared with the other `` normal '' cluster ellipticals ( see table 2 ) . another interesting finding of the more recent studies is that the globular cluster color distribution has at least two significant peaks , suggesting two or more sub populations . the photometric studies found that these peaks are separated by about 1 dex in [ fe / h ] and , taking into account the luminosity function , that the majority of the globular clusters are as old ( within a few gyr ) as the oldest globular clusters in the galaxy . the nature of these sub populations was explored spectroscopically by kissler - patig et al . the absorption line indices for the two major sub populations of globular clusters were found to be very similar to the ones observed in globular clusters around m31 and the galaxy , suggesting similar formation epochs and mechanisms . thus the majority of globular clusters around ngc 1399 appear to be `` normal '' old globular clusters . a small fraction of very red globular clusters are significantly more metal rich , and could perhaps have formed in a later merger . do ngc 1380 and ngc 1404 have globular cluster systems comparable to that of ngc 1399 ? these two galaxies are the next brightest early type galaxies in fornax ( neglecting currently star forming ngc 1316 ) . ngc 1380 and ngc 1404 are 1.4 and 1.5 magnitudes fainter , respectively , than the central cd galaxy , and only 0.3 and 0.4 magnitudes fainter if the light of the cd envelope is neglected ( see table 2 ) . these galaxies are projected only 38 ( 208 kpc ) and 9 ( 48 kpc ) from ngc 1399 and are presumably well within the embrace of the fornax cluster potential . while no spectroscopic observations have yet been made of the globular clusters surrounding these other two galaxies , a considerable amount of photometric work has been published ( ngc 1380 : kissler - patig et al . 1997b ; ngc 1404 : hanes & harris 1986 , richtler et al . 1992 , forbes et al . 1998 , grillmair et al . qualitatively , the globular cluster systems of ngc 1399 , ngc 1404 and ngc 1380 look very similar . all three systems exhibit broad color distributions with two ( or possibly three ) peaks ( ostrov et al . 1993 , kissler - patig et al . 1997a , b , forbes et al . 1998 , grillmair et al . 1999 ) . the ratio of red to blue globular clusters is roughly the same in all three cases , and between 1 and 2 when integrated over the whole system . furthermore , the two main sub populations peak at similar metallicities ( as derived from broad band colors ) in the three systems . the blue populations are distributed around a metallicity typical for the galaxy or m31 halo globular clusters , i.e. , @xmath44 [ fe / h ] @xmath45 dex . the red populations seem to have metallicities slightly higher than disk / bulge globular clusters in the galaxy ( i.e. , @xmath46 [ fe / h ] @xmath47 dex , cf.minniti 1995 , although the former could be slightly over estimated from the broad band colors , see kissler - patig et al . 1998 ) . finally , the globular cluster luminosity functions of all three systems look similar and are well represented by a gaussian with a @xmath48 mag dispersion and peak at about the same magnitude to within the errors ( see table 4 ) . assuming that the underlying globular cluster mass distributions are invariant , this result suggests that the globular clusters in the three galaxies have similar old ages . the only significant difference between the globular cluster system of ngc 1399 and those of ngc 1404 and ngc 1380 is the much higher number of clusters around ngc 1399 . quantitatively , ngc 1399 is surrounded by @xmath49 globular clusters , while ngc 1404 and ngc 1380 host only @xmath50 and @xmath51 globular clusters respectively ( see table 2 ) . in terms of specific frequency , ngc 1399 has @xmath52 ( @xmath53 for the `` galaxy '' component ) ngc 1404 has @xmath54 , and ngc 1380 has @xmath55 . using the distance in table 2 and the magnitudes and colors of the galaxies from the rc3 , the four smaller early type galaxies in fornax ngc 1387 , ngc 1374 , ngc 1379 , ngc 1427 ( kissler - patig et al . 1997a ) have specific frequencies between 2.1 and 3.2 with a mean of @xmath56 ( where the error is the dispersion around the mean ) . ngc 1399 , or rather the center of the fornax cluster , is outstanding in its relative overabundance of globular clusters , while ngc 1404 and ngc 1380 have cluster populations close to the mean expected value . in summary , the globular clusters around ngc 1399 are qualitatively very similar to the ones surrounding ngc 1380 and ngc 1404 , and appear `` normal '' when compared to the globular clusters of the galaxy . however , the central galaxy hosts a factor 10 more globular clusters than its neighbors and has a specific frequency 2 to 3 times higher than expected . based on all the evidence summarized above , we are led to two conclusions : _ i ) _ the excess globular clusters do not have unusual properties , i.e. , they are likely to have formed in the same kind of process and at a similar time as the globular clusters in ngc 1404 and ngc 1380 , and _ ii ) _ the excess number of globular clusters is most likely linked to the special location of ngc 1399 in the center of the cluster . various scenarios to explain the high @xmath42 values around central galaxies have been comprehensively discussed by blakeslee et al . these authors found several correlations of specific frequency with the galaxy cluster properties , including x - ray temperature , velocity dispersion , and galaxy distance from the center of the cluster . in summary , _ ab initio _ scenarios ( harris 1991 and references therein ) that associate the excess numbers of globular clusters to super efficient globular cluster formation in these galaxies fail to explain the correlation of @xmath42 , cluster x - ray properties and velocity dispersion . mergers as a source of the increase in @xmath42 ( e.g. ashman & zepf 1992 ) also fail to explain these correlations and have difficulty explaining the ratio of blue and red globular clusters if the red globulars are thought to be responsible for the high @xmath42 values ( forbes , brodie & grillmair 1997 ) . scenarios which involve biased formation of globular clusters in deep potential wells ( west 1993 ) and intra cluster globular clusters ( west et al . 1995 ) are largely unconstrained , and blakeslee et al . can not rule them out . tidal stripping can easily explain all of their observed correlations ; the problem , as noted by blakeslee et al . , is that the stripping simulations ( summarized in muzzio 1987 ) yield too slow an increase in @xmath42 of the central galaxy . finally , motivated by the constant luminosity of brightest cluster galaxies ( bcgs ) , blakeslee et al . propose a new scenario wherein globular clusters formed early and their numbers scaled with the available mass in the galaxy cluster , whereas the luminosity of the bcgs is relatively independent of the cluster mass and is truncated at a given luminosity . the upshot is that abnormally high @xmath42 values would be interpreted as a _ luminosity deficiency _ rather than a globular cluster over abundance ( see also harris et al . ( 1998 ) for further support to this idea ) . based on the observations presented above , we concur that excessive numbers of globular clusters are likely to be linked to the properties ( x - ray temperature ; velocity dispersion ) of the galaxy cluster and the special location of high-@xmath42 galaxies . any scenario that purports to explain the excess of globular clusters around ngc 1399 must be consistent with the finding that their properties are very similar to those of globular clusters in ngc 1404 and ngc 1380 . as discussed above , several sub populations have been identified in ngc 1399 . the two major ones , making up over 90% of the system , are old and have mean metallicities of [ fe / h]@xmath57 and @xmath58 dex respectively . the ratio of these two populations changes slowly with radius , varying from about 2:1 ( red / blue ) in the center of the galaxy to about 1:2 in the cd halo , with an overall average close to 1:1 ( see kissler - patig et al . 1997a and forbes et al . 1998 ) . in principal , the excess globular clusters could be made up by metal rich or metal poor globular clusters only . both cases would be inconsistent with ngc 1404 and ngc 1380 having also metal poor _ and _ metal rich clusters , but showing no over abundance . this suggests that the excess of globular clusters ( especially around the `` galaxy '' component ) must be made up of both old , metal poor _ and _ old , metal rich clusters . this complicates scenarios wherein all the excess clusters are formed at a very early stage , before the galaxies fully assembled ( i.e. , west et al . 1993 ; blakeslee et al . 1997 ; harris et al . 1998 ) . such scenarios would require that ngc 1399 have a specific frequency about a factor of two higher than normal , but that all of the extra clusters be metal poor . a mechanism that can explain at least some increase in the number of red globular clusters seems required . recent work by aragn - salamanca et al . ( 1997 ) puts further constraints on scenarios which require an excess number of globular clusters to have formed at very early times . these authors see no ( or negative ) passive evolution in the k band hubble diagram of bcgs . furthermore , they detect no young stellar populations . they conclude that the total mass of stars in bcgs has grown by a factor of 2 ( 4 ) since @xmath59 for @xmath60 ( @xmath61 ) , and can not be explained by new stars forming in mergers or cooling flows . bcgs must therefore have accreted or annexed mass in the form of old stars ( and presumably old globular clusters ) since @xmath59 ; either by cannibalism accreting gas poor galaxies or by stripping of material from other galaxies . irrespective of the high _ total number _ of globular clusters , the high specific frequency , when compared to the neighbor galaxies , also needs to be taken into account . if ngc 1399 gained mass simply by dissipationless mergers of typical galaxies surrounding it , then the specific frequency would _ decrease _ ( by about 30% for each doubling of the mass for ngc 1399 ) , since no neighboring galaxy has a @xmath42 value much higher than 3 . dissipationless merging therefore seems unlikely to be the main cause for the growth of the central galaxy since @xmath59 , unless the specific frequency of ngc 1399 was even higher at early times . we would then need an explanation for the original , even higher , specific frequency . it seems to us more reasonable to suppose that the mass gained by ngc 1399 consisted of accreted material with an intrinsically high specific frequency of globular clusters that matches the @xmath62 of the cd halo ( see table 2 ) . potential accretion candidates with high specific frequencies include faint dwarf galaxies and galactic halos . durrel et al . ( 1996 ) compiled a table of @xmath42 values for local group dwarfs and several dwarf ellipticals in virgo . they find that de galaxies with @xmath63 tend to have a `` normal '' specific frequency ( @xmath64 ) , whereas their two faintest de , n galaxies have very high @xmath42 values ( @xmath65 and @xmath66 respectively ) . at least two other , local group , low luminosity dwarf galaxies are also known to exhibit very high values : fornax ( @xmath67 ) and sagittarius ( @xmath68 ) . miller et al . ( 1998 ) recently enlarged the sample of de galaxies with studied globular cluster systems by 24 and extended durrel et al.s results in the sense that their de , n galaxies with @xmath69 also have @xmath42 values around 10 or above . however , brighter and non nucleated des appear to have more normal specific frequencies . simulations by ct et al . ( 1998 ) indicate that the accretion / stripping of a large number of dwarf galaxies could explain the color distributions seen around giant ellipticals , assuming that every individual galaxy forms globular clusters with a mean metallicity proportional to its size . however , the presence of a large number of such nucleated dwarf galaxies remnants at the current epoch around ngc 1399 can be ruled out ( hilker 1998 ( chapter 5 ) , hilker et al . 1998a , b and references therein ) . hilker ( 1998 ) further showed in monte carlo simulations that dwarf galaxies ( or their halos ) can not explain all the excess of globular clusters around ngc 1399 , even if one assumes an extremely steep faint end of the galaxy luminosity function in the past and efficiently accrete all the dwarf galaxies `` missing '' in todays luminosity function onto ngc 1399 . therefore , while accretion of dwarf galaxies may have played a role in the enrichment of the globular cluster system of the central galaxy , it can not be its only cause . another source of high specific frequency material may be galactic `` halos '' . extended globular cluster systems often fall off more slowly with radius than the stellar light profile , leading to gradually increasing specific frequencies in the outskirts of the distributions . recent studies have identified distinct `` halo '' globular cluster populations in early type galaxies ( kissler - patig et al . 1997b ; lee , kim & geisler 1998 ) comprised of clusters following a somewhat shallower density profile . m87 is a clear case , for which mclaughlin et al . ( 1994 ) computed an increase of @xmath42 from values around 10 inside 1 r@xmath70 to 25 at @xmath135 r@xmath6 . unfortunately , such studies do not yet exist for isolated field galaxies , but `` halo '' material can be considered a plausible candidate for the buildup of the envelope of ngc 1399 ( see also kissler - patig 1997a and forbes , brodie , & grillmair 1997 ) . we note in passing that , in the galaxy formation models of the searle & zinn ( 1978 ) type , halos are built up of fragments resembling faint dwarf galaxies . several authors have investigated the importance of stripping and harassment for the formation of central cd galaxies ( see e.g. , dressler 1984 for an early review ) . following these ideas , harris ( 1986 ) suggested that the central regions of galaxy clusters could have accumulated globular clusters stripped from their parent galaxies . white ( 1987 ) suggested that these clusters could have built up the over abundant globular cluster systems around cd galaxies . muzzio ( 1987 ) summarized the results of various simulations of globular cluster stripping in galaxy clusters . some of his conclusions were : @xmath71 ) stripping of globular clusters was most efficient before the galaxies and cluster were virialized , @xmath72 ) the initial extent of the halos is not critical , @xmath73 ) the location in the cluster and the amount of dark matter determine the number of encounters and the amount of stripping a galaxy suffers , and @xmath74 ) the brightest galaxies tend to gain globular clusters , but capture most of these from galaxies only somewhat fainter than themselves . intermediate luminosity galaxies are the ones which lose their globular clusters ; faint or dwarf galaxies are hardly affected . unfortunately , the exact properties of the resulting globular cluster systems depend too much on the initial conditions and the details of the simulations ( which had rather poor resolution ) to allow detailed predictions . we discuss the plausibility of the `` stripping '' scenario in more detail below . we first determine whether the total number of globular clusters seen in the early type galaxies in fornax requires that an additional formation process be invoked , or if it can be explained by a simple redistribution of existing clusters . in other words , do the galaxies in the cluster form a reservoir of globular clusters large enough to feed the central galaxy ? to answer this question , we assume that all galaxies had the same number of globular clusters per unit luminosity to begin with , and that the excess around ngc 1399 must be accounted for by the losses suffered by other galaxies . what , then , was the initial `` universal '' specific frequency in fornax ? table 5 summarizes the result . it shows the number of globular clusters observed today in each of brightest early type galaxies in the center of the fornax cluster , the number of globular clusters initially resident in each galaxy for an assumed `` universal '' specific frequency , and the number which had to be gained or lost . assuming an initial @xmath75 for ngc 1399 , ngc 1404 , ngc 1380 , ngc 1374 , ngc 1379 , ngc 1387 , and ngc 1427 , one can account for all the globular clusters seen in these galaxies today . if we assume that the next seven brightest early type galaxies ( ngc 1336 , ngc 1339 , ngc 1351 , ic 1963 , ngc 1380a , ngc 1381 , ngc 1389 , taken from ferguson 1989 ) , all with total magnitudes brighter than @xmath76 , suffered the same losses as the other non central galaxies , the initial specific frequency could have been as low as 3 . note that the largest losses were suffered by the brightest galaxies . including contributions from dwarf galaxies could lower the required initial mean value of @xmath42 still further . even if all globular clusters were associated with the `` galaxy '' component , and todays @xmath42 value for ngc 1399 was taken to be @xmath1310 , a similar calculation would lead to an initial @xmath42 of @xmath12 4 for all fornax galaxies . @xmath42 values of 3 to 4 are not unusual among early type , cluster ellipticals . the mean for all early type galaxies with @xmath77 in various compilations is @xmath78 ( harris 1991 ) , @xmath79 ( ashman & zepf 1998 ) and @xmath80 ( kissler - patig 1997b ) , where the quoted error is the dispersion about the mean . _ however _ , these numbers need to be confirmed for a sample of isolated field ellipticals with absolute magnitude @xmath81 before we can say with confidence that this is likely to be representative of cluster ellipticals prior to any interactions . nonetheless , the point we wish to stress is that _ the excess population around the central galaxy in fornax does not require the production of a large number of new globular clusters_. a simple redistribution , separating globular clusters from their parent galaxies through tidal encounters , is sufficient to account for the observations . ngc 1399 ( or perhaps more correctly the central potential of the fornax cluster ) would have gained about 65% of its current light ( assuming for the galaxy without envelope @xmath82 , see table 2 ) , and 75% of its current globular clusters through stripping and harassment . ngc 1380 and ngc 1404 would have lost 30% to 50% of their globular clusters , and all other intermediate size galaxies would have lost between 0% and 30% of their initial globular cluster populations . in the above , we neglected the possible existence of a population of intra cluster globular clusters ( west et al . 1995 , blakeslee 1996 ) which was _ never _ associated with a cluster galaxy . if indeed such globular clusters formed at early times , then the numbers of globulars removed from cluster galaxies needs to be even less , and the initial specific frequency could be even lower than the values we have estimated above . however , little is yet known concerning such a population , making any estimate of its importance uncertain . mergers , both gas - rich and gas - poor , must presumably have occurred during the evolution of the fornax cluster . however , rather than increasing the specific frequency as has been often suggested ( e.g. , zepf & ashman 1993 ) , such mergers could just as easily conserve @xmath42 ( e.g. , van den bergh 1995 ) . kissler - patig et al . ( 1997a ) showed that the combined color distribution of globular clusters in the intermediate luminosity galaxies surrounding the central galaxy ( ngc 1374 , ngc 1379 , ngc 1387 , and ngc 1427 ) is also consistent with the observed color distribution in ngc 1399 , and that this is compatible with the stripping scenario . unfortunately , as noted above , few simulations of the stripping of globular clusters in galaxy clusters exist . muzzio ( 1987 ) summarized a number of simulations but was unfortunately not able to make clear predictions . globular cluster systems are expected to become extended in response to the tidal stresses imposed by neighboring galaxies . however , it is not clear whether by this time we should expect to see surface density profiles which are tidally truncated , or whether they should appear extended . tidal encounters will naturally lead to the growth of tidal tails and to the possible destruction of numerous globular clusters ( gnedin & ostriker 1997 , combes et al . the onset of such tidal tails may be detectable in the form of a `` break '' in the surface density profile ( grillmair et al . 1995 ; 1999 ) . currently available information on the globular cluster system of ngc 1399 seems to favor the view that its high specific frequency mainly results from tidal stripping of relatively high@xmath42 material from neighboring galaxies . if this is the case for other nearby central cluster galaxies , it would explain the correlations of @xmath42 with galaxy cluster properties ( blakeslee et al . 1997 ) . in the case of ngc 1399 , stripping is consistent with all known properties of the globular cluster system , including the total number of globular clusters in the cluster galaxies . while this seems sensible , detailed simulations will be needed to check this suggestion . a potential remaining problem , as already pointed out by blakeslee et al . 1997 , is the inefficiency of stripping once galaxy clusters are virialized . this apparent problem would however be resolved if a significant fraction of the tidal stripping occurred during the early phases of galaxy formation , when the halos of galaxies first collapsed and the old , metal poor globular cluster population formed . however , tidal effects would continue to pull clusters from their host galaxies until the present day , although at reduced efficiency . this , as well as dissipative post collapse merger events , would allow for the addition of more metal rich globulars to the mix . all the constraints discussed in the above sections would remain satisfied . we have combined data from three separate spectroscopic investigations to study the kinematics of 74 globular clusters around ngc 1399 . the velocity dispersion of the globular clusters increases with radius , rising from a value not unlike that for the outermost stellar measurements at 2 @xmath83 , to values almost twice as high at @xmath135 @xmath83 . the outer velocity dispersion measurements are in good agreement with the temperature of the x - ray gas and the velocity dispersion of galaxies in the fornax cluster . we conclude , as already suggested by grillmair et al . ( 1994 ) , that a large fraction of the globular clusters which we associate with ngc 1399 should rather be attributed to the whole of the fornax cluster . no significant difference in the kinematics could be found between the blue and red globular cluster sub populations , but there is some evidence for rotation in the outer ( @xmath84 ) regions . a qualitative comparison of the globular cluster systems of ngc 1399 and neighboring , next brightest early type galaxies ngc 1404 and ngc 1380 indicates that these systems are indistinguishable from one another , and that there is no reason to suppose that they formed at significantly different epochs or via a different sequence of events . the ngc 1399 globular cluster system is distinguished only in being a factor of 10 more abundant than the cluster systems of either of the other two galaxies , and having a specific frequency 2 to 3 times higher . the excess is best understood if a significant fraction of the globular clusters is indeed associated with the light of the cd envelope . by association this would mean that the cd envelope around ngc 1399 should rather also be associated with the fornax cluster than with the galaxy itself . we review different scenarios to explain the high specific frequencies around the central galaxies and examine the consequences of various existing constraints for each . we come to the conclusion that tidal stripping of globular clusters from neighboring galaxies in the early history of the galaxy cluster and the consequent buildup of the cd envelope in the fornax cluster potential well is the most likely explanation . we would like to thank ann zabludoff for her improved code of the dressler & shectman test as well as useful comments , and karl gebhardt for his lowess code . we are also thankful to steve zepf and bill mathews for interesting discussions . mkp gratefully acknowledges the support of the alexander von humboldt foundation . part of this research was funded by the faculty research funds of the university of california at santa cruz , the hst grant go.06554.01 - 95a , and the u.s . department of energy by lawrence livermore national laboratory under contract w7405-eng-48 . l rl r rl rl l aat 1 & 3 35 47.3 & @xmath135 40 21 & @xmath85 & & & 21.7 & 1.31 & aat 4 & 3 35 49.2 & @xmath135 36 44 & @xmath86 & & & 22.3 & 1.04 & aat 5 & 3 35 50.1 & @xmath135 38 48 & @xmath87 & & & 21.3 & 1.32 & aat 6 & 3 35 51.4 & @xmath135 37 52 & @xmath88 & & & 21.9 & 1.13 & aat 7 & 3 35 52.0 & @xmath135 32 44 & @xmath89 & & & 21.5 & 0.98 & aat 8 & 3 35 54.8 & @xmath135 38 09 & @xmath90 & & & 22.4 & 0.84 & aat 10 & 3 35 58.4 & @xmath135 36 03 & @xmath91 & & & 21.9 & 0.91 & aat 13 & 3 36 03.6 & @xmath135 31 58 & @xmath92 & 21.15 & 0.97 & 21.9 & 1.05 & aat 15 & 3 36 05.2 & @xmath135 39 54 & @xmath93 & & & 21.9 & 1.05 & aat 16 & 3 36 05.9 & @xmath135 34 20 & @xmath94 & 21.27 & 1.17 & 22.1 & 1.16 & aat 17 & 3 36 07.8 & @xmath135 34 44 & @xmath95 & 21.57 & 0.92 & 22.3 & 0.89 & aat 20 & 3 36 17.8 & @xmath135 38 42 & @xmath96 & & & 21.6 & 1.33 & aat 21 & 3 36 18.2 & @xmath135 40 52 & @xmath97 & & & 21.9 & 1.27 aat 25 & 3 36 22.7 & @xmath135 38 12 & @xmath98 & & & 22.4 & 1.19 aat 26 & 3 36 23.5 & @xmath135 37 25 & @xmath99 & 20.80 & 1.06 & 21.8 & 1.54 aat 27 & 3 36 14.7 & @xmath135 33 54 & @xmath100 & 21.85 & 0.96 & 22.3 & 0.81 aat 30 & 3 36 24.2 & @xmath135 42 07 & @xmath101 & & & 21.8 & 1.05 aat 31 & 3 36 24.7 & @xmath135 33 24 & @xmath102 & 20.59 & 1.01 & 21.4 & 1.23 aat 33 & 3 36 30.1 & @xmath135 39 10 & @xmath103 & & & 21.6 & 1.48 aat 34 & 3 36 35.1 & @xmath135 31 15 & @xmath104 & 20.78 & 1.01 & 21.6 & 1.17 aat 36 & 3 36 41.3 & @xmath135 41 56 & @xmath105 & 21.97 & 1.43 & 22.3 & 1.17 aat 38 & 3 36 43.0 & @xmath135 33 16 & @xmath106 & 20.85 & 1.18 & 21.7 & 1.40 aat 39 & 3 36 44.5 & @xmath135 38 31 & @xmath107 & 21.19 & 1.27 & 22.0 & 1.32 aat 40 & 3 36 44.4 & @xmath135 36 49 & @xmath108 & 20.78 & 1.21 & 21.8 & 1.54 aat 41 & 3 36 45.7 & @xmath135 38 53 & @xmath109 & 20.93 & 1.22 & 21.7 & 1.31 aat 42 & 3 36 45.8 & @xmath135 37 33 & @xmath110 & 20.74 & 1.42 & 21.5 & 1.69 aat 43 & 3 36 46.0 & @xmath135 32 26 & @xmath111 & 21.23 & 1.13 & 21.9 & 1.17 aat 48 & 3 36 52.4 & @xmath135 34 33 & @xmath112 & 21.71 & 1.16 & 22.4 & 1.19 aat 49 & 3 36 46.9 & @xmath135 35 44 & @xmath113 & 21.30 & 1.28 & 22.2 & 1.55 aat 54 & 3 36 51.9 & @xmath135 33 32 & @xmath114 & 21.04 & 0.97 & 21.9 & 1.29 aat 55 & 3 36 54.0 & @xmath135 37 27 & @xmath115 & 20.94 & 1.00 & 21.6 & 1.01 aat 56 & 3 36 54.1 & @xmath135 31 25 & @xmath116 & 21.39 & 1.15 & 21.7 & 1.29 aat 57 & 3 36 55.4 & @xmath135 31 51 & @xmath117 & 20.91 & 1.12 & 22.2 & 1.26 aat 59 & 3 36 59.9 & @xmath135 41 21 & @xmath118 & 21.12 & 0.75 & 21.5 & 1.24 aat 62 & 3 37 01.2 & @xmath135 40 49 & @xmath119 & 21.25 & 0.89 & 21.2 & 1.35 aat 66 & 3 37 10.1 & @xmath135 36 36 & @xmath120 & 21.17 & 1.12 & 21.8 & 0.98 aat 67 & 3 37 10.8 & @xmath135 38 42 & @xmath121 & 21.61 & 0.83 & 22.3 & 1.09 aat 68 & 3 37 14.3 & @xmath135 37 11 & @xmath122 & & & 21.6 & 1.18 aat 69 & 3 37 15.1 & @xmath135 33 46 & @xmath123 & & & 22.4 & 0.97 aat 71 & 3 37 21.5 & @xmath135 39 41 & @xmath124 & & & 22.4 & 1.06 & & & & & & & & ntt 201 & 3 36 44.2 & @xmath135 35 40 & @xmath125 & 21.17 & 1.24 & & ntt 203 & 3 37 02.9 & @xmath135 34 40 & @xmath126 & 20.69 & 1.08 & & ntt 208 & 3 36 55.0 & @xmath135 31 31 & @xmath127 & 20.91 & 1.12 & & ntt 407 & 3 36 09.8 & @xmath135 35 02 & @xmath128 & 20.19 & 1.01 & & ntt 410 & 3 36 17.7 & @xmath135 33 45 & @xmath129 & 19.83 & 1.27 & & ntt 414 & 3 36 15.5 & @xmath135 32 40 & @xmath130 & 19.56 & 1.09 & & ntt 101 & 3 36 55.4 & @xmath135 44 05 & @xmath131 & & & & ntt 109 & 3 36 58.9 & @xmath135 41 46 & @xmath132 & 21.24 & 1.27 & & & @xmath133 , aat 58 @xmath134 ntt 113 & 3 36 40.7 & @xmath135 40 46 & @xmath135 & 21.15 & 1.26 & & ntt 119 & 3 37 01.7 & @xmath135 38 18 & @xmath136 & 21.16 & 1.19 & & & @xmath137 , aat 63 @xmath138 ntt 122 & 3 37 03.5 & @xmath135 37 43 & @xmath139 & 20.77 & 1.06 & & ntt 123 & 3 36 54.1 & @xmath135 37 32 & @xmath140 & 20.93 & 1.00 & & ntt 124 & 3 36 43.0 & @xmath135 37 14 & @xmath141 & 21.18 & 1.25 & & ntt 125 & 3 36 41.0 & @xmath135 37 00 & @xmath142 & 21.02 & 1.17 & & ntt 126 & 3 36 44.7 & @xmath135 36 52 & @xmath143 & 20.76 & 1.22 & & ntt 127 & 3 36 43.5 & @xmath135 36 32 & @xmath144 & 21.06 & 1.16 & & & & & & & & & & keck 1 & 3 36 13.8 & @xmath135 39 24.8 & @xmath145 & & & 21.8 & & keck 2 & 3 36 14.2 & @xmath135 38 51.2 & @xmath146 & & & 22.4 & 1.19&keck 3 & 3 36 09.4 & @xmath135 37 32.4 & @xmath147 & & & 22.3 & 1.02&keck 5 & 3 36 13.2 & @xmath135 37 37.8 & @xmath148 & & & 21.8 & 1.17&keck 6 & 3 36 17.8 & @xmath135 37 50.2 & @xmath149 & & & 22.3 & & keck 7 & 3 36 16.7 & @xmath135 37 01.7 & @xmath150 & 21.01 & 1.23 & 21.7 & 1.31 & @xmath151 , aat 28 @xmath152keck 9 & 3 36 19.2 & @xmath135 36 28.7 & @xmath153 & 21.04 & 1.25 & 21.8 & 1.33 & @xmath154 , aat 29 @xmath155keck 10 & 3 36 21.5 & @xmath135 36 04.4 & @xmath156 & 20.55 & 1.05 & 21.4 & 1.50 & @xmath157 , ntt 406 @xmath158 , aat 24 @xmath159keck 11 & 3 36 25.0 & @xmath135 36 28.3 & @xmath160 & 21.34 & 1.17 & 22.2 & 1.50 & keck 12 & 3 36 20.3 & @xmath135 35 15.3 & @xmath161 & 21.97 & 0.94 & 22.4 & 1.06 & @xmath162 , aat 23 @xmath104keck 13 & 3 36 23.4 & @xmath135 35 37.2 & @xmath163 & 21.51 & 0.91 & 22.2 & & keck 14 & 3 36 24.5 & @xmath135 35 36.8 & @xmath164 & 21.17 & 1.37 & 22.1 & 1.65 & keck 15 & 3 36 23.2 & @xmath135 34 39.3 & @xmath165 & 21.26 & 1.04 & 22.0 & 1.26 & keck 17 & 3 36 26.3 & @xmath135 34 20.7 & @xmath166 & 21.55 & 1.14 & 22.4 & 1.33 & keck 18 & 3 36 31.4 & @xmath135 35 05.9 & @xmath167 & 21.32 & 1.08 & 22.3 & 1.50 & keck 19 & 3 36 34.5 & @xmath135 34 52.2 & @xmath168 & 21.41 & 1.29 & 22.3 & & keck 20 & 3 36 28.5 & @xmath135 33 17.0 & @xmath169&21.63 & 1.13 & 22.3 & 0.88 & keck 21 & 3 36 35.7 & @xmath135 34 24.6 & @xmath170&21.15 & 1.09 & 22.0 & 1.37 & @xmath171 , aat 35 @xmath172 ra ( 1950 ) & 03 36 34.0 & 03 36 57.0 & 03 34 31.0 & rc3 dec ( 1950 ) & @xmath135 36 42 & @xmath135 45 18 & @xmath135 08 24 & rc3 l & 236.71 & 236.95 & 235.93 & rc3 b & @xmath153.64 & @xmath173 & @xmath154.06 & rc3 type ( t ) & @xmath15.0 & @xmath15.0 & @xmath174 & rc3 & & & & @xmath175 km s@xmath4 & @xmath176 & @xmath177 & @xmath178 & rc3 @xmath179 & @xmath180 & @xmath180 & @xmath180 & della valle et al . 1998@xmath181 & & & & @xmath182 ( mag ) & @xmath183 & @xmath184 & @xmath185 & ostrov et al . 1998 , rc3 @xmath43 ( mag ) & @xmath186 & @xmath187 & @xmath188 & using @xmath179 above @xmath189 ( mag ) & @xmath190 & @xmath191 & @xmath192 & rc3 @xmath193 ( mag ) & @xmath194 & @xmath195 & @xmath196 & buta & williams 1995 @xmath197 & @xmath198 & @xmath199 & & goudfrooij et al . 1994 & & & & n@xmath200 & @xmath49 & @xmath50 & @xmath51 & kp97a , f97,r92,kp97b s@xmath201 & @xmath202 & @xmath203 & @xmath204 & derived from this table s@xmath201(halo ) & @xmath205 & & & ostrov et al . 1998@xmath206s@xmath201(galaxy ) & @xmath207 & & & r r l l @xmath208 & @xmath209 & & aat sample @xmath181@xmath210 & @xmath211 & & ntt sample @xmath206@xmath212 & @xmath213 & & keck sample @xmath214 & & & @xmath215 & @xmath216 & & full sample & & & @xmath217 & @xmath218 & @xmath219 & 6 globular clusters within 2@xmath220 & @xmath221 & @xmath222 & 20 globular clusters within 3@xmath223 & @xmath224 & @xmath225 & 41 globular clusters within 5@xmath226 & @xmath227 & @xmath228 & 33 globular clusters outside 5@xmath229 & @xmath230 & @xmath231 & 23 globular clusters outside 6@xmath232 & @xmath233 & @xmath234 & 10 globular clusters outside 8 & & & @xmath235 & @xmath236 & @xmath237 & 16 blue globular clusters ( v@xmath1i@xmath238 ) @xmath239 & @xmath240 & @xmath241 & 36 red globular clusters ( v@xmath1i@xmath242 ) @xmath243 & @xmath244 & @xmath245 & 11 red globular clusters outside 5 arcmin @xmath246 & @xmath247 & @xmath248 & 7 blue globular clusters outside 5 arcmin r c l ngc 1404 & @xmath249 @xmath250 & richtler et al . 1992 & @xmath251 @xmath252 & blakeslee & tonry 1996 & @xmath253 @xmath254 & grillmair et al . 1999 ngc 1380 & @xmath255 @xmath256 & blakeslee & tonry 1996 & @xmath257 @xmath258 & della valle et al . 1998 ngc 1399 & @xmath259 @xmath260 & kohle et al . 1996 & @xmath261 @xmath262 & blakeslee & tonry 1996 & @xmath263 @xmath264 & bridges et al . 1991 & @xmath265 @xmath266 & geisler & forte 1990 & @xmath267 @xmath250 & madjesky & bender 1990 & @xmath268 @xmath269 & grillmair et al . 1999 & @xmath270 @xmath271 & ostrov et al . 1998 r r r r initial @xmath272 & & & + ngc 1399 & 5800 & 4527 & 1273 + ngc 1380 & 560 & 1195 & 634 + ngc 1404 & 800 & 1113 & 313 + ngc 1427 & 510 & 510 & 0 + ngc 1379 & 310 & 472 & 162 + ngc 1387 & 390 & 594 & 204 + ngc 1374 & 410 & 410 & 0 + total & 8780 & 8821 & 41 +

the importance of renewable power production is a set goal in terms of the energy turnaround . developing short - term wind speed forecasting improvements might increase the profitability of wind power . this article compares two novel approaches to model and predict wind speed . both approaches incorporate periodic interactions , whereas the first model uses fourier series to model the periodicity . the second model takes generalised trigonometric functions into consideration . the aforementioned fourier series are special types of the p - generalised trigonometrical function and therefore model 1 is nested in model 2 . the two models use an arfima - aparch process to cover the autocorrelation and the heteroscedasticity . a data set which consist of 10 minute data collected at four stations at the german - polish border from august 2007 to december 2012 is analysed . the most important finding is an enhancement of the forecasting accuracy up to three hours that is directly related to our new short - term forecasting model . daniel ambach * keywords : * time series regression ; generalised trigonometric functions ; arfima - aparch process ; high - frequency data ; wind speed forecasts . addresses : * corresponding author : * daniel ambach , european university viadrina , chair of quantitative methods and statistics , post box 1786 , 15207 frankfurt ( oder ) , germany , tel . + 49 ( 0)335 5534 2983 , fax + 49 ( 0)335 5534 2233 , [email protected] .

1509.03116

fossil fuels like oil , gas and coal are rare resources and it is well known that they are finite . the catastrophe in fukushima ( japan ) put a negative light onto nuclear energy and forced people to reconsider nuclear power plants . one of the most emergent renewable energy sources is the wind power . @xcite report that the worldwide wind capacity grew by @xmath3 from the end of 2013 to june 2014 . moreover , they state that this capacity generates approximately @xmath4 of the worldwide electricity supply . the electricity sector of the european union ( eu ) tends towards wind energy , as @xcite remark . besides , they observe that within the last 15 years , @xmath5 of the newly installed power production capacity stem from wind power . the leading wind power country in the european union is germany , which has an installed capacity of @xmath6 mw by june 2014 @xcite . @xcite find that fifteen eu countries have an installed capacity of wind power of each @xmath7 mw . poland and romania are more recent eu members which also exceed the threshold of @xmath8 mw @xcite . the installed capacity in poland amounts to @xmath9 mw @xcite . the geographical region at the german - polish border is very much alike , but the installed capacity varies significantly , as @xcite report . the profit of wind power strongly depends on several properties , as @xcite point out , but the focus of this article is the wind speed prediction . the wind speed is an essential part of the wind power . moreover , accurate wind forecasts can be used to optimise the overall electricity supply , particularly the feed - in of wind power . developing new wind forecasting models depends on the previously specified purpose and forecasting horizon . @xcite , @xcite as well as @xcite provide a broad review of differing wind speed and wind power prediction models . as @xcite point out , conventional time series models are easy to formulate and often used for short - term predictions . the wind speed possesses different properties like periodicity , persistence , heteroscedasticity as well as further characteristics . @xcite show a wind speed forecasting approach which contains an autoregressive fractionally integrated moving average ( arfima ) process . it is one of the first models that comprise a long memory process . since their contribution , the arfima model has become very common to model the wind . hence , @xcite examine extensions of the approach . they compare an arfima model with generalised autoregressive conditional heteroscedasticity ( garch ) and periodic regressors to calibrated and smoothed ensemble - based wind forecasts for hourly data . the periodic behaviour of wind speed depends on the temporal resolution of the underlying data set . frequently , such periodicities are modelled by fourier series @xcite . altogether , miscellaneous articles report that wind speed is autocorrelated and shows periodic behaviour . the wind speed characteristics are commonly combined in several ways . @xcite implement a model which is based on bayesian theory and structural breaks to implicate these properties . @xcite cover a comparison of different long range dependence models with time varying regressors and different estimation methods . a further characteristic is given by the heteroscedasticity of wind . @xcite use a combination of arma models with periodic regressors . furthermore , they introduce fourier series for the variance part to model the heteroscedasticity . @xcite find that their data set contains asymmetric conditional heteroscedasticity . certainly , this is only a brief overview of prevalent wind characteristics . @xcite incorporate spatial information within their forecasting models to extend the accuracy . they perform a comparison of a regime switching model and a bivariate skew - t model . @xcite develop a generalisation of the regime switching approach which is introduced by @xcite . the generalised regime switching space - time model includes several wind characteristics . our model covers periodicity , persistence and heteroscedasticity and thus includes information to provide suitable short - term forecasts . as @xcite point out , those forecasting horizons are beneficial for dispatch planning and for economic load decisions . this article contains two wind speed prediction approaches . both models use an autoregressive fractionally integrated moving average ( arfima ) process with periodic regressors and conditional heteroscedasticity among the variance . particularly , the two approaches incorporate an asymmetric power generalised autoregressive conditional heteroscedasticity ( aparch ) process within the conditional variance @xcite . moreover , both models take annual and diurnal interactions into account . the difference of the two approaches is linked to the periodic mean part in addition to the autocorrelation . the first model uses the classical sine and cosine fourier series . the second model contains @xmath0 generalised trigonometric functions ( see * ? ? ? * ) to model seasonal changes . the analysed data set consists of 10 minute wind speed observations . this data is used to fit the model and perform predictions up to three hours at four different stations at the german - polish border . the forecasting results are evaluated by several accuracy measures . furthermore , a loss function as introduced by @xcite is taken into consideration . the importance of this loss function is underlined by @xcite . in section 2 , the general model for our wind speed is introduced . following the theoretical content , we discuss the estimation of the model . section 3 depicts several methods to evaluate the forecasting accuracy of the selected models . finally , section 4 draws a conclusion of the results discussed in this paper . the contemplated wind speed data set is recorded at four stations in the state of brandenburg . we consider high - frequency data ( 10 minute ) which is provided by the `` deutscher wetterdienst '' ( dwd ) . brandenburg is germany s fifth biggest territorial area . the stations are situated near the polish border in manschnow ( @xmath10 @xmath11 ) , lindenberg ( @xmath12 @xmath13 ) , angermnde ( @xmath14 @xmath15 ) and grnow ( @xmath16 @xmath17 ) . the wind speed data consists of approximately @xmath18 observations within a time frame from august 2007 to december 2012 , with @xmath19 missing values . the gaps of missing data are not longer than a few hours in all cases . we take several interpolations into consideration , but they provide nearly identical results . therefore , we decide to use linear interpolation , as it is the simplest and the least computationally intensive procedure . the in - sample model calibration starts in august 1 , 2007 and reaches to december 31 , 2011 . the considered high - frequency wind data exhibit several of the aforementioned wind speed characteristics . figure [ fig : ts ] shows a sample of the wind speed observations and the corresponding histogram of station lindenberg . we observe that the wind is rapidly changing and involves a high correlation structure , as shown in figure [ fig : acf ] . this figure shows the autocorellation function up to lag @xmath20 which corresponds to more than one month . the other measurement stations provide a similar characteristics . [ figure [ fig : ts ] and [ fig : acf ] about here . ] the periodic characteristic of wind depends on the frequency of the observed data . in particular , we analyse the spectral density to find all necessary periods . the basic tool for these inferences is the periodogram , which is supposed to be the sample analogue of the spectrum . a detailed description of this is given by @xcite . figure [ fig : spectrum ] shows the smoothed periodogram by using the fast fourier transformation ( fft ) . the peaks included in this picture clearly depict a periodic behaviour for different frequencies @xmath21 . in general , we observe @xmath22 periodicities with @xmath23 and @xmath24 . our choice of the peak depends on the period @xmath25 on the one hand and on the total value of the estimated spectrum on the other hand . near the point of origin we detect periods of one year and more . bearing in mind that we analyse a data set of about five and a half years , we neglect peaks with periodic cycles of more than one year . altogether , we consider at least four periodicities . the selected peaks correspond to diurnal @xmath26 and annual @xmath27 periodicities . in addition , we consider two sub - sets of the previous determined frequencies , i.e. half - day ( @xmath28 ) and half - year ( @xmath29 ) periodicity . such frequencies reflect the day - night cycle and the summer and winter term . [ figure [ fig : spectrum ] about here . ] thus , we observe persistence , periodicity and heteroscedasticity within our wind speed data similar to the approaches by @xcite , @xcite and @xcite . apparently , these models use different data sets . for example @xcite implement an arma - garch in mean model to their 15 minute wind data set . @xcite use hourly and @xcite use three hour wind data . we apply the aforementioned approaches to our data , but they do not provide an overall satisfying result . for instance , @xcite take periodic regressors within the variance part into consideration , whereas our variance data shows an asymmetric conditional heteroscedasticity . such a result is related to the underlying data set and therefore , we have to develop another approach . we provide two novel wind speed models which use a new method to capture the periodic behaviour of wind . let @xmath30 be a high - frequency process of wind speed , measured in @xmath31 . the univariate time series of wind speed measurements is given by @xmath32 , where @xmath33 is the day , @xmath34 is the 10 minute data point within a day . we are able to generalise the model ( see * ? ? ? * ) @xmath35 where @xmath36 is the residual process with zero mean and time dependent variance @xmath37 . the residual process incorporates autoregressive coefficients and conditional heteroscedasticity within the variance . the general model ( [ general ] ) is the most simplified expression of the wind speed time series . hereafter , we explain the seasonal regressors in more detail . @xmath38 where @xmath39 and @xmath40 provide up to @xmath41 and @xmath42 time dependent periodic functions . @xmath43 represents an element of the indicator matrix @xmath44 with @xmath45 dimension and @xmath46 is the trend coefficient which is constant over time . moreover , the function set @xmath47 , where @xmath48 , provides periodic external variables given by @xmath49 \cos_{p}\left(\frac{\pi i_1 t } { s_1 } \right ) & , \ , i_1=\text{even } \\[.2 cm ] \sin_{p}\left(\frac{\pi ( i_1 - 1 ) t } { s_1}\right ) & , \ , i_1=\text{odd , } \backslash \{1\ } \end{cases } \,\ ] ] and @xmath50 \cos_{p}\left(\frac{\pi i_2 t } { s_2 } \right ) & , \ , i_2=\text{even } \\[.2 cm ] \sin_{p}\left(\frac{\pi ( i_2 - 1 ) t } { s_2}\right ) & , \ , i_2=\text{odd , } \backslash \{1\ } , \end{cases } \ , \vspace*{0.3cm}\ ] ] where @xmath51 and @xmath52 are p - generalised periodic functions or fourier , series if @xmath53 . commonly , we observe the classical way to model the periodicity by fourier series , like @xcite and @xcite show . therefore , our first model uses the classical sine and cosine decomposition ( @xmath53 ) as well . our second model uses the p - generalised sine and cosine functions of @xcite ( @xmath54 ) to obtain a more flexible periodic structure . @xcite mention that recently , the circle number @xmath55 is generalised for @xmath0 circles . furthermore , @xcite use the p - generalised radius coordinates to define p - generalised trigonometrical functions for @xmath54 . the p - generalised trigonometrical functions are given by @xcite @xmath56 and @xmath57 where @xmath58 . moreover , we have to describe the residual process @xmath59 which follows an arfima - aparch and is given by @xmath60 & z_{t}= & \sigma_{t } \eta_{t } \qquad \text{where } \quad \eta_{t } \overset{iid}{\sim } f_{\nu , \mu , \sigma , \xi } \\[.2 cm ] & \sigma_{t}^\delta = & \alpha_{0 } + \sum\limits_{l=1}^{q } \alpha_l ( | z_{t - l}| - \gamma_l z_{t - l})^{\delta } + \sum\limits_{m=1}^{p } \beta_m \sigma_{t - m}^{\delta } , \vspace*{0.3 cm } \end{array}\end{aligned}\ ] ] where @xmath61 is the differencing parameter , @xmath62 is the backward shift operator with @xmath63 . the polynomials @xmath64 and @xmath65 are @xmath66 and @xmath67 . @xmath68 is the power of the aparch process , @xmath69 is the asymmetry parameter and @xmath70 are the residuals . @xcite point out that instead of the conditional variance , the conditional standard deviation with power @xmath68 follows an aparch . wind speed data is highly volatile and different wind regimes may lead to the asymmetric conditional heteroscedasticity . thus , if the wind speed is relatively low , we observe a different conditional variance compared to high speed wind regimes . the residuals @xmath70 are described by a skew student s t - distribution @xmath71 . one definition of this distribution is given by @xcite . they combine two halves of a symmetric t - distribution which are differently scaled . the skew - t distribution is given by @xmath72^{-\frac{\nu+1}{2 } } , \vspace*{0.3cm}\ ] ] with @xmath73 degrees of freedom , @xmath74 the indicator function , expectation @xmath75 , variance @xmath76 and @xmath77 being the gamma function . moreover , @xmath78 is the skewness parameter with @xmath79 . it reduces @xmath80 to the noncentral student s t - distribution if @xmath81 . finally , the model description conducts into two approaches , although both models incorporate interaction terms @xcite . this idea is based on our data set , where we observe a high periodic structure . initially , we apply the classical additive fourier series decomposition , but there is still a periodic behaviour observable . therefore , we take the periodic interaction into account . the interaction of diurnal periodic functions with the annual ones covers a flexible and time varying explanatory variable within the mean part of . depending on the dimension of the indicator matrix , we observe different periodic functions which interact with each other . according to our wind speed models , the bayesian information criterion ( bic ) and the significance of different interactions result in the following indicator matrix @xmath82 where @xmath44 is a @xmath83 matrix . the first element ( @xmath84 ) is set to zero , because it is already covered by the intercept . obviously , our obtained indicator matrix @xmath44 excludes a few interactions . these are combinations of half - daily and half - yearly , half - yearly and diurnal as well as half - daily and annual periodic functions . thus , we obtain the following seasonal term from equation @xmath85 subsequently , we discuss the parameter estimation of both modelling approaches . the first model contains classical sine and cosine functions , which is described by , and . our first model is estimated by a quasi maximum likelihood method , whereas the p - generalisation parameter of our trigonometric functions remains constant ( i.e. , @xmath86 ) . unfortunately , the estimation procedure of the second model is more complex . here , we have to solve nonlinear equations , which is related to the p - generalised periodic function ( i.e. @xmath54 ) , to obtain suitable results . therefore , we primarily estimate the intercept , trend and the seasonal coefficients as well as the parameter @xmath54 for the generalised sine and cosine functions by generalised least squares estimation . these values are only used as starting values . in the next step we estimate the entire second model by quasi maximum likelihood . according to the complete parameter space which might be included into our model , we have to select an information criterion which punishes over - parametrisation . we choose the bayesian information criterion ( bic ) to find the best model order as it is more conservative according to the parameter selection than , e.g. , the akaike information criterion ( aic ) . as previously stated , the second wind speed prediction approach uses p - generalised sine and cosine functions ( @xmath87 ) , which are shown in and . for each periodic function @xmath88 , with @xmath48 , @xmath89 , @xmath90 and @xmath91 , we include a @xmath92 and @xmath93 function in the second model . thus , we have to estimate the parameter set @xmath94 with @xmath95 for each p - generalised trigonometrical function . bearing in mind that a special case of this function is @xmath96 and we do not estimate the parameter set @xmath97 for our first model . tables [ table : parameterestimation : m1 ] and [ table : parameterestimation : m2 ] provide the estimated parameters for each station . we show the parameter estimates for the model order which is obtained by the smallest bic as well as the smallest in - sample mean square error . to evaluate the significance of a coefficient , we use a level of significance of @xmath98 . we have to test whether the p - generalisation coefficient @xmath99 and @xmath100 . indeed , the p - coefficients are @xmath99 and @xmath100 . the estimated coefficients for all stations and both modelling approaches are significant , with few exceptions . apparently , the diurnal and annual seasons are significant as well as the biannual and half - daily ones . surprisingly , the fractional integration parameter @xmath101 is rejected at station lindenberg for both models . according to the long - memory parameter @xmath61 , we obtain different , but significant results for the other stations . the optimal model order for the arfima(j , d , q ) part in both cases is given by @xmath102 autoregressive and @xmath103 moving average parameters . the estimated aparch coefficients are significant for @xmath98 , except for the aparch intercept @xmath104 for manschnow and grnow . here , the best model order of the aparch process is given by @xmath105 and @xmath106 . moreover , @xmath69 is significant for all stations and models . according to the definition shown in , we observe negative estimates , but in fact positive shocks this means that choppy wind increases the variance of the wind speed . the estimation results of the skew - t distribution indicate a left skewed distribution @xmath107 with degrees of freedom @xmath108 . while performing a partial t - test whether @xmath81 or alternatively @xmath109 , we have to reject the null - hypothesis for each station . analysing @xmath73 for all examples provides computations of @xmath110 . this result confirms the heavy tail assumption of the t - distribution . [ table [ table : parameterestimation : m1 ] and [ table : parameterestimation : m2 ] about here . ] following the discussion of the parameter estimation , we analyse the autocorrelation function of the standardised residuals and the absolute value of the residuals with power @xmath68 . the results are illustrated in figure [ fig : acf ] and [ fig : acf2 ] . the histograms of the standardized residuals are shown and compared to the underlying skew - t distribution with skewness parameter @xmath78 and @xmath73 degrees of freedom . figure [ fig : acf ] provides satisfying results for the standardized residuals @xmath70 , but with some limitations . one problem is obviously disclosed by the autocorrelation function ( acf ) plot . the autocorrelation of the standardised residuals still show minor periodicity . in comparison to the absolute value of the lagged @xmath68 residuals , the seasonality becomes apparent . according to figure [ fig : acf2 ] , the results are very similar to figure [ fig : acf ] . the histogram of the standardized residuals @xmath70 indicates a variance reduction . moreover , the autocorrelation function supports this finding . the acf of @xmath70 provides slightly better results than the first model . nevertheless , the acf of @xmath111 shows a remaining seasonality . [ figure [ fig : acf ] and [ fig : acf2 ] about here . ] finally , we compare the model diagnostics . the improvements of the second model that is indicated by the parameter estimation are emphasized by a lower bic . in addition , the @xmath112 and the mse provide better results for the second model . the second model discloses better results related to a minor presence of autocorrelation . we perform an autocorrelation test on the standardized and the absolute values of the standardized residuals with power @xmath68 for different lags . the ljung and box q - statistics indicate a small correlation for some lags , given a level of significance of @xmath113 . nevertheless , the remaining autocorrelation should be negligible , whereas conducting a test on the underlying skew - t distribution sustains the distributional assumption . moreover , the histogram of the standardized residuals and the assumed underlying skew - t distribution point into the same direction . summing up the model diagnostics , there is a multitude of arguments for the selected time series model , but with some limitations according to the periodic correlation . the two aforementioned wind speed models are evaluated according to their out - of - sample performance . therefore , we have to examine different accuracy measures . one of those criteria is the root mean square error ( rmse ) and another one which we consider is the mean absolute error ( mae ) . in contrast to the rmse and mae , which evaluate the accuracy of the wind speed prediction , we analyse an alternative loss function . this loss function is becoming more common @xcite . the main interest is the prediction of wind power @xcite . the production of wind power strongly depends on factors like : air density @xmath114 @xmath115 , the rotor swept area @xmath116 , the wind speed @xmath117 and an efficiency or power constant @xmath118 ( see * ? ? ? the power production formula that models the relationship between wind speed and wind power is given by : @xmath119 where @xmath120 is the power production function at time @xmath121 . obviously , for each turbine type we obtain different power curves and a somehow different loss function . therefore , assuming a typical power curve which holds for each region and all wind turbines is not suitable . at the german - polish border region for instance , the fuhrlnder md 77 wind power plant is widely used . considering brandenburg altogether , we assess several different turbine types . they differ significantly with respect to their low cut - in wind speed , efficiency , rated output speed and cut - out speed . @xcite introduce a specific accuracy measure , the power curve error ( pce ) . their aim is to point out the problem of wind power prediction . we use this power curve error with a specific power function shown in figure [ fig : power ] . [ figure [ fig : power ] about here . ] a common problem of the wind power prediction is that precise output is not available at each measurement station . moreover , if wind power data is available , we only observe noisy observations . therefore , we use to approximate the real power data . the pce loss function of @xcite is given by @xmath122 where @xmath123 is a weight with @xmath124 $ ] . in contrast to @xcite , we consider cases where underestimation is penalized stronger than overestimation , @xmath125 and the reverse case @xmath126 . the main aim of the introduced models is to minimize the respective loss function . according to the pce , we want to find the quantile @xmath123 minimizing the loss function . for the out - of - sample forecasts , we select incidentally different points in time @xmath127 in the out - of - sample period @xmath128 . the previously calculated in - sample model order is used as initial estimate . a part of the information set available at time @xmath127 , namely @xmath129 is used for the out - of - sample forecasts . we calculate @xmath130 and repeat this procedure 5,000 times to obtain suitable forecasting results up to a maximum of three hours . for each model type we use the so called persistence method @xcite as a crude benchmark . such a short - term forecast model uses the observation today as predictor for the next instance in time . the forecast for such a model is quite simple , but provides practicable results for high - frequency data . for the pce , we take different values for @xmath123 . @xcite follow @xcite , who propose @xmath131 . in this case , underestimation is punished more strongly . moreover , they take different values of @xmath123 to control the overall estimation results and analyse the sensitivity of their models . according to their findings , we decide to choose three different values of @xmath132 . table [ table : pred1 ] gives an overview of predictability results of the three models . primarily , we discuss the rmse and mae and subsequently consider the pce . here , the best prediction model is obvious . in this case , the second model with generalised sine and cosine functions and arfima - aparch process is the best one for a three hour ahead prediction . we have to remark that at station lindenberg , the fractional integrated part is insignificant . even though the persistent model is hard to beat for predicting very short periods of time , the model fails to forecast up to three hours . [ table [ table : pred1 ] about here . ] apparently , table [ table : pred1 ] provides an ambiguous overall result . analysing the pce changes our previous findings . as a result from the pce comparison , we select the first model with arfima - aparch process and fourier series in the mean part , if @xmath133 . the first model has still some minor model limitations related to the remaining information within the variance of the standardised residuals . the forecasting accuracy of model two is inferior to the first model . still , it is better than the persistent model . this finding indicates that we have to choose the first model if overestimation of wind power is punished . this might be the case if the profit of a wind farm is calculated . in general , it is important for an investment into a wind farm to know how much wind on average is expected and how much wind power has to be produced . the investor is unwilling to receive losses according to an overestimation of the expected wind power . the pce for @xmath134 indicates that the second model provides the best forecasting results . here , the over- and underestimation is punished similarly . therefore , it is not surprising that after our transformation into wind power the second model remains the best one . the third row shows the pce for @xmath135 . though the underestimation is penalized , the overall pce that is shown in the last column provides only marginal differences between each model . according to our previous findings , the best prediction model changes from station to station and also from month to month . the initial best prediction model ( model 2 ) tends to overestimate the predicted wind power of the fuhrlnder md 77 . punishing underestimation of wind power is related to the loss function of a system operator . in case of underestimation , the system operator has to order too much power to match the demand . subsequently , the operator has more power than needed , the generation is reduced and he faces costs of down - regulation . @xcite point out that the costs of down - regulation are higher than those for up - regulation and lead to higher expenditures . indeed , the results of the pce are related to the underlying production function shown in figure [ fig : power ] . the fuhrlnder md 77 is a small turbine which is used onshore . we are able to include various turbines to control the sensitivity of the obtained results . while considering different types of turbines , the pce changes as well . nevertheless , the turbines onshore situated at the german - polish border are somehow comparable . most of them have a similar cut - in wind speed , but a few of them have a higher rated and cut - out wind speed . while changing the wind power production function , the pce changes in some cases . we use the power production function of a ge 1.6 mw turbine to verify our previous findings . although some results remain similar , we obtain different findings when @xmath135 . now , the second model is marginally better than the other prediction models for forecasting wind power . the second model provides the best prediction results for the german - polish border region . moreover , it is related to the weight @xmath123 which model is the best method to make predictions of the wind power . in particular , the persistent model is a good benchmark for wind power prediction , but for our wind speed prediction , it is outperformed . an onshore wind park operator who is interested in a lower bound of wind power prediction has to use the second model . recent research has covered a wide range of wind speed forecasting approaches . in this paper , two novel statistical approaches for high - frequency wind speed data are described . several extensions are implemented to cover the specific structure of the data set . the mean part takes annual and diurnal interactions into account . moreover , we distinguish between a first model with classical fourier series and a second model with p - generalised sine and cosine functions . the autocorrelation and the heteroscedasticity of both approaches is modeled by an arfima - aparch process with skew - t distributed residuals . the models are suitable to make three hour forecasts for wind speed and can be used for wind power prediction as well . according to rmse and mae , the developed approaches outperform common models . the generalised trigonometrical functions provide the smallest accuracy measures according to the three hour ahead forecasts . furthermore , we comprise the pce which provides ambiguous results . the overall pce provides the best results for the first model , if we consider a small weight . nevertheless , the second model derives the best results if the weight is @xmath136 or @xmath137 . finally , we remark that the pce is also depending on the previously specified turbine type . in this article , the models are applied to four stations situated at the german - polish border . for this reason , the power production function of a fuhrlnder md 77 is used to calculate the pce . the fuhrlnder md 77 is a small turbine which is often used onshore , especially in plain regions in germany . the model robustness which is briefly discussed in the end , shows that some changes within the power curve function lead to different best prediction models . in practice , only the wind speed data set is needed . therefore , such a statistical wind speed prediction model is cost saving compared to other modelling approaches that require a set of possibly many more variables . clearly , there are further improvements of this approach possible . an idea is to analyse the variance in more detail and to consider meteorological regressors . finally , the wind speed model might include spatial informations . tilmann gneiting , kristin larson , kenneth westrick , marc g genton , and eric aldrich . calibrated probabilistic forecasting at the stateline wind energy center : the regime - switching space time method . , 101(475):968979 , 2006 . saurabh s soman , hamidreza zareipour , om malik , and paras mandal . a review of wind power and wind speed forecasting methods with different time horizons . in _ north american power symposium ( naps ) , 2010 _ , pages 18 . ieee , 2010 . rrrrrrrrr & & & & + + & estimate & std . error & estimate & std . error & estimate & std . error & estimate & std . error + @xmath46 & -0.0000025@xmath138 & 0.00000006&-0.0000040@xmath138&0.00000006&-0.0000027@xmath138&0.00000006&-0.0000028@xmath138 & 0.00000006 + @xmath139 & 3.2941@xmath138 & 0.0075 & 4.0346@xmath138 & 0.0079 & 4.1071@xmath138 & 0.0089 & 4.6358@xmath138 & 0.0093 + @xmath140 & -0.5588@xmath138 & 0.0053 & -0.1456@xmath138 & 0.0056 & -0.5420@xmath138 & 0.0062 & -0.5464@xmath138 & 0.0065 + @xmath141 & -0.3730@xmath138 & 0.0053 & -0.1107@xmath138 & 0.0056 & -0.4179@xmath138 & 0.0062 & -0.4529@xmath138 & 0.0065 + @xmath142 & 0.0752@xmath138 & 0.0053 & 0.0313@xmath138 & 0.0056 & 0.0743@xmath138 & 0.0062 & 0.1005@xmath138 & 0.0065 + @xmath143 & 0.2098@xmath138 & 0.0053 & 0.1089@xmath138 & 0.0056 & 0.2144@xmath138 & 0.0062 & 0.1938@xmath138 & 0.0065 + @xmath144 & 0.0451@xmath138 & 0.0053 & -0.2839@xmath138 & 0.0056 & -0.3678@xmath138 & 0.0062 & -0.1986@xmath138 & 0.0065 + @xmath145 & 0.0531@xmath138 & 0.0074 & 0.0057@xmath138 & 0.0078 & -0.0888@xmath138 & 0.0088 & -0.1646@xmath138 & 0.0092 + @xmath146 & 0.1079@xmath138 & 0.0074 & -0.0334@xmath138 & 0.0078 & 0.0831@xmath138 & 0.0088 & -0.0139@xmath138 & 0.0092 + @xmath147 & 0.6339@xmath138 & 0.0053 & 0.4167@xmath138 & 0.0056 & 0.5400@xmath138 & 0.0063 & 0.6472@xmath138 & 0.0066 + @xmath148 & 0.4127@xmath138 & 0.0075 & 0.0656@xmath138 & 0.0079 & 0.3795@xmath138 & 0.0088 & 0.3441@xmath138 & 0.0093 + @xmath149 & 0.2368@xmath138 & 0.0075 & 0.1021@xmath138 & 0.0079 & 0.3114@xmath138 & 0.0088 & 0.2975@xmath138 & 0.0093 + @xmath150 & -0.0628@xmath138 & 0.0053 & -0.0418@xmath138 & 0.0056 & 0.0166@xmath138 & 0.0062 & 0.1447@xmath138 & 0.0065 + @xmath151 & -0.1060@xmath138 & 0.0053 & -0.1549@xmath138 & 0.0056 & -0.1123@xmath138 & 0.0062 & -0.2368@xmath138 & 0.0065 + + @xmath152 & 0.4310@xmath138 & 0.0064 & 0.00003 & 0.0003 & 0.0913@xmath138 & 0.0275 & 0.2323@xmath138 & 0.0164 + @xmath153 & 1.2334@xmath138 & 0.0033 & 1.4033@xmath138 & 0.0071 & 1.5345@xmath138 & 0.0168 & 1.4446@xmath138 & 0.0035 + @xmath154 & -0.2698@xmath138 & 0.0029 & -0.4104@xmath138 & 0.0077 & -0.5412@xmath138 & 0.0162 & -0.4562@xmath138 & 0.0032 + @xmath155 & -0.8065@xmath138 & 0.0067 & -0.6186@xmath138 & 0.0102 & -0.7650@xmath138 & 0.0133 & -0.7826@xmath138 & 0.0126 + + @xmath104 & 0.0012@xmath156 & 0.0004 & 0.0091@xmath138 & 0.0009 & 0.0188@xmath138 & 0.0017 & 0.0019@xmath157 & 0.0012 + @xmath158 & 0.1426@xmath138 & 0.0054 & 0.1346@xmath138 & 0.0075 & 0.1254@xmath138 & 0.0071 & 0.1580@xmath138 & 0.0064 + @xmath159 & 0.5257@xmath138 & 0.0459 & 0.7160@xmath138 & 0.0622 & 0.5523@xmath138 & 0.0533 & 0.5526@xmath138 & 0.0496 + @xmath160 & 0.3645@xmath138 & 0.0433 & 0.1618@xmath138 & 0.0569 & 0.3015@xmath138 & 0.0490 & 0.3252@xmath138 & 0.0460 + @xmath161 & -0.1208@xmath138 & 0.0179 & -0.2860@xmath138 & 0.0238 & -0.1947@xmath138 & 0.0255 & -0.1103@xmath138 & 0.0187 + @xmath68 & 0.9325@xmath138 & 0.0606 & 0.9529@xmath138 & 0.0572 & 1.3877@xmath138 & 0.0770 & 1.0750@xmath138 & 0.0543 + @xmath78 & 1.0672@xmath138 & 0.0083 & 1.0512@xmath138 & 0.0083 & 1.0534@xmath138 & 0.0084 & 1.0160@xmath138 & 0.0080 + @xmath73 & 7.8622@xmath138 & 0.3101 & 9.2202@xmath138 & 0.4505 & 8.8646@xmath138 & 0.4003 & 7.3534@xmath138 & 0.2841 + * bic * & & 1.2886 & & 1.1598 & & 1.5658 & & 1.4116 + rrrrrrrrr & & & & + + & estimate & std . error & estimate & std . error & estimate & std . error & estimate & std . error + @xmath46&-0.0000025@xmath138&0.00000006&-0.0000040@xmath138&0.00000006&-0.0000027@xmath138 & 0.00000007 & -0.0000028@xmath138&0.00000007 + @xmath139 & 3.2938@xmath138 & 0.0075 & 4.0356@xmath138 & 0.0079 & 4.1064@xmath138 & 0.0089 & 4.6355@xmath163 & 0.0093 + @xmath140 & -0.5387@xmath138 & 0.0051 & -0.1407@xmath138 & 0.0054 & -0.5227@xmath138 & 0.0060 & -0.5269@xmath164 & 0.0063 + @xmath141 & -0.3987@xmath138 & 0.0056 & -0.1183@xmath138 & 0.0060 & -0.4452@xmath138 & 0.0066 & -0.4839@xmath164 & 0.0070 + @xmath142 & 0.0666@xmath138 & 0.0047 & 0.0274@xmath138 & 0.0049 & 0.0658@xmath138 & 0.0055 & 0.0890@xmath164 & 0.0058 + @xmath143 & 0.2100@xmath138 & 0.0053 & 0.1090@xmath138 & 0.0218 & 0.2147@xmath138 & 0.0062 & 0.1941@xmath164 & 0.0065 + @xmath144 & 0.0410@xmath138 & 0.0047 & -0.2354 @xmath138 & 0.0049 & -0.3147@xmath138 & 0.0055 & -0.1672@xmath164 & 0.0058 + @xmath145 & 0.1612@xmath138 & 0.0207 & 0.0501 @xmath156 & 0.0218 & -0.2078@xmath138 & 0.0244 & -0.4095@xmath138 & 0.0255 + @xmath146 & 7.6052@xmath138 & 0.4566 & -1.6406 @xmath138 & 0.4824 & 2.7394@xmath138 & 0.5387 & -1.4781@xmath138 & 0.5633 + @xmath147 & 0.5649@xmath138 & 0.0047 & 0.3787 @xmath138 & 0.0050 & 0.4958@xmath138 & 0.0056 & 0.5853@xmath164 & 0.0058 + @xmath148 & 0.5948@xmath138 & 0.0107 & 0.0797 @xmath138 & 0.0113 & 0.5288@xmath138 & 0.0126 & 0.4836@xmath164 & 0.0132 + @xmath149 & 0.4507@xmath138 & 0.0137 & 0.1920 @xmath138 & 0.0144 & 0.5904@xmath138 & 0.0161 & 0.5665@xmath164 & 0.0168 + @xmath150 & -0.1658@xmath138 & 0.0107 & -0.0067 & 0.0113 & 0.1490@xmath138 & 0.0126 & 0.0890@xmath138 & 0.0058 + @xmath151 & -0.1030@xmath138 & 0.0050 & -0.1517 @xmath138 & 0.0053 & -0.1093@xmath138 & 0.0059 & 0.1941@xmath138 & 0.0065 + @xmath165 & 2.4709@xmath138 & 0.0104 & 4.5886 @xmath138 & 0.0029 & 1.9080@xmath138 & 0.0102 & 2.3876@xmath138 & 0.0024 + @xmath166 & 1.0264@xmath138 & 0.0058 & 1.4824 @xmath138 & 0.0039 & 1.1167@xmath138 & 0.0057 & 1.0529@xmath138 & 0.0036 + @xmath167 & 1.3801@xmath138 & 0.0432 & 51.3130 @xmath138 & 0.1569 & 1.2694@xmath138 & 0.0091 & 1.6574@xmath138 & 0.0026 + @xmath168 & 1.7776@xmath138 & 0.0189 & 1.5631 @xmath138 & 0.0019 & 1.9647@xmath138 & 0.0104 & 4.2203@xmath138 & 0.0035 + @xmath169 & 0.3945@xmath138 & 0.0030 & 0.4286 @xmath138 & 0.0009 & 0.1752@xmath138 & 0.0002 & 0.1566@xmath138 & 0.0002 + @xmath170 & 8.5535@xmath138 & 0.1220 & 31.0516 @xmath138 & 0.1011 & 30.9854@xmath138 & 0.0146 & 19.7518@xmath138 & 0.0047 + @xmath171 & 0.5056@xmath138 & 0.0060 & 76.5876 @xmath138 & 0.2387 & 0.1535@xmath138 & 0.0004 & 0.1404@xmath138 & 0.0001 + @xmath172 & 1.6385@xmath138 & 0.0485 & 15.3874 @xmath138 & 0.0409 & 6.9258@xmath138 & 0.0223 & 0.1778@xmath138 & 0.0002 + + @xmath152 & 0.4475@xmath138 & 0.0024 & 0.000001 & 0.000008 & 0.1006@xmath138 & 0.0126 & 0.2564@xmath138 & 0.0087 + @xmath153 & 1.2184@xmath138 & 0.0094 & 1.4044 @xmath138 & 0.0011 & 1.4961@xmath138 & 0.0183 & 1.4318@xmath138 & 0.0016 + @xmath154 & -0.2583@xmath138 & 0.0071 & -0.4116 @xmath138 & 0.0011 & -0.5048@xmath138 & 0.0172 & -0.4438@xmath138 & 0.0015 + @xmath155 & -0.8081@xmath138 & 0.0215 & -0.6195 @xmath138 & 0.0047 & -0.7683@xmath138 & 0.0133 & -0.7919@xmath138 & 0.0014 + + @xmath104 & 0.0012@xmath157 & 0.0006 & 1.75 & 0.0091@xmath138 & 0.0189@xmath138 & 0.0017 & 0.0020@xmath157 & 0.0011 + @xmath158 & 0.1392@xmath138 & 0.0051 & 26.79 & 0.1343@xmath138 & 0.1260@xmath138 & 0.0071 & 0.1583@xmath138 & 0.0064 + @xmath159 & 0.5188@xmath138 & 0.0463 & 11.18 & 0.7175@xmath138 & 0.5495@xmath138 & 0.0529 & 0.5516@xmath138 & 0.0494 + @xmath160 & 0.3738@xmath138 & 0.0437 & 8.54 & 0.1606@xmath138 & 0.3037@xmath138 & 0.0486 & 0.3259@xmath138 & 0.0458 + @xmath161 & -0.1372@xmath138 & 0.0131 & -10.46 & -0.2863@xmath138 & -0.1951@xmath138 & 0.0254 & -0.1086@xmath138 & 0.0188 + @xmath68 & 0.9594@xmath138 & 0.1076 & 8.91 & 0.9430@xmath138 & 1.3879@xmath138 & 0.0766 & 1.0809@xmath138 & 0.0463 + @xmath78 & 1.0693@xmath138 & 0.0083 & 128.16 & 1.0517@xmath138 & 1.0549@xmath138 & 0.0085 & 1.0158@xmath138 & 0.0079 + @xmath73 & 7.8890@xmath138 & 0.2934 & 26.88 & 9.1984@xmath138 & 8.8649@xmath138 & 0.4010 & 7.3345@xmath138 & 0.2829 + * bic * & & 1.2695 & & 1.1185 & & 1.5616 & & 1.3596 + and @xmath173.,title="fig:",width=566,height=170 ] + and @xmath173.,title="fig:",width=566,height=170 ] + and @xmath173.,title="fig:",width=566,height=170 ] + and @xmath173.,title="fig:",width=566,height=170 ] + and @xmath176 second model , title="fig:",width=566,height=170 ] + and @xmath176 second model , title="fig:",width=566,height=170 ] + and @xmath176 second model , title="fig:",width=566,height=170 ] + and @xmath176 second model , title="fig:",width=566,height=170 ] + the first zone reaches from @xmath177 to the cut - in wind speed of @xmath178 . no power is produced within this area . the sigmoidal relationship of the power function is used for the second zone . the rated output wind speed of @xmath179 and the rated output power of about @xmath180 is the beginning of the third zone . from @xmath179 to @xmath181 , the wind power plant is able to produce the maximum wind power . the rotor stops to produce power within the forth zone , beginning from the cut - out speed of @xmath181 . + model 1 rmse & 1.61 & 1.67 & 1.59 & 1.63 & 1.61 & 1.63 & 1.61 & 1.53 & 1.51 & 1.63 & 1.58 & 1.66 & 1.61 + model 2 rmse & * 1.61 * & * 1.66 * & * 1.55 * & * 1.58 * & * 1.52 * & * 1.57 * & * 1.59 * & * 1.51 * & * 1.44 * & * 1.62 * & * 1.55 * & * 1.61 * & * 1.57 * + persistent rmse & 2.26 & 1.91 & 1.67 & 1.23 & 1.42 & 1.44 & 1.65 & 1.41 & 1.36 & 1.92 & 2.17 & 2.21 & 1.72 + + model 1 mae & 1.43 & 1.51 & 1.43 & 1.45 & 1.46 & 1.45 & 1.41 & 1.37 & 1.33 & 1.44 & 1.43 & 1.51 & 1.44 + model 2 mae & * 1.42 * & * 1.47 * & * 1.38 * & * 1.41 * & * 1.36 * & * 1.41 * & * 1.39 * & * 1.34 * & * 1.27 * & * 1.43 * & * 1.41 * & * 1.45 * & * 1.40 * + persistent mae & 2.15 & 1.81 & 1.59 & 1.12 & 1.33 & 1.34 & 1.55 & 1.31 & 1.28 & 1.82 & 2.11 & 2.16 & 1.63 + + model 1 pce @xmath133 & * 26.11*&*22.36*&*16.13*&*5.83*&*8.59 * & * 6.45 * & * 15.41 * & * 12.15*&*6.92 * & * 33.19 * & * 27.79 * & * 37.15*&*16.42 * + model 2 pce @xmath133 & 34.18 & 33.21 & 25.21 & 5.95 & 9.16 & 7.12 & 23.17 & 13.24 & 7.81 & 37.97 & 37.84 & 59.15&23.128 + persistent pce @xmath133&42.15 & 44.16 & 34.39 & 6.18 & 9.76 & 7.81 & 31.75 & 14.33 & 8.69 & 42.74 & 47.91 & 81.14&30.92 + + model 1 pce @xmath182 & 33.24 & 33.18 & 21.47 & 6.41 & 9.21 & 13.17 & 14.13 & 12.37 & 11.92 & 45.91 & 38.59 & 37.88 & 23.12 + model 2 pce @xmath182 & * 31.16 * & * 23.11 * & * 14.27 * & * 5.84 * & * 5.85 * & * 8.84 * & * 9.51 * & * 8.28 * & * 9.59 * & * 26.17 * & * 25.91 * & * 28.55*&*16.42 * + persistent pce @xmath182 & 36.32 & 43.34 & 26.66 & 6.95 & 12.58 & 17.31 & 18.56 & 16.45 & 14.25 & 65.65 & 51.28 & 47.21 & 29.71 + + model 1 pce @xmath135 & 14.19 & * 19.86*&*9.68*&*4.11*&*7.38*&*11.71 * & 11.56 & 5.37 & 5.21 & 21.18 & 18.95 & * 17.65 * & 12.24 + model 2 pce @xmath135 & * 11.27 * & 21.41 & 12.24 & 6.94 & 9.37 & 11.89 & * 9.91 * & * 4.95 * & * 4.21 * & * 18.47*&*14.12 * & 19.44&*12.02 * + persistent pce @xmath135 & 12.73 & 21.64 & 11.96 & 5.52 & 8.38 & 11.81 & 11.23&5.16&4.71 & 19.32 & 16.53 & 18.55 & 12.30 + + + model 1 rmse & 1.59 & 1.56 & 1.74 & 1.54 & 1.61 & 1.73 & 1.71 & 1.66 & 1.51 & 1.57 & 1.53 & 1.58 & 1.61 + model 2 rmse & * 1.58 * & * 1.53 * & * 1.71 * & * 1.53 * & * 1.58 * & * 1.72 * & * 1.68 * & * 1.61 * & * 1.48 * & * 1.56 * & * 1.51 * & * 1.55 * & * 1.59 * + persistent rmse & 2.28 & 2.17 & 1.97 & 1.63 & 1.59 & 1.79 & 1.75 & 1.75 & 1.61 & 2.17 & 1.87 & 2.15 & 1.89 + + model 1 mae & 1.41 & 1.38 & 1.55 & 1.41 & 1.42 & 1.52 & 1.51 & 1.48 & 1.34 & 1.42 & 1.36 & 1.41 & 1.43 + m2 mae & * 1.41 * & * 1.35 * & * 1.51 * & * 1.38 * & * 1.39 * & * 1.49 * & * 1.47 * & * 1.42 * & * 1.32 * & * 1.41 * & * 1.34 * & * 1.41 * & * 1.41 * + persistent mae & 2.22 & 2.11 & 1.88 & 1.56 & 1.49 & 1.68 & 1.64 & 1.66 & 1.52 & 2.11 & 1.81 & 2.18 & 1.82 + + model 1 pce @xmath133 & * 44.38 * & * 37.36 * & * 42.47 * & * 18.56 * & * 9.68 * & * 18.77 * & * 18.11 * & * 16.13 * & * 17.96 * & * 16.35 * & * 39.82 * & * 31.21 * & * 25.90 * + model 2 pce @xmath133 & 59.64 & 51.42 & 55.58 & 28.18 & 13.42 & 21.71 & 24.96 & 27.18 & 26.91 & 25.69 & 44.76 & 38.13 & 34.80 + persistent pce @xmath133 & 74.91 & 65.47 & 68.69 & 37.59 & 17.16 & 24.63 & 31.93 & 38.14 & 35.83 & 35.14 & 49.71 & 46.16 & 43.78 + + model 1 pce @xmath182 & 63.59 & 61.31 & 69.81 & 21.39 & 15.54 & 37.45 & 28.99 & 24.59 & 25.31 & 38.47 & 51.92 & 37.91 & 39.69 + model 2 pce @xmath182 & * 51.55 * & * 47.39 * & * 41.61 * & * 14.71 * & * 14.99 * & * 22.16 * & * 21.14 * & * 15.98 * & * 16.13 * & * 25.15 * & * 32.78 * & * 32.61 * & * 28.02 * + persistent pce @xmath182 & 76.63 & 73.21 & 97.99 & 28.17 & 16.19 & 52.73 & 36.94 & 33.21 & 34.56 & 51.78 & 71.17 & 43.22 & 51.32 + + model 1 pce @xmath135 & 23.57 & 15.91 & 36.87 & 12.17 & 8.44 & 21.92 & 14.58&17.19 & 13.12 & 15.35 & 19.74 & 13.99 & 17.74 + model 2 pce @xmath135 & 25.39 & 16.74 & * 34.88 * & 13.95 & * 8.43 * & 21.76 & * 13.52 * & * 15.19 * & * 11.43 * & * 13.51 * & 21.93 & 15.17 & * 17.66 * + persistent pce @xmath135 & * 21.74 * & * 15.17 * & 38.87 & * 11.21 * & 8.46 & * 21.17 * & 15.64 & 18.98 & 14.82 & 17.21 & * 18.54 * & * 12.91 * & 17.89 + [ table : pred1 ] + model 1 rmse & 1.74 & 1.69 & 1.92 & 1.77 & 1.75 & 1.78 & 1.81 & 1.79 & 1.72 & 1.66 & 1.76 & 1.64 & 1.75 + model 2 rmse & * 1.69 * & * 1.66 * & * 1.86 * & * 1.71 * & * 1.71 * & * 1.73 * & * 1.72 * & * 1.76 * & * 1.65 * & * 1.65 * & * 1.71 * & * 1.61 * & * 1.71 * + persistent rmse & 2.15 & 2.34 & 2.19 & 1.81 & 1.98 & 2.15 & 1.88 & 1.82 & 1.88 & 2.41 & 2.41 & 2.61 & 2.14 + + model 1 mae & 1.54 & 1.49 & 1.61 & 1.56 & 1.55 & 1.53 & 1.59 & 1.55 & 1.52 & 1.47 & 1.56 & 1.43 & 1.53 + model 2 mae & * 1.48 * & * 1.45 * & * 1.55 * & * 1.49 * & * 1.49 * & * 1.49 * & * 1.46 * & * 1.53 * & * 1.46 * & * 1.45 * & * 1.51 * & * 1.41 * & * 1.48 * + persistent mae & 2.16 & 2.24 & 2.13 & 1.67 & 1.87 & 2.12 & 1.76 & 1.68 & 1.77 & 2.33 & 2.32 & 2.54 & 2.05 + + model 1 pce @xmath133 & * 65.63 * & * 52.73 * & * 53.15 * & * 21.24 * & * 21.15 * & * 28.61 * & * 17.61 * & * 21.63 * & * 22.19 * & * 38.89 * & * 45.87 * & * 41.35 * & * 35.84 * + model 2 pce @xmath133 & 85.51 & 65.99 & 74.58 & 32.76 & 24.72 & 34.72 & 23.31 & 27.18 & 34.74 & 52.16 & 61.18 & 64.31 & 48.43 + persistent pce @xmath133&115.36 & 79.25 & 96.12 & 44.29 & 29.41 & 41.84 & 28.99 & 33.73 & 47.39 & 65.24 & 74.31 & 87.26 & 61.93 + + model 1 pce @xmath182 & 55.93 & 75.21 & 72.51 & 37.57 & 28.81 & 42.21 & 27.11 & 26.66 & 29.16 & 62.25 & 66.66 & 55.79 & 48.32 + model 2 pce @xmath182 & * 53.17 * & * 58.42 * & * 49.17 * & * 31.25 * & * 25.54 * & * 28.15 * & * 21.79 * & * 18.89 * & * 21.99 * & * 37.14 * & * 41.94 * & * 42.36 * & * 35.82 * + persistent pce @xmath182 & 58.78 & 91.99 & 95.85 & 44.89 & 32.16 & 56.25 & 33.43 & 34.43 & 37.34 & 87.35 & 91.37 & 69.22 & 61.09 + + model 1 pce @xmath135 & 28.36 & 28.62 & 41.13 & 19.75 & 14.54 & 21.47 & 17.11 & 17.93 & 24.56 & 26.51 & 29.88 & 23.18 & 24.42 + model 2 pce @xmath135 & * 27.44 * & * 24.54 * & * 41.91 * & * 19.23 * & * 12.73 * & 21.66 & * 16.74 * & 18.11 & * 21.71 * & 27.78 & * 28.36 * & 25.426 & * 23.80 * + persistent pce @xmath135&29.27 & 32.69 & 41.34 & 21.26 & 16.35 & * 21.29 * & 17.27 & * 17.76 * & 28.41 & * 25.22 * & 31.41 & * 21.75 * & 25.34 + + + model 1 rmse & 1.72 & 1.74 & * 1.72 * & 1.78 & 1.75 & 1.81 & 1.78 & 1.74 & 1.67 & 1.73 & 1.73 & * 1.62 * & 1.73 + model 2 rmse & * 1.68 * & * 1.71 * & 1.73 & * 1.76 * & * 1.69 * & * 1.76 * & * 1.74 * & * 1.76 * & * 1.64 * & * 1.73 * & * 1.71 * & * 1.62 * & * 1.71 * + persistent rmse & 2.46 & 2.45 & 2.29 & 2.12 & 2.14 & 2.11 & 2.18 & 2.18 & 2.11 & 2.72 & 2.41 & 2.72 & 2.32 + + model 1 mae & 1.51 & 1.56 & * 1.47 * & 1.56 & 1.54 & 1.61 & 1.55 & * 1.52 * & 1.49 & 1.54 & 1.56 & 1.45 & 1.53 + model 2 mae & * 1.46 * & * 1.53 * & 1.47 & * 1.56 * & * 1.48 * & * 1.56 * & * 1.51 * & 1.53 & * 1.44 * & * 1.54 * & * 1.54 * & * 1.44 * & * 1.51 * + persistent mae & 2.34 & 2.32 & 2.16 & 2.15 & 1.94 & 1.89 & 1.95 & 2.16 & 2.11 & 2.65 & 2.33 & 2.66 & 2.22 + + model 1 pce @xmath133 & * 73.21 * & * 72.88 * & * 61.57 * & * 42.41 * & * 26.39 * & * 33.25 * & * 26.25 * & * 39.76 * & * 38.67 * & * 41.15 * & * 71.54 * & * 46.91 * & * 47.83 * + model 2 pce @xmath133 & 98.71 & 92.17 & 87.88 & 61.15 & 31.62 & 38.66 & 34.94 & 56.26 & 52.56 & 57.79 & 84.82 & 68.49 & 63.75 + persistent pce @xmath133&124.18 & 111.46 & 114.18 & 77.71 & 36.85 & 44.16 & 43.63 & 72.76 & 66.46 & 74.44 & 99.19 & 91.19 & 79.68 + + model 1 pce @xmath182 & 93.17 & 111.29 & 85.57 & 55.87 & 31.18 & 47.26 & 31.24 & 54.41 & 54.41 & 87.15 & 83.49 & 66.86 & 66.83 + model 2 pce @xmath182 & * 83.39 * & * 82.42 * & * 63.16 * & * 41.87 * & * 25.71 * & * 32.71 * & * 24.51 * & * 41.41 * & * 42.15 * & * 57.21 * & * 58.33 * & * 55.74 * & * 50.72 * + persistent pce @xmath182 & 112.75 & 118.16 & 118.17 & 71.86 & 36.66 & 61.81 & 37.96 & 68.41 & 66.76 & 116.91 & 118.65 & 77.97 & 83.84 + + model 1 pce @xmath135 & 33.16 & 41.77 & 34.38 & 23.17 & 19.68 & 23.63 & 22.12 & 28.56 & 28.94 & 32.98 & 33.62 & 21.95 & 28.66 + model 2 pce @xmath135 & 35.18 & 42.24 & * 31.12 * & * 21.73 * & 21.68 & * 21.54 * & * 21.97 * & * 25.79 * & 33.63 & * 32.61 * & * 31.49 * & * 21.14 * & * 28.34 * + persistent pce @xmath135&*31.94 * & * 39.29 * & 38.74 & 24.41 & * 17.69 * & 25.71 & 23.17 & 31.32 & * 24.25 * & 33.36 & 35.76 & 22.76 & 29.03 + [ table : pred2 ]

a spectral parameter power series ( spps ) representation for regular solutions of singular bessel type sturm - liouville equations with complex coefficients is obtained as well as an spps representation for the ( entire ) characteristic function of the corresponding spectral problem on a finite interval . it is proved that the set of zeros of the characteristic function coincides with the set of all eigenvalues of the sturm - liouville problem . based on the spps representation a new mapping property of the transmutation operator for the considered perturbed bessel operator is obtained , and a new numerical method for solving corresponding spectral problems is developed . the range of applicability of the method includes complex coefficients , complex spectrum and equations in which the spectral parameter stands at a first order linear differential operator . on a set of known test problems we show that the developed numerical method based on the spps representation is highly competitive in comparison to the best available solvers such as sleign2 , matslise and some other codes and give an example of an exactly solvable test problem admitting complex eigenvalues to which the mentioned solvers are not applicable meanwhile the spps method delivers excellent numerical results .

1303.3911

in the present work the equation @xmath0 , \label{i1}\ ] ] is studied , where @xmath1 is a real number , @xmath2 , @xmath3 is a complex - valued continuous function on @xmath4 $ ] satisfying a growth bound @xmath5 at the origin for some @xmath6 , @xmath7 $ ] are complex valued functions , and @xmath8 is a ( complex ) spectral parameter . denote @xmath9 . equations of the form ( [ i1 ] ) appear naturally in many real - world applications after a separation of variables and therefore have received considerable attention ( see , e.g. , @xcite , @xcite , @xcite , @xcite , @xcite ) . in preceding publications equation ( [ i1 ] ) was studied under more restrictive conditions , typically for @xmath3 and @xmath10 being real - valued and @xmath11 . the approach developed in this work does not imply such restrictions and serves both for qualitative study of solutions and spectral problems , as well as for related numerical computation . the main component in the developed approach is a spectral parameter power series ( spps ) representation for the regular solution of ( [ i1 ] ) obtained under the condition that the auxiliary equation @xmath12 possesses a regular solution which does not have zeros on @xmath13 $ ] except at @xmath14 . the spps representation for solutions of nonsingular linear differential equations and its applications in corresponding scattering and spectral problems were studied in @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and some other papers . here , in section [ sect2 ] we obtain an analogous result for the perturbed bessel equation ( [ i1 ] ) . the construction and the existence of the required particular solution are addressed in section [ sect3 ] . for example , when @xmath15 , @xmath16 $ ] such nonvanishing on @xmath4 $ ] solution exists . we give an analytic representation for it together with an estimate . let us emphasize that , in general , @xmath17 is allowed to be a complex - valued function , and the existence of such @xmath17 for a complex valued @xmath3 is an open problem . under the assumption that @xmath17 exists we obtain a dispersion ( characteristic ) relation for the sturm - liouville problem for ( [ i1 ] ) on @xmath13 $ ] , @xmath18 ( theorem [ thmspectralproblem ] ) . namely , we construct an entire function @xmath19 in the form of a taylor series such that the set of its zeros coincides with the set of all eigenvalues of the sturm - liouville problem . this immediately implies the discreteness of the spectrum and offers an efficient method for its computation . in practical applications of the spps method it is often convenient to consider not only series with the centre in the origin @xmath20 but also series with the centre at @xmath21 where @xmath22 is some complex number . in section [ sectspshift ] we show that this spectral parameter shifting technique is applicable to equation ( [ i1 ] ) and give necessary details . the spps representation allows us to obtain a result on mapping properties of the transmutation operator corresponding to the operator @xmath23 , which was studied , e.g. , in @xcite , @xcite and @xcite . we show in section [ secttransmut ] how the transmutation operator acts on certain powers of the independent variable . in the case of nonsingular schrdinger operators a result of this kind allowed us to advance in the construction of the transmutation operator itself @xcite and had applications in constructing complete systems of solutions for some partial differential equations @xcite , @xcite , @xcite . in section [ sectnumeric ] we explain the numerical implementation of the developed spps method for solving sturm - liouville problems for ( [ i1 ] ) . first , we consider several known test problems , and comparing the obtained results with the results obtained by the best available codes , as sleign2 , matslise and some others , we show that our method is highly competitive and gives better or at least comparable results on test problems to which other codes are applicable . second , we consider an example which involves a different from zero @xmath24 in ( [ i1 ] ) and a complex spectrum . meanwhile sleign2 and matslise are not applicable to problems admitting complex eigenvalues , our method delivers results which are in excellent agreement with the exact data . consider a perturbed bessel operator ( also known as a spherical schrdinger operator ) @xmath25 , \label{opsingularsl}\ ] ] where the potential @xmath3 is ( in general ) a complex - valued continuous function on @xmath4 $ ] satisfying the growth condition in the origin @xmath26 note that we understand the @xmath27-notation in the sense of inequality , i.e. , there exist a neighborhood @xmath28 $ ] of zero and a constant @xmath29 such that @xmath30 for all @xmath31 $ ] . if @xmath32 or @xmath33 $ ] , the left endpoint is singular . despite of that , the equation @xmath34 possesses a solution @xmath35 which is bounded at @xmath14 and satisfies the following asymptotics at @xmath14 @xmath36 see , e.g. , ( * ? ? ? * lemma 3.2 ) for the real - valued potential @xmath3 . we show the explicit construction of the solution with this asymptotics at zero for the general case of complex potentials in section [ sectpartsol ] meanwhile in section [ sectspectrprob ] we show that such solution is unique . together with @xmath23 consider a linear differential operator @xmath37 of order at most one , where @xmath38 $ ] are complex - valued functions , and consider the following differential equation involving a spectral parameter @xmath8 @xmath39 in order to construct a spectral parameter power series representation of a non - singular in zero solution of assume that there exists a non - vanishing on @xmath4 $ ] complex - valued solution @xmath17 of the equation @xmath40 satisfying together with its first derivative the asymptotic relations and . let us define the following system of recursive integrals @xmath41\,dt , & \text{if } n\text { is odd},\\ -\displaystyle\int_{0}^{x}\frac{\widetilde{x}^{(n-1)}(t)}{u_{0}^{2}(t)}\,dt , & \text{if } n\text { is even}. \end{cases } \end{split } \label{xtilde}\ ] ] we keep the notation @xmath42 for consistency with the notations from other publications on the spps method , see , e.g. , @xcite , @xcite , @xcite . note that for an odd @xmath43 we have@xmath44 & = \bigl(r_{1}\frac{d}{dx}+r_{0}\bigr)\bigl(u_{0}\widetilde{x}^{(n-1)}\bigr)\\ & = r_{1}u_{0}^{\prime}\widetilde{x}^{(n-1)}-r_{1}u_{0}\frac{\widetilde { x}^{(n-2)}}{u_{0}^{2}}+r_{0}u_{0}\widetilde{x}^{(n-1)}=r[u_{0}]\widetilde { x}^{(n-1)}-\frac{r_{1}}{u_{0}}\widetilde{x}^{(n-2)}. \end{split}\ ] ] hence we can write in a different form , which does not require differentiation of the functions @xmath45 , @xmath46(t)\widetilde{x}^{(n-1)}(t)-r_{1}(t)\widetilde{x}^{(n-2)}(t)\bigr)\,dt , & \text{if } n\text { is odd},\\ -\displaystyle\int_{0}^{x}\frac{\widetilde{x}^{(n-1)}(t)}{u_{0}^{2}(t)}\,dt , & \text{if } n\text { is even}. \end{cases } \label{xtildealt}\ ] ] the following lemma establishes that all the involved integrals in are well defined and provides some estimates for the functions @xmath45 . [ lemmaxtildeestimate ] let admit a solution @xmath47\cap c^{2}(0,a]$ ] ( in general , complex - valued ) which does not have other zeros on @xmath13 $ ] except at @xmath14 and satisfies the asymptotic relations and . then the system of functions @xmath48 is well defined by or and the functions @xmath45 satisfy the inequalities @xmath49 where @xmath50 is the pochhammer symbol , @xmath51 and the constants @xmath52 , @xmath53 and @xmath54 are such that for any @xmath55 $ ] the following inequalities hold @xmath56(t)\bigr|\leq c_{1}t^{2l+1},\qquad\left\vert \frac { 1}{u_{0}^{2}(t)}\right\vert \leq c_{2}t^{-2l-2},\qquad|r_{1}(t)|\leq c_{3}. \label{weightsestimate}\ ] ] the constants @xmath52 and @xmath53 in lemma [ lemmaxtildeestimate ] exist due to the fact that @xmath17 is a non - vanishing function possessing asymptotics and . the proof is by induction . indeed , for @xmath57 we have @xmath58 and @xmath59(t)\bigr|\,dt \le \int_0^x c_1 t^{2l+1}\,dt\le\frac{cx^{2l+2}}{2l+2}.\ ] ] assuming that the statement is true for some @xmath60 , for @xmath61 we obtain @xmath62 and @xmath63(t)\widetilde x^{(2k+2)}(t)\bigr|\,dt+\int_0^x\bigl| r_1(t)\widetilde x^{(2k+1)}(t)\bigr|\,dt \\ & \le \int_0^x c_1 t^{2l+1}\cdot \frac{c^{2k+2}t^{k+1}}{(2l+2)_{k+1}}\,dt+\int_0^xc_3\cdot \frac{(k+1)c^{2k+1}t^{2(l+1)+k}}{(2l+2)_{k+1}}\,dt\\ & \le \frac{c^{2k+3}x^{2l+2+k+1}}{(2l+2+k+1)\cdot ( 2l+2)_{k+1}}+\frac{c^{2k+2}\cdot(k+1)x^{2l+2+k+1}}{(2l+2+k+1)\cdot ( 2l+2)_{k+1}}\\ & \le \frac{(k+2)c^{2k+3}x^{2(l+1)+k+1 } } { ( 2l+2)_{k+2}}. \end{split}\ ] ] note that the exponents of the powers of @xmath64 in all the involved integrands are non - negative , hence all the recursive integrals are well defined . in the particular case when @xmath11 , i.e. , the right hand side of does not depend on the derivative of @xmath65 , the estimates of lemma [ lemmaxtildeestimate ] can be improved , and the following statement is valid . [ lemmaxtildeestimatezeror1 ] under the conditions of lemma [ lemmaxtildeestimate ] assume additionally that @xmath11 . then the functions @xmath45 defined by or satisfy the inequalities @xmath66 where @xmath67 is the pochhammer symbol and @xmath68 , where the constants @xmath52 and @xmath53 are such that for any @xmath55 $ ] the following inequalities hold @xmath69 the proof is analogous to that of lemma [ lemmaxtildeestimate ] . the following theorem presents the spectral parameter power series ( spps ) representation of a bounded solution of equation . [ thmsingularspps ] let admit a solution @xmath47\cap c^{2}(0,a]$ ] ( in general , complex - valued ) which does not have other zeros on @xmath13 $ ] except at @xmath14 and satisfies the asymptotic relations and . then for any @xmath70 the function @xmath71 is a solution of belonging to @xmath72\cap c^{2}(0,a]$ ] and the series converges uniformly on @xmath13 $ ] . the first derivative of @xmath65 is given by @xmath73 and the series for the first and the second derivatives converge uniformly on an arbitrary compact @xmath74 $ ] . formally differentiating the series twice with the aid of we obtain that @xmath75 should be given by and @xmath76 ( after simplification ) by @xmath77 . by lemma [ lemmaxtildeestimate ] all the involved series converge uniformly on an arbitrary compact @xmath74 $ ] , hence the formal derivatives coincide with the usual ones . since by and @xmath78,\ ] ] @xmath65 is indeed a solution of equation . the relations and follow from the corresponding asymptotics of @xmath79 because by lemma [ lemmaxtildeestimate ] we have @xmath80 and @xmath81 for @xmath82 . for a regular sturm - liouville problem the existence and the construction of the required non - vanishing solution @xmath17 present no difficulty since the equation possesses two linearly independent real - valued solutions @xmath83 and @xmath84 whose zeros alternate and one may choose @xmath85 as such solution . for the singular equation under consideration there is only one solution satisfying the asymptotic conditions and , see theorem [ thmspectralproblem ] . corollary [ corrnonvanishingps ] establishes that such non - vanishing solution @xmath17 exists in the case when @xmath86 , @xmath16 $ ] , and in remark [ rmkbddq ] we show that a modified spps representation is always possible in the case when @xmath3 is real valued and bounded from below and @xmath24 is real valued . for spps representations for solutions of nonsingular sturm - liouville equations we refer to @xcite , @xcite and @xcite . they have been applied in a number of papers to different scattering and spectral problems ( see references in the introduction ) . for the perturbed bessel equation , in the case of a real valued potential @xmath3 , @xmath11 and @xmath87 an spps representation was obtained and used in @xcite but without formulas for constructing or estimating the coefficients @xmath88 . for practical applications the partial sums of the series are of the main interest . based on lemmas [ lemmaxtildeestimate ] and [ lemmaxtildeestimatezeror1 ] the following corollary provides estimates for the difference between the exact solution and the approximate one defined as a partial sum of the series . the difference is estimated in terms of the remainders of taylor series of two special functions . [ corrsppsdiscrepancy ] under the conditions of theorem [ thmsingularspps ] consider @xmath89 , note that for @xmath90 the right - hand side is equal to @xmath17 . then @xmath91}|u_{0}(t)|\cdot\sum_{k = n+1}^{\infty } \frac{|\lambda|^{k}c^{2k}x^{k}}{(2l+2)_{k}}\leq\max_{t\in\lbrack0,x]}|u_{0}(t)|\cdot\biggl|e^{c^{2}|\lambda|x}-\sum_{k=0}^{n}\frac{(c^{2}|\lambda|x)^{k}}{k!}\biggr| , \label{eqsppssoldiscrepancy}\ ] ] where the constant @xmath92 is defined in lemma [ lemmaxtildeestimate ] . moreover , in the particular case @xmath11 the following estimate holds @xmath93}|u_{0}(t)|\cdot\biggl|\frac{2^{l+1/2}\gamma(l+\frac32)}{(|\lambda|^{1/2}cx ) ^{l+1/2}}i_{l+1/2}(|\lambda \end{split } \label{eqsppssoldiscrepancyzeror1}\ ] ] where @xmath94 is the gamma function , @xmath95 is the modified bessel function of the first kind and the constant @xmath92 is defined in lemma [ lemmaxtildeestimatezeror1 ] . for the proof one should simply compare the majorizing terms corresponding to the even indices in and with the taylor expansions for the exponential and the bessel functions appearing in and . [ examplebessel ] consider the bessel equation @xmath96 with @xmath97 and @xmath16 $ ] . the regular solution of this equation satisfying and is given by the formula@xmath98 where @xmath99 is the bessel function of the first kind . this solution may be represented as a power series in terms of the parameter @xmath8,@xmath100 see @xcite . in order to apply theorem [ thmsingularspps ] consider @xmath101 as the non - vanishing on @xmath4 $ ] solution of the equation @xmath102 , satisfying and . it is easy to verify that choosing such @xmath17 we obtain from @xmath103 i.e. , exactly the coefficients from . in this section we explain how to construct a particular solution of equation satisfying asymptotics , and present some sufficient conditions for this solution to be non - vanishing for @xmath104 . in order to construct an spps representation for the particular solution we rewrite equation in the form @xmath105 the equation @xmath106 possesses two solutions @xmath107 for @xmath97 the second solution is regular and satisfies and . since the potential @xmath3 may be singular in the origin , we can not apply theorem [ thmsingularspps ] directly . however we may construct the system of recursive integrals in the same way as in and only have to show the convergence of the integrals and obtain some estimates justifying the spps representation . consider the following system of recursive integrals @xmath108 note that since the potential @xmath109 $ ] satisfies condition for some @xmath110 there exists a constant @xmath29 such that @xmath111.\ ] ] [ lemmaytildeestimate ] suppose that the complex - valued potential @xmath112 $ ] satisfies inequality for some @xmath29 and @xmath6 . then the functions @xmath113 are well defined by and the following estimates hold @xmath114 , \label{ytildeoddestimate}\ ] ] where @xmath67 is the pochhammer symbol . the proof can be performed by induction , similarly to the proof of lemma [ lemmaxtildeestimate ] . the only difference is that it is possible that an exponent of the power of @xmath64 under the integral sign be negative , however it is always strictly greater than @xmath115 . hence all the involved integrals exist . [ proppartsol ] suppose that the complex - valued potential @xmath109 $ ] satisfies inequality for some @xmath29 and @xmath6 . then the function @xmath116 where the functions @xmath117 are defined by , is a particular solution of equation on @xmath4 $ ] satisfying asymptotics , . moreover , @xmath17 satisfies the following estimate for any @xmath16 $ ] @xmath118 due to the first and the second derivatives of @xmath79 are given by the expressions @xmath119 and @xmath120 the uniform convergence of all the involved series on an arbitrary compact @xmath74 $ ] and hence the possibility of termwise differentiation , follows from lemma [ lemmaytildeestimate ] . the asymptotic relations and for the function @xmath79 follow from the estimates @xmath121 and @xmath122 , @xmath123 for @xmath82 , see and . inequality follows from the series representation of the modified bessel functions of the first kind , see , e.g. , @xcite . the following corollary provides a sufficient condition for the particular solution constructed in proposition [ proppartsol ] to be non - vanishing . [ corrnonvanishingps ] under the conditions of proposition [ proppartsol ] assume additionally that @xmath86 , @xmath16 $ ] . then @xmath124 for any @xmath16 $ ] . we obtain from and the condition @xmath125 that @xmath126 , @xmath127 . hence , @xmath128 . the classical formulation ( see , e.g. , @xcite , @xcite ) of a spectral problem for a sturm - liouville equation of the form @xmath129 with @xmath23 from consists in finding the values of the spectral parameter for which there exists a solution @xmath130 continuous at @xmath14 and satisfying the following conditions . when @xmath131 , @xmath132 and @xmath133 for some @xmath134 such that @xmath135 . when @xmath136 and since the second linearly independent solution is also square - integrable , an additional boundary condition @xmath137 is imposed ( see , e.g. , @xcite ) . despite the fact that the right - hand side of the spectral equation may contain the derivative of the unknown function and hence not fit into the basic sturm - liouville scheme , we consider the same spectral problem for equation . the following statement gives us a characteristic equation of the spectral problem under the condition that an appropriate non - vanishing solution @xmath79 of exists . [ thmspectralproblem ] let admit a solution @xmath47\cap c^{2}(0,a]$ ] ( in general , complex - valued ) which does not have other zeros on @xmath13 $ ] except at @xmath14 and satisfies the asymptotic relations and . then the eigenvalues of the problem , coincide with zeros of the entire function @xmath138 where the functions @xmath139 are defined by or . under the condition of the theorem , the function @xmath65 defined by is a solution of satisfying the boundary conditions and . the boundary condition for the function @xmath65 coincides with @xmath140 , and @xmath19 is an entire function by theorem [ thmsingularspps ] . it is left to show that there are no more eigenvalues . for that it is sufficient to show that the second linearly independent solution @xmath141 of does not satisfy either or . we rewrite in the form @xmath142 and introduce the function @xmath143 note that @xmath144 since the solution @xmath65 defined by satisfies the asymptotic condition , there exists a constant @xmath145 such that @xmath146 for all @xmath147 $ ] . then the second linearly independent solution is given by the liouville formula ( * ? ? ? xi ) , @xcite @xmath148.\ ] ] assume first that @xmath149 . it follows from , asymptotics , and lhospital s rule that @xmath150 hence for @xmath151 the second solution @xmath141 does not satisfy the boundary condition . let now @xmath152 . recall that by abel s identity ( * ? ? ? xi ) the wronskian of @xmath65 and @xmath141 has the form @xmath153 hence using , , and we obtain @xmath154 and thus observe that @xmath141 can not satisfy the boundary condition . for @xmath155 similar reasoning shows that @xmath156 we substitute the expression for @xmath157 obtained from into and obtain @xmath158 for the first term in this expression , from and we have @xmath159 hence to prove that @xmath141 does not satisfy the boundary condition it is sufficient to show that the second term is @xmath160 as @xmath123 . due to the asymptotic relations and it is sufficient to show that @xmath161 for some @xmath162 . taking into account we have @xmath163 as can be seen from , @xmath164 , hence only the first term is relevant . since @xmath165 as @xmath123 , it is sufficient to prove that @xmath166 . similarly to we obtain that the general solution of equation can be represented as @xmath167 , where @xmath168 is given by and satisfies the asymptotic condition and @xmath169 satisfies the asymptotic condition @xmath170 , @xmath123 . since @xmath171 by the statement of the theorem , it is necessarily of the form . now using , and we obtain @xmath172 where @xmath173 participates in , and this finishes the proof for @xmath155 . the spps representation given in theorem [ thmsingularspps ] is based on a particular solution of equation for @xmath20 . in @xcite it was mentioned that for a classic sturm - liouville equation it is also possible to construct the spps representation of a general solution starting from a non - vanishing particular solution for some @xmath21 . such procedure is called spectral shift and has already proven its usefulness for numerical applications @xcite , @xcite . we show that a spectral shift technique may be applied to equation . let @xmath22 be a fixed complex number . we rewrite in the form @xmath174 where @xmath175 . suppose that @xmath17 is a solution of the equation @xmath176 such that @xmath17 does not vanish on @xmath4 $ ] . note that @xmath17 is a particular solution of for @xmath21 . then the operator @xmath177 admits the following plya factorization @xcite @xmath178 where @xmath179 based on the factorization we introduce the following system of recursive integrals @xmath180\,dt , & \text{if } n\text { is odd},\\ -\displaystyle\int_{0}^{x}\frac{\widetilde{z}^{(n-1)}(t)}{p(t)u_{0}^{2}(t)}\,dt , & \text{if } n\text { is even}. \end{cases } \end{split } \label{ztilde}\ ] ] similarly to the recurrent relation can be rewritten as @xmath181(t)\widetilde{z}^{(n-1)}(t)-r_{1}(t)\widetilde{z}^{(n-2)}(t)\bigr)\,dt , & \text{if } n\text { is odd},\\ -\displaystyle\int_{0}^{x}\frac{\widetilde{z}^{(n-1)}(t)}{p(t)u_{0}^{2}(t)}\,dt , & \text{if } n\text { is even}. \end{cases } \label{ztildealt}\ ] ] [ thmspectralshift ] let admit a solution @xmath182\cap c^{2}(0,a]$ ] ( in general , complex - valued ) which does not have other zeros on @xmath13 $ ] except at @xmath14 and satisfies the asymptotic relations and . then for any @xmath183 the function @xmath184 is a solution of belonging to @xmath72\cap c^{2}(0,a]$ ] , and the series converges uniformly on @xmath13 $ ] . the series for the first and the second derivatives converge uniformly on an arbitrary compact @xmath74 $ ] and the first derivative has the form @xmath185 note that the function @xmath186 given by is continuous and non - vanishing on @xmath13 $ ] . similarly to the proof of lemma [ lemmaxtildeestimate ] we see that the functions @xmath187 satisfy estimates with the constant @xmath188 where @xmath189}\frac{\bigl| p(t)u_0(t ) r[u_0](t)\bigr|}{t^{2l+1}},\qquad c_2 = \max_{t\in[0,a]}\frac{t^{2l+2}}{\bigl|p(t)u_0 ^ 2(t)\bigr|},\qquad c_3 = \max_{t\in[0,a]}\bigl|r_1(t)\bigr|.\ ] ] formal application of the operator @xmath190 to with the use of plya factorization and formula shows that the function @xmath65 is a solution of equation and hence of . estimates for @xmath187 justify the possibility of differentiation of the involved series and show that the function @xmath65 satisfies and . [ rmkbddq ] suppose that the functions @xmath3 and @xmath24 are real - valued and that the potential @xmath3 is bounded from below , i.e. , there exists a constant @xmath191 such that @xmath192.\ ] ] consider equation for @xmath193 . the particular solution of this equation can be constructed similarly to section [ sectpartsol ] using the generalization of formulas according to the factorization , cf . , and . since the functions @xmath186 and @xmath194 are non - negative , similarly to corollary [ corrnonvanishingps ] we deduce that the particular solution @xmath17 in such case does not have other zeros on @xmath13 $ ] except at @xmath14 . hence it is possible to construct the spps representation of the bounded solution for any equation having a real - valued @xmath24 and a real - valued bounded from below @xmath3 . in the case when @xmath11 , we do not need to introduce the new system of functions @xmath187 . the system of functions @xmath195 given by may be used directly in representation . for the difference @xmath196}|u(x)-u_{n}(x)|$ ] , where @xmath197 the estimate holds , where @xmath92 is given by . we recall a general definition of a transmutation operator from @xcite which is a modification of the definition given by levitan @xcite . let @xmath198 be a linear topological space and @xmath199 its linear subspace ( not necessarily closed ) . let @xmath200 and @xmath201 be linear operators : @xmath202 . [ deftransmut ] a linear invertible operator @xmath203 defined on the whole @xmath198 such that @xmath199 is invariant under the action of @xmath203 is called a transmutation operator for the pair of operators @xmath200 and @xmath201 if it fulfills the following two conditions . 1 . both the operator @xmath203 and its inverse @xmath204 are continuous in @xmath198 ; 2 . the following operator equality is valid @xmath205 or which is the same @xmath206 very often in literature the transmutation operators ( the term coined by delsarte and lions @xcite ) are called transformation operators . in @xcite for the case of the transmutation operator @xmath203 corresponding to the pair of operators @xmath207 and @xmath208 a mapping property of @xmath203 was found . it establishes what is the result of action of @xmath203 on the powers of the independent variable . this mapping property already found many applications in the proofs of completeness of infinite systems of solutions of some linear partial differential equations and in solving initial and spectral problems , see @xcite , @xcite , @xcite , @xcite , @xcite , @xcite . we present an analogue of the aforementioned mapping property for the transmutation operator for the pair of operators @xmath209 and @xmath210 , constructed in @xcite , @xcite . we recall some results from @xcite , @xcite . under certain additional conditions on the potential @xmath3 , discussed below , a bounded solution of the equation @xmath211 can be represented in the form @xmath212 where @xmath213 is a solution of the equation @xmath214 and @xmath99 is the bessel function of the first kind . the integral kernel @xmath215 is the solution of the partial differential equation @xmath216 satisfying the boundary conditions @xmath217 moreover , the integral kernel @xmath215 satisfies @xmath218 we denote the operator defined by as @xmath219 . the existence of such operator was established in @xcite for the case when @xmath3 is a continuous function on @xmath13 $ ] and in @xcite for the case when @xmath1 is an integer and @xmath3 is a real - valued function satisfying the condition @xmath220 for some @xmath221 . it is mentioned in @xcite that the results of @xcite allow one to extend the existence of the operator @xmath219 onto arbitrary real values of the parameter @xmath1 . the solution @xmath222 in differs from the solution @xmath223 satisfying the asymptotic condition by the factor @xmath224 , see @xcite . the series expansion of the function @xmath225 is @xmath226 similarly to ( * ? ? ? * theorem 7 ) we substitute the functions @xmath222 and @xmath225 in by their series expansions and and obtain @xmath227 . \end{split}\ ] ] the function @xmath228 is square - integrable on @xmath229 $ ] , see , and the function @xmath230 is the limit of the uniformly convergent series , hence the possibility to change the order of summation and integration in follows from the continuity of the scalar product in @xmath231 $ ] . since the equality in holds for all @xmath232 and @xmath8 , we finally obtain that @xmath233 = ( -1)^{k } 2^{2k } k ! \bigl(l+\frac32\bigr)_{k } u_{0}(x ) \widetilde x^{(2k)}(x).\ ] ] based on the results of the previous sections we can formulate a numerical method for solving spectral problems for perturbed bessel equations . 1 . find a particular solution @xmath17 of equation satisfying the asymptotic conditions and . note that due to the proof of theorem [ thmspectralproblem ] it is sufficient to check that the solution satisfies only the asymptotic condition . if an analytic expression for the particular solution is unknown , one can use a numerical approximation suggested by proposition [ proppartsol ] . 2 . check that the particular solution @xmath17 obtained in step 1 is non - vanishing for @xmath16 $ ] . under the conditions of corollary [ corrnonvanishingps ] it is always the case . if the particular solution has zeros on @xmath4 $ ] , the spectral shift technique described in section [ sectspshift ] can help , either directly as described in remark [ rmkbddq ] or by finding a suitable value of @xmath22 , complex in general . 3 . use partial sums of the series and ( or , correspondingly , and ) to obtain a polynomial @xmath234 approximating the characteristic function . find zeros of the calculated polynomial @xmath235 . 5 . to improve the accuracy of the eigenvalues located farther from the point @xmath22 , the spectral parameter corresponding to the current particular solution , perform one or several steps of the spectral shift technique . since the characteristic function @xmath19 is analytic , the rouch theorem from complex analysis , see , e.g. , @xcite , provides the stability of the numerical procedure . indeed , let @xmath236 be an arbitrary simple closed contour on which @xmath237 does not vanish . then if the absolute error of approximation @xmath238 is less than @xmath239 on @xmath236 , due to rouch s theorem the functions @xmath240 and @xmath241 possess the same number of zeros inside @xmath236 . hence the procedure described above does not produce additional ( or on the contrary less ) zeros whenever @xmath240 approximates well enough the function @xmath237 . we illustrate this point below , in example [ ex6 ] . before considering numerical examples let us explain how the numerical implementation of the spps method was done in this work . all the calculations were performed with the help of matlab 2010 in the double precision machine arithmetics . the formal powers @xmath195 were calculated using the newton - cottes 6 point integration formula of 7-th order , see , e.g. , @xcite , modified to perform indefinite integration . we choose @xmath242 equally spaced points covering the segment of interest and apply the integration formula to overlapping groups of six points . it is worth mentioning that for large values of the parameter @xmath1 a special care should be taken near the point @xmath243 , because even small errors in the values of @xmath244 after the division by @xmath245 lead to large errors in the computation of @xmath246 on the whole interval @xmath13 $ ] . to overcome this difficulty we change the values of @xmath244 in several points near zero to their asymptotic values . this strategy leads to a good accuracy . the computation of the first 100200 formal powers proved to be a completely feasible task , and even for @xmath242 being as large as several millions the computation time of the whole set of formal powers is within seconds . in the presented numerical results we specify two parameters : @xmath247 is the degree of the polynomial @xmath240 in , i.e. , the number of the calculated formal powers is @xmath248 , and @xmath242 is the number of points taken on the considered segment for the calculation of integrals . let us stress that the formal powers do not depend on the spectral parameter and once calculated can be used for computing the solution and/or the characteristic function @xmath19 for thousands of different values of the spectral parameter @xmath8 without any additional significant computation cost . there exist several computer codes for the solution of singular sturm - liouville problems . we compare our results with the results produced by sleign2 @xcite and matslise @xcite . both packages can reliably solve a variety of spectral problems for regular and singular sturm - liouville problems and are considered as basic comparison tools , the second package to our best knowledge is one of the most accurate . both packages work with double precision machine arithmetics and were used with the parameters for the highest possible accuracy goals . it should be mentioned that the applicability of both packages is restricted to the self - adjoint situation and they do not permit neither complex coefficients nor a first order differential operator at a spectral parameter . thus , to illustrate the performance of the proposed method in a situation when a comparison to the existing software is impossible , in example [ ex6 ] we consider a problem for which an exact characteristic equation can be written down and compare the obtained numerical results to those calculated from the exact equation . [ ex1]our first numerical example is the bessel equation , which we already touched on in example [ examplebessel ] . consider the following spectral problem ( * ? ? ? * example 2 ) @xmath249 in this and in all other considered examples the solution @xmath222 should also satisfy the boundary conditions and at 0 . as follows from , the eigenvalues of the problem coincide with zeros of the bessel function @xmath250 , where @xmath251 and @xmath252 . we compare the results delivered by the spps method to those from ( * ? ? ? * example 2 ) for a particular case @xmath253 . for the spps representation the exact particular solution @xmath254 was used . the results are presented in table [ ex1table1 ] together with the exact eigenvalues calculated as squares of zeros of @xmath255 and computed using the matlab routine ` besselzero.m ` by greg von winckel . [ c]cr@lr@lr@l@xmath43 & & & + 1 & 12 . & 1871394680951 & 12 . & 187139459 & 12 . & 1871394680951 + 2 & 44 . & 257559403500 & 44 . & 257558912 & 44 . & 257559403502 + 3 & 96 . & 07160483898 & 96 . & 071604502 & 96 . & 07160483884 + 4 & 167 . & 62571241787 & 167 . & 625711908 & 167 . & 62571242058 + 5 & 258 . & 91930037169 & 258 . & 919292439 & 258 . & 91930035744 + 6 & 369 . & 95220905860 & 369 . & 952209262 & 369 . & 95220926235 + 7 & 500 . & 72440133519 & 500 . & 724370471 & 500 . & 72438147579 + 8 & 651 . & 23517308180 & 651 . & 235865279 & 651 . & 23579210254 + 9 & 821 . & 50506498326 & 821 . & 486428982 & 821 . & 48642898238 + 10 & 988 . & 97560762340 & 1011 . & 476285608 & 1011 . & 47628560802 + as can be seen from table [ ex1table1 ] , the accuracy of the higher eigenvalues calculated by the spps method decreases . to improve the accuracy the spectral shift technique described in section [ sectspshift ] is applied . we choose that the values of the parameter @xmath22 change along a line in the complex plane and are given by @xmath256 , @xmath257 . on each step we compute the solution in the next point @xmath258 and use this solution as a particular solution for the next step . since the spectral problem has a purely real spectrum , among the zeros of the approximating polynomial the ones with a small imaginary part are chosen and those which are closer to the number @xmath259 than to any other of the numbers @xmath260 , @xmath261 are stored . the obtained results are presented in table [ ex1table2 ] together with the used values of @xmath22 for the spectral shift and the exact eigenvalues . in the first column of the table we combine the results produced by the matslise package with the exact eigenvalues because in this case all the presented digits coincide . as can be seen from table [ ex1table2 ] , the spectral shift technique allows us to significantly improve the obtained results for the higher eigenvalues . from now on we present only the results obtained with the help of the spectral shift technique and specify the implemented rules for choosing spectral shifts . [ c]cr@lr@lr@+lr@l@xmath43 & & & & + 1 & 12 . & 1871394680951 & 12 . & 1871394680951 & 1 & @xmath262 & 12 . & 187139459 + 2 & 44 . & 257559403502 & 44 . & 257559403500 & 50 & @xmath263 & 44 . & 257558912 + 3 & 96 . & 071604838843 & 96 . & 071604838834 & 100 & @xmath264 & 96 . & 071604502 + 4 & 167 . & 62571242058 & 167 . & 62571242056 & 150 & @xmath265 & 167 . & 625711908 + 5 & 258 . & 91930035744 & 258 . & 91930035742 & 250 & @xmath266 & 258 . & 919292439 + 6 & 369 . & 95220926235 & 369 . & 95220926232 & 350 & @xmath267 & 369 . & 952209262 + 7 & 500 . & 72438147579 & 500 . & 72438147575 & 500 & @xmath268 & 500 . & 724370471 + 8 & 651 . & 23579210254 & 651 . & 23579210250 & 650 & @xmath269 & 651 . & 235865279 + 9 & 821 . & 48642898238 & 821 . & 48642898235 & 800 & @xmath270 & 821 . & 486428982 + 10 & 1011 . & 47628560802 & 1011 . & 47628560801 & 1000 & @xmath271 & 1011 . & 476285608 + 30 & 8956 . & 5077203636 & 8956 . & 5077203638 & 8950 & @xmath272 & 8956 . & 507721371 + 50 & 24797 . & 222775294 & 24797 . & 222775296 & 24800 & @xmath273 & 24796 . & 866878938 + 75 & 55701 . & 421553437 & 55701 . & 421553167 & 50000 & @xmath274 & 55701 . & 328815911 + 100 & 98942 . & 625835 & 98942 . & 625812 & 100000 & @xmath275 & 98943 . & 253862034 + [ ex2 ] the second example is the boyd equation . consider the following spectral problem ( * ? ? ? * example 3 ) @xmath276 this equation fits into the general scheme if one takes @xmath277 . we take the function @xmath278 as a particular solution of and calculate eigenvalues using the spps method with @xmath279 and @xmath280 . for problem the characteristic equation is known and is given by @xmath281 where @xmath282 is the whittaker function and @xmath283 , see ( * ? ? ? * example 3 ) . in table [ ex2table2 ] we present the obtained results together with the exact eigenvalues calculated from and the results obtained using sleign2 and matslise software . note that some eigenvalues caused problems for the matslise package , despite the excellent accuracy of the rest of the results . to simplify the reading , in the presented numbers we truncated some digits which do not coincide with the correct ones . relative errors of the first 50 eigenvalues obtained by spps , matslise and sleign2 are presented on figure [ bezex3error ] . we emphasize a relative stability of the numerical results delivered by the spps method in comparison to those computed by matslise and sleign2 . we mention also that the same problem was considered in @xcite where less accurate results were reported : @xmath284 , @xmath285 , @xmath286 , @xmath287 and @xmath288 . [ c]cr@lr@lr@lr@lr@l@xmath43 & & & & & & + 1 & 7 . & 3739850151752 & 7 . & 3739850151751 & 7 . & 37398499 & 7 . & 3739850151751 & & + 2 & 36 . & 3360195952325 & 36 . & 3360195952318 & 36 . & 33601959522 & 36 . & 3360195952318 & & + 3 & 85 . & 292582094149 & 85 . & 292582094137 & 85 . & 29258240 & 85 . & 292582094137 & & + 4 & 154 . & 098623739770 & 154 . & 098623739767 & 154 . & 098623739742 & 154 . & 098623739767 & & + 5 & 242 . & 705559362903 & 242 . & 705559362911 & 242 . & 705559362924 & 242 . & 705559362911 & & + 6 & 351 . & 091167129407 & 351 . & 091167129418 & 351 . & 09116712937 & 351 . & 091167129418 & & + 8 & 627 . & 155044324547 & 627 . & 155044324564 & 627 . & 155033 & 627 . & 155044324564 & & + 10 & 982 . & 239093680177 & 982 . & 239093680188 & 982 . & 239069 & 982 . & 239093680188 & & + 13 & 1662 . & 98063088581 & 1662 . & 98063088578 & 1662 . & 9806308832 & 1662 . & 76 & & + 20 & 3942 . & 42966385096 & 3942 . & 42966385102 & 3942 . & 18 & 3942 . & 42966385102 & & + 28 & 7732 . & 02180519217 & 7732 . & 02180519214 & 7732 . & 021805177 & 7729 . & 4 & & + 30 & 8876 . & 82700072940 & 8876 . & 82700072941 & 8876 . & 8270009 & 8876 . & 82700072941 & & + 35 & 12084 . & 29442705883 & 12084 . & 29442705875 & 12084 . & 263 & 12083 . & 98 & & + 40 & 15785 . & 2626475018 & 15785 . & 2626475007 & 15784 . & 3 & 15785 . & 2626475007 & & + 50 & 24667 . & 683593322 & 24667 . & 683593313 & 24667 . & 683593398 & 24662 . & 5 & & + the spps representation allows one to calculate easily the approximate eigenfunctions . the eigenfunctions that were obtained for this example are shown in fig . [ bezex3fig ] . [ ex3 ] consider the following spectral problem ( * ? ? ? * example 2 ) . @xmath289 the parameter @xmath290 was chosen equal to @xmath291 . in this example we computed both partial and general solutions using the spps representations . since the exact partial solution satisfying the asymptotic condition is known and is given by @xmath292 , we compared the approximate solution with the exact one . for @xmath279 and @xmath280 the absolute error was less than @xmath293 . with the aid of maple software we found that the characteristic equation of problem has the form @xmath294 where @xmath295 is the whittaker function , and used equation to compute the exact eigenvalues . in table [ ex3table2 ] we present the results obtained by the spps method , exact eigenvalues and the results from @xcite , matslise and sleign2 . as in example [ ex1 ] here again all presented digits in the exact eigenvalues coincide with the results delivered by matslise . the performance of the spps method was considerably better than that of sleign2 and even the 50th eigenvalue was computed correctly to 9 decimal places . [ c]cr@lr@lr@lr@l@xmath43 & & & & + 1 & 2 . & 4629499739737 & 2 . & 4629499739740 & 2 . & 46295003 & 2 . & 462949030 + 2 & 3 . & 2883529299493 & 3 . & 2883529299426 & 3 . & 28835311 & 3 . & 288339398 + 3 & 4 . & 1498642187456 & 4 . & 1498642187448 & 4 . & 14986471 & 4 . & 149833151 + 4 & 5 . & 0636688237348 & 5 . & 0636688237341 & 5 . & 0636695 & 5 . & 063634795 + 5 & 6 . & 0075814581165 & 6 . & 0075814581160 & 6 . & 0075836 & 6 . & 007577378 + 7 & 7 . & 9397373768999 & 7 . & 9397373768993 & 7 . & 939745 & & + 10 & 10 . & 8861250916182 & 10 . & 8861250916173 & 10 . & 886149 & & + 15 & 15 . & 8426318195682 & 15 . & 8426318195682 & 15 . & 84275 & & + 20 & 20 . & 8202301908125 & 20 . & 8202301908124 & 20 . & 82057 & & + 30 & 30 . & 7973502195887 & 30 . & 7973502195868 & 30 . & 7989 & & + 50 & 50 . & 77867680977 & 50 . & 77867680951 & 50 . & 789 & & + [ ex4 ] consider a particular case of the hydrogen atom equation ( * ? ? ? * example 4 ) . @xmath296 we have chosen @xmath297 and found with the help of maple the characteristic equation of problem @xmath298 where @xmath299 is the whittaker function . in this example we computed both partial and general solutions using the spps representations . the procedure for computing the eigenvalues for this example is completely analogous to what was described above . in table [ ex4table2 ] we present the results obtained by the spps method , exact eigenvalues and the results from @xcite , matslise and sleign2 . again all presented digits in the exact eigenvalues coincide with the results delivered by matslise . the spps method and sleign2 performed similarly to the previous example . the next example shows that the situation may change significantly when another value of the parameter @xmath300 is chosen . [ ex4b ] consider the same problem as in example [ ex4 ] but @xmath301 which corresponds to @xmath155 in . the computation of eigenvalues for problem performed by matslise took several hours on intel i7 - 3770 microprocessor meanwhile the computation time required by sleign2 and spps did not change significantly . we present the relative error of the first 50 eigenvalues on figure [ figex4error ] . for the spps method we used @xmath279 and @xmath302 . the spectral shift was given by @xmath303 . the accuracy of the results delivered by matslise was considerably lower and once again we emphasize the stability of the accuracy of the eigenvalues computed by the spps method in comparison to sleign2 . it is worth mentioning that not only for the extreme value @xmath155 but also for the values close to @xmath304 the computation time required by the matslise package increases significantly . [ c]cr@lr@lr@lr@l@xmath43 & & & & + 1 & 1 . & 97027445061574 & 1 . & 97027445061572 & 1 . & 9702743 & 1 . & 970274439470 + 2 & 3 . & 00436042551872 & 3 . & 00436042551857 & 3 . & 0043600 & 3 . & 004360435708 + 3 & 4 . & 01515351791731 & 4 . & 01515351791736 & 4 . & 015153523 & 4 . & 015153641332 + 4 & 5 . & 0193472218607 & 5 . & 0193472218612 & 5 . & 0193459 & 5 . & 019347630098 + 5 & 6 . & 0210053515482 & 6 . & 0210053515488 & 6 . & 0210049 & 6 . & 021006315094 + 7 & 8 . & 0215089715470 & 8 . & 0215089715478 & 8 . & 0215092 & & + 10 & 11 . & 0202653559392 & 11 . & 0202653559399 & 11 . & 0202663 & & + 15 & 16 . & 0176675547289 & 16 . & 0176675547294 & 16 . & 0176656 & & + 20 & 21 . & 0155251794158 & 21 . & 0155251794156 & 21 . & 0155504 & & + 30 & 31 . & 0125189152594 & 31 . & 0125189152597 & 31 . & 01269 & & + 50 & 51 . & 00916429569 & 51 . & 00916429551 & 51 . & 0097 & & + [ ex5 ] consider the following spectral problem ( * ? ? ? * example 3 ) . @xmath305 for this problem we were unable to find an exact characteristic equation . we computed both particular and general solutions using the spps representations . the obtained results are presented in table [ ex5table1 ] . the eigenvalues computed by the spps method are very close to those delivered by matslise . [ c]cr@lr@lr@lr@l@xmath43 & & & & + 1 & 1 . & 69965392162508 & 1 . & 69965392162512 & 1 . & 69965496630163 & 1 . & 699674822427 + 2 & 2 . & 60438727880122 & 2 . & 60438727880111 & 2 . & 60439344911062 & 2 . & 604506077325 + 3 & 3 . & 56972957088682 & 3 . & 56972957088910 & 3 . & 56974717663817 & 3 . & 570068095387 + 4 & 4 . & 55232022604084 & 4 . & 55232022604096 & 4 . & 55235767643976 & 4 . & 553053525686 + 5 & 5 . & 54189892161891 & 5 . & 54189892161906 & 5 . & 54196736047125 & 5 . & 543261224280 + 7 & 7 . & 53001773432564 & 7 . & 53001773432606 & 7 . & 53019003621712 & & + 10 & 10 . & 5211087141260 & 10 . & 5211087141255 & 10 . & 5215767716971 & & + 15 & 15 . & 5141539227764 & 15 . & 5141539227760 & 15 . & 5156303086973 & & + 20 & 20 . & 5106568768322 & 20 . & 5106568768319 & 20 . & 5139986174471 & & + 30 & 30 . & 5071385063024 & 30 . & 5071385063018 & 30 . & 5174365365478 & & + 50 & 50 . & 5043027454211 & 50 . & 5043027452760 & 50 . & 5434201591972 & & + numerical tests discussed in the previous subsection show that the spps method is highly competitive with the best existing codes on their field of applicability . it delivers stable and reliable results even though there remains still plenty of room for improving different computational aspects of the developed programs which implement the spps method . moreover , the range of applicability of the spps method to the difference from the other considered codes includes complex coefficients , differential operator on the right - hand side of and complex eigenvalues . here we present one such example . [ ex6 ] consider the following spectral problem with the right - hand side of the equation involving a derivative @xmath306 using maple we found that the solution of equation satisfying the asymptotic condition is given by the expression @xmath307 where @xmath308 is the modified bessel function of the first kind . the derivative of this solution has the form @xmath309 for a numerical experiment we have chosen @xmath310 , hence the exact characteristic equation is @xmath311 [ c]cr@+lr@+l@xmath43 & & + 1 & 4.47123493365 & 6.76481747492@xmath312 & 4.47123493371 & 6.76481747480@xmath312 + 2 & 5.63553225528 & 13.37799928406@xmath312 & 5.63553225515 & 13.37799928396@xmath312 + 3 & 6.35749327967 & 19.82515033089@xmath312 & 6.35749327947 & 19.82515033081@xmath312 + 4 & 6.88515096019 & 26.20887598267@xmath312 & 6.88515095992 & 26.20887598266@xmath312 + 5 & 7.30184486323 & 32.56088281571@xmath312 & 7.30184486294 & 32.56088281579@xmath312 + 10 & 8.62739882797 & 64.14303168943@xmath312 & 8.62739882786 & 64.14303168978@xmath312 + 20 & 9.98333956705 & 127.0816376255@xmath312 & 9.98333956726 & 127.0816376257@xmath312 + 30 & 10.7844002548 & 189.9555609957@xmath312 & 10.7844002552 & 189.9555609955@xmath312 + 50 & 11.7983559318 & 315.6569255418@xmath312 & 11.7983559297 & 315.6569255437@xmath312 + 75 & 12.6055406455 & 472.7574366522@xmath312 & 12.6055406452 & 472.7574366509@xmath312 + 100 & 13.1790674123 & 629.8482784849@xmath312 & 13.1790674160 & 629.8482784850@xmath312 + when a problem admits complex eigenvalues we need to distinguish which roots of the polynomial @xmath313 correspond to the eigenvalues and which are spurious roots appearing due to the truncation procedure . contrary to the case of a purely real spectrum we can not simply discard all roots whose imaginary part is greater than some @xmath314 . instead , rouch s theorem and estimate suggest that the roots closest to the origin ( or to the current centre @xmath315 when the spectral shift is used ) can not be the spurious roots . we give an illustration of application of this theorem . according to rouch s theorem we need to find such values of the radius @xmath316 that @xmath317 establishing that the numbers of zeros of the functions @xmath313 and @xmath19 coincide inside the disk @xmath318 . we calculate @xmath319 using the spps representation . to estimate the difference @xmath320 we use . according to lemma [ lemmaxtildeestimate ] , for problem one can take @xmath321 . we present the obtained results on figure [ figcomplexrouche ] for @xmath322 and @xmath323 where it is compared to @xmath324 calculated with the aid of the exact characteristic function . as can be seen from the graphs , when we use the exact characteristic function for the estimation of the error the largest values of @xmath316 for which inequality holds are @xmath325 and @xmath326 . the estimate delivers rougher estimates , @xmath327 and @xmath328 . on figure [ ex6fig ] we show the graph of @xmath329 together with the boundary of the disk @xmath330 . the peaks on the graph correspond to the roots of the polynomial @xmath331 . only six of them are located in the disk and hence should be considered as approximations to the eigenvalues . all other roots are located outside the disk and hence should be discarded . to calculate the higher order eigenvalues , we applied the spectral shift technique described in section [ sectspshift ] . the following strategy for choosing the values of the spectral shift was implemented . let @xmath332 be the spectral shift on the @xmath43-th step and @xmath333 the already found eigenvalues . based on the representations and we calculate the roots of the polynomial @xmath334 and reorder them with respect to the distance from @xmath332 . from these ordered roots we choose the closest to the point @xmath332 and sufficiently distant from @xmath333 . we denote this root as @xmath335 and set @xmath336 , where @xmath337 is a fixed displacement . performing this procedure with @xmath338 , @xmath339 , @xmath322 and @xmath340 we obtained the list of eigenvalues presented in table [ ex6table2 ] . the eigenvalues are ordered according to their distance to the origin . as can be seen from the table the approximate eigenvalues are calculated with a remarkable accuracy . we thank our colleague r. michael porter for providing us with a first version of a 6 points newton - cottes matlab numerical integration routine . r. castillo would like to thank the support of the sibe and edi programs of the ipn as well as that of the project sip 20120438 . research of v. kravchenko and s. torba was partially supported by conacyt , mexico via the project 166141 . w. auzinger , e. karner , o. koch and e. weinmller . _ collocation methods for the solution of eigenvalue problems for singular ordinary differential equations _ , opuscula mathematica , 26 ( 2006 ) , no . 2 , 229241 . h. m. campos , v. v. kravchenko and l. m. mendez , _ complete families of solutions for the dirac equation : an application of bicomplex pseudoanalytic function theory and transmutation operators _ , advances in applied clifford algebras 22 ( 2012 ) , no . 3 , 577594 . h. campos , v. v. kravchenko and s. m. torba , _ transmutations , l - bases and complete families of solutions of the stationary schrdinger equation in the plane _ , j. math . ( 2012 ) , no . 2 , 12221238 . r. castillo , k. v. khmelnytskaya , v. v. kravchenko and h. oviedo , _ efficient calculation of the reflectance and transmittance of finite inhomogeneous layers _ a : pure and applied optics 11 ( 2009 ) , 065707 . r. castillo p. , v. v. kravchenko , h. oviedo and v. s. rabinovich , _ dispersion equation and eigenvalues for quantum wells using spectral parameter power series _ , j. math . , 52 ( 2011 ) , 043522 ( 10 pp . ) k. v. khmelnytskaya , v. v. kravchenko and j. a. baldenebro - obeso , _ spectral parameter power series for fourth - order sturm - liouville problems _ , applied mathematics and computation , 219 ( 2012 ) , 36103624 . k. v. khmelnytskaya , i. serroukh , _ the heat transfer problem for inhomogeneous materials in photoacoustic applications and spectral parameter power series_. mathematical methods in the applied sciences ( 2013 ) , doi : 10.1002/mma.2732 . v. v. kravchenko , s. m. torba , _ spectral problems in inhomogeneous media , spectral parameter power series and transmutation operators _ , in 2012 international conference on mathematical methods in electromagnetic theory ( mmet ) , ieee conference publications , pp.1822 , doi:10.1109/mmet.2012.6331232 . v. v. kravchenko , u. velasco - garca , _ dispersion equation and eigenvalues for the zakharov - shabat system using spectral parameter power series_. journal of mathematical physics , 52 ( 2011 ) , issue 6 , # 063517 , 8 pp . v. ledoux , m. van daele and g. vanden berghe , _ matslise : a matlab package for the numerical solution of sturm - liouville and schrdinger equations _ , acm transactions on mathematical software , 31 ( 2005 ) , 532554 . m. rajovi , r. stojiljkovi , _ the first , the second , and the third liouville formula and periodical solutions of linear differential equation of the second order_. kragujevac j. math . 30 ( 2007 ) , 131139 . s. m. sitnik , _ transmutations and applications : a survey _ , arxiv:1012.3741v1 , originally published in the book : advances in modern analysis and mathematical modeling ( editors : yu.f.korobeinik , a.g.kusraev , vladikavkaz : vladikavkaz scientific center of the russian academy of sciences and republic of north ossetia alania , 2008 , 226293 ) . v. v. stashevskaya , _ on the inverse problem of spectral analysis for a differential operator with a singularity at zero _ , zap . fiz .- mat . khgu i khmo 25 ( 1957 ) , no . 4 , 4986 . ( in russian )

we present results of a population synthesis of millisecond pulsars from the galactic disk . excluding globular clusters , we model the spatial distribution of millisecond pulsars by assuming their birth in the galactic disk with a random kick velocity and evolve them to the present within the galactic potential . we assume that normal and millisecond pulsars are standard candles described with a common radio luminosity model that invokes a new relationship between radio core and cone emission suggested by recent studies . in modeling the radio emission beams , we explore the relativistic effects of time delay , aberration and sweepback of the open field lines . while these effects are essential in understanding pulse profiles , the phase - averaged flux is adequately described without a relativistic model . we use a polar cap acceleration model for the-ray emission . we present the preliminary results of our recent study and the implications for observing millisecond pulsars with glast and agile .

0706.3041

gamma - ray pulsars are the brightest objects in the sky above 100 mev and the only identified gev galactic sources . although there are presently around 1700 known radio pulsars , only a relatively small fraction of these are detected at higher energies . the egret detector on the compton gamma - ray observatory ( cgro ) detected six @xmath0-ray pulsars with high confidence @xcite , five of which were known radio pulsars . in addition , there were a few @xmath0-ray pulsars detected with lower confidence , one of which was the millisecond psr . the large area space telescope ( glast ) , set for launch in late 2007 , will have a sensitivity about 30 times better than that of egret and is expected to detect many more @xmath0-ray pulsars . population synthesis can predict the number of expected @xmath0-ray pulsar detections , but the results are highly model - dependent . as a consequence of this sensitive model dependence , the number of @xmath0-ray pulsars that glast does detect , especially the ratio of radio - loud to radio - quiet pulsars , will be an excellent discriminator between @xmath0-ray models . using polar cap models to describe the @xmath0-ray emission , population synthesis studies of @xcite have estimated that glast should detect several hundred pulsars . population synthesis studies for outer gap models @xcite generally predict fewer radio - loud @xmath0-ray pulsars and a much higher ratio of radio - quiet to radio - loud pulsars . these studies evolved neutron stars in the galaxy from birth distributions of period , magnetic field , position and space velocity , but excluded the population of millisecond pulsars ( msps ) . since msps are thought to be recycled through spin - up by accretion from a binary companion , their evolution is somewhat more complicated . because of their very short periods , msps can have spin - down luminosities that are comparable to those of young pulsars and therefore should be sources of high - energy photons . indeed , a much higher fraction of the total radio msp population ( around 225 ) have been detected as x - ray sources ( around 60 , including globular cluster pulsars ) compared to normal pulsars . the pulsed x - ray spectra of most msps in the galactic plane are dominated by non - thermal emission @xcite , so these sources must be accelerating particles to high energy . polar cap models predict that msps should produce a high - energy emission component due to curvature radiation up to energies around 50 gev @xcite . outer gap models also predict that msps could produce high - energy emission from particles moving downward toward the stellar surface from the outer gap @xcite . predictions for the number of @xmath0-ray msps in globular clusters have been discussed by @xcite . studies of the radio characteristics of msps indicate some differences with those of the normal pulsar population . @xcite and also @xcite found that msps are somewhat less luminous than normal pulsars . however , these studies compare the pseudo - luminosities , i.e. the observed radio flux times the square of the distance , which are dependent on solid angle and observer viewing angle , and do not address real luminosities . these studies also found that the beam radii of msps , although larger on average than those of normal pulsars , are smaller than expected by an extrapolation of beam radii with a @xmath1 dependence of the normal pulsars population . the average msp beam widths seem to reach a saturated value of around @xmath2 , whereas the extrapolated value is greater than @xmath3 . we will argue however that relativistic effects @xcite may strongly influence the shape of msp profiles . msp pulse profiles have considerably larger duty cycles than normal pulsars and the beaming fractions are thought to be around 75% @xcite . the profiles were originally thought to have more complex structure , a feature that was attributed to the presence of multipole field components , possibly induced by the accretion phase @xcite . but the observations of @xcite do not support this claim and show that msp profiles display in general only marginally more complexity . the more observable details afforded by the larger duty cycles make the profiles seem more complex with the large beams increasing the chance of detecting emission from both poles . additionally , the emission height dependence on frequency ( radius - to - frequency mapping ) is less strong , and this is understandable since the msp magnetospheres are much smaller . in general , msp radio profiles are consistent with an expanded version of normal pulsar emission on open dipole field lines , with the smaller widths being attributed to emission occurring only on a subset of the open field lines or to relativistic distortions of the beam . we present in this paper new results of a population synthesis of msps born within the galactic disk . in this study we define msps as those pulsars with period derivatives @xmath4 . this criterion includes the binary pulsars j1744 - 3922 and b0655 + 64 with periods of 172.4 ms and 195.7 ms , respectively , but excludes the binary pulsar j1711 - 4322 with a period of 102.6 ms . by means of a monte carlo population code , we treat them as point particles and evolve their trajectories , periods and period derivatives from their birth ( at time of their last spin - up phase ) forward in time to the present . we exclude msps in globular clusters from our study . after determining a present - day spatial equilibrium distribution , we assign radio and @xmath0-ray characteristics to each msp and then filter its properties for detection through a select group of radio surveys and the @xmath0-ray instruments egret , agile and glast . we normalize the number of simulated radio msps to the number detected by the group of radio surveys , allowing us to predict the number of radio - loud and radio - quiet @xmath0-ray msps detected by egret , agile and glast . we recently added the swinburne intermediate latitude survey to our selected group of radio surveys used in previous studies . of the ten surveys , six of them , arecibo 2 & 3 , greenbank 2 & 3 , molongo 2 and parkes 2 , have an observing frequency around 400 mhz , and four surveys at 1400 mhz include parkes 1 , jodrell bank 2 , parkes multibeam , and swinburne intermediate latitude . in our computer code , we use the characteristics of these ten surveys to determine the minimum flux threshold for each simulated pulsar ( for details see @xcite ) at the observing frequency of the survey . while we recognize that some of these radio surveys were not very sensitive to the detection of msps , we include them to confirm null detections . when the simulated radio flux is below the flux threshold @xmath5 of all the radio surveys that potentially are viewing the simulated pulsar , we refer to the pulsar as being radio - quiet . otherwise , the pulsar is radio - loud . these characterizations are relative to the sensitivity of the surveys and are not addressing the nature of the emission process . we then are restricted to comparing the characteristics of the pulsars simulated by our computer code to those msps detected by this select group of radio surveys , excluding msps detected only by other surveys . we assume exponential spatial distributions of the form described in the study of @xcite , but using a scale height parameter above the galactic plane of 200 pc for msps in contrast to 75 pc used in our simulations for nps . we use a maxwellian distribution with a characteristic width of @xmath6 = 70 km / s to describe the supernova kick velocity at birth from the study of @xcite , resulting in an average velocity of 110 km / s . we assume a uniform birth rate back in time to 12 gyr . to determine an equilibrium spatial distribution , we evolved a group of 40,000 neutron stars to the present without making any selections . as in our recent studies , @xcite and @xcite , we use the galactic potential ( mass model 2 ) of @xcite ( w. dehnen private communication ) to evolve the trajectories of msps using a 5th order cash - karp runge - kutta routine @xcite . while msps are born in binary systems , we have not attempted to simulate the evolution of the binary system , but rather treat the system as a point particle and evolve it within the galactic potential . with low kick velocities , the msps remain bound to the galaxy and achieve normalized equilibrium distributions in galactic radius ( r ) and galactic height ( z ) shown in figure 1 as solid curves with the initial distributions represented by the dotted curves . fitting the equilibrium distributions results in radial and out - of - plane scale heights of 4.2 kpc and 0.50 kpc , respectively . the z scale height of 0.50 kpc is in good agreement with the one obtained by @xcite of 0.50 kpc and with the recent determinations of the scale heights of low mass x - ray binaries ( lmxb ) by @xcite of 0.41 kpc . in subsequent calculations , we use these equilibrium distributions to randomly select the present - day spatial distribution of msps within the galaxy . in choosing the r and z values of msps from such an evolved distribution rather than evolving each msp from its birth location , we ignore any dependence of evolved r and z on msp age . according to @xcite , the transverse motion of a pulsar causes a change in the pulsar s detected period . due to the doppler effect , an increase in the pulsar s period derivative is observed . the contribution to @xmath7 due to this effect is given by @xcite @xmath8 where @xmath9 is the pulsar period , @xmath10 is the proper motion , and @xmath11 is the distance . for msps , whose period derivatives are extremely small , this effect can increase period derivatives significantly , in some cases by as much as 90% as noted by @xcite who have also included the smaller effects of galactic rotation and vertical acceleration . in this study , we only include the shklovskii effect due to the proper motion of the simulated pulsar . while it may be that the magnetic field of msps decays slowly , we have assumed a constant field during their lifetime . however , due to selection effects , the observed field distribution will differ from the initial distribution . we explored power laws with various indices to characterize the distribution of pulsar surface magnetic field at the birth line . while @xcite preferred a power law with an index of -2 , the group of 22 msps used in their study had smaller magnetic fields . for the group of 56 msps in this study , we find a preferred index of -1 with a normalized distribution of the form @xmath12 where @xmath13 is in units of @xmath14 g. we choose a @xmath15 and a @xmath16 that provides the best agreement between various simulated and detected distributions . having the magnetic field for a simulated msp , we can obtain its initial period with a spin - up relation . @xcite and similarly @xcite found that for recycled pulsars spun up by accretion at the eddington limit from a binary companion , the initial period is estimated by the expression @xmath17 which can be rewritten as a birth line in the @xmath18 diagram by the form @xmath19 where the period is in seconds and the period derivative is @xmath20 . having the magnetic field , the initial period , and the age , we can determine the present period and period derivative of the msp assuming a pure dipole spin down . however , recent studies of lmxbs where the spin periods of accreting neutron stars were measured have allowed the estimation of the surface magnetic fields . @xcite conclude that the general properties of lmxbs can be understood if the accretion rates range from the eddington critical accretion rate to @xmath21 times that rate . these different accretion rates lead to different birth lines in the @xmath22 diagram . to incorporate a distribution of accretion rates , we have included an approximate procedure in which we dither the intercept of the following birth line @xmath23 where the dithering parameter @xmath24 varies from 0 to 2.8 . the birth line can be reformulated by the expression @xmath25 the extremes of the dithering parameter @xmath24 represent the approximate eddington accretion critical rate and @xmath21 times that rate ( see figure 4 in @xcite ) . while we explored a gaussian distribution of the dithering parameter , we obtain better agreement with a ramp distribution that increases by a factor of 4 between 0 and 2.8 . with the magnetic field randomly selected from the distribution of equation 2 , we then use equation 6 to obtain the initial period of the simulated msp . however , we impose a minimum initial period of @xmath26 = 1.3 ms in accord with the rxte studies of lmxbs by @xcite . having randomly determined the pulsar s age assuming a uniform birth rate , we spin the pulsar down to obtain the present period and period derivative assuming constant magnetic surface field . a constant magnetic field used in this study of msps is in contrast to the assumption of magnetic field decay in our simulations of nps @xcite . in our previous studies of nps @xcite , we adopted the intrinsic radio luminosity model given by the general form @xmath27 from @xcite ( acc ) with @xmath28 , @xmath29 and @xmath30 , but reduced the luminosity @xmath31 by a factor of 60 @xcite and by 73 @xcite to obtain adequate agreement between the simulated birth rate , flux , and distance distributions and those detected , particularly by the parkes multibeam pulsar survey ( pmbps ) ( see @xcite ) . in acc , the predicted birth rate was 0.13 neutron stars per century . in order to obtain a birth rate near 2 neutron stars per century , we had to decrease the luminosity constant @xmath31 . the assumed radio luminosity is directly related to the simulated neutron star birth rate . various studies propose neutron star birth rates that span somewhat of a range . for example , the a range of birth rates from 0.9 to 1.9 per century was obtained from the analysis of pmbps by @xcite . in a recent population synthesis , @xcite estimate a larger birth rate of about 2.8 per century , whereas @xcite find @xmath32 per century from their analysis . more recently @xcite determined the reduction factor of the radio luminosity by normalizing the simulated neutron star birth rate to rate of type ii supernovae of 2.1 per century , following the work of @xcite , using only the parkes multibeam survey , as we may have the best description of the flux threshold @xmath5 for that survey ( crawford , private communication ) . while the neutron star birth rate of 2.1 per century provides a constraint on the luminosity of nps , msps appear in a very different region in the @xmath22 diagram , requiring careful consideration of the period and period derivative dependence of the radio luminosity used by the population synthesis study . often , radio astronomers refer to the radio luminosity in terms of a pseudoluminosity " ( @xmath33 ) determined from the radio flux @xmath34 observed at a specific frequency and the pulsar distance @xmath11 . one can not generally determine the intrinsic radio luminosity of a given pulsar from the @xmath33 , even at a given frequency , as the detected average flux is strongly dependent on the viewing geometry . however , if the number of detected pulsars in a group is large , one might be justified in assuming that the emission region of the group of pulsars is completely sampled in a random fashion . therefore , the detected @xmath33 would be proportional to the intrinsic luminosity of the group of pulsars if the surveys represented an unbiased sample of the true flux distribution . however , since radio surveys necessarily sample the high end of the flux distribution , inferring the intrinsic luminosities of pulsars is difficult at best . in addition , selection effects of the radio surveys may be different for nps and msps , and these effects do influence the sampling of the emission region . for example , the width of the pulse profile is used in the determination of the flux threshold @xmath35 for each radio survey ; therefore , one would expect that narrow pulse profiles or larger inclination angles are preferably detected . regardless of these effects , such comparisons are important in characterizing differences between nps and msps . @xcite compared the average @xmath36 of a group of msps with that of a group of nps and found that the mean @xmath37 values for their groups of 31 msps and of 369 nps of @xmath38 and @xmath39 , respectively . the authors conclude that msps and nps have very different radio luminosity distributions . since this study was performed , the ne2001 distance model @xcite was developed . we use the pulsar distance given by the ne2001 model unless the distance has been determined by other more accurate methods . after this distance correction and using the 1400 mhz fluxes in the atnf catalogue that have a recorded uncertainty @xcite , we find average @xmath37 values of @xmath40 for our select group of 39 msps ( 24 with s1400 fluxes ) and @xmath41 for a group of 1102 nps ( 869 with s1400 fluxes ( references in atnf catalogue ) ) detected by parkes multibeam and the swinburne intermediate latitude surveys . the uncertainties are just the statistical error in the mean and do not take into account the error in the fluxes and distances . subtle selection effects may be introduced within the context of a population statistics study with a group of surveys that have different sensitivities . as a result , it may be inappropriate to compare these mean @xmath37 values with those from @xcite because of the different samples of pulsars . the msps recently discovered by the parkes multibeam and swinburne surveys are not only more distant but also weaker with effectively smaller @xmath37 . we have chosen to reproduce the @xmath37 values of the normal pulsars and msps detected by these surveys to constrain our luminosity model . we adjust our intrinsic radio luminosity model such that the simulated mean @xmath37 values approximately match both the mean values of @xmath40 for msps and @xmath41 for nps detected by both the parkes multibeam and the swinburne intermediate latitude surveys . we find that the exponents of the period ( @xmath42 ) and period derivative ( @xmath43 ) dependence of the luminosity are not uniquely constrained to specific values but are correlated with each other . we find better agreement between the various simulated distributions for nps with values of @xmath44 and @xmath45 . the coefficient of the luminosity is adjusted to give a birth rate of 2.1 neutron stars per century using the parkes multibeam and the swinburne intermediate latitude surveys . we were able to find a common radio luminosity that provides adequate results for both nps and msps , given by the expression @xmath46 in this study , we assume that the radio emission originates from a single core beam and a single cone beam axially aligned with the magnetic field from each pole of the star and isotropically distributed in the sky . as indicated by @xcite , the profiles of msps display similar structure to those of nps , suggesting that the core or central beam and the conal beam make similar contributions to np and msp profiles . however , as discussed later , due to the rapid rotation of msps , relativistic effects are more evident in the pulse profiles of msps than in those of nps @xcite . we use the same description for the core beam as in our past study of @xcite , but we replace the geometry of the cone beam from the @xcite with the description of @xcite having the form @xmath47 which describes the characteristic width of the conal beam at 0.1% of the peak intensity of the profile assuming that the edges of the pulse are along the last open field line . they find that the altitude of the emission @xmath48 in stellar radii is given by @xmath49 where @xmath50 is in units of @xmath51 . in this formulation , the overall period dependence of the characteristic width of the conal beam is @xmath52 , reproducing well the trend observed for nps and msps @xcite . in the formulation of a gaussian conal beam , the characteristic width @xmath53 requires the specification of the radius @xmath54 and width @xmath55 of the annulus of the beam . the characteristic width of @xcite is for 0.1% of the peak intensity making @xmath56 . within such a model , one is free to specify some form for either @xmath57 or @xmath58 . we adopted @xmath59 in our study of both normal and msps . we find that the population statistics are not very sensitive to the choice of this value , which we have arrived at from our study of three peak pulse profiles of radio pulsars . in our study of the characteristics of core and cone beams of about 20 radio pulsars whose profiles have three peaks , we found ( @xcite and in preparation ) that the ratio of core - to - cone peak fluxes does not follow the @xmath60 dependence suggested in the work of acc . in the acc model , the profiles of msps are dominated by the central core beam , so that even at small impact angles the conal beam would not be seen . as seen in the study of @xcite , similar complexity is manifested in the msps profiles as in profiles nps , indicating that the conal beam is quite prominent . recent polarimetry studies of young radio pulsars also call the acc core - to - cone ratio into question . @xcite and @xcite measured the radio polarization of nine pulsars , six of which had characteristic ages less than 100 kyr . they found that the profiles of all the young pulsars indicate a significant degree of linear polarization and a low degree of circular polarization , which is a traditional feature of conal emission . in this study , the two older pulsars with characteristic ages greater than 1 myr do not show any degree of polarization . they conclude , as did @xcite , that the profiles of young short period pulsars tend to be characterized by partial cones . similarly , in a more recent study @xcite measured polarization profiles of 14 young pulsars with ages @xmath61 75 kyr . they found that the profiles are dominated by linear polarization , suggesting that the core beam is weakly manifested , if at all . because of these inconsistencies , we have reassessed the relationship between the core and cone beams ( @xcite ) , using three - peak normal pulsars and a few three - peak msps . in our three - peak pulsar study , we find for the ratio of peak fluxes a power - law dependence with a break in period at about 0.7 s having the approximate form @xmath62 hence in this model , short period pulsars are much less core dominated than in the acc model . the model does not distinguish between young short - period pulsars and msps , although there is evidence ( @xcite ) that the ratio of core - to - cone peak fluxes may be different for a young pulsar and a msp with the same period . the shapes of msp profiles are complicated by relativistic effects , but due to the wide emission beams , contributions from both poles are often seen in the profiles , making the analysis of the viewing geometry and conal widths difficult . with this model , we do accommodate more complex pulse profiles for short period pulsars than in the acc model . in describing the core and cone angle - integrated spectra , we assume indices of -2.36 and -1.72 , respectively . while these indices are not measurable , they are related in a complex fashion , due to selection effects , to the measured spectral indices . using these values , our simulation gives a flux - averaged spectral index of @xmath63 for msps . within our select group of surveys , we find 20 msps within the atnf catalogue whose spectral indices have recorded uncertainties @xcite with a mean of @xmath64 . @xcite found a mean spectral index of @xmath65 for a group of 32 msps . they also point out that msps have significantly steeper mean spectral indices than nps . for a group of 346 nps , they find a mean spectral index of @xmath66 . in our simulation , we found it necessary to have softer angle - integrated spectral indices for the core and cone beams of nps , resulting in a mean spectral index of @xmath67 . this value agrees well with the mean spectral index of @xmath68 for the 245 nps with indices that have recorded uncertainties in the atnf catalogue ( references therein ) . adopting this radio luminosity and beam geometry model , we attempt to describe radio emission properties from both nps and msps @xcite . the success of this radio model is remarkable in describing the physical luminosities of both nps and msps , given the range of three decades in period and seven decades in period derivative between nps and msps . however , it is clear that there are many complexities that may not modeled correctly and that the population synthesis study may not be very sensitive to various characteristics of the beam geometry . because of their rapid rotation , msps can display very different features in their pulse profiles than those observed in the profiles of nps . these features can be partly understood in terms of the relativistic effects of aberration , time delay and the sweep back of the retarded dipole field @xcite . the shorter the pulsar period , the more dramatically these effects are manifested in the profiles . according to equation 10 , the conal emission of pulsars with short periods occurs at high altitude relative to the light cylinder radius . for msps the altitude of emission for the cone beams ( @xmath48 ) is @xmath69 . the combined effects of time delay and aberration result in a shift in phase of the conal beam . deformations in the shape of some profiles of msps are a result of the distortion of the open field volume of the retarded dipole @xcite . while relativistic effects are important in understanding the shape of the profile and in studying correlation with high - energy profiles , in a recent study @xcite , we find that the phase - averaged flux is very similar for relativistic and nonrelativistic emission . since in population statistics studies , the average flux of the profile is the determining parameter that is compared to the threshold of each radio survey , we have neglected these effects in this study . because of the lack of a theoretical understanding of the mechanism that produces the radio emission , the radio emission model is phenomenological by nature . on the other hand , the @xmath0-ray emission models are motivated by theory . two competing @xmath0-ray emission models , the polar cap and outer gap models , assume different locations in the magnetosphere where the acceleration of charge occurs . in the outer gap model , the acceleration of charge takes place in vacuum gaps formed along the last open field line , between the null surface and the light cylinder . while we are beginning to incorporate an outer gap in our simulations @xcite , treating it on the same footing as the polar cap model using the same group of evolved neutron stars and selection criteria , we use only the polar cap model in this study of msps . recent outer gap models for msps @xcite actually assume that the high - energy emission originates not from the outer gap but near the neutron star , from particles flowing down from the gaps , and thus would have a very different high - energy emission geometry from that of normal pulsars . two regions in the @xmath22 diagram , separated by the curvature radiation ( cr ) pair death line , differentiate the @xmath0-ray luminosities and beam geometries within the polar cap model . pulsars above the cr line can produce cr photons that are energetic enough to pair produce in the strong magnetic field above the neutron star surface . the cr - initiated pair cascades have sufficient multiplicity to fully screen the parallel electric field everywhere except in the slot gap along a narrow region near the last open field line where the primary particles are accelerated along unscreened electric field lines . recent studies @xcite of the emission above the cr pair death line in the polar cap model have provided a framework for understanding @xmath0-ray emission from the slot gap . the electrons in the low altitude slot gap initiate pair cascade emission , which was incorporated in our simulations of @xcite . in addition , @xcite find that the primary particles along the last open field lines can be accelerated to high altitudes as the potential in the slot gap remains unscreened and the emission beam forms a characteristic caustic component @xcite . in recent population studies of nps , we have begun to include both the low and high altitude emission geometries @xcite and both are included in this study for those simulated msps above the cr line . however , as one can see in figure 2 , all the msps with the exception of one lie below the cr line where curvature radiation no longer produces pairs . these pulsars can produce much weaker pair cascades through inverse compton scattering radiation . when pulsars are no longer able to produce abundant pair cascades , they do not form slot gaps above the polar cap @xcite as the parallel electric field is not screened on all open magnetic field lines . as the @xmath0-ray energies are below the pair threshold , the curvature radiation from the primary electrons escapes out to infinity along all open field lines . in this pair - starved polar cap model , the primary particles continue to accelerate and radiate to high altitudes above the polar cap out to the light cylinder along all the open field lines . however , cr emitted by the particles can be inhibited by resonant cyclotron absorption of radio emission @xcite where synchrotron radiation is more efficient particle energy loss mechanism than cr . since synchrotron photons usually have lower energies than curvature photons , we have not included this mechanism in the pair - starved model in this study . the cr emission model below the cr pair death line used in this study is developed from the work by @xcite . the inclination angle @xmath42 and the viewing angle @xmath70 determine which open field lines are sampled by the line of sight . as illustrated in figure 3 , the line of sight intersects tangentially a particular open field line with a radius of curvature @xmath71 at a radial distance r and polar angle @xmath72 . the open field lines are defined within the polar cap angle given by the expression @xmath73 where p is the pulsar period in seconds , @xmath74 is the neutron star radius ( @xmath75 cm ) and c is the speed of light . we partition the open field lines through a dimensionless parameter , @xmath76 , that varies between 0 and 1 and is defined as @xmath77 where @xmath78 is the polar angle of the open field line at the intersection with the surface of the star . the particle emits curvature photons along the line of sight at an angle @xmath79 given by @xmath80 where @xmath81 is the phase angle , @xmath42 is the magnetic inclination and @xmath70 is the viewing angle . we approximate the accelerating electric field from equation 14 of @xcite using the expression @xmath82(1 - \xi ^2 ) { \rm{\ statvolt } } \cdot { \rm{cm}}^{{\rm { - 1 } } } , \ ] ] where @xmath83 = 0.15 is the general relativistic inertial frame dragging factor , @xmath84 is the surface magnetic field in units of @xmath85 g , @xmath86 = r / r is the dimensionless radius and @xmath87 is the azimuth angle around the magnetic pole given by the expression @xmath88 the gain in energy of the accelerating primary electron is compensated by the cr losses @xcite , so that the electron lorentz factor @xmath0 becomes radiation - reaction limited to @xmath89 where @xmath71 is the classical radius of curvature of the field line given by @xmath90 at low altitudes near the stellar surface , @xmath91 , the near surface electric field is approximated from equation 21 in @xcite by the expression @xmath92\ ] ] and goes to zero at the stellar surface . the cr spectrum therefore has a spectral index of -2/3 with a high - energy cutoff @xmath93 given by @xmath94 in @xmath95 units where @xmath96 is the electron charge and @xmath97 is the electron compton wavelength . the instantaneous curvature radiation power spectrum has the form @xcite @xmath98 where @xmath99 is the photon energy , @xmath0 is the particle lorentz factor , @xmath100 is the fine structure constant , and the @xmath101 function is defined as @xmath102 we find that the low energy asymptotic form provides a sufficiently accurate description of the power spectrum above @xmath103=100 mev and the instantaneous curvature photon radiation spectrum per primary is given by @xmath104 which is similar to equation 7 in @xcite . this expression can be integrated from @xmath103=100 mev to the high - energy cut off @xmath93 to give the expression @xmath105 which provides the number of curvature photons emitted along the line of sight @xmath106 from an open field line @xmath76 at a radial distance @xmath107 per primary particle , where @xmath107 is given by the expression @xmath108 the goldreich - julian current of primary particles from the polar cap is uniformly distributed over the polar cap and is given by the expression @xmath109 the number of primary particles in a particular patch on the stellar surface at r , @xmath78 , and @xmath87 , which gives the number of primaries along a particular field line @xmath76 , is provided by the expression @xmath110 the total photon luminosity from one pole can be obtained by integrating over the open field volume to give @xmath111 with @xmath112 the integral over @xmath107 can be written as an integral over @xmath72 to give @xmath113 where the integral over @xmath76 ( the @xmath114 has been included in the integrand ) is over the open field lines and @xmath115 is the angle at the footpoint of the field line on the stellar surface . the expression in parentheses represents the photon emission from a particular field line at @xmath76 and @xmath86 , integrated above @xmath103=100 mev , along the line of sight into a solid angle @xmath116 and is given by the expression @xmath117 the total emission from the polar cap for a particular phase bin @xmath81 is obtained by integrating over parameter @xmath76 to include the contributions of all the open field lines along a line of sight defined by the viewing angle @xmath70 . the total emission in the phase bin is given by @xmath118 having calculated the @xmath0-ray pulse profile for each msp for a given viewing geometry , we average the profile to obtain the average photon flux , and compare it to the instrument threshold . our resulting @xmath0-ray efficiency ( 7% ) for the total spectrum and profiles for the nearby msp psr are in agreement with those obtained by @xcite . from the simulated @xmath0-ray pulse profile , we obtain an average flux that we compare to all sky threshold maps for egret , agile and glast large area telescope ( lat ) . we use the recently revised egret map that includes the dark clouds @xcite , which has led to a radical reassessment of the egret unidentified sources . the glast threshold has been improved and updated ( grenier & casandjian private communication ) as a 1 year glast lat threshold map . the all sky map for agile ( pellizzoni private communication ) has not been recently updated in our computer code . the detection of radio and @xmath0-ray point sources within the code are independent of each other , allowing the tagging of radio - quiet ( below the survey flux thresholds ) and radio - loud @xmath0-ray msps . to improve the simulated statistics , we run the simulation to obtain ten times the number of detected msps and then normalize the distributions accordingly . in figure 4 , we present aitoff projections for the detected ( a ) and simulated ( b ) msps . since msps are closer to us and are much older than nps , the graphs indicate larger out - of - plane distributions than those of nps . of the ten radio surveys that we have included in our simulation , the surveys most sensitive to msp detection are parkes 2 , parkes multibeam and swinburne intermediate latitude which together account for majority of the 56 msps , as seen in table 1 . within the limited statistics , there is good agreement between the number of msps detected and simulated among the ten surveys used in our selected group . the few detections in radio surveys that are not very sensitive to the detection of msps are also well reproduced by the simulation . the distributions in the @xmath22 diagram are compared in figure 5 for detected ( a ) and simulated ( b ) msps . the dotted broken lines represent the pair death lines for curvature radiation ( cr ) and for nonresonant inverse compton scattering ( nrics ) . below the cr death line , cr no longer is able to produce pairs . however , a limited number of pairs can still be produced via inverse compton scattering above the nrics death line . below the nrics death line , pair production is no longer possible , presumably inhibiting radio emission mechanisms . the upper and lower msp birth lines indicated by the solid lines represent the approximate eddington critical accretion rate ( upper line ) and @xmath119 times that rate ( lower line ) discussed earlier . with the @xmath120 magnetic field distribution and the uniform distribution of birth lines between the upper and lower msp birth lines , we reproduce the observed broadness of the pulsar distribution in the @xmath22 diagram . in figure 6 , we present histograms of the indicated measured and derived pulsar characteristics for the detected ( shaded histograms ) , and the simulated ( unshaded histograms ) msps . although the poor statistics make it difficult to judge the agreement , the simulation appears to reproduce the observed distributions fairly well . the clump of detected msps observed in figure 5a is exhibited as narrow peaks in the period and period derivative distributions . the narrow peak in the period derivative distribution is not present in the simulated distribution . in the comparisons of the age distributions , we calculated the traditional characteristic age @xmath121 for the simulated pulsars . in the comparisons of the distances and dispersion measures of the msps , we have used the new distance model of @xcite . the distance of the detected pulsar is established typically by its measured dispersion measure and location in the sky . the new distance model is used to determine the simulated dispersion measure from the simulated distance and location in the sky for each msps , which is then used in the calculation of the flux threshold @xmath5 for each radio survey . both the simulated dispersion measure and distance distributions indicate good agreement with the corresponding distributions of detected msps with a few more simulated closer pulsars . the simulated radial and height distributions within the galaxy are consistent with those distributions of the detected pulsars . however , the simulated spectral index distribution is narrower than that of the detected pulsars . for the pulsar characteristics in figure 6 , we have performed a kolmogorov - smirnov ( ks ) statistics test of the histograms assuming a significance level ( @xmath42 ) of 5% . we indicate in figure 6 the resulting p - values for each of the comparisons . sometimes the d statistic and the critical value are quoted in the literature and compared to each other to decide on the null hypothesis . the critical value is derived from @xmath42 and the number of samples . equivalent comparisons can be made between the p - value and @xmath42 . if the p - value statistic is larger than @xmath42 , the null hypothesis is not excluded at the level of significance determined by @xmath42 . our results indicate that in all comparisons , the distributions of simulated pulsars are consistent with those detected with the exception of the distributions of the galactic height z and s400 whose p - value is less that 0.005 . in the case of the s400 distributions , the statistics are insufficient in the number of measured flux densities at 400 mhz for an adequate comparison . we simulate slightly more pulsars with larger s1400 fluxes . we applied the two dimensional ks test to the unbinned period and period derivative distributions using the prescription of @xcite and obtained a p - value of only of 0.03 at the same significance level of 5% indicating that the detected and simulated @xmath122 distributions are not entirely consistent with each other . while the ks tests might be helpful in such comparisons , the numerous assumptions within the simulation , such as the radio luminosity , the pulsar distance , and birth distributions , contain systematic uncertainties . therefore , we do not want to overemphasize the importance of these statistical tests . the detection of a radio pulsar in the computer code is independent of the detection of a @xmath0-ray pulsar , allowing the identification of radio - quiet and radio - loud @xmath0-ray pulsars . by radio - quiet , we mean that the radio flux of the simulated pulsars is below the flux threshold of the surveys that could potentially observe it ; otherwise , the pulsar is flagged as radio - loud . in figure 7 , we show the simulated radio - quiet and radio - loud @xmath0-ray msps for the instruments egret ( upper right ) , agile ( lower left ) and glast lat ( lower right ) . for comparison , we also show the @xmath0-ray pulsars that were detected by egret ( upper left ) , including the msp . in the case of geminga ( cross in upper right ) , it is not clear whether its radio silence results from a misalignment of the @xmath0-ray and radio beams or from its intrinsic radio weakness . while it is unclear how many of the egret unidentified sources might be msps detected as point sources , the simulation predicts that a few may be radio - quiet @xmath0-ray msps . all the simulated @xmath0-ray pulsars that are detected by egret , agile and glast lie below the curvature radiation pair death line in the @xmath22 diagram ( figure 7 ) . as discussed above , in this pair - starved polar cap model , curvature radiation escapes along all open field lines . however , as indicated by @xcite , curvature radiation can be limited when the accelerated particles undergo resonant cyclotron absorption of radio photons under special circumstances . the excited particles emit synchrotron radiation typically at lower energies than curvature radiation . while we are in the process of developing a full three - dimensional study of these processes , in the present calculation we have not limited the curvature radiation in altitude . therefore , our results may be somewhat optimistic . in figure 7 , the @xmath0-ray msps cluster in the lower left of the @xmath22 diagram because the @xmath0-ray luminosity increases with decreasing @xmath9 . in table 2 , we present the @xmath0-ray msps statistics for radio - loud and radio - quiet pulsars . due to the uncertainties in the radio luminosity , we vary the coefficient of the radio luminosity in equation 8 by @xmath123 to obtain a range in radio - quiet @xmath0-ray pulsars . since the simulation is normalized to the detected number of radio pulsars , there is no variation in the number of radio - loud @xmath0-ray pulsars . of the 33 - 40 radio - quiet @xmath0-ray msps predicted to be detected by glast , about 6 have fluxes above the @xmath124 photons/@xmath125/s level ( grenier , private communication ) , that glast may be able to detect through blind period searches . however , since 70% of the msps in the atnf catalogue are in binaries , many of these radio - quiet @xmath0-ray pulsars are expected to be in binary systems . it will be difficult for glast to detect pulsed emission from these msps in blind period searches due to timing uncertainties associated with the orbital motion . without radio ephemerides , glast will detect binary msps only as point sources . in order to explore the nature of the radio - quietness of the glast pulsars , we present in figure 8 histograms of the radio flux at 1400 mhz of the detected ( solid dark gray histograms ) and simulated ( open thin histograms ) radio pulsars in ( a ) and the radio flux of glast radio - quiet ( solid light gray histograms ) and radio - loud ( open thick histograms ) in ( b ) . as expected , glast radio - loud pulsars have large radio fluxes . however , the glast radio - quiet pulsars form two groups , one with a flux distribution very similar to the radio detected and simulated pulsars in figure 8a and the other with a much lower flux distribution that would be difficult to detect without deep radio pointings . the parkes multibeam and swinburne intermediate latitude surveys are the two most sensitive surveys in our select group of radio surveys . however , they cover a rather narrow strip along the galactic disk , and msps seem to be rather scattered in space and are not as correlated with the galactic disk as are nps . as a result , many of the glast detected pulsars are radio - quiet because they happen to be outside the sensitive survey regions or have radio fluxes that are slightly lower than the survey thresholds . in figure 9 , we plot the impact angle @xmath126 as a function of the radio flux at 1400 mhz for the simulated group of glast radio - quiet pulsars . we can see that the two groups of fluxes correspond to two groups of impact angles . the msps in the low flux group with large impact angles will not be observed as radio sources without deep radio observations . however , the simulation suggests that they can be seen as @xmath0-ray point sources . in the pair - starved model of @xmath0-ray emission , curvature radiation occurs along all open field lines all the way out to the light cylinder . while the intensity of curvature radiation decreases at higher altitudes for nearby pulsars , we predict that glast should be able to detect them as point sources , yet their radio emission will be very difficult to detect as the radio beam and the @xmath0-ray beam are separated by large angles . however , the higher flux group with smaller impact angles would be detected by glast as point sources and potentially are detectable with pointed radio observations . as a result , our simulations suggest that it will be important to follow up glast detections with radio observations in order to discover a greater number of radio - loud @xmath0-ray pulsars . the predicted energy spectra of msps are very different than those of either nps or agns . in this pair - starved polar cap model , the energy cutoffs of the msp @xmath0-ray spectra are predicted to be in the 10 - 50 gev energy region @xcite suitable for glast . clearly the number of detected radio - quiet and radio - loud @xmath0-ray msps depends strongly on the detection thresholds of radio surveys and @xmath0-ray instruments . in figure 10 , we present the log(n)-log(s ) for the radio flux at 400 mhz ( a ) and 1400 mhz ( b ) and for egret ( c ) and glast ( d ) . the light shaded histograms represent the simulated undetected source counts . the dark histograms represent the simulated detected source counts by the select group of ten radio surveys at 400 mhz ( a ) and 1400 mhz ( b ) . for egret ( c ) and glast ( d ) , the dark and medium dark histograms represent the simulated source counts for radio - loud and radio - quiet @xmath0-ray msps , respectively . the appearance of fractions of a pulsar in the figure is a result of the re - normalizing the distributions because the simulation was run for ten times the number of detected msps to improve the simulated statistics . assuming a uniform distribution of sources in space with isotropic emission of equal flux , one expects to a slope in the log(n)-log(s ) diagram of -3/2 . indeed the tails of these distributions have slopes very similar to -3/2 . while for egret and glast the log(n)s of simulated detections have a sharp decrease with decreasing flux , the log(n)s of simulated radio detections have a more gentle roll over , which may reflect the larger number of factors that determine the minimum flux threshold for the radio surveys . both the number of undetected radio and @xmath0-ray sources follow a simple power law far below the sensitivity of the radio surveys and @xmath0-ray instruments . we have presented a monte - carlo simulation of the galactic - plane population of msps and corresponding predictions for the numbers of both radio - quiet and radio - loud msps that are detectable as @xmath0-ray sources . in order to accomplish this , we made significant improvements in our population statistics monte carlo code by adding the more realistic description of the galactic potential of @xcite with a more accurate 5th order cash - karp runge - kutta trajectory integration routine @xcite , using improved all - sky threshold maps for egret @xcite and glast ( grenier & casandjian , private communication ) , and including the swinburne intermediate latitude radio survey @xcite . we have modified the intrinsic radio luminosity model of acc to describe the total radio luminosity of both normal and msps . from our study of radio pulsars with three - peak profiles @xcite , we find that the ratio of the core - to - cone peak fluxes for short period pulsars is much smaller for short period pulsars than the ratio predicted by acc model . although msps have on average larger pulse widths than nps , this resulting radio beam geometry model provides a reasonable agreement to the average pseudoluminosities and pulse widths at 1400 mhz and pulse widths of both nps and msps . we include a @xmath0-ray beam geometry and luminosity model of the curvature radiation from all open field lines within a pair - starved polar cap model @xcite that accounts for @xmath0-ray emission below the curvature radiation pair death line , which is applicable to msps . these improvements in our computer code have allowed us to study the population statistics of radio and @xmath0-ray msps to complement our previous studies of nps from the galactic disk . as a first order study , we believe that we have improved the simulation by including a beam geometry for msps and an intrinsic radio luminosity whose functional form allows for the description of the population statistics of normal pulsars . the actual beam geometry of msps may turn out to be significantly different and require a different model than the one proposed here from the work of @xcite . the population statistics study is not sensitive enough to discriminate between beam geometry models as the average flux of the profile is the quantity that is compared to the @xmath5s of the radio surveys . a detailed study including relativistic effects will be necessary along with high quality polarization data may enable establishment of systematic trends that allows for the development of an adequate beam model for msps . in this study , we do not attempt to describe the numerous msps within globular clusters . limiting our study to the msps born in the galactic disk , we treat them as point particles , evolving their location within the galactic potential and their spin - down , after spin - up by mass accretion from a binary companion star has ended . this group of old , short period pulsars with low magnetic fields are given radio and @xmath0-ray beam and luminosity characteristics and filtered through a set of ten radio surveys and the @xmath0-ray instruments egret , agile and glast . our simulation is run until the number of simulated radio pulsars is equal to the number detected by the same group of radio surveys . the radio luminosity is adjusted in our study of nps @xcite to give a birth rate of type ii supernovae of 2.1 per century to agree with the study of @xcite and to reproduce the detected mean pseudo - luminosities and spectral indices of msps and nps , after which there are no further adjustments performed in the simulation of msps . allowing for a @xmath127 10% variation in the radio luminosity , we predict a birth rate of @xmath128 msps per century , which is in good agreement with the studies of @xcite ( @xmath129 per century ) and @xcite ( @xmath130 per century ) . while the population statistics studies of lmxbs contain numerous uncertainties , our simulated birth rate agrees with the estimate of the birth rate of lmxbs from the preferred model ( f ) of @xcite ( @xmath131 per century ) . we are also in agreement with @xcite as our description of the spectral properties of the core and cone beams for msps requires steeper angle - integrated spectral indices than those for the simulated nps . in outer gap accelerator models @xcite , msps are not expected to produce @xmath0-ray emission above 1 - 2 gev ( @xcite , hirotani private communication 2005 ) due to the smaller electric field and larger radius of curvature in the gap . thus , detected emission from msps above 5 gev would imply an origin near the polar cap . @xcite propose that particles accelerated in the outer gaps could be the origin of msp radiation at super - gev energies , but only if the cascade emission from the particles as they travel down toward the neutron star surface is visible . however , it seems unlikely that both the upward - going radio emission and downward - going @xmath0-ray emission could be visible to the same observer . in that event , one would expect the radio and high - energy peaks to be out of phase , and x - ray and radio peaks are often in phase for msps ( e.g. , , ) . in any case , glast observations of msps will be an excellent model discriminator . our simulations predict the number of radio - loud and radio - quiet @xmath0-ray pulsars observed by the instruments egret , agile and glast . egret was only able to detect one msp , psr , during its nine - year mission . egret did not have the capability of performing blind period searches of pulsars and required reliable radio ephemerides to detect @xmath0-ray pulsations . however , these observations were prior to the parkes multibeam pulsar survey and the swinburne intermediate latitude survey that tripled the number of detected radio msps in galaxy ( not counting those in globular clusters ) . we predict that egret should have detected as point sources eight radio - loud and @xmath132 radio - quiet @xmath0-ray msps , some of which could be associated with egret unidentified sources . however , the recently revised all - sky egret threshold map with dark clouds @xcite , used in our simulations , has significantly affected the catalogue of egret unidentified sources . as the all - sky threshold map for agile has not been recently updated , our predictions for agile might be somewhat optimistic . however , the trend in the statistics for egret , agile and glast suggests that the number of radio - quiet and radio - loud @xmath0-ray msps predicted to be detected by agile might be reasonable . the shklovskii effect is a significant , positive contribution to the intinsic @xmath7 of the pulsar resulting in much larger measured @xmath7s and needs to be taken into account in population statistics studies . the effect of having to decrease the intrinsic @xmath7 to accomodate this effect results is pulsars with lower @xmath0-ray luminosities . we predict @xmath133 of radio - quiet @xmath0-ray pulsars for glast with the ratio of radio - quiet to radio - loud increasing from 1.5 for egret to 3.2 for glast . curvature radiation of @xmath0-rays below the curvature pair death line occurs along all open field lines all the way out to the light cylinder , as the weak production of pairs through nonresonant inverse compton scattering is insufficient to screen the electric field . the increased sensitivity of glast results in a large number of detections at higher altitudes where the @xmath0-ray beam is at large angles relative to the radio beam , resulting in radio fluxes below the flux thresholds of our select group of radio surveys . however , in many cases , the radio fluxes are relatively large but are outside of the sky region of the parkes multibeam and swinburne intermediate latitude surveys . msps are much less confined to the plane of the galactic disk than are nps and therefore often appear at high latitudes , far above the narrow belt covered by these more sensitive surveys . our simulations suggest that glast will detect many msps as point sources , and only a handful as radio - loud pulsars . the point sources can be identified as pulsar candidates by their spectra . @xcite indicate that the curvature radiation energy spectra of msps are very hard , with cutoffs in the region of 10 to 50 gev , well within the glast range . such detected candidates can easily be followed up with pointed radio observations to obtain the periods of the pulsars . thus , glast with radio follow - up observations offers the potential to discover many new msps in the galaxy . we are very grateful to isabelle grenier and jean - marc casandjian for providing a glast - lat 1 year point source threshold map . we express our appreciation to the anonymous referee for the insightful comments , which led to a much - improved manuscript . we are indebted to dunc lorimer who pointed out our neglect of the shklovskii effect in an earlier version of the manuscript . we express our gratitude for the generous support of the national science foundation ( reu and ast-0307365 ) , the michigan space grant consortium and the nasa astrophysics theory program . alpar , m.a . , cheng , a.f . , ruderman , m.a . , & shaham , j. 1982 , nature , 300 , 728 arzoumanian , z. , chernoff , d.f . , & cordes , j.m . 2002 , , 568 , 289 bailes et al . 1997 , , 481 , 386 bhattacharya , d. & van den heuvel , e.p.j . 1991 , phys . , 203 , 1 bulik , t. , rudak , b. & dyks , j. 2000 , , 317 , 97 crawford , f. manchester , r.n . & kaspi , v.m . , 2001 , , 122 , 2001 crawford , f. & keim , nc . 2003 , , 590 , 1020 dehnen , w. & binney , j. 1998 , , 294 , 429 dyks , j. , & harding a.k . , 2004 , , 614 , 869 dyks , j. , harding , a.k . , & rudak , b. 2004 , , 606 , 1125 dyks , j. & rudak , b. 2003 , , 598 , 1201 edwards , r. t. et al . 2005 , , 326 , 358 faucher - gigure , c. & kaspi , v. m. 2006 , , 643 , 332 faulkner et al . 2005 , , 618 , l119 faulkner et al . 2004 , , 355 , 147 ferrario , l. & wickramasinghe , d. 2006 , , 375 , 1009 gonthier , p.l . , ouellette , m.s . , berrier , j. , obrien , s. , & harding , a.k . 2002 , , 565 , 482 gonthier , p.l . , van guilder , r. & harding , a.k . 2004 , , 604 , 775 gonthier et al . 2006 , chin . j. astron . astrophys . , 6 , 97 gonthier , p.l . , story , s.a . , clow , b.d . , & harding , a.k . 2007 , , in press , astro - ph/0702097 grimm , h .- j . , gilganov , m. , & sunyaev , r. 2002 , , 391 , 923 harding , a.k . & muslimov , a.g . 1998 , , 508 , 328 harding , a.k . , muslimov , a.g . , & zhang , b. 2002 , , 576 , 366 harding , a.k . & muslimov , a. 2001 , , 556 , 987 harding , a.k . , greiner , i.a . & gonthier , p.l . 2006 , , in press , astro - ph/0703019 harding , a.k . , usov , v.v . , & muslimov , a. 2005 , , 622 , 531 hobbs et al . 2004 , , 352 , 1439 hobbs et al . 2005 , , 360 , 963 jackson , j.d . 1975 , classical electrodynamics ( new york : wiley ) jiang , z.j . & zhang , l. , 2006 , , 643 , 1130 johnston , s. & weisberg , j.m . 2006 , , 368 , 1856 kiel , p. d. & hurley , j. r. 2006 , , 369 , 1152 kijak , j. , & gil , j. 1998 , , 299 , 855 kijak , j. , & gil , j. 2003 , , 397 , 969 kramer , m. et al . 1997 , a&a , 322 , 846 kramer , m. et al . 1998 , , 501 , 270 kramer , m. et al . 1999 , , 526 , 957 kramer , m. 2002 , in neutron stars , pulsars and supernova remnants , proceedings of the 270 . we - heraeus seminar held in bad honned , germany , ed . by w. becker , h. lesch , & j. trmper , 278 , 177 kramer et al . 2003 , , 342 , 1299 krolik , j. 1991 , , 373 , l69 kuiper , l. , hermsen , w. , & stappers , b. 2003 , in pulsars , axps and sgrs observed with bepposax and other observatories , proceedings of the international workshop held in marsala , ed . by g. cusumano , e. massaro , & t. mineo , 31 lamb , f.k . , & yu w. , in proc . asp conference series , _ binary radio pulsars _ , ed . by f.a . rasio & i.h . stairs , 2005 , 328 , 299 lommen et al . 2000 , apj , 545 , 1007 lorimer , d.r . yates , j.a . , lyne , a.g . & gould , d.m . 1995 , mnras , 273 , 411 lorimer , d. 2005 , living rev . rel . 8 , 7 lorimer et al . 2006 , , 372 , 777 lou , q. , shibata , s. & melrose , d. b. 2000 , , 318 , 943 manchester , r. n. 1996 , in iau colloq . 160 , _ pulsars : problems and progress _ , ed . s. johnston , m. bailes , & m. walker ( asp conf . 105 ; san francisco : asp ) , 193 manchester , r.n . 1999 in pulsar timing , general relativity and the internal structure of neutron stars , edited by z. arzoumanian , f. van der hooft , and e. p. j. van den heuvel . koninklijke nederlandse akademie van wetenschappen , amsterdam , the netherlands , isbn 90 - 6984 - 247 - 5 , 53 manchester et al . 2001 , , 328 , 17 manchester , r. n. , hobbs , g. b. , teoh , a. & hobbs , m. 2005 , 129 , 1993 mitra , d. & deshpande , a. a. 1999 , , 346 , 906 morris et al . 2002 , , 335 , 275 muslimov , a.g . & harding , a.k . 2004 , , 606 , 1143 muslimov , a.g . & harding , a.k . 2003 , , 588 , 430 paczyski , b. 1990 , , 348 , 485 press , w.h . 1992 , _ numerical recipes in c the art of scientific computing _ , cambridge university press new york press , w.h . 1992 , _ numerical recipes in c & c++ source code v 2.11 _ , cambridge university press new york shklovskii , i.s . , 1970 , soviet astronomy , 13 , 562 story , s.a . , gonthier , p.l . , clow , b.d . , & harding , a.k . , 2006 , aas , 209 , # 159.07 stairs et al . 2005 , , 632 , 1060 tammann , g. a. , lffler , w. , & schrder , a. 1994 , , 92 , 487 thompson , d.j . in _ cosmic gamma - ray sources cheng & g.e . romero , ( kluwer ) p. 149 ( 2004 ) toscano , m. bailes , m. manchester , r. , & sandhu , j. 1998 , , 506 , 863 toscano , m. et al . 1999 , , 307 , 925 venter , c. & dejager , o. c. 2005 , , 619 , l167 vranesevic et al . 2004 , , 617 , l139 wang , w. , jiang , z. j. , & cheng , k. s. 2005 , , 358 , 263 xilouris , k.m . et al . 1998 , , 501 , 286 zhang , l. & cheng , k. s. 2003 , , 398 , 639 cccc survey & frequency ( mhz ) & detected & simulated + arecibo 3 & 430 & 4 & 8 + arecibo 2 & 430 & 2 & 1 + greenbank 3 & 390 & 0 & 3 + greenbank 2 & 390 & 1 & 0 + molongo 2 & 408 & 0 & 1 + parkes 2 & 436 & 18 & 23 + parkes 1 & 1520 & 0 & 0 + jodrell bank 2 & 1400 & 0 & 0 + parkes mb & 1374 & 28 & 20 + swinburne il & 1374 & 12 & 9 + cccc instrument & radio - loud & radio - quiet & ratio rq / rl + egret detected & 1 & ? & + egret simulated & 4 & 5 - 6 & 1.5 + agile simulated & 7 & 11 - 13 & 1.9 + glast simulated & 12 & 33 - 40 & 3.2 +

we present stellar metallicities in leo i , leo ii , ic 1613 , and phoenix dwarf galaxies derived from medium ( f390 m ) and broad ( f555w , f814w ) band photometry using the wide field camera 3 ( wfc3 ) instrument aboard the hubble space telescope . we measured metallicity distribution functions ( mdfs ) in two ways , 1 ) matching stars to isochrones in color - color diagrams , and 2 ) solving for the best linear combination of synthetic populations to match the observed color - color diagram . the synthetic technique reduces the effect of photometric scatter , and produces mdfs 30 - 50 narrower than the mdfs produced from individually matched stars . we fit the synthetic and individual mdfs to analytical chemical evolution models ( cem ) to quantify the enrichment and the effect of gas flows within the galaxies . additionally , we measure stellar metallicity gradients in leo i and ii . for ic 1613 and phoenix our data do not have the radial extent to confirm a metallicity gradient for either galaxy . we find the mdf of leo i ( dwarf spheroidal ) to be very peaked with a steep metal rich cutoff and an extended metal poor tail , while leo ii ( dwarf spheroidal ) , phoenix ( dwarf transition ) and ic 1613 ( dwarf irregular ) have wider , less peaked mdfs than leo i. a simple cem is not the best fit for any of our galaxies , therefore we also fit the ` best accretion model ' of lynden - bell 1975 . for leo ii , ic 1613 and phoenix we find similar accretion parameters for the cem , even though they all have different effective yields , masses , star formation histories and morphologies . we suggest that the dynamical history of a galaxy is reflected in the mdf , where broad mdfs are seen in galaxies that have chemically evolved in relative isolation and narrowly peaked mdfs are seen in galaxies that have experienced more complicated dynamical interactions concurrent with their chemical evolution .

1504.07264

dwarf galaxies are important constituents of the universe because they are both the most numerous type of galaxy and test beds for examining galactic evolution on small scales . additionally , dwarf galaxies show varying and extended periods of star formation ( sf ) over the age of the universe @xcite . one method of examining these populations is through metallicities , specifically through metallicity distribution functions ( mdf ) . metallicity measurements of large numbers of stars over the full range of metallicities are crucial in studying the populations of dwarf galaxies because they allow us to construct robust mdfs , examine any structure indicating sub - components , and measure gradients across the galaxy . the star formation history ( sfh ) , accretion , outflows via supernova and stellar winds , and galaxy interactions that cause tidal or ram - pressure stripping all play a part in shaping the mdf . the shape of the mdf offers clues to the galaxy s evolution , which can be characterized by fitting them to analytic chemical evolution models . the most basic analytic chemical evolution model is a simple closed box , where gas turns into stars , the stars evolve and return enriched gas to the interstellar medium ( ism ) , which is again formed into stars . in this model no material enters or leaves the system . however , the simple model is often an inaccurate description of the chemical evolution in dwarf galaxies because the simple model overpredicts the number of metal poor stars ( i.e. the g dwarf problem ) . semi - analytic chemical evolution models and hydrodynamic simulations utilize the galactic sfhs of dwarf galaxies , and nucleosynthetic yields from type i and ii supernova ( sn ) , to predict the gas inflow and outflow necessary to match abundance patterns from multiple elements and the overall mdfs . in lieu of these complex models the simple analytic chemical evolution models offer a concise way to quantify and compare the mdfs of different galaxies , providing insight into the enrichment history of galaxies , especially when only modeling the overall mdf of the dwarf galaxy without information on @xmath1 abundances . recent works have examined the stellar metallicities found in dwarf galaxies by using medium and high resolution spectra on 8 - 10 meter telescopes . these works examine variations in @xmath1 elements , abundance patterns in the lowest metallicity stars , abundance patterns in r and s process elements , and the overall mdfs to better understand the processes that are important in the chemical evolution ; processes that include the sfhs , the imf , stellar nucleosynthesis , and the time - scales for the formation of galaxies . one of the largest samples of stellar metallicities ( @xmath2 3000 stars ) comes from @xcite who present spectroscopic mdfs for 15 dsph and 7 dirs . they also fit their mdfs to analytical chemical evolution models , extended the stellar - mass stellar - metallicity relation and determined that the mdf shapes vary depending on the morphology such that dsph tend to have narrower mdfs than dirs . while individual spectra of stars in dwarf galaxies provide abundance and kinematic information , they are still difficult to obtain . spectra of stars in more distant local group ( lg ) objects are limited to small sample sizes by the long exposure times and large telescopes required to make the observations . while spectral targets are limited to the few brightest stars , photometric metallicities probe deeper and thus sample stars that are typically more representative of the metallicity distribution . photometric metallicities , though not as accurate as spectra , provide measurements for every star in the field , allowing us to increase samples sizes by an order of a magnitude . this allows us to probe galaxies farther out in the lg , especially when using the resolving power of the hubble space telescope ( hst ) . we obtained hst images in metallicity sensitive filters ( f390 m ) , and 2 wide band filters ( f555w and f814w ) for four lg galaxies . by measuring every star in the field we can build up larger samples of stellar metallicities in these dwarf galaxies . in this work we present photometric metallicities of individual stars in four lg dwarf galaxies : leo i , leo ii , ic 1613 and phoenix . we chose galaxies that span different morphologies , masses , sfhs and distances from the milky way ( mw ) . section 2 describes our two methods of measuring the photometric mdfs , including a new synthetic color - color diagram method , which is similar to the synthetic color magnitude diagram ( cmd ) method of deriving sfhs . in section 3 we compare our metallicities to literature values , including recent work with overlapping spectroscopic targets @xcite . in section 4 we fit the mdfs with chemical evolution models to quantify the enrichment of the galaxy . in section 5 we present metallicity gradients from the central regions of each galaxy . in section 6 we discuss implications of our results on theories of galaxy evolution , and conclude in section 7 . llcccc[h ] leo i & 2011 , jun 02 @xmath3 mar 25 & 21,024 & 1,760 & 1,500 + leo ii & 2012 , mar 24 @xmath3 25 & 10,080 & 1,760 & 1,500 + & 2013 , mar 30 & 10,080 & & + ic 1613 & 2011 , dec 16 @xmath3 20 & 15,720 & 1,200 & 1,224 + phoenix & 2012 , jan 30 @xmath3 31 & 16,320 & 1,300 & 1,340 [ obsdat ] the wfc3 observations were obtained between march , 2011 and march 2013 as part of hst proposal 12304 @xcite . the dates and exposure times of the observations are given in table [ obsdat ] . we used the reduced images from the stsci pipeline , which performs bias , flat - field , and image distortion corrections . we additionally used a charge transfer efficiency ( cte ) correction module provided by stsci . magnitudes were measured using the point spread function fitting photometry package dolphot , which is a modified version of hstphot @xcite . the photometric catalog was culled to reject objects based on goodness of fit and profile sharpness as recommended by the dolphot manual . we adopt reddening and distance values reported in @xcite to calculate the absolute magnitudes . all magnitudes are reported in the vegamag system . table [ prop ] lists some basic observable quantities for these dwarf galaxies . lllcclclllll ic 1613 & 01 04 47.8 s & + 02 07 04 & 721 @xmath4 5 & -14.6 & 6.81 , 1496 & 100 & 79.5 & ... & 65 & 25 & dir + phoenix & 01 51 06.3 & @xmath544 26 41 & 406 @xmath4 13 & -10.1 & 3.75 , 454 & 0.77 & 3.3 & ... & 0.12 & 10 & dtrans + leo i & 10 08 28.1 s & + 12 18 23 & 254 @xmath4 17 & -11.9 & 3.40 , 251 & 5.5 & 2.2 & 9.2 & n / a & n / a & dsph + leo ii & 11 13 28.8 & + 22 09 06 & 233 @xmath4 15 & -9.8 & 2.60 , 176 & 0.74 & 0.97 & 6.6 & n / a & n / a & dsph the general technique to measure photometric metallicities relates color to metallicity . the stellar properties that control observed colors are metallicity , effective temperature and surface gravity . however , effective temperature and surface gravity are dictated by the mass , metallicity and the current evolutionary stage of a given star . for populations of comparable age the color is directly related to the metallicity . however , an individual giant branch star can be redder either because it has higher metallicity or because it is older , leading to a color degeneracy between age and metallicity . for a mixed aged population , the younger giants are hotter and more massive than older giants of the same metallicity so the color - metallicity relation no longer holds . using specifically designed hst wfc3 filters which separate the effects of metallicity and temperature on color , we can break the age - metallicity degeneracy . the metallicity sensitivity of the ( f390m f555w ) color comes from the f390 m medium band filter which covers the ca h @xmath3 k spectral features , one of the strongest metal absorption features in the visible spectrum . the temperature sensitive color , ( f555w f814w ) , uses two broad band filters which cover mostly continuum . the color - color diagram ( f390m f555w , f555w f814w ) , has been shown to effectively separate the competing color changes due to metallicity and temperature @xcite , breaking the age - metallicity degeneracy . in the color - color diagram the color change between a 12 and 4 gyr isochrone of the same metallicity is on order of a few hundredths of a magnitude , ( demonstrated in figure 5 of ross et al . therefore , these two colors can be used to measure individual stellar metallicities of populations of mixed ages and abundances , like dwarf galaxies . we used only giant branch stars with errors less than 0.05 mag in all filters . these error cuts resulted in samples of 3,449 stars , 444 stars , 896 and 578 stars for leo i , leo ii , ic 1613 and phoenix respectively . stellar metallicities are assigned in the ( f390m f555w , f555w f814w ) color - color diagram where each giant branch star is matched to the closest dartmouth isochrone @xcite in a grid ( spaced by 0.05 dex in [ fe / h ] and assuming solar [ @xmath1/fe ] ) . the isochrones were empirically corrected to align with globular clusters of known metallicities ( m92 , ngc 6752 , ngc 104 , ngc 5927 and ngc 6791 with [ fe / h ] @xmath6 @xmath7 and @xmath8 respectively ) . the empirical corrections were derived in @xcite using the same hst filters used in this study . this method of deriving photometric stellar metallicities from hst photometry was tested against the listed globular clusters and shown to produce metallicities with errors of 0.15 - 0.3 dex , with larger errors occurring at lower metallicities . see @xcite for the full details on the metallicity derivations from colors . although each giant branch star is matched to an isochrone grid of [ fe / h ] assuming a solar abundance ratio , many stars are known to have @xmath1 enhancement , especially low metallicity ones . the colors we measure are a product of [ fe / h ] and the @xmath1 enhancement ; therefore , it is more accurate to think of our measurement as an indicator of the total metallicity , [ m / h ] , that represents the intrinsic combination of [ fe / h ] and [ @xmath1/fe ] , even though we are matching each star to an isochrone grid of [ fe / h ] . total metallicity , [ m / h ] , and the iron abundance , [ fe / h ] , are often used interchangeably , because they are usually very similar ; only in the cases of large abundance variations ( e.g @xmath1 enhancements ) do the two values differ . in @xcite we quantified the amount of color change expected from @xmath1 enhancements and provide a means of calculating the [ fe / h ] if the @xmath1 enhancement is known . in addition to measuring metallicities of individual stars based on color , we have also developed a synthetic color - color diagram technique to derive a mdf from photometry . the technique uses isochrones and an initial mass function ( imf ) to make a model color - color diagram of the synthetic populations in order to reproduce the observed color - color diagram . the technique is similar to the synthetic cmd method used to derive sfhs ( see tolstoy et al . in the cmd the position and number density of stars depend upon the imf , age , and metallicity . as pointed out by @xcite , the cmd of any complex stellar system will be a composite of all the individual stellar populations that comprise the system . therefore it is possible to model a large number of simple stellar populations ( ssps ) in various combinations to reproduce the observed data , including the observational errors . the most likely combination of populations of various ages and metallicities that best match the observations will give the sfh . in the color - color diagram the position and number density of stars depend upon metallicity and the imf of a population , and is independent of age , unlike a cmd . the synthetic color - color diagram method of deriving mdfs solves for the weights of a given linear combination of populations that equals the observed color - color diagram ; the weights from the linear combination give the metallicity distribution of the population . the advantage is that a given population will occupy a locus where color and number density of stars found at each location within the color - color diagram is dictated by stellar evolution and the imf . to create the synthetic color - color diagrams a kroupa imf @xcite and colors from the dartmouth isochrones are used to initially populate a grid of hess diagrams of ssps . the isochrone colors were empirically corrected following the calibration described in @xcite . the grid spans a metallicity range @xmath9 } < + 0.5 $ ] in steps of 0.05 dex . each ssp hess diagram was blurred using the measured photometric errors from each dwarf galaxy . the linear combination of ssps that reproduces the observed color - color diagram are calculated from the array of synthetic hess diagrams . the isochrone color spacing and the bin spacing of the hess diagrams ( 0.04 by 0.04 mag ) require that the metallicity spacing ( and subsequent color change ) be larger than the hess bins . in some regions of the color - color diagram isochrones spaced 0.05 dex of [ fe / h ] apart have a color change less than 0.04 mag . if the color change is less than the bin size only one of the weights for the linear combination of isochrones will be positive the rest will have negative and nonphysical weights . additionally , in the synthetic color - color diagram , when the synthetic ssps are blurred by the photometric errors ( up to @xmath100.07 mag near the bottom of the giant branch ) , isochrones closer than 0.15 dex produce nonphysical negative solutions . to account for the issues arising from the isochrone color spacing and the bin spacing in the hess diagrams , we combine the weights from various metallicity combinations to ensure we sample all metallicities equally . we start with the full range of metallicities ( -2.5 @xmath11 [ fe / h ] @xmath11 0.5 , and spacing of 0.05 dex ) , calculate the linear combination , eliminate the metallicities that have negative , nonphysical solutions and recalculate the linear weights . this leads to uneven spacing in metallicity due to the smaller color separations at low metallicity and larger color spacing at high metallicity . we repeat the calulation with different metallicity spacings ( 0.1 , 0.15 , 0.2 , and 0.25 dex ) and shifts of 0.05 dex in order to fill in the metallicity spacings . the weights from various metallicity combinations are combine to ensure we sample all metallicities equally . the mdf weights are checked by creating a synthetic color - color diagram to compare to the observations . we examine the residuals between our observed hess diagram and a synthetic hess diagram to ensure that @xmath12 of the residuals deviate by less than 3@xmath13 of the overall residual within the hess diagram . the residual hess diagram is created using the following equation : @xmath14 to test the synthetic color - color diagram method of deriving mdfs we performed this analysis on globular clusters of known metallicity , specifically m92 , ngc 6791 , ngc 104 , and ngc 5927 , with [ fe / h]@xmath15 respectively . for these gcs we adopted all of the parameters reported in @xcite , where the individual mdfs were derived . here after we will refer to the mdfs derived using the synthetic color - color diagram method as ` synthetic mdfs ' , and the mdfs created from individually measured metallicities as ` individual mdfs ' . the synthetic mdfs recovered peak metallicities that are all within 0.05 dex of the literature values . additionally we find the synthetic distributions to be over two times as narrow as the distributions found from fitting each star individually ( @xmath160.15 , 0.19 , 0.24 , and 0.19 for m92 , ngc 6752 , ngc 104 and ngc 5927 , respectively ) . the narrower mdfs are expected because this method accounts for the photometric errors , while the individually measured metallicities do not . figure [ clusters ] shows the process and final results of the synthetic color - color diagram method of deriving mdfs . the top panels display the observed sequence of giant branch stars in the color - color diagram . the middle panels show the hess diagram of the residuals as described in equation ( 1 ) . the bottom panels show the resulting mdfs from the synthetic method ( solid line ) as well as the individual mdfs ( dashed line ) . the mdfs in the bottom panels illustrate the utility of using the synthetic method over the individual star method , wherein the synthetic mdfs are 30 - 50 @xmath0 narrower than the individual mdfs . the synthetic method is statistical in nature , therefore it does not provide individual metallicities for each star , but rather gives the relative number of stars at each metallicity for the population as a whole . we directly compare the subset of our individually measured stellar sample that overlap with spectroscopic measurements from @xcite in figure [ z_comp ] . @xcite used medium resolution spectra and spectral synthesis of fe i absorption lines to measure metallicities for 814 , 256 , and 125 stars in leo i , leo ii , and ic 1613 , respectively . due to the greater spatial extent of their sample we were only able to compare a total of 170 stars ; 108 from leo i , 54 from leo ii , and 8 from ic 1613 . the standard deviation of metallicity differences is 0.38 dex , a wider spread ( by 0.1 to 0.2 dex ) than found in the globular clusters from @xcite . the larger spread between photometric and spectroscopic metallicity can be partially attributed to the difference in photometric errors between the dwarf galaxies and the globular clusters . the photometric errors along the dwarf galaxies giant branch are 2 to 4 times larger than equivalent absolute magnitudes found in the globular clusters , which translates to a @xmath17[fe / h ] of @xmath100.1 dex . the larger spread might also be partially accounted for by @xmath1 abundance variations . @xcite find that variations in @xmath1 abundance cause color changes analogous to metallicity changes of a few tenths of a dex ( specifically @xmath18[fe / h]/@xmath19 @xmath10 @xmath20 across the metallicity range ) . without knowing the intrinsic @xmath1 abundance , and assuming it to be solar , naturally our metallicity measurement will be lower than the actual [ fe / h ] . in general dwarf galaxies tend to show less @xmath1 enhancements than globular clusters of similar metallicity . @xcite find that the @xmath1 abundance distributions slowly evolve from large @xmath1 enhancement at low metallicity to roughly solar ratios at [ fe / h ] close to a tenth solar . any @xmath1 enhancement will cause an underestimate of the metallicity using the photometric method . in addition to the greater difference between photometric and spectroscopic metallicities , the photometric metallicities deviate more towards lower metallicity . this is not unexpected as we assign metallicities with an isochrone grid assuming no @xmath1 enhancements . for any star that has an enhanced @xmath1 abundance the assigned metallicity from isochrones will be lower than the intrinsic stellar metallicity because the color of a star of solar @xmath1 abundance will have the same color as an @xmath1 enhanced star with lower [ fe / h ] . the stars with photometric metallicities measured to be [ m / h ] @xmath212 tend to be the most discrepant , with differences @xmath22 1 dex . the main difference is due to the decreasing color change as a function of decreasing metallicity , which means that a small random variation in color for a bluer ( metal poor ) star will produce a larger discrepancy in the reported metallicity than the same amount of variation in a redder ( metal rich ) star . we used the synthetic color - color diagram technique to derive mdfs for the four dwarf galaxies in addition to the individually measured metallicities that were compiled to make mdfs . figure [ dwarfs ] , following the same layout as figure [ clusters ] , shows the results from the synthetic mdf derivation method . the top panels show the color - color diagrams for the four dwarfs , the middle panels show the residual hess diagrams , and the bottom panels show the resulting synthetic ( solid line ) and individual ( dash - dotted line ) mdfs . the individual and synthetic mdfs both show similar shapes and peaks , although the synthetic mdfs are narrower , as was noted in section 3.3 from the mdfs of globular clusters . for the dwarf galaxies the average metallicities of the synthetic mdfs are 0.1 to 0.2 dex more metal rich than the individual mdfs , although the mdf peaks from the two methods are more closely aligned . additionally , the globular cluster mdfs from the two methods are not offset . this leads us to believe that the underlying population distribution is causing the offset . for the dwarf galaxies the difference in average metallicity from the two methods can be attributed to the differences in the metal poor tails of the mdfs , where the individual mdf has photometric errors that propigate into a wider spread in metallicity . additionally , the fact that the color change due to metallicity is smaller at low metallicities means that for the same amount of color error there will be a larger spread in metallicities at low values . the large extent of the metal poor tail in the individual mdf is the main driver in the offset of the average metallicities for the 2 methods . despite the offset in average metallicity we believe the synthetic method provides a more accurate mdf shape than the mdf produced by individually measuring metallicities with isochrone matching . the synthetic method includes stellar sequence information from the color - color diagram , and it systematically accounts for the photometric errors , whereas the individual method relies only on the photometry . in figure [ dwmdf_comp ] we compare our two methods of deriving mdfs to the spectroscopic mdfs available from @xcite . table [ mdfparam ] reports the average metallicities , widths , and number of stars measured using the three different mdf derivation methods for each galaxy . the spectroscopic metallicities from @xcite are more spatially extended than our photometric data . since we are probing a smaller area within each galaxy , it is expected that our mdfs will have some differences compared to the spectroscopic mdfs . the results for each dwarf galaxy are discussed in the following paragraphs . for leo i we find @xmath11[m / h]@xmath23 @xmath24 . the synthetic mdf is narrow ( @xmath25 ) , mostly gaussian with an extension on the metal poor end , and shows an abrupt cutoff on the metal rich side of the distribution . the shapes , peaks and widths are consistent with metallicities found in the literature . @xcite found the median [ fe / h ] @xmath261.42 ( @xmath13 = 0.33 ) . @xcite measured a mdf ( for 101 rgb stars ) peaked at [ ca / h ] @xmath261.34 ( @xmath13 = 0.21 dex ) , using the infrared ca - triplet method of measuring stellar metallicities . using a different ca - triplet calibration found [ fe / h ] @xmath27 ( @xmath13 = 0.18 dex ) using 54 stars . for leo ii we find @xmath11[m / h]@xmath23 @xmath28 with the mdf showing a more extended tail at low metallicities . the peak value and overall mdf shape are consistent with those reported in the literature . @xcite measured a mdf with a peak [ fe / h ] @xmath29 from the medium resolution spectra of 256 stars . a near ir photometric study of rgb stars in leo ii by @xcite found a mdf peak , [ m / h ] @xmath30 , when they account for the mean age ( 9 gyr ) . @xcite measured stellar metallicities of 74 stars using the infrared ca - triplet of rgb stars , the mdf they measured peaked at [ ca / h ] @xmath261.65 ( @xmath13 = 0.17 dex ) . @xcite also measured spectroscopic metallicities using the ca ii infrared method and found the mean metallicity of leo ii to be [ fe / h ] @xmath31 from 52 stars . for ic 1613 we find @xmath11[m / h]@xmath23 @xmath32 , which is more metal rich than the average spectroscopic metallicity measured by @xcite , @xmath11[fe / h]@xmath2 @xmath33 . another average metallicity measurement from @xcite finds @xmath11[m / h]@xmath2 @xmath34 from a sfh study of a non - central field . the discrepancy between the higher average metallicity that we measure compared to the values reported by @xcite and @xcite , is partially accounted for by the fact that we are looking at different regions of the galaxy , and thus a younger population of stars . sfhs of ic 1613 show the star forming regions move continuously inward over the lifetime of the galaxy @xcite , where sf is associated with higher average metallicity . additionally , @xcite find the stellar metallicities of ic 1613 to progress from [ fe / h ] @xmath35 to @xmath36 over the age of the galaxy . our field of view looks at the inner @xmath105 @xmath0 of the galaxy , while the spectroscopic metallicities are from stars spanning the whole galaxy and the sfh comes from a field 5.5 from the center . while the average metallicity we find is more metal rich than most studies , we are more metal poor than the [ fe / h ] @xmath37 from the high resolution study based upon 3 supergiants by @xcite . we believe the average metallicity we measure is reasonable for the young central region of ic 1613 that our field of view covers ( @xmath38 ) even though it is more metal poor than the average galactic metallicity measured in other studies . for phoenix we find @xmath11[m / h]@xmath23 @xmath39 . upon first examination this average metallicity is in conflict with metallicities reported in the literature , although none of the literature values are spectroscopically derived . most of the reported metallicities are from sfh studies , like the one by @xcite who found @xmath11[fe / h]@xmath2 @xmath40 . the radial sfh study by @xcite found the old population to have [ fe / h ] @xmath41 , and the younger populations to have [ fe / h ] @xmath42 . additional studies find [ fe / h ] @xmath43 , @xmath44 a study of variable rr lyrae stars in phoenix predicts @xmath11[fe / h]@xmath2 @xmath45 or @xmath46.75 based on two different period metallicity relations @xcite . according to the mass - metallicity relationship @xcite , phoenix , which is the same stellar mass as leo ii , should have a similar metallicity ( [ fe / h ] @xmath47 ) ; yet , the average metallicity we measure is higher than the galactic average metallicity reported in other studies . however , the field of view for our observations covers the innermost region ( @xmath48 ) of phoenix , where sf has occurred as recent as 100 myr ago and higher metallicities are expected due to the continuous star formation occurring in the central region . additionally , the radial sfh study by @xcite find the central regions of phoenix to have a mean metallicity of @xmath42 , which is in agreement with our results . as with ic 1613 , we believe the higher average metallicity we measure reflects the greater amount of enrichment that occurs in the central regions of dwarf galaxies . lccccccccc[h ] leo i & -1.43 & 0.43 & 3,449 & -1.34 & 0.21 & 4,296 & -1.45 & 0.32 & 814 + leo ii & -1.77 & 0.72 & 444 & -1.54 & 0.27 & 780 & -1.63 & 0.40 & 256 + ic 1613 & -0.99 & 0.54 & 896 & -0.84 & 0.30 & 2,111 & -1.19 & 0.37 & 125 + phoenix & -1.17 & 0.67 & 578 & -1.14 & 0.28 & 1,399 & & & [ mdfparam ] the chemical enrichment within galaxies provides information on the gas accretion , gas expulsion , the interaction history and star formation history . matching analytic models of chemical evolution to observations will constrain different evolutionary scenarios . it stands to reason that dwarf galaxies could be represented by simple models due to their isolation and relatively less complicated dynamics . therefore we begin by fitting a simple model of chemical evolution to our observed mdfs . the simple model is essentially a leaky box , where gas is allowed to leave the galaxy , but no accretion is occurring . additionally we assume @xmath49 is the effective yield , not the true nucleosynthetic yield , where the effective yield is a product of both the amount of metals created , and the amount blown out of the galaxy through winds . the effective yield @xmath49 is in units of z@xmath50 . the shape of the mdf for the simple model follows the functional form as defined by @xcite : @xmath51 } } \propto \left ( 10^{\rm{[fe / h ] } } \right ) { \rm exp } \left ( - \frac{10^{\rm{[fe / h]}}}{p } \right)\ ] ] the simple model has a tendency to overproduce metal poor stars , whether in dwarf galaxies or the solar neighborhood . we find that the simple model is often unable to completely reproduce the mdf of dwarf galaxies , in agreement with the findings of @xcite . it has been theorized that the paucity of metal poor stars can be explained by gas infall at later times . therefore we also tested an analytical model which includes infalling gas . we fit the ` best accretion model ' defined by @xcite which allows for the gas mass to start small ( or at zero ) , rise to a maximum , then approach zero again as all the mass is accumulated in stars . the accreted gas is assumed to be metal free . lynden - bell found a relation between the gas mass , @xmath52 , stellar mass , @xmath53 , and total final mass , @xmath54 , that permits an analytic solution to the differential metallicity distribution ( also see pagel 1997 ) , defined as : @xmath55 where all quantities are in units of the initial gas mass . when @xmath56 the equation reduces to a closed box where @xmath57 . when @xmath54 is larger than 1 it essentially becomes a measure of the total amount of gas that has entered the system over the lifetime of the galaxy so @xmath54 can be thought of as an accretion parameter . however , it should be noted that while equation ( 3 ) produces the desired mdf shapes , there is no physical rationale behind the actual form of the equation . equation ( 3 ) simply assumes a quadratic relation between gas mass and the stellar mass that peaks then decays as a function of stellar mass . the mdf of the accretion model , as defined by @xcite , follows the form : @xmath58 } } \propto \frac{10^{\rm{[fe / h]}}}{p } \frac{1 + s(1-\frac{1}{m } ) } { \left ( 1-\frac{s}{m } \right)^{-1 } -2\left ( 1-\frac{1}{m } \right ) \frac{10^{\rm{[fe / h]}}}{p } } \ ] ] where @xmath53 must be solved for numerically from the equation : @xmath59}(s)= & \\ \log \nonumber & \left\ { p \left ( \frac{m}{1+s -\frac{s}{m } } \right)^2 \left [ \ln \frac{1}{1-\frac{s}{m } } - \frac{s}{m } \left ( 1-\frac{1}{m } \right ) \right ] \right\ } \end{aligned}\ ] ] for both the simple and the accretion model , the metallicity peak of the model directly increases with @xmath49 , as long as instantaneous mixing of the ism is assumed . the analytic form of the differential metallicity distribution for these models requires the instantaneous recycling and mixing approximations , where all sn yields are immediately and uniformly returned to the ism , respectively . the instantaneous recycling approximation does not reproduce the characteristic patterns seen in @xmath1 element abundances . however @xmath1 abundances can not be determined with photometric data alone , therefore we could not constrain more sophisticated models that account for time dependent recycling . the analytic models used in this work neglect some of the physics known to be important for galactic chemical evolution , however , these models provide insight into the difference in evolution between galaxies . for dsph , one leading hypothesis is that their chemical evolution was interrupted by the removal of gas . this follows from the morphology - density relationship where most dwarf galaxies close to a large primary are dsph , and devoid of gas @xcite . environmental mechanisms such as tidal stirring and ram pressure stripping have been invoked to transform close satellites from a star forming , gas rich , dir , into a quiescent , gas poor dsph that are found close to a much larger primary @xcite . @xcite used simulations to shows that gas rich dwarfs can be stripped of their gas over a few pericentric passages as they travel through the hot halo of a larger primary . ram pressure stripping effectively cuts off all star formation , stopping the chemical evolution prematurely , likely producing a steep truncation on the metal rich end of the mdf . @xcite show that a sharp metal rich cutoff of the mdf is primarily seen in dsph type galaxies , not dirs . additionally , they added a ram pressure stripping term to the simple chemical evolution model to fit the metal rich cutoff , however , this model was found to not have the best match to the dsphs in their sample . since the accretion models fit all the dsph galaxies better than the simple models we would like to incorporate a metallicity cutoff into the accretion models , that would simulate ram pressure stripping . a more basic approach to testing a metal rich cutoff is achieved by changing the upper metallicity bounds to a more metal poor value when integrating the equations of chemical evolution . we ran the simple and accretion chemical evolution models with an imposed metal rich cutoff , truncating the cem at a lower metallicity than the model would naturally evolve to , in order to approximate the effects of ram pressure stripping . a sharp cutoff such as this is somewhat nonphysical , since ram pressure stripping does not immediately remove all gas , additionally the sf would not be immediately cutoff at a particular metallicity across the galaxy . however , this simplistic approach allows us to quickly and efficiently test our models for signs of truncation . besides the dsph galaxies , dir are expected to have a steeper mdf on the metal rich side than the chemical evolution models predict . the mdf of a star forming galaxy should not look like a completed chemical evolution model , rather , the mdf should show a more abrupt cutoff on the metal rich side because dir are still forming stars and enriching their ism and by definition are not at the end of their chemical evolution . fitting cem with an imposed metal rich cutoff will also simulate an incomplete chemical evolution , in addition to a cem that has been halted due to ram pressure stripping . to find the best model parameters we perform 10@xmath60 trials of the simple model while varying the effective yield , @xmath49 . we calculate the least squares value between each model and mdf to determine the best parameters for the individual and synthetic mdfs for each galaxy . we also calculate the two sided confidence interval to find the 1 @xmath13 range of model parameters . for the accretion model we preform 10@xmath61 trials while varying the effective yield , @xmath49 , and the extra gas parameter , @xmath54 . for the accretion + truncation models we run 10@xmath62 trials varying the effective yield , @xmath49 , the extra gas parameter , @xmath54 , and the cutoff metallicity . the best fit parameters for each model and mdf are listed in table [ cemparam ] . to fit the individual mdfs the three chemical evolution models have been convolved with a gaussian with a dispersion equal to the dispersion of the difference between the photometric and spectroscopic metallicities ( @xmath13=0.38 dex ) to account for the spreading of the mdf from the photometric errors . the synthetic mdf method models the photometric errors , therefore we do not convolve the chemical evolution models with any additional errors when fitting to the synthetic mdfs . the individual mdfs do not show an abrupt metal rich cutoff , mostly due to the scatter in metallicity that comes from the photometric technique as can be seen in figure [ cem ] . however , in the narrower synthetic mdfs we do see truncation of the mdfs on the metal rich side , indicative of ram pressure stripping . in the accretion model the larger the m parameter the more the cem takes on a gaussian shape . while this is good for fitting the narrow peak of the synthetic mdfs , it does not account for the asymmetry of the extended metal poor tail and the sharper truncation of the mdf on the metal rich side as shown in figure [ cem_syn ] . the accretion + truncation model allows for a narrower mdf peak , and keeps the asymmetry by truncating the metal rich side of the distributions , which produces a better model fit for the synthetic mdfs in all four dwarf galaxies , as shown in figure [ cem_syn_cut ] . leo i is one of the more distant mw companion dwarf spheroidal galaxies ( 254 kpc , mcconnachie 2012 ) ; additionally it is receding quickly from the mw ( @xmath63=167.9 km s@xmath64 , @xmath65101 km s@xmath64 , sohn et al . 2013 ) , making it one of the more unusual dsph galaxies . since its discovery in the 1950 s @xcite leo i has been studied by many groups to better understand the sfh , dynamic properties , metallicities and chemical enrichment history . here we compare our chemical evolution models to results from previous studies . we find the accretion + truncation model is the best fit model for the synthetic mdf . the synthetic mdfs shows a narrow peak at [ m / h ] @xmath66 , a slight metal poor tail , and an abrupt cutoff on the metal rich side of the distribution . the simple model underpredicts the number of stars in the peak , and overpredicts the amount of metal poor stars . the accretion model fits the mdf better than the simple model , however this model does not fit the truncation on the metal rich side of the mdf , and subsequently overpredicts the number of metal rich stars . we find that the narrowness of the synthetic mdf , and the metal rich truncation is best fit by the accretion + truncation models , with a large accretion parameter of @xmath67 , yield of @xmath68 and a cutoff metallicity of [ fe / h ] @xmath261.0 . @xcite also modeled the mdf of leo i , with a simple , pre - enriched and accretion model . they found the accretion model was the best fit , with @xmath69 and @xmath70 , which is within errors but smaller than our accretion model parameter of @xmath71 and the yield we derive , @xmath72 , is higher than the one they derive , but our mdf has a higher average metallicity ( see discussion in section 4.2 ) . created detailed chemical evolution models for leo i and ii , that use nucleosynthesis for type i and ii sn , an exponentially decreasing accretion parameter , the sfh , and @xmath1 abundances and the mdf as inputs to solve for the outflowing wind , and star formation efficiency . the detailed chemical evolution model that best matches the leo i abundances predicts that the galaxy started with a low star formation efficiency ( 0.6 gyr@xmath64 ) for the first 5 gyr , then had another 7 gyr episode of sf starting at 9 gyr with substantial galactic winds throughout . the evolution model of agrees with the color magnitude diagram ( cmd ) based sfhs showing that leo i had slow ancient sf , with an increase in the sfr around 7 - 8 gyr ago , and another increase to its highest rate 2 - 3 gyr ago , after which the galaxy stopped forming stars 1 gyr ago @xcite . leo ii is also one of the more distant mw companion dwarf spheroidal galaxies ( 233 kpc , mcconnachie 2012 ) . unlike leo i , the proper motion study of leo ii reveals the galactocentric velocity is mostly tangential ( @xmath73 = 265.2 @xmath74 ) , with only 21.5 @xmath75 in the radial direction . subsequent dynamic modeling of the large velocity and large distance of leo ii suggest that leo ii is most likely passing into the mw halo for the first time @xcite . the synthetic mdf of leo ii is best fit by the accretion + truncation model with an accretion parameter of @xmath76 , yield of @xmath77 and a metallicity cutoff of [ fe / h ] @xmath261.0 . the simple model underpredicts the number of stars found in the mdf peak , and overpredicts the number stars in the metal poor tail . the accretion model fits the peak and metal poor tail , but over predicts the number of stars on the metal rich end of the distribution , explaining why the best fit model includes the metallicity truncation term . our chemical evolution results are in agreement with @xcite who found the accretion model to be the best fit for leo ii , with an accretion parameter of @xmath78 and yield of @xmath79 . for our accretion model we measured @xmath80 and @xmath81 . modeled the detailed chemical evolution of leo ii based on the overall mdf and the @xmath1 element abundance ratios measured by @xcite . their best fit model predicts one long ( 7 gyr ) episode of sf starting at 14 gyr , with a lower sf efficiency than seen in leo i ( 0.3 gyr@xmath64 ) with very high wind efficiency throughout . the chemical evolution of is in good agreement with the sfh derived from the wide field photometric survey of @xcite . they found that leo ii evolved with a low sf rate up to 8 gyr ago when sf stopped in the outer regions , from 8 to 4 gyr ago the central star forming region continually shrank until sf essentially stopped . lccccccll[h ] leo i & indiv & 0.054 @xmath82 & 0.054 @xmath83 & 8.2@xmath84 & 0.050 @xmath85 & 7.7@xmath86 & -0.10 & a+t + leo ii & indiv & 0.038 @xmath87 & 0.038 @xmath88 & 3.0@xmath89 & 0.038 @xmath90 & 3.0@xmath91 & -0.40 & s + ic 1613 & indiv & 0.228 @xmath92 & 0.184 @xmath93 & 1.5@xmath94 & 0.182 @xmath95 & 1.7@xmath96 & -0.30 & a+t + phoenix & indiv & 0.118 @xmath97 & 0.112 @xmath98 & 1.5@xmath99 & 0.112 @xmath100 & 1.5@xmath101 & -0.20 & a + + leo i & synth & 0.054 @xmath102 & 0.068 @xmath103 & 9.8@xmath104 & 0.070 @xmath105 & 6.6@xmath106 & -1.0 & a+t + leo ii & synth & 0.036 @xmath107 & 0.038 @xmath108 & 4.1@xmath109 & 0.038 @xmath110 & 4.0@xmath111 & -1.0 & a+t + ic 1613 & synth & 0.182 @xmath112 & 0.196 @xmath113 & 4.1@xmath114 & 0.200 @xmath115 & 3.7@xmath116 & -0.40 & a+t + phoenix & synth & 0.086 @xmath117 & 0.092 @xmath118 & 3.4@xmath111 & 0.094 @xmath119 & 3.3@xmath120 & -0.60 & a+t [ cemparam ] as one of the nearest gas rich dwarf irregular galaxies at a distance of 721 kpc there have been numerous studies of the gas and stars in ic 1613 @xcite . ic 1613 is similarly distant from m31 ( 517 kpc ) however it is not considered to be a satellite of either the mw or m31 @xcite , but without data on the proper motion the membership of ic 1613 to either is not definitive . various h i studies have measured the total h i mass as 3 - 6 x@xmath121 . additionally , the studies show complicated morphology including numerous h i arcs , holes and shells produced from the ongoing sf within ic 1613 as evidenced by the presence of ob associations and h ii regions . for ic 1613 we find the accretion + truncation model with an accretion parameter of @xmath122 , a yield of @xmath123 and a cut off metallicity of [ fe / h ] @xmath1240.40 is the best fit to the synthetic mdf . ic 1613 s mdf shows a broad peak at [ m / h ] @xmath125 , an asymmetric gaussian shape with an extended metal poor side of the distribution . the simple model underpredicts the number of stars in the peak , overpredicts the metal poor tail , and slightly overpredicts the number of stars on the metal rich side of the mdf . the difference between the accretion model and the accretion + truncation model is small , and almost entirely due to the metal rich side of the mdf . since ic 1613 is an isolated galaxy that has a substantial gas mass and shows no sign of interaction , ram pressure stripping can not be the process truncating the mdf . instead , the steeper metal rich side of the mdf reflects the continuing chemical enrichment of ic 1613 . comparing our work with @xcite we find that our accretion model with @xmath126 and @xmath127 has a similar accretion parameter but a higher effective yield compared to their @xmath128 and yield of @xmath129 . the higher effective yield we measure is potentially due to the different regions of the galaxy sampled . we cover a small ( @xmath38 ) central field while the metallicities from @xcite cover the entire galaxy . even though our accretion models find similar parameters , @xcite find that a pre - enriched model has a more likely fit than the accretion model . the pre - enriched model is a generalization of a closed box model that starts with some initial metallicity and does not accrete additional gas over its lifetime . phoenix is a transition type galaxy at a distance of 415 kpc and an absolute v magnitude of -9.9 @xcite . there is no h i emission detected within phoenix itself , however , h i gas has been detected at 4.5 to 9 from the center of the galaxy @xcite indicating recent gas expulsion . young et al . also state that the expelled gas is likely linked with the most recent episode of sf ( @xmath10100 myr ago ; bianchi et al . 2012 ) . we find the accretion + truncation model to be the best fit model for the mdf , with an accretion parameter of @xmath130 , a yield of @xmath131 , and a cut off metallicity of [ fe / h ] @xmath132 . the simple model underpredicts the number of stars in the mdf peak and overpredicts the amount of metal poor stars in the tail of the distribution . the difference between the accretion model and the accretion + truncation is small ; both models slightly overpredict the number of metal poor stars while on the metal rich side of the mdf the accretion + truncation model has a slightly better agreement . ground - based photometry of phoenix by @xcite noted the presence of a structurally distinct inner population with mostly young stars in the east west direction , while isophots of the older component are aligned in the north - south direction . another ground based wide - field ( 26 x 26 arcmin@xmath133 ) photometric study shows the young ( @xmath11 1 gyr ) inner population to have a disk like distribution @xcite . while the accretion + truncation model is found to be the best fit , evidence for ram pressure stripping is not seen in any of the population studies , nor would it be expected in an isolated galaxy . in a study of the radial sfh properties of dwarf galaxies @xcite additionally modeled the chemical evolution for phoenix . their best fit model predicts an initial episode of sf lasting 3 - 4 gyr with very little or no metal enrichment that accounts for 75@xmath0 of the total cumulative stellar mass . additionally , the sf episode is longest in the central regions with shorter sf episodes in the outer regions . to account for the minimal amount of enrichment ( [ fe / h ] @xmath134 ) in this first phase their model predicts a high sfr that would expel most of the metals formed in this episode before they can mix with the interstellar medium . the second phase of sf produces most of the metal enrichment , and mainly occurs in the central regions where the gas density is still sufficiently high to support continued sf , they find the metallicity of this burst to be [ fe / h ] @xmath135 . we find the mdf peak to be [ m / h ] @xmath136 , which is consistent with the enrichment values @xcite found in the central regions of phoenix from their radial sfh study . however , much like the initial metallicity @xcite found , other photometric studies measured the overall metallicity in phoenix as @xmath11[fe / h]@xmath2 @xmath10 -1.7 , @xcite , [ fe / h ] @xmath43 , @xcite , @xmath11[fe / h]@xmath137 . metallicity gradients found in stellar populations offer clues about past sf , as well as interaction history since mergers are thought to remove gradients . many dwarf galaxies have detectable metallicity gradients ( e.g. fornax , cetus , carina , sculptor , sextans , tucana , leo i , leo ii , and draco , andromeda i - iii , v and vi . from harbeck et al . 2001 ; tolstoy et al . 2004 ; battaglia et al 2006 ; 2011 , 2012 ; monelli et al 2012 ; kirby et al . 2011 ) , while others like ursa minor and canes venatici do not show measurable gradients ( kirby et al . 2011 ) . metallicity gradients can be produced by increased sf ( and thus enrichment ) in the central regions due to increased gas supply there . the gas density increases with the gravitational potential and thus accumulates towards the centers of galaxies . additionally , star formation is proportional to the gas density thus producing increased sf and enrichment towards the insides of the galaxy . environmental interactions are expected to modify the stellar distributions , diluting any gradients , and possibly inducing morphological transformations @xcite . additionally , recent simulations of the chemo - dynamical evolution of dwarf galaxies have shown tidal interactions to be an efficient mechanism to remove metallicity gradients . also , observations by @xcite have shown that radial sfhs and metallicity gradients are consistent with the sf in dwarf galaxies being quenched in the outer regions due to limited gas availability . gradients in age and metallicity have been seen for a wide range of physical characteristics of dwarf galaxies . the trend of younger and higher metallicity stars increasing towards the centers of galaxies spans many galactic properties such as total mass , gas mass , metallicity , velocity dispersion and environment . the fact that gradients are consistently found even across a wide range of physical characteristics hints that population gradients are intrinsic to dwarf galaxy formation and evolution . however , at large radii , many of these differences tend to disappear , suggesting some physical processes may not affect all galactocentric radii equally ( e.g. the uv background could affect the inner and outer regions differently ) . lccc[h ] + leo i & -1.52 & -0.34 & @xmath138 + leo ii & -11.55 & -3.02 & @xmath139 + ic 1613 & -2.48 & -0.20 & @xmath140 + phoenix & -1.84 & -0.26 & @xmath141 [ grad ] the radial metallicity gradients we measured are reported in table [ grad ] , and shown in figure [ gradients ] . we find metallicity gradients in leo i and leo ii . in phoenix and ic 1613 the errors on the gradients are consistent with no slope . much of the error is due to the small radial extent the wfc3 field of view covers in these galaxies , thus we can not confirm or rule out a metallicity gradient in phoenix or ic 1613 . in leo i age and metallicity gradients have been seen in various star formation histories , variable star populations , and metallicity studies . @xcite found evidence for a radial gradient in the 1 - 3 gyr stellar population using near ir photometry . found a shallow metallicity gradient of -0.27 dex kpc@xmath64 as part of their metallicity study of leo i , which is shallower than the gradient we measure , -0.34 dex kpc@xmath64 , in the same units . in the work by @xcite gradients for leo i are reported as @xmath142[fe / h]/@xmath142(@xmath143 ) @xmath144 dex per @xmath145 , while we find a similar gradient for leo i of @xmath146 dex per @xmath147 . where the core radius ( @xmath145=240 pc ) or the half light radius ( @xmath147=251 pc ) both give the same gradient value to 2 significant figures . our photometry covers leo i out to @xmath10 0.7 @xmath147 , while the work by @xcite extends @xmath104 times farther for leo i. in leo ii , the wide field photometric survey by @xcite found evidence of a radial gradient in the hb morphology , where red hb stars are more centrally concentrated than blue hb stars . in the work by @xcite the gradient for leo ii is reported as @xmath142[fe / h]/@xmath142(@xmath143 ) @xmath148 dex per @xmath145 , while we find a steeper gradient of @xmath149 dex per @xmath147 . where the core radius ( @xmath145=180 pc ) and the 1/2 light radius ( @xmath147=176 pc ) both give the same gradient value to 2 significant figures . our photometry covers leo ii out to about 0.6 @xmath147 , while the work by @xcite extends @xmath103 times farther . in ic 1613 the wide field photometric study of @xcite looked at radial differences in the cmds and concluded that there is an age gradient , however , they could not confirm a metallicity gradient in ic 1613 from their broadband ground based photometry . the radial sfh study of phoenix by @xcite measured the radial sf in phoenix while simultaneously deriving metallicities . they found that phoenix had longer periods of sf in the central regions , and decreasing average metallicities as function of radius , with the central regions having an average [ fe / h ] @xmath10 @xmath150 , decreasing outwards to @xmath151 , although they do not quantify a metallicity gradient for the galaxy . as the name suggests , a dwarf transition galaxy is considered to be in a state of morphological transition between a dir and dsph type galaxy @xcite . this idea is supported by the presence of the morphology - density relationship , where dsph are mainly found in dense environments close to a much larger primary while dir and dtrans are found in isolation @xcite . the dwarf galaxy morphology segregation has led to an environment dependent explanation invoking tidal stirring and ram pressure stripping to transform a dir close to a larger galaxy into a dtrans then a dsph over subsequent orbital passages around the primary @xcite . however , some dwarf galaxy properties conflict with the morphological transition scenario . dtrans have similar gas fractions @xcite , and uv fluxes @xcite as dirs while the h@xmath1 emission in dtrans is markedly less and the typical stellar mass of dtrans is @xmath10 3 - 4 times lower than that of dirs . dtrans are mostly located in isolated regions ( like dirs ) , but dtrans luminosities , like dsphs , are lower than dirs . this led @xcite to suggest that dtrans appear to be lower mass versions of dirs that lack recent sf , and not an evolutionary bridge between dsphs and dirs . additionally , should dtrans be progenitors of dsph then they should be seen in similar locations as both dsph and dirs , but they are found only in the same regions as dirs . further evidence against a morphological transition theory comes from @xcite who found systematic differences in the mdf shapes of dirs and dsphs . they found the dsph mdf shapes were systematically narrower and more peaked while dirs tend to have wider mdfs that more closely resemble a simple chemical evolution model . the narrower dsph mdf is best fit by chemical evolution models with large amounts of accretion which is at odds with the proposed mechanism to transform dirs into dsphs , i.e. gas stripping . accretion has been shown to be an important component in dwarf galaxy evolution @xcite . also , the analytical ` best accretion ' model of @xcite imply that dsph tend to require large accretion parameters ( m values ) to fit the observed mdfs @xcite . however , the morphology - density relation points to gas stripping and tidal disruption as important factors in shaping dsph . simulations have also shown that gas stripping and tidal disruption is integral in creating many of the the dsph seen today @xcite . further examination of the best accretion model of @xcite shows that there is no physical basis behind the form of equation ( 3 ) other than that it produces the desired mdf shapes . equation ( 3 ) simply assumes a non - linear relation between gas mass and the stellar mass that follows a quadratic relation that peaks then decays as a function of stellar mass . due to the contradictory implications from the accretion model and the gas stripping caused by gravitational interactions known to be integral in shaping dsphs , in addition to the lack of physical meaning behind the functional form of the accretion model we suggest that the interpretation of the accretion portion of the accretion model should be left for more sophisticated chemical evolution models . in lieu of drawing interpretations from the models themselves , we can use the differences in the model parameters in the context of the dynamic histories of each galaxy to inform our discussion . examining the dsph type galaxies in our sample ( leo i and ii ) , we find that the mdf of leo ii is more similar to the mdfs from ic 1613 ( dir ) and phoenix ( dtrans ) than to leo i s . leo ii is also one of the more distant dsph associated with the mw ( 233 kpc ) , although it is slightly closer than leo i ( 254 kpc ) @xcite . dynamical studies of leo ii @xcite and wide field photometric studies @xcite show no significant signs of tidal distortion . @xcite studied leo ii s proper motion and found nearly all of its velocity is in a tangential component ( @xmath152 km s@xmath64 ) with only a small radial component ( @xmath153 km s@xmath64 ) . @xcite take this to mean that leo ii either has a low - eccentricity orbit , or is near perigalacticon or apogalacticon , and the lack of evidence for tidal disruptions @xcite likely rules out the latter . currently the only known mechanism to transform a rotating dir into a dsph is through gravitational influence from a larger primary @xcite . consequently , it can be assumed that leo ii has experienced some tidal interactions with the mw in the past . however , leo ii shows no signs of dynamical interactions , beside the fact that it is a dsph . moreover the similarity of the mdf of leo ii , ic 1613 and phoenix suggest these galaxies experienced comparable conditions during their chemical evolution . since ic 1613 and phoenix are isolated and evolved without outside influence , it could be inferred that leo ii also chemically evolved without dynamical influence from the mw . the sfh of leo ii shows that 95@xmath0 of the stars were formed over 6.3 gyr ago @xcite ; with so many of the stars formed at early times leo ii has had ample time to be transformed from a disk to a spheroid after most of the chemical evolution occurred . we do find the accretion + truncation model to be the best fit for leo ii , where there the cutoff on the metal - rich side of leo ii s mdf is shallower than predicted from the accretion model . we interpret this as an indication of ram - pressure stripping ; however , in leo ii , the cutoff is not nearly as drastic as leo i , indicating that most of the chemical evolution had occurred before the mw began stripping and truncating leo ii s sf . while this is by no means direct proof , the evidence supports the idea that leo ii chemically evolved without strong dynamical influence from the mw . leo i on the other hand shows evidence of strong tidal interactions with the mw . the proper motions and dynamic modeling of leo i show it passed into the mw s potential 2.33 gyr ago , reaching its pericentric approach 1.05 gyr ago @xcite . additionally , the dynamic results correlate well with the cmd based sfhs . leo i shows slow ancient sf , with an increase to its highest rate 2 - 3 gyr ago , after which the galaxy stopped forming stars 1 gyr ago @xcite . @xcite suggest that the increased sf at 3 gyr and the abrupt stop at 1 gyr could have been caused by ram pressure compression or gravitational torques exerted by the mw as the galaxy passed into the mw s potential . for leo i , the dynamical history has had a significant impact on the sfh , and by extension on the chemical evolution . we interpret the sharp difference in the m parameters along with the dynamic information in the literature on leo i , leo ii , ic 1613 and phoenix , as possible evidence that the galaxies chemically evolved under different conditions . the first is passive chemical evolution , where gas rich , star forming galaxies ( i.e. dir ) gradually truncate their star formation as the gas supply is exhausted , evolving into a dtrans , producing a broad mdf in the process . this type of evolution is what is most likely what is seen in leo ii ( dsph ) , phoenix ( dtrans ) and ic 1613 ( dir ) , at various stages in the process . leo ii formed the majority of its stars over 6 gyr ago and the mdf suggests it chemically evolved without significant interactions before it was transformed into a dsph by the mw . phoenix appears to be in the process of passively evolving ; the dynamics and sfh show no indication of interaction . instead the sfh shows a slow decrease in sf up to a few 100 myr ago when sf stopped . additionally , ic 1613 is isolated from both the mw and m31 , and does not show dynamical evidence of interactions , and the mdf of ic 1613 is similar in shape to leo ii and phoenix . it is known that dsph are created through the gravitational influence of a larger primary , if the interaction occurs concurrent with the chemical evolution it will be reflected in the mdf . previous studies have shown that the mw s influence on leo i is reflected in the sfh and dynamics ; we are suggesting it is also shown in the narrow mdf . one understated assumption in analytic chemical evolution models is that each model represents the end point of that galaxy s evolution . however , dir and dtrans are not at the end of their chemical evolution since they are still forming stars and enriching the ism . therefore we must take care when comparing the mdfs of active galaxies such as ic 1613 and phoenix , to the dsphs like leo i and ii , which arguably , have completed their chemical evolution 1 to 6 gyr ago respectively . additionally , for dsph the morphology - density relation suggest that the removal of gas interrupted the chemical evolution altering the chemistry to no longer reflect the end state of chemical evolution models . these various endpoints can be seen in our galaxy sample . leo i shows a mdf truncated on the metal rich side indicative of a halted chemical evolution . leo ii shows signs of mild truncation , as the cem slightly overpredicts the number of metal rich stars and the accretion + truncation model was the best fit model for leo ii . as we discussed previously , leo ii is within the virial radius of the mw , but has a mostly circular orbit at a large distance , therefore it has been weakly affected by the mws gravity , as the mild truncation shows . as a dir , ic 1613 is still actively forming stars and enriching its ism , therefore we would not expect the mdf to reflect a fully completed chemical evolution . the ongoing enrichment in ic 1613 manifests itself with fewer metal rich stars than predicted by the cem . phoenix , on the other hand , is the only galaxy to nearly match the metal rich end of the cem . phoenix is a transition galaxy , that expelled its last remaining gas reserves with the final burst of star formation @xmath10 100 myr ago @xcite . the isolated location and apparent exhaustion of gas imply that phoenix has completed its chemical evolution . one other consideration is that our field of view is small for all of the galaxies in our sample ( covering 5 - 20@xmath0 of the various galaxies ) . given that many dwarf galaxies show metallicity gradients , and variations in sfh as a function of radius , the mdfs will also change as a function of radius , and the solutions to the cems can not be applied to the entire galaxy . we measured the metallicity distribution functions of leo i , leo ii , ic 1613 and phoenix dwarf galaxies by measuring individual stellar metallicities , and by modeling the stellar population to create synthetic mdfs for each galaxy . we find the synthetic mdfs to be a better representation of the overall metallicity distribution , because this method cuts down the photometric scatter propagated into the photometric metallicities . we fit each mdf with chemical evolution models . we find the simple model to be a poor fit for all of our dwarf galaxies , while the accretion + truncation model is the best fit for the synthetic mdfs for all four of the galaxies . the fact that all of our galaxies are best fit by the accretion + truncation model reflects the fact that most galaxies do not make it to the natural completion of the chemical evolution , either because they are still actively forming stars ( dirs ) or because they were truncated by other processes such as ram pressure stripping ( dsphs ) . we find a similar accretion parameter , ( m@xmath104 ) , for leo ii , ic 1613 and phoenix despite the fact that they all have different masses , sfh , morphologies and average metallicities . we interpret the resemblance in their mdfs as an indication that their chemical evolutions occurred under similar conditions , which indicates that leo ii completed most of its chemical evolution in isolation before it was significantly tidally disrupted and transformed into a dsph type galaxy by the mw . leo i has the narrowest mdf , a much larger accretion parameter , and shows significant evidence for interactions with the mw at the same time as the galaxy was forming stars and chemically evolving . we suggest that the mdfs can reveal dynamical interactions if they occur in concert with sf and chemical evolution , or if the galaxy chemically evolved in relative isolation . the narrower mdfs are indicative of interactions shaping the galaxy s present morphology , while a broader mdf indicates a passive evolution . the differences in the mdfs could be a way to distinguish between the two formation pathways . to further this theory we would need to examine the mdfs and dynamic histories of many lg dwarf galaxies . additionally , we measured metallicity gradients for leo i and leo ii . we see some evidence of gradients in phoenix and ic 1613 , however our data does not cover the radial extent required to determine a metallicity gradient greater than the error bars . support for program 12304 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 .

the classical - field formalism has been widely applied in the calculation of normal correlation functions , and the characterization of condensation , in finite - temperature bose gases . here we discuss the extension of this method to the calculation of more general correlations , including the so - called anomalous correlations of the field , without recourse to symmetry - breaking assumptions . our method is based on the introduction of-symmetric classical - field variables analogous to the modified quantum ladder operators of number - conserving approaches to the degenerate bose gas , and allows us to rigorously quantify the anomalous and non - gaussian character of the field fluctuations . we compare our results for anomalous correlation functions with the predictions of mean - field theories , and demonstrate that the nonlinear classical - field dynamics incorporate a full description of many - body processes which modify the effective mean - field potentials experienced by condensate and noncondensate atoms . we discuss the role of these processes in shaping the condensate mode , and thereby demonstrate the consistency of the penrose - onsager definition of the condensate orbital in the classical - field equilibrium . we consider the contribution of various noncondensate - field correlations to the overall suppression of density fluctuations and interactions in the field , and demonstrate the distinct roles of phase and density fluctuations in the transition of the field to the normal phase .

1011.6289

the experimental realization of bose - einstein condensation ( bec ) in dilute atomic gases @xcite has lead to a resurgence of interest in the theory of weakly interacting bose gases @xcite . theories of weakly - interacting bec were initially developed @xcite in the hope of obtaining insight into the physics of the strongly interacting superfluid helium , and can only give a qualitative account of the physics of the liquid superfluid phase @xcite . by contrast , the advent of dilute , weakly interacting bose gases in the laboratory provides for the direct comparison of theories of weakly interacting bec with experiments ( for a review see ref . @xcite ) . more generally , these systems offer an unprecedented opportunity for the quantitative experimental evaluation of quantum - field models at finite temperatures , and away from equilibrium , as noted by other authors @xcite . the analysis of the weakly interacting bose gas at finite temperature is significantly complicated by the necessity of considering interactions between excitations of the condensate . a self - consistent mean - field approach , based on approximate factorizations of field - operator products @xcite leads to the so - called hartree - fock - bogoliubov ( hfb ) formalism for the finite - temperature field , which is also obtained in a variational approach to the problem @xcite . this description , however , violates known exact constraints on the excitation spectrum @xcite , due to the inconsistent introduction of ( leading - order ) many - body effects @xcite into the description of atomic interactions . the simplest resolution of this problem is provided by the so - called popov approximation @xcite to the hfb theory ( hfb - popov ) , in which interactions between excitations at the hartree - fock level are retained @xcite , while many - body corrections are neglected in the treatment of all interatomic interactions . a systematic perturbative treatment of the beyond - quadratic terms of the bose - field hamiltonian @xcite ( see also refs . @xcite ) demonstrates that the failure of the hfb method results from an inconsistent treatment of hamiltonian terms cubic in the bose field operator in the mean - field factorization approximation . this provides support for an extension of the conventional hfb approach @xcite in which the popov theory is augmented by spatially dependent _ effective _ interaction strengths , which serve to upgrade the hartree - fock interactions between atoms to interactions mediated by a many - body @xmath1 matrix , which accounts for all ladder - diagram processes contributing to interactions in the finite - temperature medium @xcite . however , in this theory the zero - energy limit of the @xmath1 matrix is substituted for the interaction strength in all condensate - excitation interactions , neglecting the dependence of the @xmath1 matrix on the collision energies . more complicated ( bubble - diagram @xcite ) effects which appear at the same order of perturbation theory @xcite as the ladder - diagram corrections are also neglected . kinetic theories which include processes beyond those contained in the hfb theories , such as the exchange of atoms between the condensate and thermal cloud , have been considered by several authors @xcite . an alternative approach to modeling the finite - temperature bose gas is provided by the so - called classical - field ( or c - field ) formalism @xcite . in this approach , one considers the dynamics of a classical dynamical system corresponding to the high mode - occupation limit of the formal second - quantized field theory @xcite for the dilute bose gas . the classical - field model provides a leading - order description of the long - wavelength physics which dominate the critical behavior associated with the bose - condensation phase transition @xcite . more generally , the _ dynamical _ classical - field equations of motion arise as the `` classical '' component of the atomic - field evolution in the wigner representation of the second - quantized field @xcite . the classical - field model is thus expected to give a good description of the most highly occupied modes of the system , including the condensate and its low - lying excitations , for which quantum fluctuations can be safely neglected . the introduction of an explicit single - particle energy cutoff restricts the system to these low - energy modes @xcite . in closely related stochastic - field methods @xcite , the description of the low - energy modes includes explicit damping and noise terms arising from the coupling of these modes to the eliminated high - energy component of the bose field . the classical wave - mixing dynamics of the resulting ( hamiltonian or stochastic ) field equation of motion then provide an intrinsic nonperturbative description of all interaction processes in the low - energy field , neglecting only the quantal nature of excitations , which is essentially irrelevant in the high - temperature regime . the inclusion of many - body effects beyond bogoliubov theory in the hamiltonian classical - field theory , and agreement of this theory with the second - order perturbative treatment of ref . @xcite was demonstrated for a homogeneous field in refs . @xcite . in this article , we present a comprehensive , explicit demonstration that the equilibrium classical - field dynamics provide an intrinsic description of many - body interactions in the finite - temperature , harmonically confined bose field . introducing appropriate @xmath0-symmetric field variables , analogous to the modified ladder operators of number - conserving bogoliubov theories @xcite , we quantify the anomalous and non - gaussian nature of the field fluctuations . these correlations reveal signatures of many - body processes neglected in ( e.g. ) the hfb theories . we discuss the importance of these processes in shaping the condensate orbital , and demonstrate the consistency of the penrose - onsager definition of condensation in the classical - field equilibrium . in particular , we show that the penrose - onsager condensate appears as an effective eigenfunction of the total mean - field potential it experiences , provided that the contributions of anomalous averages ( both pair and triplet ) to the latter are taken into account . we also consider the temperature dependence of the many - body effects , and discuss the relation between condensation and the overall suppression of density fluctuations in the system . our results reveal that the partially condensed bose gas exhibits non - trivial correlations as a result of purely thermal ( classical ) fluctuations of the atomic field , which dominate over the quantum fluctuations of the field in realistic experimental systems @xcite . this article is organized as follows : in sec . [ sec : formalism ] we briefly review the derivation of the projected gross - pitaevskii equation ( pgpe ) formalism , and give the parameters of the system we investigate . in sec . [ sec : correlation_functions ] we describe the microcanonical interpretation of the pgpe , and define @xmath0-symmetric classical - field variables appropriate for the evaluation of anomalous correlation functions of the classical field . in sec . [ sec : correlations ] we discuss the local correlation functions of the noncondensed component of the field and their relation to many - body interaction effects described by the classical - field model . in sec . [ sec : e_dependence ] we consider the dependence of the noncondensate correlations and interaction effects on the energy ( and thus temperature ) of the classical - field equilibrium . in sec . [ sec : suppression ] we consider the various contributions to the suppression of density fluctuations in the field which are neglected in mean - field theories . in sec . [ sec : conclusions ] we summarize and present our conclusions . a detailed review of the formalism of ( projected ) classical - field methods has recently been given in ref . @xcite , but for the reader s convenience , we briefly describe the relevant details of the formalism here . formally , the physics of the harmonically trapped dilute bose gas is governed by the second - quantized hamiltonian @xmath2 where the single - particle hamiltonian is @xmath3,\ ] ] and @xmath4 is the exact interatomic potential . we introduce a single - particle subspace @xmath5 spanned by eigenmodes @xmath6 of the single - particle hamiltonian [ @xmath7 with energies @xmath8 less than some cutoff @xmath9 , and a complementary subspace comprised of the remaining high - energy modes . provided @xmath9 is chosen such that the high - energy modes are essentially unoccupied , the dynamics of these modes can be integrated out to obtain an effective hamiltonian for the low - energy ( coarse - grained ) bose field @xmath10 , as shown by morgan @xcite . atomic interactions described by the effective hamiltonian are mediated by an approximate 2-body @xmath1 matrix , and the interaction can thus be rigorously approximated by a contact " potential , with a renormalized coupling constant @xmath11 . the low - energy hamiltonian then takes the form @xmath12 which defines an effective field theory @xcite for the coarse - grained field @xmath13 . we then further divide the low - energy region @xmath5 into a coherent region ( or condensate band ) @xmath14 , spanned by single - particle eigenmodes @xmath15 with energies below some classical - field cutoff @xmath16 , and a complementary incoherent region @xmath17 . introducing the projector @xmath18 onto the coherent region @xmath19 , we define a coherent - region field operator @xmath20 neglecting the coupling of this field operator to modes of the field in the incoherent region @xmath21 , we find that @xmath22 is governed by a hamiltonian obtained from eq . ( [ eq : h_l ] ) by the replacement @xmath23 . the classical - field approximation is then made by demoting the operators @xmath24 in eq . ( [ eq : c_region_opr ] ) to classical variables @xmath25 , thereby defining the classical field @xmath26 . the evolution of the field @xmath27 is governed by the classical - field hamiltonian @xmath28 = \int\!d\mathbf{x}\,\psi^*(\mathbf{x})h_\mathrm{sp}\psi(\mathbf{x } ) + \frac{u_0}{2}|\psi(\mathbf{x})|^4,\ ] ] and hamilton s equations for the classical variables @xmath25 obtained from eq . ( [ eq : hcf ] ) can be expressed concisely as the field equation of motion @xmath29 which is termed the projected gross - pitaevskii equation @xcite . equation ( [ eq : pgpe ] ) describes the classical ( thermal ) fluctuations of the coherent region , while neglecting the contribution of quantum ( vacuum ) fluctuations , which is a valid approximation in regimes of significant thermal excitation , where mode occupations are large , and thermal fluctuations dominate . for an appropriately chosen cutoff @xmath16 , one can augment the classical - field description of the coherent region @xmath19 with a mean - field ( hartree - fock ) description of the incoherent region @xmath21 , allowing for accurate quantitative predictions for the full bose - field system @xcite . in this article , we consider only the correlations of the coherent region @xmath19 that arise from the kinetics of eq . ( [ eq : pgpe ] ) . for the purpose of the numerical implementation of the pgpe , we rescale eq . ( [ eq : pgpe ] ) by introducing units of distance @xmath30 , time @xmath31 and energy @xmath32 , with @xmath33 the transverse trapping frequency of a cylindrically symmetric system . with these choices , ( [ eq : pgpe ] ) becomes @xmath34 where we have chosen the classical field @xmath27 to be unit - normalized , and absorbed the total classical - field atom number @xmath35 into the nonlinear coefficient @xmath36 . we follow ref . @xcite in choosing parameters @xmath37 ( a typical three - dimensional trap geometry ) , @xmath38 , and @xmath39 . the corresponding ground ( gross - pitaevskii ) eigenstate of the system has energy @xmath40 . we analyze equilibrium configurations of this system with energies in the range @xmath41n_\mathrm{c}\hbar\omega_r$ ] . we note that our choice of parameters makes no reference to the physical number of atoms in an experimental system . the classical - field approximation becomes asymptotically exact in the `` classical '' limit @xmath42 , with @xmath43 fixed @xcite , and our results correspond to this idealised classical - limit system . nevertheless , for realistic experimental systems in moderate and high - temperature regimes , the magnitude of quantum fluctuations is small compared to that of thermal fluctuations , and the pgpe provides a good description of the thermal bose gas @xcite . a central feature of the pgpe formalism is the microcanonical ( ergodic ) interpretation of field trajectories at equilibrium . it is well known that the solutions of the nonlinear schrdinger ( gross - pitaevskii ) equation exhibit _ stochastization _ , leading to the ergodic equilibration of the system and the equipartition of the system energy @xcite . this property is also shared by trajectories of the pgpe for homogeneous @xcite and harmonically trapped @xcite systems . in the ergodic interpretation , the pgpe trajectories provide a sampling of the microcanonical density @xmath44 = \left\ { \begin{array}{rl } \mathrm{const } & \quad\quad h_\mathrm{cf}[\psi ] = e \\ 0 & \quad\quad h_\mathrm{cf}[\psi ] \neq e , \end{array } \right . \ ] ] defined by the conserved first integrals of the system . as the trajectories of @xmath45 cover the density @xmath46 $ ] densely , averages in the microcanonical density can be approximated by time averages along the trajectories of @xmath45 . we therefore define correlation functions of the classical field as averages ( or expectation values ) of functionals of the field @xmath45 in the density eq . ( [ eq : mu_density ] ) , which we approximate by time averages of the field ( denoted by @xmath47 ) . applying the microcanonical thermodynamic formalism of rugh @xcite to the pgpe trajectories , one finds that the pgpe system evolves over time to an equilibrium characterized by a well - defined chemical potential and temperature @xcite , providing strong support for this ergodic interpretation of the classical - field dynamics . a correlation function of particular importance in the application of the pgpe to partially condensed bose systems is the covariance matrix @xmath48 which is the classical - field analog of the quantum one - body density matrix @xcite . as @xmath49 is hermitian , we can diagonalize it to obtain a complete basis of eigenvectors @xmath50 ( the eigenmodes of the one - body density matrix ) with real eigenvalues @xmath51 ( the mean occupations of the eigenmodes at equilibrium ) , such that @xmath52 . by analogy to the penrose - onsager definition of condensation @xcite , we identify the largest eigenvalue @xmath53 as the condensate occupation , and the corresponding eigenvector @xmath54 as the condensate orbital . an important feature of classical hamiltonian systems is the relation between symmetries of the hamiltonian and the conservation of quantities during the system s evolution ( see , e.g. , refs . the hamiltonian ( [ eq : hcf ] ) governing the classical - field dynamics is invariant under the @xmath0 ( gauge ) transformation @xmath55 , and this symmetry ensures that the normalization @xmath56=\int\!d\mathbf{x}\,|\psi(\mathbf{x})|^2 $ ] of the classical field is conserved under the action of eq . ( [ eq : pgpe ] ) . moreover , as eq . ( [ eq : hcf ] ) has no explicit time dependence , the classical - field energy @xmath57 $ ] is also conserved under the pgpe evolution . we note that microcanonical density [ eq . ( [ eq : mu_density ] ) ] of the pgpe system inherits the symmetries of the hamiltonian ( [ eq : hcf ] ) ; in particular , @xmath58 $ ] is symmetric under the @xmath0 gauge transformation @xmath55 , and thus only the averages of quantities which are invariant under such transformations are nonzero in the microcanonical density . the @xmath0 phase symmetry of the classical - field hamiltonian @xmath59 is of course also a symmetry of the fundamental " second - quantized hamiltonian ( [ eq : h_fundamental ] ) , in which case it corresponds to the conservation of particle number under the action of the corresponding heisenberg equation of motion for @xmath60 @xcite . however , in traditional mean - field theories of bose condensation , the condensate is assumed to acquire a definite phase , _ breaking _ this symmetry @xcite . theories built on this assumption do not strictly conserve particle number , and the grand - canonical formalism is typically employed to ensure conservation of the _ mean _ particle number @xcite . condensation in the field is then associated with the appearance of anomalous moments ( moments of the field operator which are strictly zero in a state of fixed particle number ) such as the condensate mean - field @xmath61 , and the anomalous thermal density @xmath62 ( where @xmath63 ) . in ref . @xcite , it was demonstrated that although the @xmath0 symmetry of the hamiltonian @xmath59 is formally inherited by the microcanonical density eq . ( [ eq : mu_density ] ) , the classical - field solutions do exhibit a symmetry - breaking aspect , which allows for the calculation of anomalous averages such as @xmath64 and @xmath65 in analogy to symmetry - breaking approximations to the second - quantized field theory . an alternative approach to the theory of bose condensates that respects the @xmath0 symmetry of the quantum - field hamiltonian , and therefore conserves the number of atoms in the system , was presented by girardeau and arnowitt @xcite , and later rediscovered by gardiner @xcite and castin and dum @xcite . this approach is distinguished from the symmetry - breaking approaches in that no spurious phase is assumed for the condensate , and the fluctuations of the quantum field about the condensed mode are described in terms of modified ladder operators @xmath66\hat{a}_i$ ] ( and their hermitian conjugates ) @xcite , which demote a field quantum from a noncondensate mode @xmath67 into the condensate mode @xmath68 ( and _ vice versa _ ) . in effect , these operators serve to describe the excitations of the system in a picture in which the ( indeterminate ) phase of the condensate mode is canceled , providing a rigorous basis for the construction of a theory of the fluctuations around the condensate . such an approach has been used by several authors in developing theories of bose - einstein condensation at zero and finite temperature @xcite . in this article , we take an analogous approach to characterizing fluctuations of the classical field about the condensate mode . we introduce the fluctuation ( or noncondensate ) field @xmath69 where @xmath70 is the classical - field amplitude corresponding to the condensate mode @xmath54 , and @xmath71 is the component of the classical field orthogonal to the condensate mode . in terms of the poisson brackets @xcite defined as @xmath72,\ ] ] where the projected functional derivative operator @xmath73 @xcite , we easily find @xmath74\big\ } = 0.\ ] ] the field @xmath75 is therefore formally invariant under global rotations of the classical - field phase . thus , whereas moments of the noncondensate field which are not invariant under such rotations ( e.g. , @xmath76 ) necessarily vanish in the microcanonical ensemble due to the @xmath0 phase symmetry , the analogous moments of @xmath75 ( e.g. , @xmath77 ) , which we will refer to as _ anomalous _ moments of @xmath75 in the remainder of this article , may legitimately acquire nonzero values in the ensemble . in the limit of a perfectly coherent condensate , for which @xmath78 , it is clear that the first moment @xmath79 , as is appropriate for a fluctuation variable . it should be noted that more generally the vanishing of @xmath80 is not guaranteed _ a priori _ @xcite ; nevertheless we find @xmath81 to good accuracy in our simulations . we note that normal correlation functions [ those which are not anomalous ; e.g. , @xmath82 are _ manifestly _ real , i.e. , it is clear _ a priori _ that complex conjugation has no effect on such correlation functions . anomalous correlation functions , by contrast , are not fundamentally constrained to be real in this manner . however , we find that the anomalous correlation functions we consider in this article are predominantly real quantities ( relative to a real condensate orbital ) , as a result of detailed balance in the equilibrium field . our procedure for calculating correlation functions is as follows : we form random initial states @xcite with prescribed classical - field energies . after evolving these initial states to equilibrium , we construct the one - body density matrix [ eq . ( [ eq : obdm ] ) ] by ergodic averaging of the field trajectory . the bulk features and simplest correlations of the field equilibrate on a time scale of @xmath83 @xcite ; however , we do not begin our ergodic averaging of the field trajectory until after an equilibration period of @xmath84 , to ensure that all moments of the field have settled down to their equilibrium values . we then form ergodic averages from @xmath85 equally spaced samples of the classical field taken over a subsequent period of @xmath86 of the field evolution . we diagonalize @xmath49 to find the condensate orbital @xmath54 and its mean occupation @xmath53 . for the irrotational system we consider here , the condensate orbital has ( aside from small numerical fluctuations ) a uniform phase . before proceeding we absorb the overall phase of @xmath54 into a global ( time - independent ) phase shift of the field trajectory used to construct the microcanonical ensemble . this amounts to making a choice of gauge such that the condensate orbital is real and positive , and in terms of this orbital we define the condensate wave function @xmath87 . from each member of the ensemble we then form @xmath75 according to eq . ( [ eq : lambda_defn ] ) , and we average products of @xmath75 over the ensemble to form its moments . we note that eq . ( [ eq : lambda_defn ] ) is singular when the overlap of the classical field @xmath45 with the condensate orbital @xmath54 vanishes . however , the configurations of the field for which @xmath88 are a set of measure zero in the microcanonical density , and in practical simulations we always find a finite value for @xmath89 , even in the high - temperature regime where the mode @xmath54 is incoherent and undergoes large number fluctuations @xcite . in this section we characterize the fluctuations of the noncondensate field by calculating the local second and third moments , and fourth cumulants @xcite , of the fluctuation field @xmath75 . in terms of the @xmath0-symmetry preserving approach to calculating correlations we adopt in this article , we find that the field @xmath75 exhibits fluctuations which are both anomalous ( representing pairing effects induced by the condensate ) , and non - gaussian ( exhibiting deviations from the gaussian ansatz for fluctuations assumed in the hfb theories @xcite ) . these nontrivial correlations of the field can be related to corrections to the mean - field potentials experienced by the condensed and noncondensed atoms , due to many - body effects ( see e.g. , refs . @xcite ) . we focus here on a representative pgpe equilibrium , with energy @xmath90 , and corresponding _ condensate - band _ condensate fraction @xmath91 . this equilibrium is therefore a reasonably high - temperature state of the field , well above the validity regime of the simple bogoliubov description @xcite of the noncondensate , while remaining far from the critical regime associated with the transition to the normal state @xcite . there are two independent quadratic moments of the noncondensate field : the normal covariance matrix ( or noncondensate density matrix ) @xmath92 , and the anomalous covariance matrix ( pair matrix ) @xmath93 . all other quadratic moments can be related to these matrices by transposition and complex conjugation . the spectral representation of the classical field allows us to calculate the full off - diagonal structure of these matrices , however , in this article we consider only their diagonal elements : the normal thermal density @xmath94 and the anomalous thermal density @xmath95 we note that in fact @xmath96 , i.e. , this quantity can be defined perfectly well without recourse to the considerations of sec . [ subsec : gauge_invariance ] . by contrast , the anomalous density @xmath97 is distinct from the moment @xmath98 , which of course vanishes in the microcanonical ensemble . the physical origin of @xmath97 can be inferred by assuming that a factor @xmath99 can be factored out of the expectation value eq . ( [ eq : kappa_defn ] ) [ which should be a very good approximation for a coherent condensate mode @xmath54 ] , whereby we find that @xmath101 . the anomalous density thus quantifies the correlations of pairs of noncondensate atoms with pairs of condensate atoms due to the ( classical wave - mixing analog of the ) bogoliubov pair - promotion process in which two condensate atoms scatter each other out of the condensate ( and the time - reversed process ) , which is responsible for the well - known bogoliubov particle - hole structure of excitations in the system @xcite . the anomalous density can also be interpreted as a measure of the squeezing of the noncondensate field fluctuations @xcite . in fig . [ fig : cumu_spatial](a ) we plot azimuthally averaged values of the thermal density @xmath102 ( solid line ) and the anomalous thermal density @xmath97 ( dashed line ) on the @xmath103 plane . ( color online ) correlations of the fluctuation field @xmath75 in a representative pgpe equilibrium with energy @xmath90 : ( a ) second moments of @xmath75 ( normal and anomalous thermal densities ) . ( b ) third moments and ( c ) fourth cumulants of @xmath75.,scaledwidth=45.0% ] the condensate density @xmath104 ( dot - dashed line ) is plotted for comparison [ the peak condensate density @xmath105 . the thermal density reaches its maximum in the periphery of the condensate , as is well known from mean - field theories ( cf . , for example , ref . the anomalous thermal density @xmath97 is real and negative to within statistical uncertainty due to the finite size of the ensemble @xcite , as expected , since in equilibrium the bogoliubov pair - promotion process must be balanced by the corresponding time - reversed process . we note that the anomalous density is almost entirely located within the extent of the condensate mode , and its magnitude exhibits a small dip around the trap center , in agreement with the results of previous studies @xcite . we now turn our attention to the third moments of @xmath75 . there are two independent third moments ( or triplets @xcite ) of @xmath75 . in this article we only consider their diagonal elements , and we define @xmath106 and @xmath107 in fig . [ fig : cumu_spatial](b ) we plot the functions @xmath108 ( solid line ) and @xmath109 ( dashed line ) , evaluated on the @xmath103 plane and azimuthally averaged . like the function @xmath97 , both these moments are anomalous , but we find that they are real to within statistical uncertainty . both functions reside primarily in the central region where the condensate exists and , like @xmath102 and @xmath97 , each exhibits its maximal absolute value at some distance from the origin . although similar correlations have been discussed in the literature previously @xcite , we are not aware of any calculated forms for these functions with which to compare the results presented here . we stress that the expectation values @xmath108 and @xmath109 vanish implicitly in the self - consistent mean - field ( hfb and hfb - popov ) theories , and their appearance here is a signature of the non - gaussian nature of the fluctuation field @xmath75 . the physical origin of the correlator @xmath108 can be inferred similarly to that of @xmath97 : assuming a factor @xmath110 can be factored out of the expectation value eq . ( [ eq : lambda_defn ] ) , we find that @xmath111 . the function @xmath108 is thus associated with processes in which two thermal atoms collide and one of them is scattered into the condensate ( i.e. , condensate _ growth _ processes @xcite ) . in equilibrium this process is balanced by the time - reversed process , which is consistent with the real value of @xmath108 we find here , and represents the detailed balance of beliaev and landau processes ( see , e.g. , ref . @xcite ) in the noncondensate , which we emphasize are not included in the hfb and hfb - popov treatments @xcite . the correlation function @xmath109 is less straightforward to interpret @xcite ; however , analogous correlations are found to enter into the equations of motion for the pair matrix and the correlation function @xmath108 in generalized mean - field treatments of the finite - temperature bose - gas dynamics @xcite . the appearance of the nonzero third moments @xmath108 and @xmath109 shows that the fluctuations of the noncondensate field are not strictly gaussian . however , as the classical field we consider is weakly interacting , the fluctuations of @xmath75 should be reasonably _ close _ to gaussian . we therefore expect the fourth moments of the noncondensate field to be approximately equal to their naive gaussian ( wick ) factorizations @xcite in terms of the second moments of @xmath75 ; e.g. , @xmath112 the differences between the actual fourth moments and their approximate gaussian factorizations are quantified by the fourth cumulants of the field @xcite . there are three independent fourth cumulants of @xmath75 , corresponding to the three independent fourth moments of @xmath75 . we consider here only the diagonal elements of these cumulants : @xmath113 in fig . [ fig : cumu_spatial](c ) we plot the azimuthally averaged values of @xmath114 ( solid line ) , @xmath115 ( dashed line ) , and @xmath116 ( dot - dashed line ) on the @xmath103 plane . we note that these cumulants are small compared to the corresponding fourth moments . for example , from fig . [ fig : cumu_spatial](a ) it can be inferred that the maximal value of the moment @xmath117 is @xmath118 ; i.e. , @xmath119 times the maximal value of @xmath120 . we note that the fourth cumulants are somewhat more `` noisy '' than the lower correlations of the field ; it is of course intuitively clear that the statistical demands for the accurate evaluation of cumulants will increase with the order of the cumulant . the cumulants presented here do , however , indicate that the fluctuations of @xmath75 exhibit a clear deviation from gaussianity . of particular interest is the cumulant @xmath120 , as this measures the deviation of the fourth moment @xmath117 from its wick factorization [ eq . ( [ eq : wick ] ) ] . this factorization becomes exact in the limit that the noncondensate fluctuations are gaussian , and motivates the approximate factorizations of field - operator products made in mean - field theories in order to reduce the second - quantized hamiltonian to a ( self - consistent ) quadratic form @xcite . in fact , morgan @xcite has shown that the treatment of the hamiltonian term quartic in field operators in his second - order self - consistent calculation is equivalent to the wick factorization employed in the hfb theories . effects which lead to the appearance of a nonzero fourth cumulant @xmath114 at equilibrium thus enter at a higher order of perturbation theory . we note that the cumulant @xmath120 found here is negative , which can be understood intuitively : the field admits non - gaussianity of its fluctuations so as to lower the interaction energy associated with a product of four noncondensate operators below the level obtained in self - consistent mean - field approaches . we now relate the anomalous and non - gaussian correlations of the classical field discussed in sec . [ subsec : cumulants ] to many - body interaction effects in the field which provide corrections to the mean - field potentials experienced by the condensate and noncondensed atoms . it is well known that the appearance of the anomalous average in the generalized gross - pitaevskii equation of the hfb theory introduces many - body interaction effects which modify the collisions between condensate atoms @xcite . the origin of this effect is the bogoliubov pair - excitation process ( sec . [ subsec : cumulants ] ) , which introduces the possibility that two colliding condensate atoms are both scattered into ( possibly occupied ) noncondensate modes , whereafter they may scatter off one another an arbitrary number of times before returning to the condensate . in general , the entire series of such ladder diagram " processes @xcite contributes to the effective interaction between condensate atoms in the finite - temperature system . in the hfb theory @xcite , it is found that the condensate @xmath121 obeys a generalized gross - pitaevskii equation of the form @xmath122\phi_0 + u_0\kappa\phi_0^ * = 0 $ ] , and indeed this form is also obtained in the formalism of ref . the analyses of refs . @xcite show that the term @xmath123 corresponds to the introduction of the many - body @xmath1 matrix in the description of condensate - condensate interactions . we therefore reformulate this term as an additional potential experienced by the condensate , which then obeys the nonlinear eigenvalue relation @xmath124\phi_0=0 $ ] . the mean - field potential experienced by the condensate due to its own self - interaction @xmath125 thus receives a correction @xcite @xmath126 ^ 2\langle\lambda(\mathbf{x})\lambda(\mathbf{x})\rangle\big\}.\ ] ] in fig . [ fig : mf_potls](a ) we plot the bare mean - field potential @xmath127 ( blue / black solid line ) , the many - body correction @xmath128 ( dashed line ) , and their sum , the corrected condensate mean - field potential @xmath129 ( dot - dashed line ) . ( color online ) many - body interaction effects in the classical - field equilibrium : ( a - b ) mean - field potentials experienced by the condensate and their many - body corrections . ( c ) local chemical potential of the classical - field condensate , in varying degrees of approximation ( see text ) . ( d ) effective interaction strengths deduced from the correlations of @xmath75 . gray solid lines in ( b ) and ( c ) indicate the spatial density profile of the condensate.,scaledwidth=45.0% ] we indeed find that the presence of the anomalous average corresponds to a noticeable reduction of the mean - field potential @xmath130 . similarly to the mutual interaction of condensate atoms , the scattering of condensate atoms by noncondensate atoms is also affected by many - body processes in the finite - temperature bose field : upon its interaction with a noncondensed atom , a condensate atom may be scattered out of the condensate , and undergo an arbitrary number of ( repeated ) interactions with the noncondensate atom before returning to the condensate . the consideration of all such processes amounts to the inclusion of all ladder diagrams in the condensate - noncondensate interaction . in an analysis beyond the usual mean - field ( hfb ) approach , proukakis _ et al . _ have shown @xcite that the condensate obeys an equation of motion of the form @xmath131 , where @xmath132 is the ( time - dependent ) triplet correlator corresponding to eq . ( [ eq : lambda_defn ] ) . the adiabatic elimination of @xmath132 from this equation of motion leads to the introduction of the many - body @xmath1-matrix in the description of the condensate - noncondensate interactions , similar to the role of @xmath97 in the generalized gpe of the hfb theory . the term @xmath133 can therefore be reformulated as an additional mean - field potential experienced by the condensate , yielding the equation of motion @xmath134\phi_0 $ ] . the bare mean - field potential experienced by condensate atoms , due to the presence of the noncondensate , @xmath135 thus receives a many - body correction given by @xcite @xmath136 it is important to note that while the correction eq . ( [ eq : dvc ] ) arises due to the bogoliubov processes which are accounted for to all orders by the bogoliubov particle - hole structure of excitations in the hfb theories @xcite , the correction of eq . ( [ eq : dvnc ] ) corresponds to processes in which atoms are exchanged between the condensate and its excitations , which are not described in such theories . in fig . [ fig : mf_potls](b ) we plot the bare potential @xmath137 ( blue / black solid line ) , the correction @xmath138 ( dashed line ) and the corrected potential @xmath139 ( dot - dashed line ) . we find that the correction is localized near the boundary of the condensate , where the noncondensate density is maximal , however , it is nonzero at smaller radii also . again , we see the effect of the many - body correction is to weaken the mean - field potential experienced by the condensate . the properties of a superfluid are intimately connected to the _ phase _ of the order parameter ( i.e. , the condensate ) , as the superfluid velocity is proportional to the gradient of the order - parameter phase : @xmath140 @xcite . the chemical potential of the superfluid is , in general , a spatially varying quantity , corresponding to ( @xmath141 times ) the phase - rotation frequency of the condensate @xcite . a gradient in the chemical potential therefore corresponds to an acceleration of the superflow , and a stationary superfluid should exhibit a spatially uniform chemical potential @xcite . here we define a local condensate chemical potential corresponding to the action of the aforementioned ( generalized ) mean - field operator @xmath142 on the penrose - onsager condensate : @xmath143/\phi_0(\mathbf{x } ) + u_0\big[|\phi_0(\mathbf{x})|^2 + 2\langle|\lambda(\mathbf{x})|^2\rangle\big ] \nonumber \\ & + \frac{u_0}{|\phi_0(\mathbf{x})|^2}\mathrm{re}\big\{[\phi_0^*(\mathbf{x})]^2\langle \lambda^2(\mathbf{x})\rangle + \phi_0^*(\mathbf{x})\langle \lambda^*(\mathbf{x})\lambda^2(\mathbf{x})\rangle\big\}. \end{aligned}\ ] ] we note that this local chemical potential is analogous to that derived by zaremba _ et al . _ @xcite within a symmetry - breaking framework . by evaluating @xmath144 and approximations to it that neglect the many - body correction terms appearing in eq . ( [ eq : local_mu ] ) , we can assess the importance of these terms . specifically , in addition to the full form of @xmath144 , we consider the chemical potential obtained by evaluating only those quantities on the first line of eq . ( [ eq : local_mu ] ) [ which we will refer to as the popov " chemical potential , @xmath145 , and that obtained by evaluating all but the last term of eq . ( [ eq : local_mu ] ) [ the hfb " chemical potential , @xmath146 . we calculate the action of @xmath147 on the condensate @xmath121 in the spectral representation , and azimuthally average the resulting quantity [ @xmath148 , and the individual factors of the remaining terms on the rhs of eq . ( [ eq : local_mu ] ) , on the @xmath103 plane in order to improve the estimates of these quantities and obtain radial representations of @xmath144 , @xmath149 , and @xmath150 . the results are presented in fig . [ fig : mf_potls](c ) , along with the effective eigenvalue @xmath151 of the condensate orbital , obtained from the temporal analysis of ref . we note first that the popov - level chemical potential @xmath149 ( thick solid black line ) is somewhat nonuniform and overestimates the condensate eigenvalue @xmath151 ( thin solid black line ) over the extent of the condensate mode [ where the concept of @xmath144 is meaningful ] . the hfb chemical potential @xmath150 ( dashed line ) is smaller than @xmath149 throughout the extent of the condensate mode , and approaches @xmath151 much more closely at the smallest radii , while increasingly overestimating @xmath151 at increasing radii @xcite . by contrast , the full chemical potential @xmath144 ( dot - dashed line ) is much more uniform and in good agreement with the eigenvalue @xmath151 aside from small discrepancies at the inner and outer extremes of the condensate orbital . this result shows that the condensate orbital obtained from the penrose - onsager analysis is consistent with a mean - field picture of the condensate and the complementary thermal component of the field , _ provided _ that the corrections to the mean - field potentials due to many - body effects are taken into account . this explains the essentially uniform phase rotation of the ( penrose - onsager ) condensate mode discussed in ref . @xcite , and moreover demonstrates the consistency of the equilibrium classical - field formalism with other formulations of the finite - temperature bose - gas problem ( e.g. , ref . we stress that the quantities @xmath149 and @xmath150 do not correspond to the condensate chemical potentials that would be obtained in formal calculations using the corresponding self - consistent mean - field formalisms ; such calculations would of course describe a condensate mode with a uniform chemical potential . nevertheless , the results presented here give an indication of the order of the error involved in calculations employing such approaches . we now reconsider the corrections to the mean - field potentials experienced by the condensate ( and noncondensate ) in terms of effective interaction strengths , which feature in the theory introduced in refs . @xcite , and reviewed in ref . briefly , this so - called gapless - hfb " ( ghfb ) theory is based on the identification that the inconsistency of the hfb theory arises because the theory introduces many - body corrections to the condensate - condensate scattering , but that effects of this order ( i.e. , the many - body @xmath1-matrix approximation @xcite ) are not included in the description of interactions between the condensate and its excitations . the ghfb theory ( of which there are two variants @xcite ) thus introduces spatially dependent _ effective _ interaction strengths into the hfb - popov theory , so as to approximate the effects of the many - body @xmath1-matrix corrections . in fig . [ fig : mf_potls](d ) we plot three effective interaction potentials deduced from the results of our classical - field simulation . we define a condensate - condensate effective interaction by grouping the condensate - condensate mean - field potential @xmath127 with its many - body correction @xmath128 , and absorbing the effect of the latter into a spatially dependent redefinition of coupling constant @xcite : @xmath152 making the substitution @xmath153 in the bare condensate mean - field potential @xmath127 [ eq . ( [ eq : vc0 ] ) ] we obtain ( by definition ) the corrected potential @xmath129 . the effective interaction @xmath154 plotted in fig . [ fig : mf_potls](d ) ( solid line ) thus provides an alternative visualization of the correction to the mean - field potential presented in fig . [ fig : mf_potls](a ) , and is qualitatively consistent with the effective interactions presented in refs . @xcite . next , we define an effective interaction appropriate to condensate - noncondensate scattering by absorbing the many - body correction @xmath138 to the mean - field potential @xmath137 experienced by the condensate , due to the presence of the noncondensate , into a local coupling constant @xmath155 making the substitution @xmath156 in @xmath137 [ eq . ( [ eq : vnc0 ] ) ] clearly yields the corrected mean - field potential @xmath139 . we observe that @xmath157 [ dashed line in fig . [ fig : mf_potls](d ) ] , like @xmath158 , is suppressed below the uniform value of the bare interaction potential @xmath11 , and that this suppression is most pronounced around @xmath159 ( i.e. , the location of the edge of the condensate and the peak of the thermal - cloud density ) . however , the suppression exhibited by @xmath157 is significantly less than that of @xmath158 , suggesting that the so - called _ g2 _ variant of the ghfb theory @xcite , in which @xmath11 is replaced by @xmath158 everywhere it appears in the gross - pitaevskii and bogoliubov - de gennes equations , significantly overestimates the suppression of the noncondensate mean - field potential experienced by the condensate . by contrast , the _ g1 _ variant of the theory retains the unmodified coupling constant @xmath11 in the mean - field potential @xmath160 , and thus neglects any suppression of this potential . finally in this section , we introduce an effective interaction strength for noncondensate - noncondensate interactions . to define this quantity , we recall that the appearance of a negative fourth cumulant @xmath161 implies that the noncondensate - noncondensate interaction energy @xmath162 is suppressed somewhat below the value obtained by the gaussian factorization of the moment @xmath163 assumed in the hfb theory ( sec . [ subsec : cumulants ] ) . by assuming that the difference between @xmath164 and the corresponding approximate gaussian expression can be accounted for by an effective ( spatially - dependent ) interaction strength associated with the mean - field potential due to noncondensate atoms , we identify @xmath165 we observe that this quantity [ dot - dashed line in fig . [ fig : mf_potls](d ) ] , like @xmath158 and @xmath157 , describes interactions which are suppressed most strongly around the condensate periphery . however , the maximal level of suppression here is of the order of @xmath166 , smaller again than that in the case of condensate - noncondensate scattering . the small value of this correction is to be expected , as eq . ( [ eq : uncnc ] ) represents an average over all ( coherent - region ) excited - state collisions , many of which take place at high energies , where the many - body @xmath1 matrix reduces to the two - body @xmath1 matrix @xcite . in this section , we consider the dependence of the noncondensate correlation functions and interaction potentials described in sec . [ sec : correlations ] on the total energy ( and thus temperature ) of the classical - field equilibrium . in fig . [ fig : cumu_vs_e ] we plot the spatially integrated values of correlation functions of the noncondensate field , obtained from pgpe equilibria with energies in the range @xmath41n_\mathrm{c}\hbar\omega_r$ ] . ( color online ) dependence of thermodynamic parameters and noncondensate correlations on the classical - field energy . ( a ) temperature and chemical potential of the field . magnitudes of spatially integrated ( b ) second moments , ( c ) third moments , and ( d ) fourth cumulants of @xmath75 . inset : dependence of the binder cumulant @xmath167 on the field energy . the solid line interpolates smoothly between the numerical data points ( circles ) . the dashed lines indicate the critical value @xmath168 and the corresponding critical field energy ( see text).,scaledwidth=45.0% ] the temperature and chemical potential of the field , obtained using the microcanonical thermodynamic formalism of rugh @xcite , are presented in fig . [ fig : cumu_vs_e](a ) for reference . in the inset to fig . [ fig : cumu_vs_e](a ) , we plot the quantity @xmath169 as a function of the field energy . bezett and blakie @xcite have suggested this quantity as the appropriate generalization of the _ binder cumulant _ associated with the phase transition of the homogeneous gas @xcite to the harmonically trapped case . et al . _ @xcite found the critical value @xmath168 at the phase transition of the classical three - dimensional xy model , and this value was used to identify the critical temperature in pgpe simulations of the homogeneous bose gas in ref . @xcite . the solid line in the inset to fig . [ fig : cumu_vs_e](a ) interpolates between the values of @xmath167 obtained from our simulations ( circles ) , and intersects the critical value @xmath170 ( horizontal dashed line ) at @xmath171 ( vertical dashed line ) , which we take as an estimate of the critical field energy for this system . the critical energy is indicated by a vertical dashed line in fig . [ fig : cumu_vs_e](a - d ) . figure [ fig : cumu_vs_e](b ) shows the spatially integrated normal thermal density @xmath172 ( circles ) and anomalous thermal density @xmath173 ( squares ) . the results are similar to those obtained in @xcite , in which the anomalous correlation function @xmath97 was defined in terms of a symmetry - breaking interpretation of the classical - field trajectories . it is important to note that the noncondensate population increases approximately linearly with energy ( and thus temperature ) up to the transition of the field to the noncondensed phase in the idealized pgpe system " consisting of a fixed field population distributed over the modally finite coherent region @xcite . upon the consideration of the above - cutoff fraction of atoms , one obtains the expected , geometry - dependent scaling of the ( non-)condensate fraction with temperature ( see , for example , ref . @xcite ) . by contrast , the integrated anomalous density reaches its maximum absolute value at intermediate energies ( temperatures ) , and vanishes as the system approaches the phase transition , in agreement with previous studies @xcite . the integrated third moments @xmath174 and @xmath175 are plotted in fig . [ fig : cumu_vs_e](c ) . the quantity @xmath175 reaches its maximum at intermediate energies , and vanishes along with @xmath173 at @xmath176 . the quantity @xmath174 , however , remains nonzero at the transition , and tends to zero at a somewhat higher energy . we reiterate that although the definition [ eq . ( [ eq : lambda_defn ] ) ] of @xmath75 is singular when @xmath88 , this condition does not occur in practical simulations @xcite . nonzero values for @xmath174 are obtained at the phase transition as @xmath108 represents the scattering of atoms into and out of the condensate " mode @xmath54 , which persists in the critical regime , even though the mode @xmath54 is no longer coherent . at higher temperatures , where critical fluctuations subside and modes recover gaussian fluctuations , @xmath108 vanishes . in fig . [ fig : cumu_vs_e](d ) we plot the integrated values of the fourth cumulants of the noncondensate field @xmath75 . the cumulant @xmath116 behaves similarly to @xmath97 , as expected , given that this cumulant quantifies the difference between the purely anomalous ( condensate - induced ) moment @xmath177 and its naive factorization in terms of @xmath97 [ see eq . ( [ eq : cumu_def3 ] ) ] . similarly to @xmath108 , the integrated value of @xmath178 vanishes at a slightly higher temperature . from the definition [ eq . ( [ eq : cumu_def2 ] ) ] of this cumulant , it is clear that this behavior reflects that of the moment @xmath179 , which can be understood in a similar manner to the behavior of @xmath108 . by contrast , the normal " cumulant @xmath161 reaches its _ maximum _ absolute value around the transition to the normal phase . this shows that many - body interaction effects are most important in the noncondensate field in this regime , in which the condensate is surrounded by a much larger region of the field which exhibits suppressed density fluctuations but no phase coherence ( i.e. , a _ quasicondensate _ @xcite ) , as we discuss further in sec . [ sec : suppression ] . we now consider the variation of the many - body interaction effects identified in sec . [ subsubsec : effective_int ] as functions of the field energy . we introduce three interaction energies : @xmath180 corresponding to the condensate - condensate , condensate - noncondensate and noncondensate - noncondensate interactions , respectively . we compare these three energies with the corresponding uncorrected energies , obtained from eqs . ( [ eq : ecc])-([eq : encnc ] ) by replacing the spatially dependent interaction strengths [ @xmath154 , etc ] with the bare interaction strength @xmath11 , which we denote by @xmath181 , etc . in fig . [ fig : corrections_vs_e ] we plot the corrected interaction energies eqs . ( [ eq : ecc])-([eq : encnc ] ) as fractions of the corresponding uncorrected energies . the approximate location of the phase transition determined from the analysis of the binder cumulant @xmath167 ( sec . [ subsec : e_dep_c_fns ] ) is again indicated by a vertical dashed line . the absolute magnitudes of the corrected energies are shown in the inset to fig . [ fig : corrections_vs_e ] . ( color online ) dependence of the corrected interaction energies on the energy of the classical - field equilibrium . interaction energies are plotted as fractions of uncorrected interaction energies ( see text ) . the absolute values of the interaction energies are shown in the inset.,scaledwidth=45.0% ] considering the condensate - condensate interaction energy @xmath182 ( circles in fig . [ fig : corrections_vs_e ] ) , we note that it continues to decrease ( relative to @xmath181 ) as @xmath183 increases towards the transition , and reaches its minimum at @xmath184 . we contrast this with the behavior of @xmath173 [ fig . [ fig : cumu_vs_e](b ) ] , which reaches its maximum ( absolute ) value at @xmath185 . at energies @xmath186 , where no condensate is present [ i.e. , the mode @xmath54 is completely incoherent ] , the expression eq . ( [ eq : ecc ] ) is meaningless , and so we do not present data for @xmath182 at these highest energies . we find that the total condensate - condensate interaction energy is suppressed by @xmath187 at the transition , which is consistent with previous mean - field studies of harmonically trapped condensates @xcite , and should be contrasted with the complete vanishing of the condensate - condensate interaction at the transition temperature in the homogeneous case @xcite . turning our attention to the energy @xmath188 of the condensate - noncondensate interaction ( triangles in fig . [ fig : corrections_vs_e ] ) , we observe that @xmath189 , like @xmath190 , reaches its minimum ( @xmath191 ) around the transition to the normal phase . at higher energies ( temperatures ) , the interaction returns to its bare value ; in this limit , the condensate " mode @xmath54 is simply a thermal ( gaussian ) mode ( i.e. , @xmath192 ) , and eq . ( [ eq : ecnc ] ) reduces to the hartree - fock interaction energy between this mode and the rest of the field . the energy of the noncondensate - noncondensate interaction ( squares ) is similarly suppressed most strongly at around @xmath184 , but is subject to less suppression than the other two interaction energies . we note the reasonable agreement between the critical energy estimated from the consideration of the binder cumulant , and the energy at which maximal suppression of the field interactions occurs , which provides an independent estimate of the critical point ( see ref . @xcite and references therein ) . this suggests that the critical value @xmath168 appropriate to the phase transition of the homogeneous bose gas does indeed yield a good estimate of the transition energy ( temperature ) in the harmonically trapped system . this is in contrast to the results of ref . @xcite which suggested , based on an analysis of the field correlation length , that the critical value of @xmath167 in the harmonically trapped system may be significantly lower than that in the homogeneous case . [ indeed the results presented in figs . [ fig : cumu_vs_e ] and [ fig : corrections_vs_e ] of this work suggest that @xmath170 slightly _ underestimates _ the correct critical value ] . finally in this section , we note that the interaction energies considered here constitute an increasingly small fraction of the total field energy as the field energy ( temperature ) increases ( inset to fig . [ fig : corrections_vs_e ] ) . in particular , in the critical region , the noncondensate - noncondensate interaction energy is @xmath193 of the total field energy , and the remainder is mostly comprised of the kinetic energy of the noncondensate fraction . at the transition , the contribution of beyond - gaussian fluctuations to the total field energy is therefore only of order @xmath194 ( cf . the agreement of hfb - popov and pgpe calculations of the critical temperature to about this level reported in ref . we have seen in sections [ sec : correlations ] and [ sec : e_dependence ] that the classical - field equilibria exhibit a negative fourth cumulant @xmath161 , corresponding to the suppression of density fluctuations in the noncondensate field @xmath75 below the gaussian level assumed in the hfb and hfb - popov formalisms . we now consider the suppression of density fluctuations in the total classical field @xmath27 , and identify and compare the various contributions to this suppression . we follow ref . @xcite in considering the correlator @xmath195 and defining the _ quasicondensate _ density @xmath196^\frac{1}{2}\langle|\psi(\mathbf{x})|^2\rangle,\ ] ] where the classical coherence function @xmath197 @xcite . the quasicondensate density @xmath198 has been used by bisset _ _ @xcite to characterize the superfluid transition of the quasi - two - dimensional bose gas ( see also ref . we recall that the local correlation function @xmath199 adopts values of @xmath200 and @xmath201 for normal " thermal fields ( those without anomalous correlations ) and purely coherent fields , respectively @xcite . in fig . [ fig : quasicond](a ) we plot the ( azimuthally averaged ) quasicondensate density @xmath202 for the pgpe equilibrium with energy @xmath203 ( blue / black solid line ) , along with the condensate density @xmath204 ( gray solid line ) and total field density @xmath205 ( dashed line ) . ( color online ) suppression of density fluctuations in the classical field . ( a ) condensate , quasicondensate , and total density profiles in a representative classical - field equilibrium with energy @xmath203 . ( b ) dependence of the total condensate and quasicondensate populations on the field energy . ( c ) dominant contributions to the quasicondensation correlator @xmath206 in the case @xmath203 . ( d ) dependence of the dominant contributions to @xmath206 on the field energy.,scaledwidth=50.0% ] we observe that the quasicondensate density @xmath202 is everywhere greater than the condensate density , and in particular , @xmath202 remains significant at radii @xmath207 , for which the condensate density essentially vanishes . in fig . [ fig : quasicond](b ) we plot the integrated condensate density , i.e. , the condensate population @xmath208 ( circles ) , and the integrated quasicondensate density @xmath209 ( squares ) , as functions of the classical - field energy @xmath183 . we find that the quasicondensate population is greater than the condensate population for all energies considered , and that the difference between the two populations increases with increasing @xmath183 . most notably , in the highest energy equilibria considered ( @xmath210 ) , in which no condensation remains , the quasicondensate population is still significant . the presence of this quasicondensate at high energies ( i.e. , in the absence of condensation ) clearly corresponds to the significant fourth cumulant @xmath161 found in this regime ( sec . [ subsec : e_dep_c_fns ] ) . by expressing @xmath205 and @xmath211 in terms of @xmath212 and @xmath213 , substituting @xmath214 and assuming that the condensate can be factored out of these moments [ e.g. @xmath215 and @xmath216 , we obtain the approximate form @xcite @xmath217 where the binder cumulant @xmath167 accounts for the amplitude fluctuations of the condensate mode ( see sec . [ subsec : e_dep_c_fns ] ) . we note that only by neglecting all but the first term on the rhs of eq . ( [ eq : q_approx ] ) do we obtain ( in the limit @xmath218 ) @xmath219 ( cf . ref . @xcite ) . the second term on the rhs of eq . ( [ eq : q_approx ] ) represents the _ enhancement _ of density fluctuations in the noncondensate component relative to the normal gaussian [ @xmath220 level , due to the anomalous nature of its fluctuations ( consider that @xmath221 ) @xcite , and thus subtracts from @xmath206 . in practice we find that the contribution of this term is small . the remaining terms on the rhs of eq . ( [ eq : q_approx ] ) all represent reductions in the density fluctuations of the field @xmath27 , and thus add to the quasicondensate density . in fig . [ fig : quasicond](c ) we plot the ( azimuthally averaged ) quantity @xmath206 for the case @xmath203 ( solid blue / black line ) , along with the three dominant contributions to this quantity : @xmath222 ( solid gray line ) , @xmath223 ( dashed line ) and @xmath224 ( dash - dot line ) . the contributions of @xmath225 and @xmath161 are essentially negligible for this field energy , and the suppression of density fluctuations here is therefore primarily associated with the presence of a condensate in the field . however , the extent to which density fluctuations are suppressed is somewhat greater than that due to the absence of density fluctuations in the condensate itself . this suggests that an estimation of the condensate density based on an experimental measurement of @xmath226 could overestimate the size of the condensate considerably . we plot the spatially integrated values of the correlator @xmath206 and the dominant contributions to it as functions of the field energy in fig . [ fig : quasicond](d ) . we observe that the contribution of @xmath227 ( triangles ) is the dominant beyond - condensate contribution to @xmath206 at the lowest energies , but reaches its maximum magnitude at @xmath228 , and is overtaken by the contribution of @xmath224 ( pluses ) at higher energies . this contribution is itself overtaken soon after by the contribution of the fourth cumulant @xmath161 ( crosses ) , which is of course the dominant contribution in the critical regime . the results presented here should be compared with the behavior of the homogeneous bose gas in two dimensions , for which it is found that the quasicondensate density is in general greater than the superfluid density , and that the quasicondensate correlations appear at temperatures well above the critical temperature for the appearance of superfluidity in the system @xcite . classical - field calculations for the harmonically trapped quasi - two - dimensional bose gas appear to be consistent with a local - density version of this picture @xcite . it is perhaps not surprising that we observe similar behavior in the three - dimensional case considered here : the comparatively large transverse susceptibility of the field @xcite leads to the ( local ) destruction of order by phase fluctuations before size fluctuations become significant @xcite . we have presented a methodology for the calculation of general spatial correlations of the finite - temperature bose gas , including anomalous correlations , in a classical - field approach . we introduced @xmath0-symmetric classical - field variables , analogous to number - conserving quantum ladder operators considered by other authors @xcite , in order to rigorously quantify the anomalous correlations in the microcanonical pgpe ensemble . we have demonstrated that the finite - temperature bose field exhibits non - gaussian correlations which are classical in origin , and by relating these correlations to many - body corrections to the mean - field interaction potentials @xcite , we have explicitly demonstrated that the classical - field theory provides an intrinsic nonperturbative description of many - body processes in the field . we discussed the role of these processes in determining the condensate mode , and demonstrated the consistency of the penrose - onsager @xcite definition of condensation in the classical - field equilibrium . finally , we elucidated the contribution of the anomalous and non - gaussian correlations of the noncondensate to the overall suppression of density fluctuations in the field , and discussed the distinction between the quasicondensate defined by the suppression of density fluctuations , and the true condensate defined by the suppression of phase fluctuations . our demonstration of the intrinsic description of many - body interaction effects provided by the classical - field method underlines the utility of these techniques in describing low - dimensional systems @xcite , in which such effects can have profound consequences for the structure of the field equilibrium @xcite . indeed our results illustrate the complexities introduced by the independence of phase and density fluctuations @xcite in the inhomogeneous finite - temperature bose gas even in the comparatively straightforward three - dimensional case . moreover , although we have only considered systems at equilibrium , our results demonstrate that the classical - field model of the low - energy coherent region does describe the higher - order processes associated with quasiparticle damping , and the exchange of atoms between the condensate and its excitations , which are essential for the description of nonequilibrium dynamics of the finite - temperature bose field @xcite . while making revisions to this article , the authors were saddened to learn of the passing of professor allan griffin , and would like to acknowledge discussions and correspondence with him that have helped to shape the ideas presented here . n.p.p . would like to thank the university of queensland node of the arc centre of excellence for quantum - atom optics for hospitality during his visit , and the uk epsrc for financial support . this work was supported by the australian research council through the arc centre of excellence for quantum - atom optics ( ce0348178 ) . m.j.d . is the recipient of an arc qeii fellowship ( project dp1094025 ) .

the aim of this paper is to review a new perspective about decoherence , according to which formalisms originally devised to deal just with closed or open systems can be subsumed under a closed - system approach that generalizes the traditional account of the phenomenon . this new viewpoint dissolves certain conceptual difficulties of the orthodox open - system approach but , at the same time , shows that the openness of the quantum system is not the essential ingredient for decoherence , as commonly claimed . moreover , when the behavior of a decoherent system is described from a closed - system perspective , the account of decoherence turns out to be more general than that supplied by the open - system approach , and the quantum - to - classical transition defines unequivocally the realm of classicality by identifying the observables with classical - like behavior .

1402.3525

since more than two decades ago , the environment - induced decoherence ( eid ) approach is considered a new orthodoxy in the physicists community ( @xcite , @xcite ) . it has been fruitfully applied in many areas of physics and supplies the basis of new technological developments . in the foundations of physics community , eid has been viewed as a relevant element for the interpretation of quantum mechanics ( @xcite , @xcite ) and for the explanation of the emergence of classicality from the quantum world ( @xcite , @xcite , @xcite , @xcite ) . the great success of eid has given rise to the idea that decoherence necessarily requires the interaction between an open quantum system and an environment of many , potentially infinite , degrees of freedom . however , the historical roots of the decoherence program can be found in certain attempts to explain the emergence of classicality in closed systems . in turn , at present other approaches have been proposed , and in several of them the openness of the system is not an essential factor . these new approaches are usually conceived as rival to eid , or even as dealing with different physical phenomena . on the basis of our previous works on the subject , the aim of this paper is to supply a comprehensive presentation of a new perspective , according to which formalisms originally devised to deal just with closed or open systems can be subsumed under a closed - system approach that generalizes the traditional account of the phenomenon . with this purpose , in section ii we will briefly review the historical development of the decoherence program , and in section iii we will stress some conceptual difficulties of the eid program . in section iv we will recall the orthodox approach to decoherence , emphasizing its open - system character . section v will be devoted to introduce a closed - system perspective , and to use it to reformulate the orthodox open - system eid approach . on this basis , in section vi we will formulate the closed - system approach from a generic viewpoint : in this framework , the account of decoherence turns out to be more general than that supplied by the orthodox open - system approach . in order to illustrate this theoretical claim , in section vii we will consider the description of the well - known spin - bath model in the context of the closed - system framework , as well as the description of a generalization of that model . this task will allow us to argue , in section viii , that decoherence is relative to the decomposition of the whole closed system into a decoherent system and its environment . on this basis , in section ix we will cast a new look at the phenomenon of decoherence by considering how the new theoretical framework dissolves the conceptual difficulties of the eid approach , and by discussing the relationship between decoherence and dissipation . section x will be devoted to describe the quantum - to - classical transition from the new closed - system approach , which will allow us to unequivocally define the realm of classicality by identifying the observables with classical - like behavior . in section xi we will compare our development with other approaches to decoherence . finally , in section xii we will resume the main results of our paper and draw our conclusions . from a historical perspective , the decoherence program finds its origin @xmath0though , of course , not under this name@xmath0 in the attempts to explain how a coherent pure state becomes a final decohered mixture with no interference terms . three general periods can be identified in the development of this program : * * * first period ( * * closed systems ) . in the fifties and the early sixties , some authors directed their attention to the emergence of classical macroscopic features from quantum microscopic descriptions ( van kampen @xcite , van hove @xcite , @xcite , daneri _ et al . _ @xcite ) . in this period , the issue was treated in the context of the study of irreversibility and , therefore , closed systems were considered . on this basis , the states indistinguishable from the viewpoint of certain gross observables were described by the same coarse - grained state , whose evolution was proved to reach equilibrium in a certain relaxation time . the main problem of this period was that the relaxation times so obtained turned out to be too long to account for experimental results ( see @xcite ) . * * second period * ( open systems ) . in the seventies , the emergence of classicality begun to be conceived in terms of quantum measurement and , then , was addressed from an open - system perspective ( zeh @xcite , @xcite , @xcite ) . on the basis of these precedents , the eid approach was systematized and developed mainly by zurek and his collaborators in a great number of works ( @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) . in this context , an open system is considered in interaction with its environment , and the evolution of its reduced state is studied . eid proves that , in many physical models , the interference terms of the reduced state rapidly vanish and the system decoheres in an extremely short decoherence time . this result solves the main problem of the first period ; however , the foundations of the eid program are still threatened by certain conceptual problems derived from its open - system perspective ( we will return on this point in the next section ) . * * third period * ( open and closed systems ) . although ` eid ' is still considered almost as a synonym for ` decoherence ' , in the last times other approaches have been proposed to face the conceptual difficulties of eid ( diosi @xcite , @xcite , milburn @xcite , penrose @xcite , casati and chirikov @xcite , @xcite , adler @xcite ) . some of these accounts are clearly non - dissipative ( bonifacio _ et al . _ @xcite , ford and oconnell @xcite , frasca @xcite , sicardi schifino _ @xcite , gambini _ et al_. @xcite ) , that is , not based on the dissipation of energy from the system to the environment . among them , the self - induced decoherence ( sid ) approach shows that a closed quantum system with continuous spectrum may decohere by destructive interference ( @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) . in spite of the fact that , at present , formalisms for open and closed systems coexist , in the literature both kinds of approaches are often presented as alternative scenarios for decoherence , and even as theories dealing with different physical phenomena . in the next sections we will challenge this common view by showing that the different formalisms can be understood in the context of a general closed - system framework in spite of the great success of eid , and of the fact that it is conceived as the orthodox approach for decoherence , it still has to face two conceptual difficulties . according to eid , decoherence is a consequence of the interaction between an open system and its environment ; this process is what einselects the quantum states that become the candidates to classical states . therefore , decoherence must always be accompanied by other manifestations of openness , such as the dissipation of energy into the environment . zurek even considers that the prejudice which seriously delayed the solution of the problem of the emergence of classicality is itself rooted in the fact that the role of the openness of the system was traditionally ignored ( @xcite , @xcite ) . if only open systems may decohere , from this perspective the issue of the emergence of classicality in closed systems , in particular , in the universe as a whole , can not even be posed ( see @xcite ) . zurek expresses the criticism to eid in the following terms : _ the universe as a whole is still a single entity with no ` outside ' environment , and , therefore , any resolution involving its division into systems is unacceptable _ ( @xcite , p.181 ) . this objection has led to the development of the non - dissipative approaches to decoherence which , for this reason , are usually viewed as alternative or rival to the eid approach . when the eid approach is applied to cosmology , the universe is split into some degrees of freedom representing the system , and the remaining degrees of freedom that are supposed to be non accessible and , therefore , play the role of the environment ( see , e.g. , @xcite ) . the same strategy is followed in the case of internal environments , such as collections of phonons or other internal excitations . the possibility of internal environments shows that eid supplies no general criterion for distinguishing between the system and its environment : the partition of the whole closed system is decided case by case , and usually depends on the previous assumption of the observables that will behave classically ( see discussion in @xcite ) . the absence of a general criterion for deciding where to place the cut between system and environment is a serious difficulty for an approach that insists on the essential role played by the openness of the system in the emergence of classicality . zurek recognizes this problem as a shortcoming of his proposal : _ in particular , one issue which has been often taken for granted is looming big as a foundation of the whole decoherence program . it is the question of what are the ` systems ' which play such a crucial role in all the discussions of the emergent classicality . this issue was raised earlier , but the progress to date has been slow at best _ ( @xcite , p.22 ) . as we will see , these problems , which seem to be serious conceptual obstacles for the eid approach , loose their original strength when decoherence is understood from a new general perspective . for this purpose , in the next two sections we will begin by comparing the open - system perspective and the closed - system perspective . as it is well - known , given a closed system @xmath1 represented in the hilbert space @xmath2 , its state @xmath3 evolves according to the liouville - von neumann equation , @xmath4 $ ] , where @xmath5 is the system s hamiltonian . if the task is to describe the behavior of @xmath1 as a whole , there is no theoretical difference between using its state @xmath6 and using the expectation values @xmath7 of its observables @xmath8 : given the state , we can compute the expectation values of all the observables of the system , and given the expectation values of all the observables ( or of the projectors corresponding to a basis of the hilbert space ) , we can compute the system s state . therefore , in this case both descriptions are equivalent : although the state - based description is the commonly used one , all the information of physical interest is also given by the expectation values of the system s observables . if the system is composite , its initial state can be obtained as the tensor product of the states of its subsystems . for example , in the case of two subsystems @xmath9 and @xmath10 , represented in the hilbert spaces @xmath11 and @xmath12 such that @xmath13 @xmath14 , and in initial states @xmath15 and @xmath16 respectively , the initial state of the composite system @xmath17 is computed as @xmath18 and , conversely , the initial states of the subsystems can be obtained from the initial state of the composite system by means of the operation of partial trace:@xmath19 the question is how to describe the behavior of the subsystems _ at any time_. the usual strategy consists in generalizing the partial trace procedure for all times by considering that the reduced states , obtained as@xmath20 are the quantum states of the subsystems @xmath9 and @xmath10 in the same sense as @xmath21 is the quantum state of the whole closed system @xmath1 ( a strategy that is at least controversial : see @xcite , @xcite , @xcite ; we will come back to this point below ) . however , unlike @xmath21 , the reduced states @xmath22 and @xmath23 do not always evolve according to the unitary liouville - von neumann equation : in the case of interacting subsystems , they evolve according to non - unitary master equations , whose specific forms depend on the particular features of the involved subsystems and , as a consequence , must be constructed case by case . nevertheless , if @xmath24 is any observable of @xmath9 and @xmath25 is any observable of @xmath10 , the expectation values of these observables can be computed as@xmath26 in fact , the reduced state of a subsystem of a larger system is defined precisely as the density operator by means of which the expectation values of all the observables belonging to that subsystem can be computed . this means that if we focus only on , say , @xmath9 , we can rely on the description given by its reduced state @xmath22 , since it supplies all the information that we can obtain when we have experimental access only to that subsystem . therefore , analogously to the closed - system case , in this situation it is also true that there is no difference between describing the open system @xmath9 by means of its reduced state @xmath22 and describing it by means of the expectation values @xmath27 of its observables . the orthodox eid approach considers the system @xmath28 under study @xmath0represented in a hilbert space @xmath29 in interaction with an environment @xmath30 @xmath0represented in a hilbert space @xmath31 that induces decoherence . this approach is based on the analysis of the evolution of the reduced state @xmath32 of @xmath28 represented in a certain pointer basis of @xmath33 . either by explicitly computing @xmath32 or by analyzing the master equation , it can be determined whether , under certain conditions , the reduced state operator becomes diagonal or not in that basis . the diagonalization of its reduced state is viewed as a manifestation of the decoherence of the open system @xmath28 . the eid approach proves that , in many physical models , the non - diagonal terms of the reduced state rapidly tend to vanish after an extremely short decoherence time @xmath34 : @xmath35 where @xmath36 is diagonal in the pointer basis . thus , it is usually said that the system @xmath28 decoheres as a consequence of its interaction with the large number of degrees of freedom of the environment @xmath30 . we know that the state @xmath32 is a hermitian operator , so it can always be diagonalized in the smith basis . so , what is special about the state diagonalization in eid ? according to eid , the diagonalization of the state must be studied in the preferred basis or pointer basis . pointer states have some very special proprieties namely @xcite : 1 . they are single out by the dynamics , remaining basically unchanged and 2 . they result in the smallest entropy increase . although the open - system perspective is what underlies the usual practice in physics , it is not unavoidable . the behavior of the open subsystems can also be described from a closed - system perspective by considering the expectation values of the observables relevant in any case . for example , if we still focus only on the subsystem @xmath9 , the relevant observables @xmath37 of the whole closed system @xmath17 are those that act only on that subsystem:@xmath38 where the @xmath24 are the observables of @xmath9 , and @xmath39 is the identity of the space of observables of @xmath10 . the expectation values of these relevant observables can be computed as@xmath40 eq.([6 ] ) clearly shows that , if we want to describe only @xmath9 , its reduced state is not indispensable . the physically relevant information about that subsystem can also be obtained by studying the state @xmath21 of the whole closed system @xmath1 and its relevant observables @xmath41 . this means that there is no difference between describing the open system @xmath9 by means of its reduced state @xmath22 and describing it from a closed - system perspective by means of the expectation values of the relevant observables @xmath41 of the closed composite system @xmath1 in the state @xmath21 . although the eid approach relies on the diagonalization of the reduced state , the same phenomenon can also be described from a closed - system perspective ( we will consider the discrete case , but analogous arguments can be developed in the continuous case ) . in fact , analogously to the case of the previous subsection , if we call the whole closed system @xmath42 , represented in the hilbert space @xmath43 and whose state is @xmath44 , in this case the relevant observables of @xmath45 are of the form@xmath46 where the @xmath47 are the observables of @xmath28 , and @xmath48 is the identity of the space of observables of @xmath30 . the expectation values of these relevant observables can be computed as ( see eq.([6]))@xmath49 it is clear that the evolution of the reduced state @xmath32 ( see eq.([4 ] ) ) has its counterpart in the evolution of the expectation values : @xmath50 where the @xmath51 and the @xmath52 are the diagonal components of @xmath36 and of @xmath53 in the pointer basis , respectively . but since @xmath54 ( see eq.([8 ] ) ) , from the closed - system perspective we can study the time - evolution of the expectation value @xmath55 : if , for any @xmath56 , such a function tends to settle down , in an extremely short time , in a value @xmath57 , where @xmath58 and @xmath59 , then it can be said that @xmath60 ( see eq.([9 ] ) ) . as a consequence,@xmath61 and one can also say that the system has decohered in the pointer basis ( see @xcite ) . moreover , any expectation value @xmath62 can also be expressed as the expectation value of the corresponding relevant observable @xmath63 in a coarse - grained state @xmath64:@xmath65 the state @xmath66 can be obtained as @xmath67 , where @xmath68 is a normalized identity operator with coefficients @xmath69 . it is quite clear that @xmath66 is not the quantum state of the closed system @xmath45 ; nevertheless , if we trace off the degrees of freedom of the environment , we recover the diagonalized reduced state of the system : @xmath70 this means that the coarse - grained state @xmath66 supplies the same information about the open system @xmath28 as the reduced state @xmath36 , but now from the viewpoint of the closed system @xmath45 . therefore , from eq.([10 ] ) we can conclude that@xmath71 by following the eid approach , up to this point we have analyzed the case based on a particular set of relevant observables . in the next section we will see that the conception of the phenomenon of decoherence can be generalized when different sets of relevant observables are considered . as emphasized by omns @xcite , decoherence is just a particular case of the general problem of irreversibility in quantum mechanics . the problem of irreversibility can be roughly expressed in the following terms . since the quantum state @xmath72 follows an unitary evolution , it can not reach a final equilibrium state for @xmath73 . therefore , if the non - unitary evolution towards equilibrium is to be accounted for , a further element has to be added to the unitary evolution . from the most general viewpoint , this element consists in the splitting of the maximal information about the system into a relevant part and an irrelevant part : whereas the irrelevant part is disregarded , the relevant part is retained and its evolution may reach a final equilibrium situation . this broadly expressed idea can be rephrased in operators language . the maximal information about the system is given by the set of all its potentially possible observables . by selecting a particular subset @xmath74 of this set , we restrict the maximal information to a relevant part : the expectation values @xmath75 of the observables @xmath76 express the relevant information about the system . of course , the decision about which observables are to be considered as relevant depends on the particular purposes in each situation ; but without this restriction , irreversible evolutions can not be described . as we have said above , when eid is described from the closed - system perspective , the phenomenon can be viewed as a particular kind of time - evolution of the expectation values of certain relevant observables of the closed system @xmath45 . in that particular case , the relevant observables are of the form @xmath77 . the question is whether that account of decoherence can be generalized ( see @xcite for detalis ) . let us begin by considering the expectation values of certain relevant observables @xmath78 , which now are not necessarily of the form @xmath77 . let us suppose that , for any @xmath56 , those expectation values tend to settle down , in an extremely short time , in a value @xmath79 , where @xmath58 and @xmath59 ( see the analogous case in eid in eq.([10])):@xmath80 of course , the @xmath81 are not the diagonal elements of a time - independent @xmath82 , since the state @xmath44 of the closed system evolves unitarily . nevertheless , the sum of eq.([14 ] ) can also be expressed as @xmath83 where the @xmath84 can be conceived as the components of a kind of coarse - grained state @xmath85 , diagonal in a basis @xmath86 of @xmath43 , which plays the role of pointer basis . in other words , @xmath87 converges , in an extremely short time , to a value that can be computed as if the system were in a state @xmath85 represented by a diagonal density operator : @xmath88 if eq.([13 ] ) and eq.([16 ] ) are compared , it is easy to see that in both cases the expectation values of certain relevant observables of the whole closed system tend very rapidly to certain time - independent values that can be computed as expectation values of those relevant observables in a time - independent diagonal state . therefore , if eq.([13 ] ) describes decoherence , there is no reason to deny that eq.([16 ] ) is also a description of the same phenomenon . this means that , although the off - diagonal terms of @xmath44 never vanish through the unitary evolution , decoherence obtains because it is a coarse - grained process : the system decoheres from the observational viewpoint given by any observable belonging to the space @xmath74 . it might be objected that , whereas in the case of eid the time - independent diagonal state @xmath66 is defined in terms of the reduced state @xmath36 , which is the quantum state of the open system , in the general approach the time - independent diagonal @xmath85 is not the quantum state of the closed system . however , this objection is based on the assumption that the reduced state of an open system is its quantum state , a position that is seriously challenged by the difference between proper and improper mixtures ( see @xcite , @xcite ) : reduced states are improper mixtures , they are only _ a calculational tool _ for computing expectation values ( @xcite , p.48 ) . for this reason , even in the particular context of decoherence , schlosshauer warns us _ against a misinterpretation of reduced density matrices as describing a proper mixture of states _ ( @xcite , p.69 ; see also @xcite ) . therefore , although it is true that @xmath85 is not the state of the closed system @xmath45 but a calculational tool for computing expectation values after the extremely short decoherence time , the same can be said of the reduced state @xmath36 regarding the open system @xmath28 . it is interesting to emphasize in what sense this closed - system perspective for understanding decoherence is more general than the orthodox open - system perspective . the eid approach always selects a set of relevant observables that correspond to a subsystem , with its associated hilbert space and its reduced state . from the closed - system perspective , by contrast , the set of the relevant observables that decohere is completely generic and may not correspond to a subsystem . for example , one might be interested in the decoherence of a single observable of the closed system @xmath45 , which certainly does not define a subsystem but , nevertheless , may manifest a classical - like behavior . summing up , the closed - system perspective has allowed us to conceptualize decoherence from a viewpoint more general than that supplied by the eid approach : decoherence induced by the environment is a particular case of a phenomenon that involves the evolution of the expectation values of certain relevant observables towards a specific time - independent value . there are , then , certain processes that fall under the concept of decoherence from this general viewpoint , but are not conceptualized as decoherence from the eid approach . it might be thought that this is a merely semantic issue ; however , the fact that the difference has physical content will become clear when the problem of the transition from quantum to classical will be considered . in this section we will analyze the spin - bath model , both in the traditional and in the generalized versions , as presented in previous works , from the open - system approach . we will show how this model leads to conclusions that sound paradoxical when read from an open - system perspective . this is a very simple model that has been exactly solved in previous papers ( @xcite ) . here we will study it from the closed - system perspective presented in the previous section . let us consider a closed system @xmath89 , where ( i ) @xmath90 is a spin-1/2 particle represented in the hilbert space @xmath91 , and ( ii ) each @xmath81 is a spin-1/2 particle represented in its hilbert space @xmath92 . the hilbert space of the composite system is , then,@xmath93 in the particle @xmath90 , the two eigenstates of the spin operator @xmath94 in direction @xmath95 are @xmath96 and @xmath97 , such that @xmath98 and @xmath99 . in each particle @xmath81 , the two eigenstates of the corresponding spin operator @xmath100 in direction @xmath95 are @xmath101 and @xmath102 , such that @xmath103 and @xmath104 . therefore , a pure initial state of @xmath45 reads@xmath105 where the coefficients @xmath106 , @xmath107 , @xmath108 , @xmath109 are such that satisfy @xmath110 and @xmath111 . if the self - hamiltonians @xmath112 of @xmath90 and @xmath113 of @xmath81 are taken to be zero , and there is no interaction among the @xmath81 , then the total hamiltonian @xmath5 of the composite system is given by the interaction between the particle @xmath90 and each particle @xmath81 . for instance ( see @xcite ) , @xmath114 where @xmath115 is the identity operator on the subspace @xmath116 and the @xmath117 are the coupling constants . in the typical situation studied by the eid approach , the open system @xmath28 is the particle @xmath90 and the remaining particles @xmath81 play the role of the environment @xmath30 : @xmath118 and @xmath119 . then , the decomposition for this case is@xmath120 and the relevant observables @xmath56 of @xmath45 are those corresponding to the particle @xmath90:@xmath121 the expectation value of these observables in the state @xmath122 is given by ( see @xcite)@xmath123={\textstyle\sum\nolimits^{d } } + { \textstyle\sum\nolimits^{nd } } ( t ) \label{22}\ ] ] where@xmath124 \label{23}\ ] ] by means of numerical simulations it is shown that , for @xmath125 , in general @xmath126 and , therefore , @xmath127 : the particle @xmath90 decoheres in interaction with a large environment @xmath30 composed by @xmath128 particles @xmath81 ( see @xcite ; for larger values of @xmath128 and realistic values of the @xmath117 in typical models of spin interaction , see @xcite ) . although in the usual presentations of the model the system of interest is @xmath90 , there are different ways of splitting the whole closed system @xmath45 . for instance , we can decide to observe a particular particle @xmath129 of what was previously considered the environment , and to consider the remaining particles as the new environment : @xmath130 and @xmath131 . the total hilbert space of the closed composite system @xmath45 is still given by eq.([17 ] ) , but now the decomposition is@xmath132 and the relevant observables @xmath56 of @xmath45 are those corresponding to the particle @xmath129:@xmath133 the expectation value of these observables in the state @xmath134 is given by ( @xcite)@xmath135={\textstyle\sum\nolimits^{d } } + { \textstyle\sum\nolimits^{nd } } ( t ) \label{26}\ ] ] in this case , numerical simulations are not necessary to see that the time - dependent term of eq.([26 ] ) is an oscillating function which , therefore , has no limit for @xmath73 . this result is not surprising , but completely reasonable from a physical point of view . in fact , with the exception of the particle @xmath90 , the remaining particles of the environment @xmath30 are uncoupled to each other : each @xmath81 evolves as a free system and , as a consequence , @xmath30 is unable to reach a final stable state . let us now consider a closed system @xmath136 where 1 . the subsystem @xmath137 is composed of @xmath138 spin-1/2 particles @xmath139 , with @xmath140 , each one represented in its hilbert space @xmath141 : in each @xmath139 , the two eigenstates of the spin operator @xmath142 in direction @xmath95 are @xmath143 and @xmath144 . the subsystem @xmath145 is composed of @xmath128 spin-1/2 particles @xmath146 , with @xmath147 , each one represented in its hilbert space @xmath148 : in each @xmath146 , the two eigenstates of the spin operator @xmath149 in direction @xmath95 are @xmath150 and @xmath151 . the hilbert space of the composite system @xmath136 is , then@xmath152 and a pure initial state of @xmath45 reads@xmath153 with @xmath154 and @xmath155 . as in the original spin - bath model , the self - hamiltonians @xmath156 and @xmath157 are taken to be zero , and there is no interaction among the particles @xmath139 nor among the particles @xmath146 . as a consequence , the total hamiltonian of the composite system is given by @xmath158 where @xmath159 and @xmath160 are the identity operators on the subspaces @xmath161 and @xmath148 respectively . let us notice that the eq.([19 ] ) of the original model is the particular case of eq.([29 ] ) for @xmath162 . we can consider the decomposition where @xmath137 is the open system @xmath28 and @xmath145 is the environment @xmath30 . this is a generalization of decomposition 1 in the traditional spin - bath model : the only difference is that here @xmath28 is composed of @xmath163 particles instead of only one . then , the decomposition is@xmath164 and the relevant observables @xmath56 are those corresponding to @xmath137 : @xmath165 when the expectation value @xmath166 of the observables @xmath56 in the state @xmath134 is computed , two cases can be distinguished : * * case ( a ) * : @xmath167 + numerical simulations show that @xmath127 very fast for increasing time ( see figure 2 of @xcite ) . this means that , as expected , a small open system @xmath168 of @xmath138 particles decoheres in interaction with a large environment @xmath169 of @xmath170 particles . * * case ( b ) * : @xmath171 or @xmath172 + numerical simulations show that @xmath173 exhibits an oscillating behavior and , then , it does not approach zero for increasing time ( see figures 3 and 4 of @xcite ) . this means that , when the environment @xmath169 of @xmath128 particles is not large enough when compared with the open system @xmath168 of @xmath138 particles , @xmath28 does not decohere . in this case we decide to observe only one particle of @xmath137 . this amounts to splitting the closed system @xmath45 into two new subsystems : the open system @xmath28 is , say , the particle @xmath174 , and the environment @xmath30 is @xmath175 . let us notice that the decomposition 2 of the traditional spin - bath model is a particular case of this one , for @xmath176 ( where @xmath128 plays the role of the @xmath138 of this case ) . the decomposition here is@xmath177 and the relevant observables @xmath56 of @xmath45 are those corresponding to @xmath174:@xmath178 when the expectation value @xmath166 is computed , numerical simulations show that , if @xmath125 , @xmath127 very fast for increasing time ( see figures 5 , 6 and 7 of @xcite ) . this means that the particle @xmath174 decoheres when @xmath125 , independently of the value of @xmath138 . but since the particle @xmath174 was arbitrarily selected , the same argument holds for any particle @xmath139 of @xmath137 . then , when @xmath125 and independently of the value of @xmath138 , any particle @xmath139 decoheres in interaction with its environment @xmath30 of @xmath179 particles . on the other hand , the symmetry of the whole system @xmath45 allows us to draw analogous conclusions when the system @xmath28 is one of the particles of @xmath145 : when @xmath180 and independently of the value of @xmath128 , any particle @xmath146 decoheres in interaction with its environment @xmath30 of @xmath179 particles . let us consider the generalized spin - bath model when @xmath181 . in this case , the subsystem @xmath182 does not decohere ( decomposition 1 ) , but the particles @xmath139 , considered independently , do decohere ( decomposition 2 ) . in other words , in spite of the fact that certain particles decohere and may behave classically , the subsystem composed by all of them retains its quantum nature . we have also seen that , since @xmath180 , all the particles @xmath146 , considered independently , decohere . then , in this case not only all the @xmath139 , but also all the @xmath146 decohere . this means that all the particles of the closed system @xmath183 may become classical when considered independently , although the whole system @xmath45 certainly does not decohere and , therefore , retains its quantum character . the fact that certain particles may be classical or quantum depending on how they are considered sounds paradoxical in the context of an approach that explains decoherence as the result of an interaction between open systems . this difficulty can also be seen as a manifestation of the looming big problem of defining the open systems involved in decoherence . the irony of this story is that such a problem is the consequence of what has been considered to be the main advantage of the eid program , its open - system perspective , according to which particles interacting with other particles are well - defined open systems , and the collections of those particles are open systems too . so , the problem is to decide which one of all these open systems is the system @xmath28 that decoheres or , in other words , where to place the cut between the system @xmath28 and its environment @xmath30 . the open - system approach not only leads to the looming big problem , but in a certain sense also disregards the well - known holism of quantum mechanics : a quantum system is not the mere collection of its parts and the interactions among them . in order to retain its holistic nature , a quantum system has to be considered as a whole : the open subsystems are only partial descriptions of the whole closed system . it is on the basis of this closed - system perspective that a different conceptual viewpoint for understanding decoherence can be developed . from the closed - system perspective introduced in the previous sections , the discrimination between system and environment turns out to be the selection of the relevant observables . by following harshman and wickramasekara ( @xcite ) , we will use the expression tensor product structure ( tps ) to call any factorization @xmath184 of a hilbert space @xmath185 , defined by the set of observables @xmath186 , such that the eigenbases of the sets @xmath187 and @xmath188 are bases of @xmath189 and @xmath190 respectively . if @xmath185 corresponds to a closed system @xmath45 , the tps @xmath184 represents the decomposition of @xmath45 into two open systems @xmath191 and @xmath192 , corresponding to the hilbert spaces @xmath189 and @xmath190 respectively . in turn , given the space @xmath193 of the observables of @xmath45 , such a decomposition identifies the spaces @xmath194 and @xmath195 of the observables of the open systems @xmath191 and @xmath192 , such that @xmath196 and @xmath197 . in particular , the total hamiltonian of @xmath45 , @xmath198 , can be expressed as @xmath199 , where @xmath200 is the hamiltonian of @xmath191 , @xmath201 is the hamiltonian of @xmath192 , and @xmath202 is the interaction hamiltonian , representing the interaction between the open systems @xmath191 and @xmath192 . in general , a quantum system @xmath45 admits a variety of tpss , that is , of decompositions into @xmath191 and @xmath192 . among all these possible decompositions , there may be a particular tps that remains _ dynamically invariant _ ( see @xcite ) . this is the case when there is no interaction between @xmath191 and @xmath192 , @xmath203 , and , then,@xmath204 = 0\quad\rightarrow \quad\exp\left ( -iht\right ) = \exp\left ( -ih_{a}t\right ) \exp\left ( -ih_{b}t\right ) \label{34}\ ] ] therefore,@xmath205 this means that , even if the initial state @xmath206 of @xmath45 is an entangled state with respect to the tps @xmath207 , @xmath191 and @xmath192 are _ dynamically independent _ : each one evolves unitarily under the action of its own hamiltonian . as a consequence , the subsystems @xmath191 and @xmath192 resulting from this particular tps do not decohere . once we have excluded the dynamically invariant tps , all the remaining tpss of @xmath45 define subsystems @xmath191 and @xmath192 such that @xmath208 . as a result of the interaction , @xmath191 and @xmath192 evolve non - unitarily ; then , depending on the particular @xmath209 , they may decohere . but the point to stress here is that there is no privileged non - dynamically invariant decomposition of @xmath45 : each partition of the closed system into @xmath191 and @xmath192 is just a way of selecting the spaces of observables @xmath210 and @xmath211 . once these concepts are considered , the selection of the space @xmath74 of relevant observables in the eid approach amounts to the selection of a particular tps , @xmath212 , such that @xmath213 . from this closed - system perspective , it turns out to be clear that there is no essential criterion for identifying the open system and its environment . given the closed system @xmath45 , that identification requires two steps : ( i ) to select a tps @xmath184 such that @xmath214 , and ( ii ) to decide that one of the systems resulting from the decomposition , say @xmath191 , is the open system @xmath28 , and the other , @xmath192 , is the environment @xmath30 . since the tps is defined by the spaces of observables @xmath210 and @xmath211 , the decomposition of @xmath45 is just the adoption of a descriptive perspective : the identification of @xmath28 and @xmath30 amounts to the selection of the relevant observables in each situation . but since the split can be performed in many ways , with no privileged decomposition , there is no need of an unequivocal criterion for deciding where to place the cut between the system and the environment . decoherence is not a yes - or - no process , but a phenomenon _ relative _ to the chosen decomposition of the whole closed quantum system . from this perspective , the perplexities derived from the generalized spin - bath model vanish . in fact , when we consider the whole closed system @xmath45 , there is no difficulty in saying that from the viewpoint of the space of observables , say @xmath215 , ( corresponding to the particle @xmath216 ) there is decoherence , but from the viewpoint of the space of observables @xmath210 ( corresponding to the open subsystem @xmath182 ) there is no decoherence . moreover , even if there is decoherence from the viewpoint of all the @xmath217 , this does not imply decoherence from the viewpoint of @xmath210 since , as it is well - known , @xmath210 is not the mere union of the @xmath218 . in other words , in agreement with quantum holism , the open subsystem @xmath137 is not the mere collection of the particles @xmath139 ; then , it is reasonable to expect that the behavior of @xmath137 can not be inferred from the behavior of all the @xmath139 . in the same sense , it is not surprising that there is no decoherence from the viewpoint of the total space of observables @xmath219 of @xmath45 , in spite of the fact that there is decoherence from the viewpoint of anyone of the @xmath217 and @xmath220 , corresponding to the particles @xmath139 and @xmath146 respectively . and since the privileged viewpoint does not exist , the conclusions about decoherence have to be relativized to the particular observational perspective selected in each case . when decoherence is understood in the new general framework , the conceptual difficulties of the eid program turn out to be not as serious as originally supposed . in fact : 1 . with respect to the open - system problem , it does not make sense to say that closed quantum systems may not decohere : a closed quantum system may decohere from certain observational perspectives , that is , from the viewpoint of certain spaces of observables . furthermore , in spite of the fact that eid focuses on open systems , it can also be formulated from the perspective of the composite closed system and , in this case , meaningful relationships between the behavior of the whole system and the behavior of its subsystems can be explained ( see @xcite , @xcite ) . the defining systems problem is simply dissolved by the fact that the splitting of the closed system into an open subsystem and an environment is just a way of selecting the relevant observables of the closed system . since there are many different sets of relevant observables depending on the observational viewpoint adopted , the same closed system can be decomposed in many different ways : each decomposition represents a decision about which degrees of freedom are relevant and which can be disregarded in any case . since there is no privileged or essential decomposition , there is no need of an unequivocal criterion for deciding where to place the cut between the system and the environment . as said above , decoherence is not a yes - or - no process , but a phenomenon relative to the chosen decomposition . therefore , zurek s looming big problem is not a real threat to the eid approach : the supposed challenge dissolves once the relative nature of decoherence is taken into account ( see @xcite , @xcite ) . although the new framework neutralizes the conceptual difficulties of the eid approach , also points to some warnings about the way in which the proposal is usually presented . from the new perspective , the insistence on the essential role played by the openness of a system and its interaction with the environment in the phenomenon of decoherence sounds rather misleading . the essential physical fact is that , among all the observational viewpoints that may be adopted to study a closed system , some of them determine a subset of relevant observables for which the system decoheres . as pointed out in section ii , certain presentations of the eid approach suggest the existence of a certain relationship between decoherence and dissipation , as if decoherence were a physical consequence of or , at least , were strongly linked to energy dissipation . some particular models studied in the literature on the subject tend to reinforce this idea by describing the behavior of a small open system @xmath0typically , a particle@xmath0 immersed in a large environmental bath . on this basis , the eid approach has been considered a dissipative approach , by contrast to non - dissipative accounts of decoherence that constitute the heterodoxy in the field ( @xcite , @xcite , @xcite , @xcite , @xcite ) . the fact that energy dissipation is not a condition for decoherence has been clearly emphasized by schlosshauer , who claims that _ decoherence may , but does not have to , be accompanied by dissipation , whereas the presence of dissipation also implies the occurrence of decoherence _ ( @xcite , p.93 ) . this fact is explained by stressing that the loss of energy from the system is a classical effect , leading to thermal equilibrium in the relaxation time , whereas decoherence is a pure quantum effect that takes place in the decoherence time , many orders of magnitude shorter than the relaxation time : _ if dissipation and decoherence are both present , they are usually quite easily distinguished because of their very different timescales _ ( @xcite , p.93 ) . according to the author , it is this crucial difference between relaxation and decoherence timescales that explains why we observe macroscopic objects to follow newtonian trajectories @xmath0effectively created through the action of decoherence@xmath0 with no manifestation of energy dissipation , such as a slowing - down of the object . schlosshauer recalls an example used by joos ( @xcite ) : the planet jupiter has been revolving around the sun on a newtonian trajectory for billions of years , while its motional state has remained virtually unaffected by any dissipative loss . this explanation , although correctly stressing the difference between decoherence and dissipation , seems to present both phenomena on the same footing : an open system would first become classical through decoherence , and would then relax due to energy dissipation . following this view , dissipation involves the loss of energy from the system to the environment , while decoherence amounts to a sort of dissipation of coherence that leads the open system , in a very short time , to the classical regime : the environment plays the role of a sink that carries away the information about the system ( @xcite , p.85 ) . the results obtained in the generalized spin - bath model show that the coherence - dissipation or information - dissipation picture has to be considered with great caution , as a mere metaphor . in fact , to the extent that decoherence is a relative phenomenon , no flow of a non - relative quantity from the open system to the environment can account for decoherence . in particular , although energy dissipation and decoherence are in general easily distinguished because of their different timescales , the very reason for their difference is that energy dissipation is not a relative phenomenon but results from the effective flow of a physical entity , whereas decoherence is relative to the observational partition of the whole closed system selected in each situation . on the other hand , decoherence can be explained in terms of the flow of information from the open system to the environment if information is also conceived as a relative magnitude ( @xcite , @xcite ) . as it is well known , one of the main aims of the decoherence program is to explain _ the transition from quantum to classical _ ( @xcite , p.36 ) . for this purpose , it is expected that the phenomenon of decoherence cancels the quantum features that preclude classicality . in this section we will see how this may happen . let us begin by recalling that , in the quantum domain , the expectation values are the experimentally accessible magnitudes : we do not measure states but expectation values . therefore , the purpose must be to obtain quantum expectation values with a classical - like form . the expectation value of a classical observable @xmath221 , whose possible values are @xmath222 , is computed as @xmath223 where @xmath81 is the probability corresponding to the value @xmath222 . in quantum mechanics , by contrast , the expectation value of an observable @xmath221 in the state @xmath224 is computed as @xmath225 where the @xmath226 and the @xmath227 are the components of @xmath221 and @xmath224 in a given basis . it is important to note that in this approach , the terms of the mean values must be written in the same preferred basis of eid ( however it is also possible to define the baseline from the mean values @xcite @xcite ) . therefore , if the purpose is that quantum expectation values behave as in the classical case , we need to identify the situation in which they acquire a form that resembles the classical one , that is , a situation in which the non - diagonal sum @xmath228 vanishes and only the diagonal sum @xmath229 remains . since the non - diagonal terms of a quantum state do not have a classical analogue , in the eid approach the diagonalization of the reduced state of the system is viewed as a manifestation of the quantum - to - classical transition : the states of the pointer basis , einselected by decoherence , are the candidates to classical states ( strictly speaking , in order to obtain a classical - like description it is necessary to apply the wigner transformation and the limit @xmath230 , see @xcite ) . in most models , the system is a particle and , as a consequence , it is usually said that the particle becomes @xmath0or , at least , begins to behave as if it were@xmath0 classical . in fact , the diagonalization of the reduced state @xmath32 of @xmath28 in the pointer basis leads to a situation in which certain expectation values acquire a classical - like form . since @xmath36 is diagonal , it can be conceived as representing a classical - like probability that makes the interference terms represented by @xmath228 to be zero ( see eq.([9 ] ) ) : @xmath231 as a consequence , the expectation value of any observable @xmath232 of @xmath28 that is diagonal in the pointer basis ( the same of eid ) will have the classical form of eq.([37 ] ) : @xmath233 where the diagonal components @xmath234 of @xmath36 play the role of the classical - like probabilities corresponding to the possible values @xmath0eigenvalues@xmath0 @xmath235 of the observable @xmath232 . in summary , the decoherence of the observables corresponding to the system @xmath28 selects the pointer basis in which the reduced state @xmath32 becomes diagonal after the decoherence time and , in turn , that pointer basis identifies the observables of @xmath28 whose expectation values acquire a classical form . as a consequence , instead of saying that the system @xmath28 behaves as if it were classical or that the states of the pointer basis become classical , it would be more accurate to say that _ the observables _ @xmath232 of @xmath28 behave as if they were classical observables . moreover , on the basis of the arguments of the previous sections , from the closed - system perspective one could also say that the observables @xmath236 of the whole closed system @xmath42 behave as if they were classical observables . as we have said in section vi , from our viewpoint decoherence is a process that leads @xmath55 to converge , in an extremely short time , to a value that can be computed as if the system were in a state represented by a diagonal density operator ( see eq.([16 ] ) ) : @xmath237 now @xmath0analogously to the eid case@xmath0 the expectation value of any observable @xmath238 of @xmath45 that is diagonal in the pointer basis will have the classical form of eq.([37 ] ) : @xmath239 where the diagonal components @xmath84 of @xmath85 play the role of the classical - like probabilities corresponding to the possible values @xmath0eigenvalues@xmath0 @xmath240 of the observable @xmath238 . in complete analogy with the quantum - to - classical transition in eid , here the decoherence of the relevant observables of @xmath45 selects the pointer basis in which the state @xmath85 is diagonal and , in turn , this pointer basis identifies , among all the relevant observables , those whose expectation values acquire a classical form . but in this case , the relevant observables do not need to correspond to a subsystem of @xmath45 : it is possible to focus attention on any group of observables of the closed system , or even on a single observable , in order to know if they behave in a classical - like manner . since according to the eid approach only open systems may decohere , from this perspective the issue of the emergence of classicality in closed systems , in particular , in the universe as a whole , can not even be posed since _ the state of a perfectly isolated fragment of the universe _ @xmath0__of , for that matter , of the quantum universe as a whole__@xmath0 _ _ would evolve forever deterministically _ _ ( @xcite , p.181 ) . nevertheless , this difficulty can be overcome from the closed - system perspective . in fact , once the idea of the decoherence and the classicality of systems or states is replaced by the view that certain observables decohere and may acquire a classical - like behavior , the restriction to open systems loses its strength . moreover , the perplexities resulting from the generalized spin - bath model can also be solved . in fact , they derive from supposing that the open subsystems become classical when interacting with their environments . from the open - system perspective , on the contrary , one can univocally identify , among all the observables of the closed system , which acquire a classical - like behavior : the classical world is unambiguously defined by the set observables whose expectation values acquire a classical form . current works indicate that the most suitable scenario for early cosmology is the one of inflation @xcite . this is because the whole structure of the universe can be traced to the primordial fluctuations during an accelerated phase of the early universe . several works show that eid can solve the quantum - classical transition of the universe , by separating the different degrees of freedom in system and environment . since there is no a standard way to divide the universe , some authors take the inflaton as a system and the gravitons as the environment @xcite , while other authors divide the universe between long and short frequencies @xcite . however there is an alternative approach proposed by kiefer , polarski and starobinsky @xcite , it studies the problem of quantum - classical transition without the need to attend to decoherence . the predictions obtained through mean values are indistinguishable from stochastic averaging . according to paper @xcite , inflation can be appears in energy scales where spacetime can be described as a classic curved spacetime where fluctuations are defined . inflaton fluctuations @xmath241 can be studied as scalar fields without mass . it is convenient to consider the rescaled quantity @xmath242 and the conformal time @xmath243 . the symbol @xmath244 will be used to denote a derivative with respect to @xmath245 . the quantization of the perturbation @xmath246 is the usual quantization , and @xmath247 is the momentum conjugate to @xmath248 , @xmath249 where @xmath250 is the lagrangian that corresponds to the hamiltonian : @xmath251 \label{dsd5}\ ] ] @xmath252 is the time dependent annihilation operator@xmath253 @xmath254 and @xmath255 satisfy the canonical commutation relation @xmath256 = i\delta^{(3)}(k - k^{\prime } ) \label{dsd4}\ ] ] this equation expresses the quantum character of the perturbation @xmath254 . if we compute the time evolution in the heisenberg representation we have that the following : @xmath257 where @xmath258 and the modes @xmath259 satisfy the equation @xmath260 we can write the equation ( [ dsd11 ] ) in a more suggestive way @xmath261 where @xmath262 and @xmath263 , @xmath264 , @xmath265 . analogously @xmath266 this formulation is relevant to understand the quantum - classical transition . for example , kiefer and polarski @xcite assume that there is a limit in which @xmath267 and @xmath268 ( or @xmath269 and @xmath270 respectively ) behave as@xmath271 then from ( [ dsd12 ] ) and ( [ dsd13 ] ) we can see that the non - commutativity of operators @xmath272 and @xmath273 is no longer relevant@xmath274 = 2\left ( f_{k1}(\eta)g_{k1}(\eta)+f_{k2}(\eta)g_{k2}(\eta)\right ) \left [ y_{k},p_{k}\right ] \longrightarrow0 \label{dsd13b}\ ] ] to understand the physical meaning of this limit , we can consider a classical stochastic system where the dynamics is still described by equations of the form ( [ dsd12 ] ) , but with @xmath275 and @xmath276 representing random initial values . if @xmath267 and @xmath268 vanish , we get @xmath277 this equation is valid when @xmath248 and @xmath247 are operators and also if they are classical stochastic variables ( @xmath278 and @xmath279 ) . in this case , for a given realization there are a pair @xmath280 . for this reason kiefer and polarski claim _ then the quantum system is effectively equivalent to the classical random system , which is an ensemble of classical trajectories with a certain probability associated to each of them _ ( @xcite pp . 4 ) a concrete example in which @xmath267 and @xmath268 tend to zero , is the perturbation on de sitter space @xcite . when the decaying mode becomes vanishingly small ; when this mode is neglected we are in the limit of a random stochastic process . kiefer and polarski present their explanation of the quantum - classical transition of the universe as a pragmatic view . this is because although we can obtain classical statistics from quantum statistics , there is not an associated interpretation . the main characteristic of the pragmatic view is that it focuses on the mean values and it does not use any reduced state . most interpretations of quantum mechanics emphasize the quantum state , assigning to it some kind of ontology . in fact , in eid the focus is in the reduced state . for this reason the authors believe that their point of view , that does not consider the states , is simply a pragmatic view . and to understand the phenomenon is necessary to introduce eid . however , we believe that this formalism can be given in a proper interpretive framework that does not consider to the state , then eid is not necessary the pragmatic view basically consists of studying the mean values directly , without taking into account the state , and it determines when these values do take the form of classical statistics and when they do not . this way of studying decoherence is similar to the approach proposed by the closed - system perspective . as it is known , a quantum system evolves under the schrdinger equation . as the evolution is unitary , the system can not reach equilibrium or decoherence . as explained in previous sections , if we choose appropriate relevant observables we can describe the phenomenon of decoherence in the mean values of the closed system . therefore , the quantum - classical transition of the pragmatic view has to ignore some kind of decoherence . this is difficult to determine if the decoherence is understood from the point of view of eid , since it divided the whole system into the system and the environment , computes the reduced state of the system and considers its diagonalization . however , according to the closed - system perspective , decoherence is understood as the vanish of the interference terms in the mean values , then the problem is simplified . in fact , the pragmatic view selects the observable ad ignores evanescent modes . the expression ( [ dsd13b ] ) is misleading because from ( [ dsd4 ] ) , ( [ dsd12 ] ) and ( [ dsd13 ] ) we know that@xmath281 then @xmath274 \nrightarrow0\ ] ] then kiefer and polarski say _ clearly , when _ @xmath282 _ _ , _ _ @xmath283 _ are unobservable , this coherence becomes unobservable as well . this is the case when the decaying mode is so small that we have no access to it in observations _ ( @xcite pp . 5 ) this means that coherence becomes unobservable from the observational point of view , as in the closed - system perspective . since @xmath267 and @xmath268 are undetectable , it requires a gross observable that is not capable to detect the presence of @xmath267 and @xmath268 . that is , we need to select a relevant observable @xmath56 . if , for example , we have only access to the observable@xmath284 then@xmath285 in this case we can say that the system decoheres from the observational point of view . then the pragmatic view of kiefer , polarski and starobinsky is fully compatible with the closed - system perspective . the aim of this paper has been to argue that decoherence can be viewed from a closed - system perspective , which improves the understanding of the phenomenon . after recalling the historical development of the decoherence program and stressing some conceptual difficulties of the eid program , we have introduced the closed - system approach from which the orthodox eid approach can be reformulated . in this theoretical framework we have analyzed the spin - bath model , both in its traditional and in a generalized version : by considering different partitions of the whole closed system in both cases , we have shown that decoherence depends on the way in which the relevant observables are selected . on this basis , the following conclusions can be drawn : 1 . decoherence is a phenomenon relative to which degrees of freedom of the whole closed system are considered relevant and which are disregarded in each situation . since there is no privileged or essential decomposition of the closed system , there is no need of an unequivocal criterion for identifying the systems involved in decoherence . therefore , the looming big problem @xmath0which , according to zurek , poses a serious threat to the whole decoherence program@xmath0 dissolves in the light of the relativity of decoherence . 3 . due to its relative nature , decoherence can not be accounted for in terms of dissipation of energy or of any other non - relative magnitude . the open - system perspective leads to the unambiguous identification , among all the observables of the closed system , of those that will have a classical - like behavior because their expectation values acquire a classical form . once the phenomenon of decoherence is de - substantialized in this way , one might ask in what sense it can be still understood as the result of the action of an environment that destroys the coherence between the states of an open quantum system . we are very grateful to roland omns and maximilian schlosshauer for many comments and criticisms . this research was partially supported by grants of the university of buenos aires , conicet and foncyt of argentina . w. h. zurek , preferred sets of states , predictability , classicality and environment - induced decoherence , in j. j. halliwell , j. prez - mercader and w. h. zurek ( eds . ) , _ physical origins of time asymmetry _ , cambridge : cambridge university press , 1994 . o. lombardi , s. fortin and m. castagnino , the problem of identifying the system and the environment in the phenomenon of decoherence , in h. de regt , s. okasha and s. hartmann , _ epsa philosophy of science : amsterdam 2009 _ , amsterdam : springer , 2011 .

in this paper we present a new , essentially empirical , model for the relation between the mass of a dark matter halo / subhalo and the luminosity of a galaxy hosted in it . to estimate this , we replace the assumption of linearity between light and mass fluctuations with the assumption of monotonicity between galaxy light and halo or subhalo mass . we are enabled to proceed with this less restrictive ansatz by the availability of new , very high resolution dark matter simulations and more detailed and comprehensive global galactic luminosity functions . we find that the relation between halo / subhalo mass and hosted galaxy luminosity , is fairly well fit by a double power law . that between halo mass and group luminosity has a shallower slope for an intermediate mass region , and is fairly well fit by a two branch function , with both branches double power laws . both relations asymptote to at low , while at high mass the former follows and the latter . in addition to the mass - luminosity relation , we also derive results for the occupation number , luminosity function of cluster galaxies , group luminosity function and multiplicity function . then , using a prescription for the mass function of haloes in under / overdense regions and some further assumptions on the form of the mass density distribution function , we further derive results for biasing between mass and light and mass and galaxy number , light distribution function and the void probability distribution . our results for the most part seem to match well with observations and previous expectations . we feel this is a potentially powerful way of modelling the relation between halo mass and galaxy luminosity , since the main inputs are readily testable against dark matter simulation results and galaxy surveys , and the outputs are free from the uncertainties of physically modelling galaxy formation . galaxies : haloes cosmology : theory dark matter large - scale structure of the universe

astro-ph0402500

in recent years , n - body numerical simulations have given us a good understanding of dark matter structure for standard cosmological scenarios , while large scale observational surveys have done the same for the distribution of galaxies . in this way , we are now developing a good picture of how mass and luminosity are distributed in the universe . however , it is still not well known how to connect the two pictures . while it is well established that dark matter haloes are the hosts of the observed galaxies , it is still poorly understood how the former are related to the latter . further , the picture is complicated by the fact that what is usually taken as a halo in simulations would often host multiple galaxies , especially for higher masses . to analyse the issue fully , it is necessary to look at the halo substructure , since each subhalo can host a galaxy . establishing such a link between halo mass and galaxy luminosity would be important because , first of all , it would allow us to have a direct connection between theory and observation , dark matter haloes and galaxies . further , it could also shed some light into the theory of galaxy formation . there are several ways in which this problem can be studied . the more direct ones involve either numerical simulations including gas dynamics @xcite , or semi - analytical models of galaxy formation @xcite , but , while they explicitly give the properties of galaxies located in a given halo , they have the added difficulty that many of the mechanisms involved in galaxy formation are poorly understood , and difficult to compute . their complexity could also mask any fundamental relations that might be present between halo and galaxy properties . more indirect approaches have also been studied . the halo occupation distribution ( hod ) model @xcite is based on the probability @xmath4 that a halo of mass m is host to n galaxies . by specifying the @xmath4 function , along with some form for the distribution of dark matter and galaxies within each halo , it is then possible to relate different statistical indicators of the dark matter and galaxy distributions , such as correlation functions , to each other . this fully specifies the bias between the galaxy and the underlying matter distributions . a recent paper by @xcite , has done a detailed study of results from simulations and related them to the hod model , and has concluded that the form of @xmath5 for the subhaloes is approximately universal , where @xmath6 is the subhalo mass scaled to an appropriate minimum mass . this paper also give results for the relation between galaxy absolute magnitude and halo circular velocity . other work @xcite has taken this approach one step further by studying not only the number of galaxies associated with each halo , but also their luminosity , by building the conditional luminosity function , @xmath7 . this gives the number of galaxies with luminosities in the range @xmath8 contained in a halo of mass @xmath1 . while this work directly relates the halo mass to the galaxy luminosity , it does so only to the average values , lacking the full statistical treatment which is analysed in the hod models . other authors have used a slightly different method . instead of trying to specify the number of galaxies in each halo , they treat the halo as a whole and identify it with a galaxy group . then , by comparing the group luminosity function with the halo mass function , they obtain the luminosity associated with each halo @xcite , and also develop ways to estimate the number of galaxies hosted in a halo , thus coming back partly to the @xmath4 estimate of the hod models . in the present paper , we follow a new and conceptually clear approach based on one simplified and testable hypothesis : there is a one to one , monotonic correspondence between halo / subhalo mass and resident galaxy luminosity . we might call this the empirical ( rather than the semi - analytical ) approach because there is no attempt to physically model the galaxy formation process . instead we take from observations the galaxy luminosity distribution and match it with the theoretical halo / subhalo distribution . this has the additional advantage of naturally giving a lower mass threshold for haloes that host luminous galaxies , as the luminosity decreases sharply with mass for less massive haloes . it also implicitly gives rise to galaxy systems , if one identifies a system of a massive halo and its subhaloes with the central galaxy and its satellites in groups and clusters . we find that the single assumption is very powerful and allows us to compute , and compare to observations many quantities from bias to the void distribution function to the spatial correlation function . this paper is organized as follows : in section 2 we present our model for the subhalo mass distribution in a parent halo , and build the global subhalo mass function . in section 3 , we derive the relation between mass and luminosity , as well as some other functions such as the luminosity function of cluster galaxies , the group luminosity function and the multiplicity function . in section 4 we study how to apply the relation we obtain to get the light density and the number density of galaxies as a function of mass density , and also obtain results for the distribution function of light density and the void probability function . finally , we conclude in section 5 . throughout we have used a concordance cosmological model , with @xmath9 , @xmath10 , @xmath11 and @xmath12 @xcite . the first step in building our model is to specify the mass distribution of subhaloes for a given parent halo . we adopt the following function : @xmath13 which gives the number of subhaloes with masses in the range @xmath14 to @xmath15 , for a parent halo of mass @xmath1 . the normalisation @xmath16 is such that the total mass in these subhaloes , @xmath17 , is a fraction of the parent halo mass , @xmath18 ( where the factor @xmath19 accounts for the added mass of the original , unstripped , subhaloes ) . with this definition , we can write @xmath16 as @xmath20 this expression is motivated by recent analysis of high resolution dark matter simulations of @xcite , and we use for the parameters the values @xmath21 , @xmath22 and @xmath23 . its results are similar to those obtained by the simulations of @xcite , who find a power law fit to the subhalo mass function with a slope close to -2 ( that is , in terms of equation [ nsubhalo ] , @xmath24 ) . however , they also find that if they only include the lowest mass bins where statistical errors are smallest , this slope is reduced to values around -1.9 , very similar to what we use here . it further agrees well with the cumulative mass function derived analytically by @xcite ( see section below on the occupation numbers for further discussion ) , and it is also similar to the power law form for the subhalo number given in @xcite , only instead of having a sharp cutoff at the parent halo mass , we introduce a smooth exponential cutoff from a mass @xmath25 . since it is a schechter function , it is also similar to the halo mass function , and the slope @xmath26 is close to the expected value of the slope of the halo mass function . it gives a total mass in subhaloes of 18% , close to but slightly higher than the values of @xmath27 obtained in different studies ( see , for example , @xcite ) . @xcite derive a lower value of around 6% at radius @xmath28 ( where the average overdensity is 200 times the critical ) , though they also find that this fraction can be as high as 10 - 15% in some cases . however , @xmath28 is smaller than the virial radii typically measured from the simulation results by @xcite , which helps to explain the difference . it is worth noting that , like previous results from simulations and analytical modelling ( e.g. , @xcite ) , the shape of the subhalo mass function given by equation [ nsubhalo ] is independent of parent halo mass . this does not mean , however , that one should automatically expect the same to be true of satellite galaxy distribution in galaxies like the milky way and clusters . in fact , as can be see from our results further below , the mass luminosity relation has very different behaviour depending on the mass of the host halo / subhalo ; this leads to parent halos of very different masses ( such as a cluster and a galaxy ) having very different satellite luminosity distributions , even though the scaled mass functions of their subhaloes are identical . a very important point when analysing this expression is to note exactly what mass is being accounted by @xmath14 . in fact , the distribution is valid for the present mass of the subhalo satellites , obtained when @xmath29 , which is measured after the tidal stripping of the outer parts of the subhalo in the halo potential well . however , to use this function to build the host distribution and to subsequently compare it to the galaxy luminosity function , it is necessary to use the original , unstripped mass of the subhalo , since only then can we establish a monotonic correspondence between host halo mass and galaxy luminosity . the factor @xmath19 put into the expression takes this mass loss into account , since we can then treat the mass @xmath14 in equation ( [ nsubhalo ] ) as the original mass , with the stripped mass being then @xmath30 with @xmath31 . obviously , the actual factor for a given subhalo will be highly variable , but for reasons of simplicity and also because it is not well known ( see e.g. @xcite for a treatment of the profile of stripped subhaloes ) , we will be treating this as an average applicable to all subhaloes , and using @xmath32 in the present work . it should be noted that , since we are taking the total subhalo mass to be 18% of the parent halo mass , this factor can not be more than 5 , assuming that the majority of the mass in the parent halo was built up by stripping of the subhaloes . a recent paper by @xcite includes a comparison between maximum and present mass of subhaloes in the dwarf mass range . their results ( see their figure 4 ) seem to support that the average change in mass does not depend on the mass of the subhalo , and an average mass stripping factor of 3 as we adopt here seems a fair agreement . all plots and further expressions ( where we will drop the factor @xmath19 ) presented refer to this original , unstripped mass ; to invert our approximation is simply a case of dividing the subhalo mass by a factor of 3 to represent the actual stripped mass which would be measured in a simulation . ideally , we should use instead a distribution for the maximum circular velocity of the subhaloes . even though this quantity also shows a large scatter between the maximum value and that at present after stripping @xcite , its relative change should be smaller than what would be expected for a given change in mass . this should be the case especially for subhaloes massive enough to host galaxies , which would have only their outer parts stripped off , and where therefore the inner parts which determine the peak circular velocity are left relatively undisturbed . this would mean that a calculation based on the cumulative number function like we use here would incurr a smaller error , even with a large scatter , and in later work we will transform to that variable . using the expression for the subhalo number ( [ nsubhalo ] ) together with the halo mass function , @xmath33 , it is then possible to build the global subhalo mass function , that is , the number density of subhaloes in a given mass range , by @xmath34 if we assume that the halo mass function has a schechter form , @xmath35 it is possible to write an analytical expression for it : @xmath36 } \big(\frac{m}{\beta m_*}\big)^{-\frac{a+\alpha}{2 } } k_0\big(2 \sqrt{\frac{m}{\beta m_*}}\big ) \ , , \ ] ] where @xmath37 stands for the modified bessel function . with more general mass functions , the calculation has to be done numerically . in the present work , we use the more accurate sheth tormen mass function @xcite , @xmath38 with @xmath39 , @xmath40 , @xmath41 and @xmath42 ; as usual , @xmath43 is the variance on the mass scale @xmath1 , @xmath44 is the growth factor , and @xmath45 is the linear threshold for spherical collapse , which in the case of a flat universe is @xmath46 . the result for the global subhalo mass function is shown in figure [ massffig ] , where the sheth tormen mass function is also shown . ) , while the subhalo one is the one obtained using our model for the subhalo number distribution with parent halo mass.,width=317 ] with the expression for the subhalo number , equation ( [ nsubhalo ] ) , it is also possible to calculate the halo occupation number , that is , the number of subhaloes in a halo of mass @xmath1 : @xmath47 because the integral diverges , it is necessary to set a minimum threshold for the subhaloes . this can be set as the minimum mass for a halo to host a galaxy , in which case this occupation number corresponds to the number of galaxies contained in a given halo . this is one of the basic ingredients of hod models , where it is taken as the average number of galaxies in a halo . figure [ haloocupfig ] shows the occupation number as a function of the parent halo mass , in units of the minimum mass considered . the most prominent feature is the cutoff for @xmath48 ; this is inherent to our model , since we consider the total mass in subhaloes to be only 18% of the halo mass , and we also have a cutoff in the number of subhaloes , given by the parameter @xmath49 . at the high mass end , the function is well fit by a power law , with @xmath50 for @xmath51 , which is similar to what is predicted by most hod models ( see e.g. , @xcite ) . this behaviour is also analogous to that found in the analytically derived halo mass function of @xcite , who also find a slope of roughly 0.9 at high @xmath52 , but a steeper value close to 1 in the lower range , similar to our results . the values they obtain are also similar to ours , with one caveat : they correspond to those we would get if we took the values for the stripped mass of the subhaloes , instead of what is plotted in figure [ haloocupfig ] ; this would correspond roughly to diving the numbers we get by a factor of 3 . our result is also similar to what has been observed for the case of galaxies in a cluster , with @xcite obtaining a relation for the number of galaxies with luminosity greater than @xmath53 , @xmath54 , where all quantities are normalised to the usual mass overdensity of 200 . if we assume that their @xmath53 galaxy would have a mass of around @xmath55 ( see below for the results we obtain ) , we also obtain approximately the same number of galaxies in a @xmath56 parent halo . the overall shape of our occupation number is also similar to the results obtained by @xcite in their simulations , although in their case the cutoff is less pronounced , and in fact is only noticeable for @xmath57 . as referred , however , our cutoff around @xmath58 is in fact a feature of our model , and something similar should be present as long as a cutoff in subhalo mass is introduced and there is an upper limit to the total mass in subhaloes . the normalisation of the curves is also similar , with a value of @xmath59 for @xmath60 . it should however be cautioned that the way in which the subhalo mass is accounted may not be exactly the same in both cases ( see discussion above ) . , solid line).the halo mass is scaled to the minimum mass considered for the subhaloes . the solid line shows the results for the model used in this paper ( @xmath22 ) . the two dotted lines show the result of changing the subhalo cutoff ( respectively for @xmath61 and @xmath62).,width=317 ] once we have the total mass function for haloes and their subhaloes , it is then possible to compare it to the galaxy luminosity function to obtain a mass luminosity relation . as noted , a few prior assumptions go into this : first , that the galaxies are hosted individually in the haloes or subhaloes , and that each contains a single one ( or , in the case of the parent haloes , they have one in their centre , plus the ones in their subhaloes ) ; and second , that the luminosity is a monotonic function of the halo mass . the first assumption is supported by previous studies of the group mass function . for example , @xcite obtain a group mass function from the 2df catalogue which matches well with theoretical mass functions like the sheth tormen one , equation ( [ stmf ] ) , indicating that each of their identified groups corresponds to a halo . as for the second , @xcite have shown that subhaloes identified in a set of simulations have a correlation function and power spectrum that matches the galaxies in the pscz survey , which shows it is possible to understand the spatial distribution of galaxies by identifying galaxies brighter than a given luminosity with haloes larger than a certain mass . this is further justified by studies which show that the only halo property dependent on the large scale environment is the mass distribution ( e.g. , @xcite ) , which coupled with the current understanding of galaxy formation theories should imply that we have captured most of the environmental dependence of the galaxy luminosity . we take the galaxy luminosity function to have the usual schechter form : @xmath63 the values of the different parameters are taken from the @xmath64 band 2df galaxy luminosity function @xcite , with @xmath65 , @xmath66 and @xmath67 . although this fit was determined for the magnitude range @xmath68 , here we will be extrapolating its result to higher or lower luminosities as necessary . the mass luminosity relation is then calculated by setting the luminosity @xmath69 of a galaxy hosted in a halo of mass @xmath1 to be such that the number of galaxies with luminosity greater than @xmath69 equals the number of haloes plus subhaloes with mass greater than @xmath1 : @xmath70 dm \ , .\ ] ] it should be noted that , because the values for the subhalo mass function are generally lower than those of the halo one , the subhaloes make only a small contribution to this expression , only being important at lower mass values . this guarantees that a possible second generation of subhaloes ( subhaloes of the subhaloes considered here ) would not influence this result much , since their mass function would have values that much lower than the original halo mass function , and then only at the lowest masses considered . it is also possible to obtain the group luminosity associated with each halo . in order to do this , we first assume that a galaxy system ( which would be either an isolated galaxy , a group or a cluster , depending on the mass of the halo containing it ) can be represented by a halo and its associated subhaloes . the group luminosity for a halo of mass @xmath1 is then simply @xmath71 refers to the parent halo mass only . also shown as a dashed line is the result when the mass is taken to be the actual stripped mass of the subhaloes ( which is obtained by dividing the original mass by a factor of 3 ) . the different points are estimates from different observational studies ( see discussion in text for references).,width=317 ] there is an important point that should be considered when analysing this expression , and that is the possibility of putting a lower limit in the integral , since there may be a minimum mass for a halo to be host to a luminous galaxy . however , as can be seen in figure ( [ masslightfig ] ) , which shows the results for the luminosity and group luminosity associated with a halo , the calculated luminosity decreases sharply with mass at the low end , becoming negligible for haloes with masses lower than approximately @xmath72 . therefore , instead of putting in a cutoff , the value of which is not clear from what is known of galaxy formation , we simply use the natural drop off in the derived relation . this lower limit seems interestingly to be roughly the mass which is usually considered for the host haloes of dwarf galaxies ( for example , @xcite give a lower limit of @xmath73 to the maximum circular velocity of halo capable of hosting a dwarf galaxy ; see also @xcite for their simulation results on the halo hosts of dwarf galaxies , where they give similar values for the less massive of them ) , but it is quite low when compared with the values used in some studies ( for example , in the hod model in @xcite ) , although this may be related to the fact that most dark matter simulations used in these studies do not reach such low masses . another important aspect that can be seen in the figure is that the group luminosity only starts to depart from the halo luminosity for masses above @xmath74 . this is to be expected given the estimated mass of the host haloes of rich groups and clusters . this difference is the contribution of the subhaloes , and it is clearly seen that , although their numbers may not be very significant to the total mass function , they are very important for the luminosity of high mass haloes , and therefore to the luminosity function . due to their small numbers when compared with subhaloes , it is to be expected that subsubhaloes make only a small contribution to this group luminosity , and that this would only be noticeable for the most massive parent haloes , where the subsubhaloes can be massive enough to be luminous . noticeably , the group luminosity has a middle region of intermediate mass with a shallower slope , and then an upturn for high mass , which was also present in similar studies which used the group luminosity function ( e.g. , @xcite ) , while the single halo luminosity has a much shallower high end slope . we find that the mass luminosity relation we obtain can be fairly well fit by a double power law of the type : @xmath75 for the luminosity of individual galaxies ( the solid line in figure [ masslightfig ] ) , the parameters are @xmath76 , @xmath77 , @xmath78 , @xmath79 , @xmath80 and @xmath81 . therefore , at the low mass end , we have @xmath0 . in fact , if we go to very low luminosities , the slope will actually be slightly steeper ; nonetheless , the region of interest essentially begins for luminosities above a few times @xmath82 ( which corresponds to the fainter dwarf galaxies ; however , note that such low luminosities are outside the range of the schechter fit to the 2df luminosity function , so this is based on an extrapolation of the 2df results to this luminosity range ) , and for these the slope is approximately 4 . such a slope is actually what you would expect from a straightforward comparison between the halo mass function ( assuming a slope of -1.8 at the low mass end ) and the galaxy luminosity function . since at the low mass end the subhaloes give an important contribution to the total number of hosts ( see figure [ massffig ] ) , this is most likely a coincidence , arising from the fact that the total host distribution ( haloes+subhaloes ) has a slope similar to the halo mass function . at the high mass end , we essentially obtain the relation between the halo mass and the luminosity of the brightest cluster galaxy , which has a much shallower slope ( @xmath83 ) . this is most likely due to the fact that , by construction , the mass term refers to the mass of the entire halo hosting the cluster , and not just to the mass in the region of the galaxy itself . the halo mass / group luminosity relation is not really well fit by a double power law , since it has a third region for middle values of mass with a shallower slope than either of the asymptotic values . however , we find that it is possible to describe it as two different double power law branches , which provide a fair fit to the results . thus , for @xmath84 , the group luminosity is essentially the same as the luminosity of the parent halo ( as these haloes do not have subhaloes massive enough to be luminous ) , so we fit it also with equation [ fit ] , and use the same parameters as before . for parent halo masses higher than @xmath85 , we find that a simple double power law of the form @xmath86 , with @xmath87 is a good fit . such a behaviour , of a curve with a relatively flat slope in the intermediate mass range and steeper slopes at both the low and high mass ends is similar to what was observed by @xcite , who compared the ags group luminosity function with the halo mass function , although the actual shape of the curve is somewhat different from what we find here . at the high mass end , the cluster luminosity is almost directly proportional to halo mass ( in fact , @xmath3 ) . this means that the resulting mass to light ratio will be almost constant , rising only very slowly with halo mass , which matches well with previous results for the mass to light ratio of clusters ( e.g. , @xcite ) . however , the values we obtain for the group luminosity seem to be smaller than the observational results . further , the value derived by @xcite for the cluster blue mass to light ratio is @xmath88 , which is roughly consistent with the value we obtain around @xmath89 , but smaller than what we get at @xmath56 . since the mass luminosity relation for a single halo is dominated at the high end by the parent haloes , it seems unlikely that this result could be much modified in the scope of our model , since both the halo mass function and the luminosity function are well known . this then means that the problem would lie in the subhalo distribution for these massive parent haloes ; it would be necessary either to have more of them , or else for them to be more luminous . the two are in fact related , since an increase in the number of subhaloes causes an increase in the total number of hosts , and thus a decrease in the luminosity of a halo ( especially in the low mass range , where subhalo number is more significant ) . however , our results for the occupation number and the luminosity of lower mass haloes seem to be in good agreement with observations and prior theoretical models , which leads us to believe that the most likely cause of this result is the simplistic way in which we treated subhalo mass stripping . we can also compare our results with those in @xcite , who fit the mass to light ratio in different models they study to a double power law . in general , their results are in fair agreement with what we obtain . at the low mass end , the results are quite similar , with these authors obtaining a minimum of the mass to light ratio at a slightly lower mass . at the high mass end , they obtain a steeper mass to light ratio as a function of mass , although this is due to the flattening off we find in our results , since in the intermediate range @xmath90 we obtain higher mass to light ratios and a steeper relation . this discrepancy is most likely due to the factor that these authors are fitting the mass to light ratio to a double power law , which as was discussed above does not provide a good fit to our results . in fact , when they adopt a different model with a fixed mass to light ratio at high mass , their results agree with ours slightly better in this intermediate region . overall , our results seem a fairly good match to estimates of mass taken from a range of observational results across the entire mass range . shown are points for the three most luminous of the dwarf spheroidals in the local group , where the luminosity was taken from the review by @xcite , and the mass was estimated from the results of @xcite ( see the results in their figure 13 for bounds to the peak circular velocity function of unstripped nfw haloes estimated to be possible hosts of the local group dwarf spheroidals ) , by assuming that the relation between mass and luminosity is monotonic . the dashed box represents the relation obtained from the weak lensing study of @xcite , for galaxies around @xmath91 ( where we have taken the values applicable to a nfw halo ) . the results for poor groups are taken from @xcite , while those for clusters are from @xcite , where the bounds come from the two different methods the authors use for estimating fore / background corrections . also included is a point for the milky way @xcite . the major difficulty in obtaining a luminosity function for galaxies in clusters lies in distinguishing what haloes should be treated as rich clusters in our model , and which are simply groups . we adopt the abell definition for rich cluster , namely that it must have upwards of 30 objects brighter than @xmath92 , where @xmath93 is the magnitude of the third brightest galaxy in the cluster . using our derived mass luminosity relation , it is possible to transform this magnitude threshold into a mass one . in our model , this would correspond to the mass of the second most massive subhalo ( since we consider the brightest galaxy to be in the parent halo ) . in order to calculate this , we assume that its mass is given by a distribution which is the product of the poisson probability that on average there exists a single subhalo more massive than it , by the probability of finding a subhalo at that mass . that is , its average value for a parent halo of mass @xmath94 is given by @xmath95 where @xmath96 is the probability of the second most massive subhalo having mass @xmath1 , @xmath97 where @xmath98 is the mass distribution function of the subhaloes , equation [ nsubhalo ] , and @xmath99 is the average number of subhaloes more massive than @xmath1 in a parent halo of mass @xmath94 . since this mass threshold depends on the mass of the parent halo , in the end we obtain a condition on halo mass for it to be treated as a host to a rich cluster . using our results , we find this to be @xmath100 . if we now combine the distribution of all haloes more massive than this , together with their subhaloes , with the mass luminosity relation , we can obtain the luminosity function of galaxies in clusters . in fact , it is possible to make a rough prediction for its shape at the high end by comparing the luminosity of the brightest galaxy in the cluster ( given by the solid line in figure [ masslightfig ] , which represents the luminosity of the galaxy associated with the parent halo itself ) with the total cluster luminosity . the average luminosity of the brightest galaxy can be estimated as @xmath101 , where @xmath102 is the cluster galaxy luminosity function , and @xmath103 is such that on average there is only one galaxy more luminous than @xmath103 , i.e. @xmath104 ( see @xcite ) . with @xmath102 normalized to the total cluster luminosity , @xmath105 , it is then possible to extract a relation between @xmath106 and @xmath105 , which will depend on the shape of the luminosity function . taking our results from figure [ masslightfig ] that @xmath107 and @xmath108 , it is possible to work out that , at the high end , @xmath102 is given by a power law with slope -4.21 . for the mass of a halo associated with a rich cluster . the dotted line shows the 2df global luminosity function , the short dashed one the luminosity function of cluster galaxies also from 2df , and the long dashed line is a double power with the same parameters as the previous function , but with @xmath109 at high @xmath69.,width=317 ] we show the results we obtain in figure [ clusterlffig ] , where we have excluded the first brightest galaxies ( in effect , we are just accounting for the galaxies in subhaloes ) . for comparison , we include the 2df global luminosity function of @xcite , as a dotted line . we also include the luminosity function for galaxies in clusters in the 2df survey , as derived by @xcite , as the short dashed line . this is also given by a schechter function , with parameters @xmath110 and @xmath111 . since the normalisation is not given , we adjusted this to have values in the same range as the ones we obtain . also shown , as the long dashed line , is a double power law with the same low end slope , normalisation , and @xmath53 as the ones used to plot the previous curve , but where the exponential cutoff has been replaced with a power law with the same slope as calculated above , namely @xmath112 . our results show a good agreement to those obtained from 2df , which is even better if a double power law is considered instead of a schechter function . another way to check our results is to build the group luminosity function . this is the equivalent of the galaxy luminosity function , only applied to groups , and it is usually obtained from galaxy catalogues by building groups of gravitationally bound galaxies . in our case , we start with the same assumptions for groups described above , and use the halo mass function and the relation between halo mass and group luminosity given by equation ( [ grouplum ] ) so that @xmath113 ) ; the two other lines are observational group luminosity functions extracted from galaxy catalogues , respectively the ags @xcite ( dotted ) and the vslf @xcite ( dashed).,width=317 ] we show our result in figure [ grouplumfig ] , along with the ags for the cfa survey @xcite and the vslf @xcite group luminosity functions . while at higher luminosities our model would seem to underpredict the abundance of groups , it reproduces the slope of the ags luminosity function quite well . this is undoubtedly related to the fact that our calculated group luminosity is somewhat lower than the observed values ( see figure [ masslightfig ] ) ; an increase in group luminosity would cause a shift to the right of our curve , providing a better match for the observational results . it should also be noted that the inclusion of subsubhaloes would probably tend to slightly increase the values at the high end , relatively to the low end , since most of these subsubhaloes would be dark for low mass parent haloes , but be more massive and therefore contribute a small luminosity to the group total in the case of massive parent haloes . it is also curious to note that , since the fits shown in the figure are a double power law in the case of the ags and a schechter function in the case of the vslf , our result would probably not be well fit by a schechter type function . it is also possible to derive the multiplicity function , the number of groups / clusters as a function of their richness , by a process similar to the one used for the group luminosity function . only in this case , we use the occupation number ( plus one to account for the central galaxy , which is hosted by the parent halo itself ) instead of the group luminosity . then , using an expression similar to ( [ grouplf ] ) with the subhalo number @xmath114 put in place of the group luminosity @xmath115 , we obtain the result shown in figure [ multfig ] . the main point to take into account in this calculation is , as was referred in our above discussion of the occupation number , the need to introduce a minimum mass for the subhaloes . in the case of the figure shown , this was taken to be the mass equivalent to @xmath116 , but in general , and to compare to results from observational studies , this should be set to equal the minimum luminosity considered for objects in the observations . there is also a sharp upturn at @xmath117 due to all the haloes that are massive enough to be considered to host a galaxy , but not enough so that they can have subhaloes with galaxies in them . figure [ multfig ] shows our results in comparison with some observational data . the points were taken from the analysis of @xcite , while the two lines were constructed from the two group luminosity functions shown in figure [ grouplumfig ] , by using the luminosity functions of the two surveys to relate total group luminosity to the number of galaxies above the magnitude threshold . the galaxy numbers for both the points and the results derived from the group luminosity functions were then multiplied by a factor of 0.66 to take into account the difference in radius between friends of friends estimates and the usual definition of virial radius ( see @xcite ) . even though our results lie in the range between those estimated from the vslf and the ags group luminosity functions , they seem to be a bit lower than the observational points . since the group abundance is directly related to the halo mass function , which is well known , this discrepancy must be caused by the occupation number . a slightly higher occupation number would shift our curve further to the right in the figure , bringing it into better agreement with the observational data . incidently , a higher occupation number would also bring our results for the group luminosity and the related group lumininosity function into better agreement with observational values , so we believe that this is where the problem lies . our calculated occupation numbers seem to be in fair agreement with expectations ; however , they depend sensitively on our prescription for subhalo stripping , where we adopted a simplistic approach . a more detailed and correct model for this effect would most likely produce better results . . the different points are taken from the results derived by @xcite from the cfa survey , while the two additional lines are derived from the group luminosity functions shown in figure [ grouplumfig ] , where the magnitude limit for all of these is also @xmath116.,width=317 ] to derive a relation between the mass and light densities , we first need to obtain the mass function of haloes for regions of different density . in order to do this , we follow the method outlined in @xcite . in essence , we treat the evolution of the matter distribution in a region as if it derived from a universe with the same cosmological parameters as that region , namely the same average mass density as is found in the local region . we start by labelling each region by its average mass density , @xmath118 , to which corresponds a given value of the parameter @xmath119 , with @xmath120 the critical density . overdense regions will have @xmath121 , where the barred quantities refer to the background universe , while voids will have the opposite . the growth of perturbations will be affected by the different matter content , and this is reflected by changing the normalisation of the power spectrum according to the differential of the growth factor , @xmath44 , relative to the background : @xmath122 where @xmath123 is some initial redshift for which the fluctuations were equal in the background and in this region ; in the present case , we use @xmath124 . we also use the usual normalisation for the growth factor , @xmath125 . to obtain the mass function for a region with a particular average density , @xmath126 , we then apply this new normalisation to the sheth tormen mass function given in equation ( [ stmf ] ) , changing the @xmath127 term as appropriate , and also the value for @xmath45 , which has a small dependence on @xmath128 ( see for example @xcite ) . an important point is that this prescription does not utilise an explicit smoothing radius for the region considered . instead , these are labelled by their average density . however , when regions of limited size are considered , there exists a maximum mass for objects in them , given by the total mass they contain . as the above prescription does not take this into account , we introduce an additional term to compensate for this effect . therefore , after we obtain the mass function , we put in a further cutoff of the form @xmath129 $ ] , where @xmath14 is the mass of the halo and @xmath130 the total mass in the region of radius @xmath131 and average density @xmath132 . the parameter @xmath133 allows for some tuning of the actual cutoff , and in the present case we use @xmath134 . once this cutoff is introduced , it becomes necessary to renormalise the mass function , so that it still gives the appropriate density when the mass of all haloes is calculated . two examples of the resultant mass functions are shown in figure [ voidmffig ] , with curves for regions with the average density , @xmath135 of the average and @xmath136 times the average , for two different radii . , the lower for @xmath137 . in both , the solid curves are the background sheth tormen mass function , the dotted ones are for densities of(top to bottom ) : @xmath138 , @xmath139 , @xmath140 . , title="fig:",width=317 ] , the lower for @xmath137 . in both , the solid curves are the background sheth tormen mass function , the dotted ones are for densities of(top to bottom ) : @xmath138 , @xmath139 , @xmath140 . , title="fig:",width=317 ] once the mass function for an over or under dense region has been determined , it is then simple to find the equivalent light density , @xmath141 , by using the group luminosity correspondent to each halo , given by equation ( [ grouplum ] ) : @xmath142 our result is shown in figure [ mldensityfig ] . the results for @xmath143 and @xmath144 are quite similar , with the curve for @xmath145 being different . this is due to the large suppression of high mass haloes , even at high overdensities , which can be seen in figure [ voidmffig ] , which alters the proportion of luminous to non - luminous haloes in favour of the latter . the curve we obtain for @xmath146 is similar to the one in @xcite , where the authors have used a mass function for different densities derived from simulation results , to which they then apply the conditional luminosity function . the agreement is quite good with our result , even though we used a theoretical model for the mass function instead of taking it from simulations . , as a function of mass density , @xmath147 . both axis are scaled to the average background values , so regions with @xmath148 are overdense and regions with @xmath149 are underdense . the different lines are for different smoothing lengths : @xmath145 ( solid ) , @xmath150 ( dotted ) , @xmath146 ( short dashed ) . the long dashed line marks @xmath151 . the dashed dotted line is the fit to the same relation mentioned in @xcite.,width=317 ] the most striking feature in these results is the sharp decline in the light density in underdense regions . this would mean that the majority of void regions would be quite dark . another important feature is that for the most part @xmath152 , and the slope of the curve at high density is lower than 1 . the overall results ( e.g. , darkness of the voids ) are similar to those from hydrodynamical simulations by @xcite . but the simulation results show a sharper cutoff at low mass density and light being more overdense than mass for overdense regions and a slope to the relation greater than one . this difference is the more remarkable since the results of @xcite are similar to ours while using a mass function derived from simulations , which otherwise might be considered the most likely origin of the discrepancy . however , previous results shown in @xcite , also taken from simulations , seem to be more in line with what we obtain in the present work . these authors in fact give an explanation for the apparent increasing antibias at higher mass densities : at low redshift these regions usually consist of rich clusters and superclusters , whose stellar population tends to be old . therefore , young blue stars tend to be rare and consequently the total luminosity in the blue band is lower than what could be expect from their high mass , giving rise to the slight antibias . it is therefore likely that this difference is coming from the way in which light is being counted in the two methods . in fact , one problem with the approach presented here is that it breaks down for haloes whose size is comparable to the size of the region being considered , since it is then possible for a sphere to encompass an outer region of the halo where a subhalo and a galaxy lie , and which is therefore luminous but is considered dark in our prescription . since @xcite study the light distribution directly , they would account for the light present in such a situation . this effect would be the more noticeable for high density regions and small smoothing lengths , and in fact the results presented in @xcite are for @xmath145 , while the major discrepancy with our results occurs for high density regions . the differences in underdense regions can most likely be explained by limited resolution effects . we also derive results for the bias between galaxy numbers and the dark matter mass density . the procedure is similar to the adopted for the luminosity , replacing the @xmath153 term in ( [ rholight ] ) with the number of galaxies in a halo of mass @xmath1 , i.e. @xmath154 , corresponding to the central galaxy hosted in the halo and the ones present in the subhaloes . the integral should also be taken with a lower limit at @xmath155 . the minimum mass @xmath155 corresponds to the intended cutoff in mass ( or luminosity ) for the galaxies to be counted . our result is shown in figure [ mndensityfig ] . the general behaviour is similar to what was observed for the luminosity . according to these results , the galaxy number distribution would be a fairly unbiased tracer of the underlying dark matter mass in moderately overdense regions , as would be expected from previous studies ( e.g. , @xcite ) , but the galaxy numbers severely underestimate the mass density in underdense regions and consequently assuming a direct proportionality can lead to a significant underestimate of @xmath128 @xcite . ( curves have same label ) , but for @xmath156 , the overdensity in the number of galaxies . the minimum threshold was set at halo mass corresponding to a luminosity @xmath157 . the average number of galaxies brighter than this limit in the background , taken from the luminosity function , is @xmath158.,width=317 ] using the relations obtained in the previous section , it is also possible to derive a distribution function for the light density ( i.e. , the probability of a region of a given size having a certain average light density , the light pdf ) . to make this calculation , besides the relation between the mass and light densities it is also necessary to have the mass distribution function . in general , this can be obtained from dark matter simulations . for the present work , we will use the approximation of treating this as a lognormal , with an appropriate dispersion dependent on the radius being considered and on the cosmological model . it has been known for some time that such a function is a fair approximation to the real mass distribution function , at least in regions of moderate over / under density ( @xcite ; see also @xcite and @xcite for some recent analysis on this subject ) . the form of the lognormal is given by : @xmath159 ^ 2}{2\omega^2}\big ) \ , , \ ] ] where @xmath160 and the parameter of the distribution @xmath161 is related to the smoothing length and the variance @xmath162 of the mass density field by @xmath163 here , the variance @xmath162 refers to the full , nonlinear spectrum . to relate this to the linear variance , and hence to the cosmological parameters we use , we follow the prescription outlined in @xcite to convert the linear power spectrum of mass density fluctuations to the nonlinear one . to obtain the light distribution function , we then follow a similar procedure to the one used for the group luminosity function : @xmath164 where we use the notation @xmath165 and @xmath166 is the light distribution function . such a transformation assumes a monotonic dependence of the light density with the mass density , which is true of our results ( see figure [ mldensityfig ] ) . another important point to note is that this function is not normalised to one , since some of the underdense areas ( in terms of mass ) are non luminous ( see discussion in @xcite ) . our result is shown in figure [ distfuncfig ] . the general behaviour is what would be expected : the distribution functions are similar in the overdense regions , but in general differ substantially in the underdense regions , with the light distribution function in general not well fit by a lognormal . , with peak height increasing with radius . the x - axis represents either mass or light overdensity , depending on the curve . the distribution functions are shown scaled appropriately for logarithmic binning.,width=317 ] the void probability function is defined as the probability of having no galaxies in a sphere of radius @xmath131 . although voids are a particularly striking feature in the galaxy distribution , their study is impaired by their large size and low number of galaxies present in the observational case , and by resolution difficulties when the study is conducted through numerical simulations including galaxy formation . due to these factors , in the past studies of the voids in the galaxy distribution have been relatively few ( see , for example , @xcite for observational results , and @xcite and @xcite for comparisons with theoretical results ; there are also two recent papers with observational results for the 2df survey , @xcite ) . despite this , the void probability is a powerful tool in the analysis of a particular model of galaxy formation , since it probes a highly non - linear regime and is not derivable from the low order correlation functions ( in fact , it depends on all of the n - point correlation functions ) . in the case of our model , we can determine the void probability function by combining the dark matter distribution function with the expected galaxy number in regions of a given size . thus the number density of galaxies in a region of radius @xmath131 and average density @xmath118 , given a minimum mass cutoff of @xmath167 will be : @xmath168 where @xmath169 is the mass function in a region of size @xmath131 and average density @xmath118 ( see discussion in section 4.1 , also figure [ voidmffig ] ) , and @xmath170 is the number of subhaloes in a halo of mass @xmath1 . the number is then obtained by multiplying this density by the region volume , @xmath171 . we treat this as the average number and assume a poissonian distribution around this average ( see @xcite , where the authors have shown that the full hod is consistent with a poisson distribution for large host masses ) , so that the probability of having 0 galaxies will in fact be @xmath172 . finally , the void probability function is obtained by integrating this number times the probability of a region having this given density , @xmath173 where @xmath174 is the probability of a region of radius r having density @xmath118 , which we take to be given by the lognormal distribution , equation ( [ lognormal ] ) . our results are shown in figure [ voidpffig ] , where we also show observational results for the 2df survey taken from @xcite , for both the ngp and the sgp . at low radius , we obtain a good agreement , but our model would seem to overestimate the probability at large radii . our results are also in better agreement with the values observed for the sgp than for the ngp , since @xcite have found the former to be somewhat emptier than the latter . there is an additional effect that may go some way towards explaining this discrepancy , and that is corrections for the peculiar velocity distortions . these would lead , especially for the larger radii , to the inclusion in voids of galaxies that are not actually there , due to the smearing effect caused by peculiar velocities to the positions of galaxies in redshift space . this effect is not symmetric , since the voids are areas of low mass density and therefore the galaxies in them will have low velocity dispersions . the final result of this would be to overpopulate the voids and consequently to underestimate the measured void probability function , which would lead to a better agreement with our results . , and the corresponding mass was taken as the minimum mass limit in the calculations ( equation [ avgno]).,width=317 ] in this paper we present a new model for relating halo mass to hosted galaxy luminosity , based on the dark matter substructure . we feel this is a potentially powerful way to approach this problem , since it is based on two main inputs , the halo / subhalo distribution and the galaxy luminosity function , which can be well tested and adjusted to results from simulations and surveys , respectively . additionally , the model requires only one further assumption : that there is a one to one and monotonic relation between imbedded halo / subhalo mass and galaxy luminosity . this is more explicit but in general agreement with the general assumptions made in past studies and general assumptions made in similar work of the same subject . it is a much less restrictive assumption than the still sometimes utilized ansatz of a linear `` bias '' between galaxy numbers and dark matter density . we have shown how , starting with a prescription we describe for the subhalo mass distribution in a parent halo , it is possible to obtain a relation for the luminosity of a hosted galaxy , as well as the group luminosity when the system of a halo and its subhaloes is identified with groups or clusters of galaxies . the subsequent results appear to match well with general assumptions of the behaviour of such a relation , as well as results for the mass and luminosity at scales of dwarf galaxies , @xmath91 galaxies and massive clusters . from this model , it is then possible to derive many quantities that can be directly compared with further simulation or observational results . we have shown four examples of this , namely the occupation number , the luminosity function of cluster galaxies , the group luminosity function and the multiplicity function . using further assumptions on how to calculate the mass function for regions of different average densities , and on the shape of the dark matter distribution function , we have also obtained a relation between mass and light and number densities for different smoothing lengths close to what is expected from previous bias studies . we also obtain the distribution function of light density , and the void probability function . the latter is a powerful additional test , since it probes a highly non - linear regime , and our results seem to match well with previous observations . the major difficulty with the model as it is presented here has to do with the identification of the mass of the subhaloes . to be able to apply the monotonic correspondence to the galaxy luminosity , the original mass in the subhaloes has to be taken into account . however , the mass distribution we use is measured for the stripped mass of the subhaloes in their parent halo . in order to build the model , we took the approximation of taking an average for the stripping factor . this is , however , a not wholly satisfactory approach , since in general the stripping history will be highly variable from subhalo to subhalo , and also in different parent haloes . ideally , we would like to use the maximum circular velocity instead of the mass to identify the subhaloes , since this quantity should less sensitive to stripping for the relatively massive subhaloes which can host galaxies . unfortunately , we do not have at present a good distribution for subhalo abundance as a function of the maximum circular velocity , although we wish to study this further in the future . nonetheless , although somewhat crude , the approximation we took is reasonable , and since this work deals with statistical averages for most of the quantities , it does fit somewhat well into the whole structure . more importantly , the results we obtain seem for the most part to match well with observational data . in future work which takes into account the statistical variation , however , this factor will certainly be of great importance . we wish to thank michael strauss and jochen weller for useful discussions . we are also very grateful to the anonymous referee for a most helpful report . av acknowledges financial support from fundao para a cincia e tecnologia ( portugal ) , under grant sfrh / bd/2989/2000 . 99 bahcall n. a. , ostriker j. p. , perlmutter s. , steinhardt p. j. , 1999 , sci , 284 , 1481 bahcall n. a. , cen r. , dav ' e r. , ostriker j. p. , yu q. , 2000 , apj , 541 , 1 benson a. j. , 2001 , mnras , 325 , 1039 benson a. j. , baugh c. m. , cole s. , frenk c. s. , lacey c. g. , 2000a , mnras , 316 , 107 benson a. j. , cole s. , frenk c. s. , baugh c. m. , lacey c. g. , 2000b , mnras , 311 , 793 benson a. j. , frenk c. s. , baugh c. m. , cole s. , lacey c. g. , 2001 , mnras , 327 , 1041 benson a. j. , frenk c. s. , baugh c. m. , cole s. , lacey c. g. , 2003 , mnras , 343 , 679 benson a. j. , hoyle f. , torres f. , vogeley m. s. , 2003 , mnras , 340 , 160 berlind a. a. , weinberg d. h. , 2002 , apj , 575,587 berlind a. a. et al . , 2003 , apj , 593 , 1 bullock j. s. , wechsler r. h. , somerville r. s. , 2002 , mnras , 329 , 246 coles p. , jones b. , 1991 , mnras , 248 , 1 cox a. n. , ed . , 2000 , allen s astrophysical quantities 4th edn . , aip press , new york croton d. j. et al . , 2004 , astroph/0401406 , submitted to mnras de lucia , g. , kauffmann , g. , springel , v. , white , s. d. m. , lanzoni , b. , stoehr , f. , tormen , g. , & yoshida , n. 2004 , mnras , 348 , 333 de propris r. et al . , 2003 , mnras , 342 , 725 fukugita m. , hogan c. j. , peebles p. j. e. , 19 98 , apj , 503 , 518 ghigna s. , moore b. , governato f. , lake g. , quinn t. , stadel j. , 2000 , apj , 544 , 616 girardi m. , borgani s. , giuricin g. , mardirossian f. , mezzetti m. , 2000 , apj , 530 , 62 gottl " ober s. , okas e. l. , klypin a. , hoffman y. , 2003 , mnras , 344 , 715 governato f. , baugh c. m. , frenk c. s. , cole s. , lacey c. g. , quinn t. , stadel j. , 1998 , natur , 392 , 359 hayashi e. , navarro j. f. , taylor j. e. , stadel j. , quinn t. , 2003 , apj , 584 , 541 hoekstra h. , yee h. k. c. , gladders m. d. , 2004 , apj , 606 , 67 hoyle f. & vogeley m. s. , 2003 , astro - ph/0312533 , submitted to apj kauffmann g. , nusser a. , steinmetz m. , 1997 , mnras , 286 , 795 kauffmann g. , colberg j. m. , diaferio a. , white s. d. m. , 1999a , mnras , 303 , 188 kauffmann g. , colberg j. m. , diaferio a. , white s. d. m. , 1999b , mnras , 307 , 529 kayo i. , taruya a. , suto y. , 2001 , apj , 561 , 22 kochanek c. s. , white m. , huchra j. , macri l. , jarrett t. h. , schneider s. e. , mader j. , 2003 , apj , 585 , 161 kravtsov a. v. , berlind a. a. , wechsler r. h. , klypin a. a. , gottlber a. , allgood b. , primack j. r. , 2003 , astro - ph/0308519 , submitted to apj kravtsov a. v. , gnedin o. y. , klypin a. a. , 2004 , astro - ph/0401088 , submitted to apj lemson g. , kauffmann g. , 1999 , mnras , 302 , 111 magliocchetti m. , porciani c. , 2003 , mnras , 346 , 186 marinoni c. , hudson m. j. , 2002 , apj , 569 , 101 marinoni c. , hudson m. j. , giuricin g. , 2002 , apj , 569 , 91 martnez h. j. , zandivarez a. , merch ' an m. e. , domnguez m. j. l. , 2002 , mnras , 337 , 1441 mateo m. l. , 1998 , ara&a , 36 , 435 mathis h. , white s. d. m. , 2002 , mnras , 337 , 1193 mo h. j. , yang x. , van den bosch f. , jing y. p. , astro - ph/0310147 , submitted to mnras moore b. , frenk c. s. , white s. d. m. , 1993 , mnras , 261 , 827 moore b. , ghigna s. , governato f. , lake g. , quinn t. , stadel j. , tozzi p. , 1999 , apj , 524 , l19 nagamine k. , fukugita m. , cen r. , ostriker j. p. , 2001 , apj , 558 497 navarro j. f. , frenk c. s. , white s. d. m. , 1997 , apj , 490 , 493 neyrinck m. c. , hamilton a. j. s. , gnedin n. y. , 2004 , mnras , 348 , 1 norberg p. et al . , 2002 , mnras , 336 , 907 oguri m. & lee j. , 2004 , astro - ph/0401628 , submitted to mnras ostriker j. p. , nagamine k. , cen r. , fukugita m. , 2003 , apj , 597 , 1 peacock j. a. , dodds s. j. , 1996 , mnras , 280 , l19 peacock j. a. , smith r. e. , 2000 , mnras , 318 , 1144 pearce f. r. , jenkins a. , frenk c. s. , white s. d. m. , thomas p. a. , couchman h. m. p. , peacock j. a. , efstathiou g. , 2001 , mnras , 326 , 649 ramella m. , geller m. j. , pisani a. , da costa l. n. , 2002 , aj , 123 , 2976 seljak u. , 2000 , mnras , 318 , 203 sheth r. k. , diaferio a. , 2001 , mnras , 322 , 901 sheth r. k. , jain b. , 2003 , mnras , 345 , 529 sheth r. k. , tormen g. , mnras , 308 , 119 somerville r. s. , lemson g. , sigad y. , dekel a. , kauffmann g. , white s. d. m. , 2001 , mnras , 320 , 289 spergel d. n. et al . , 2003 , apjs , 148 , 175 stoehr f. , white s. d. m. , tormen g. , springel v. , 2002 , mnras , 335 , l84 thoul a. a. , weinberg d. h. , 1996 , apj , 465 , 608 tormen g. , diaferio a. , syer d. , 1998 , mnras , 299 , 728 van den bosch f. c. , yang x. h. , mo h.j . , 2003 , mnras , 340 , 771 vogeley m. s. , geller m. j. , park c. , huchra j. p. , 1994 , aj , 108 , 745 wechsler r. h. , somerville r. s. , bullock j. s. , kolatt t. s. , primack j. r. , blumenthal g. r. , dekel a. , 2001 , apj , 554 , 85 weller j. , ostriker j. p. , bode p. , astro - ph/0405445 , submitted to mnras white m. , hernquist l. , springel v. , 2001 , apj , 550 , l129 yang x. h. , mo h. j. , van den bosch f. c. , 2003 , mnras , 339 , 1057 yoshikawa k. , taruya a. , jing y. p. , suto y. , 2001 , apj , 558 , 520 zheng z. , tinker j. l. , weinberg d. h. , berlind a. a. , 2002 , apj , 575 , 617

confinement and screening are investigated in susy gauge theories , realized by an m5 brane configuration , extending an approach applied previously to sym theory , to other models . the electric flux tubes are identified as m2 branes ending on the m5 branes and the conserved charge they carry is identified as a topological property . the group of charges carried by the flux tubes is calculated and the results agree in all cases considered with the field theoretical expectations . in particular , whenever the dynamical matter is expected to screen the confining force , this is reproduced correctly in the m theory realization . ri-5 - 98 + cern - th-98 - 184 + hep - th/9805123 + 255= 255 by 60 255 by-60 255 by // + + shmuel elitzur , oskar pelc and eliezer rabinovici + _racah institute of physics , the hebrew university + jerusalem , 91904 , israel _ + and + _ theory division , cern + ch-1211 geneva 23 , switzerland _ + e - mail : [email protected] , [email protected] , + [email protected] +

hep-th9805123

the advance in the ability to analyze strong coupling effects in supersymmetric field theories @xcite led to a verification of some qualitative ideas about the behavior of gauge field theories . in particular , confinement in @xmath0 supersymmetric qcd ( sqcd ) was shown @xcite to be related , as expected , to monopole condensation . in a confining phase , the electric field is confined to flux tubes strings carrying a definite energy density ( string tension ) . electrically charged sources are connected by such strings and this leads to a confining force between them , , a force that does not vanish as the distance between them is increased . an explicit demonstration of such a situation , for @xmath0 sym theory was suggested recently , in the framework of a realization of gauge theories in m theory on @xmath4 , which is dual to type iia string theory . the 4d gauge theory is realized as the dimensionally reduced low energy effective field theory on an m5 brane wrapped on a riemann surface . when all the characteristic distances , including the radius of @xmath5 , are large with respect to the planck length , the semi - classical approximation is used @xcite . strictly speaking , the effective gauge theory is identified when the radius @xmath6 of @xmath5 is vanishingly small , corresponding to perturbative type iia string theory . as explained in @xcite , a change in @xmath6 has , in general , an effect on the world - volume field theory , so the results that one obtains for large @xmath6 are for a theory which can be different from the gauge theory one attempts to study . however , there are indications that these theories are in the same universality class and , therefore , have the same qualitative properties . a weaker expectation would be that qualitative confining features , will be shared by these theories . it is those features which we wish to study in this paper . a candidate for flux tubes in @xmath0 sym theory was suggested in @xcite : an m2 brane with one spatial dimension extended in a physical direction and the other in an internal direction , extending between points on the m5 brane . this is called _ an mqcd string _ ; it was explored further in @xcite . considering the realization of @xmath0 sqcd supersymmetric @xmath7 gauge theory with fundamental quarks the quark states were also identified as m2 branes ending on the m5 brane , these with their full extension in internal space , and it was shown that when confinement is expected , these quarks can not exist in isolation and must be either grouped in multiples of @xmath8 , forming baryons , or connected to mqcd strings , forming mesons . moreover , for the @xmath9 model , weakly broken to @xmath0 by a mass for the adjoint , the authors of @xcite identified @xmath10 types of strings with different tensions and reproduced the field theoretical results of @xcite . the string connecting two oppositely charged sources , carries flux which is determined by the sources . in the absence of dynamical matter , this flux protects the string from breaking , by charge conservation . however , when there is dynamical matter carrying an appropriate charge , a pair may be created from the vacuum , cutting the string in two . physically , this means that the potentially confining force between the external sources is screened by the dynamical matter . because of this screening , the string itself is expected to be charged only under those elements of the gauge group that act _ trivially _ on the dynamical fields . in particular , when there are no such group elements , one does not expect stable strings . instead , the force between any external sources is expected to vanish as the distance between them increases . this leads to a distinction between two physically different possibilities , depending on the algebraic structure of the matter sector . when the dynamical matter can screen any external charge , , when there are no group elements acting trivially on the dynamical fields , the forces felt by these charges are qualitatively the same as in the higgs phase and there is no actual phase boundary between the higgs and confining phases ; the corresponding branches are smoothly connected @xcite . on the other hand , when there are external charges that can not be screened by the dynamical matter , the force between them is qualitatively different in the higgs and confining branches and , therefore , these branches are expected to be separated by a boundary . to determine which of these possibilities is realized , one should find the subgroup @xmath11 of the gauge group that acts trivially on the dynamical fields . confinement and higgs phases are expected to be distinct iff @xmath11 is non - trivial . a realization of the subgroup @xmath11 was suggested in @xcite . considering @xmath0 sym theory , @xmath11 was identified as the homology group of the mqcd string and this group was , indeed , shown to be isomorphic to @xmath12 . in this work we elaborate on the geometric manifestation of the kinematic algebraic screening considerations , implied by the above identification of @xmath11 : the confining phase should be distinct from the higgs phase , when the later exists , iff there is a stable mqcd string , carrying a non - trivial @xmath11 charge . to investigate confinement and screening , it is useful to realize the external probes as dynamical , but very heavy , particles , as is done , for example , in @xcite . at energies small compared to the mass of the probes , when the string connecting them is not too long , it is energetically protected from being cut by a pair of the massive particles . therefore , such a string probes the model without the heavy particles : if in this model confinement and higgs phases are distinct , this will be manifested by the stability of the string . when the string is long enough , it should become unstable . in the present work we use m theory to study confinement and screening for systems with various local and global symmetries . we identify the mqcd string , find its conserved charge the homology group and compare to the field - theoretical expectation the subgroup @xmath11 . we then introduce external quarks and demonstrate confinement , when they are heavy and screening , when they are light . all our results are in agreement with the field theoretical analysis . the outline of this work is as follows . in section [ sec - su ] we consider @xmath7 models . we start with @xmath0 sqcd , the model considered in @xcite , as the simplest example , to explain the relevant concepts and methods . we then add an adjoint matter field with a polynomial superpotential . section [ sec - lsu ] is devoted to models with products of @xmath7 gauge factors and section [ sec - o4 ] , to models with @xmath13 and @xmath14 gauge groups . the @xmath15 and @xmath16 models are realized here by configurations that include an orientifold 4-plane , and we find ourselves on an unpaved way as far as brane dynamics is involved . guided by field theoretical expectations , we suggest some rules for m2 brane configurations in the presence of the orientifold . these rules are then used to obtain predictions for the confining behavior of all the @xmath15 and @xmath16 models considered . we end in section [ sec - disc ] with a summary and concluding remarks . in the appendix we prove some properties of m2 branes used in section [ sec - o4 ] . in this section we consider models with @xmath7 gauge group . we start with a brief review of the brane realization and then analyze the @xmath0 sym model . we use this model to introduce and explain the concepts and methods that are used in the other parts of this work . in subsection [ sec - screen ] , we show that the dynamical fundamental quarks screen the confining force and in subsection [ sec - adj ] , we add an adjoint matter field with a polynomial superpotential . we start with the brane configuration in weakly coupled type iia string theory . it is illustrated in figure [ f - cfg ] . all branes are extended in the @xmath17 directions , and are at a fixed value of @xmath18 , so these directions are suppressed . there are two ns5 branes at fixed values of @xmath19 while the d4 branes are extended in this direction . @xmath8 of them are suspended between the two ns5 branes and @xmath20 additional d4 branes end on the right ns5 brane and extend directions . in the description used here , these d6 branes are taken very far away in the @xmath19 direction . ] to @xmath21 . when the ns5 branes are parallel ( , extend in the @xmath22 direction ; see figure [ f - cfg]a ) , the low energy effective field theory on the world - volume of the _ finite _ d4 branes , , those with a finite extent in the @xmath19 direction , is an @xmath9 4d @xmath7 gauge theory with @xmath20 hypermultiplets in the fundamental representation @xcite . the vector multiplets , in the adjoint representation , originate from strings between the finite d4 branes , while the hypermultiplets originate from strings between a semi - infinite d4 brane and a finite one . the finite d4 branes are free to move in the @xmath23 direction , with the ` center of mass ' frozen by a non - perturbative effect @xcite , and this motion realizes the coulomb branch of the field theory : the @xmath23 locations of these branes are identified with the eigenvalues of the adjoint scalar s vev . a rotation of one of the ns5 branes in the @xmath24 space , where @xmath25 , corresponds to a mass for the adjoint scalar , which lifts the coulomb branch . indeed , such a rotation restricts the finite d4 branes to the unique ( @xmath24 ) point shared by both ns5 branes . generically , such a rotation will break all remaining supersymmetry , but there are rotations that preserve half of the remaining supersymmetry @xcite , corresponding to an @xmath0 supersymmetric mass term for the adjoint hypermultiplet @xcite . in particular , when the right ns5 brane is extended in the @xmath26 direction ( see figure [ f - cfg]b ) , the adjoint scalar and its spinor superpartner decouple get infinite mass and one obtains @xcite @xmath0 sqcd . as already noted , the above description is in the framework of weakly coupled type iia string theory , which is a limit of m theory compactified on @xmath4 , with vanishingly small radius @xmath6 of the circle @xmath5 . when @xmath6 is finite , the above brane configuration turns into a single m5 brane @xcite and for large @xmath6 , compared to the 11d planck scale , one uses m theory . the m5 brane is of the form @xmath27 , where @xmath28 corresponds to the @xmath17 directions and @xmath29 is a two dimensional surface embedded in the internal space @xmath30 , parameterized by the coordinates @xmath24 and @xmath31 . in the @xmath9 configuration @xcite , the ns5 branes contribute to @xmath29 two surfaces in the @xmath23 direction and each d4 brane contributes a tube , wrapping once the @xmath32 direction ( degenerating to a line in the @xmath33 limit ) . the resulting curve is the seiberg - witten curve @xcite , encoding the low energy effective gauge coupling . the various bps states are realized as minimal membranes m2 branes with their full extension in @xmath30 , , point - like in @xmath34 , and having boundaries on the m5 brane . when the ns5 branes are not parallel , as in figure [ f - cfg]b , the finite d4 branes merge and form a single tube wrapping the compact dimension @xmath8 times . for the configuration described in figure [ f - cfg]b , when the ns5 branes are orthogonal to each other , @xmath29 assumes the following form @xcite : t = v^-n(m - v)/m vw=:=[^bm]^ . here @xmath35 ( @xmath36 being a dimension - full constant ) , @xmath37 is the instanton factor , @xmath38 and @xmath39 is the quark mass matrix . geometrically , the eigenvalues of @xmath39 are the @xmath23-locations of the semi - infinite d4 branes . the curve ( [ sg - su ] ) is only valid for @xmath40 , but we indeed consider only massive quarks . in @xmath0 sym theory ( @xmath41 ) , all the dynamical fields are in the adjoint representation . the kernel @xmath11 of this representation ( . the set of group elements acting trivially ) is the center @xmath42 , so one expects the existence of strings ( flux tubes ) with a conserved @xmath42 charge . this means that a single string ( with unit charge ) should be stable ( protected by its charge ) , but @xmath8 parallel strings should be able to annihilate . a realization of such a string , called mqcd string , was studied in @xcite . it is an m2 brane with its boundary on the m5 brane . in its simplest form , its world - volume is of the form @xmath43 , where the factor @xmath44 is a string in the physical spacetime @xmath28 and @xmath45 is a curve in the internal space @xmath30 , extending between points on @xmath29 . its @xmath42 charge was identified with the homological classification of such surfaces . in the following we recall this analysis , relying on a pictorial approach , which will be useful in the later derivations . the relevance of the homological classification to conserved charges can be motivated as follows : the homological equivalence of two brane configurations means that there is a world - manifold that can be interpreted as describing an evolution between them , so a property which is the same for equivalent configurations will be conserved under such an evolutions . this is illustrated in figure [ f - toy ] , describing the evolution of a string in @xmath46 dimensions . when one considers a closed @xmath47-brane ( with no boundary ) , the above reasoning leads to the usual homology group @xmath48 : _ two @xmath47-branes - brane ' , possibly disconnected and/or ` wrapped ' a few times . ] @xmath49 , @xmath50 ( in the space @xmath30 ) are equivalent if the difference @xmath51 is a boundary of a @xmath52-brane . _ for example , figure [ f - toy](a ) describes the evolution of a closed string from the initial configuration @xmath49 ( consisting of a single string ) to the final one @xmath50 ( consisting of two strings ) and @xmath49 and @xmath50 are homologically equivalent . to see what this equivalence means , assume that the space @xmath30 in which the string evolves is a punctured plane : the line @xmath53 in figure [ f - toy ] describes the world - line of the excluded point . a closed string has a conserved winding number around this point ( which is an element of @xmath54 ) and @xmath49 and @xmath50 have _ the same _ winding number , the winding number is conserved . when the @xmath47 brane is allowed to end on some manifold @xmath29 , the boundary of its world - manifold @xmath55 can be more complicated , as illustrated in figure [ f - toy]b : in addition to the initial and final configurations ( @xmath49 and @xmath50 ) there may be another part @xmath56 which describes the evolution of the @xmath47-brane s boundary . this part is constrained to be on is restricted to the _ world - manifold _ of @xmath29 , however , assuming @xmath29 is static , this world - manifold is @xmath57 and the @xmath58 factor is trivial topologically . ] @xmath29 , so one is led to the _ relative homology group _ , denoted by @xmath59 : _ two @xmath47-branes @xmath49 , @xmath50 are equivalent if the difference @xmath51 is either a boundary or can be complemented by another @xmath47-brane @xmath56 , which lies in @xmath29 , to form a boundary . _ the significance of this change in classification is demonstrated in figure [ f - toy]b : the string is allowed to end on a circular d1-brane @xmath29 , which surrounds the excluded point . as a result , the winding number is no longer conserved ( @xmath49 and @xmath50 in figure [ f - toy]b have _ different _ winding number ) . this is expressed mathematically by the fact that the relative homology group @xmath60 is trivial in this case . one can see clearly that this lack of conservation is closely related to the fact that @xmath29 itself carries a non - trivial winding number . we return now to our specific case : the mqcd string . it is an m2 brane which looks in the physical space as a string , extending between external charges at infinity . in its simplest form , it has the structure @xmath61 , where the factor @xmath58 represents the coordinate @xmath62 along the string and @xmath45 is a curve in the internal space @xmath30 . more generally , @xmath63 can depend on @xmath62 . we look for a conserved property of this string that can be identified with the flux of the flux tube that this string represents , so ( 1 ) this should be a property of the _ cross - section _ @xmath64 of the string , which is independent of @xmath62 and ( 2 ) it should governs the stability of the string . the homological classification of the cross - section obeys both these requirements : ( 1 ) different cross - section are , by definition , homologically equivalent and ( 2 ) to break the string , the cross - section must be deformed to a point , which is possible iff the cross - section is trivial homologically . therefore , one is led to identify the charge group @xmath12 of the flux tube with the homology group @xmath60 . to show ( following @xcite ) that @xmath60 is indeed isomorphic to @xmath42 , we first state a general result concerning @xmath60 , which is true whenever @xmath29 is connected ( as will be the case in all the models we will consider ) . * to investigate @xmath60 , it is enough to consider _ closed curves _ , since an open curve @xmath45 ( ending on @xmath29 ) can be complemented to a closed one by a curve in @xmath29 connecting the endpoints of @xmath45 . * to obtain @xmath60 , one can start from @xmath65 and find the elements ( equivalence classes ) that contain ( closed ) curves in @xmath29 . these elements form a subgroup @xmath66 of @xmath65 . the curves in @xmath66 are exactly the curves which are trivial in @xmath60 , therefore , h_1(/)h_1()/k . this result can be derived @xcite using the exact sequence @xmath67 now we turn to the specific form of @xmath30 and @xmath29 and apply eq . ( [ hr ] ) . recall that @xmath30 is of the form @xmath68 , where @xmath5 is the 10th ( compact ) spatial dimension . @xmath30 is parameterized by @xmath24 , and @xmath69 , so arg(@xmath70 ) parameterizes the compact dimension and the point @xmath71 ( corresponding to @xmath72 ) is excluded . therefore , @xmath54 , its elements being the winding numbers around @xmath71 . the embedding of @xmath29 in @xmath30 is given by t = v^-n vw= , so @xmath29 can be identified with the punctured @xmath23 plane ( with the origin deleted ) . therefore , closed curves in @xmath29 are classified by their winding number around @xmath73 and from the relation @xmath74 in eq . ( [ sg - su0 ] ) we learn that each loop around @xmath73 is accompanied by @xmath8 loops around @xmath71 . this means that the group @xmath66 of elements in @xmath65 , which are trivial in @xmath60 are the multiples of @xmath8 so , using eq . ( [ hr ] ) , we obtain @xmath75 . to summarize , the mqcd string is an m2 brane which is represented in the internal space @xmath30 by a curve @xmath45 , extending between points in @xmath29 . as such , it has a ( topological ) @xmath42 charge , that is identified with the center of the gauge group . to determine the charge of a given string @xmath45 one should ` close ' it by a curve in @xmath29 and find the winding number of the resulting curve around @xmath71 . such a string is described in figure [ f - sus ] . this is a projection on the @xmath23 plane . each point in the figure corresponds to a unique point on @xmath29 but we want to consider also points in @xmath30 which are _ not _ on @xmath29 . to distinguish between them we use dashed lines for curves on @xmath29 and full lines for curves not on @xmath29 . the circle is on @xmath29 and corresponds to @xmath76const . , which implies @xmath77const .. the emphasized points have also common @xmath32 and there are @xmath8 such points . now we connect two adjacent points @xmath78 by a curve @xmath45 with _ constant @xmath70 _ ( which is , therefore , not in @xmath29 ) . to find its charge , we close it by adding to it the arc @xmath56 above it ( which is in @xmath29 ) . the resulting loop winds once around @xmath32 , so @xmath45 has unit charge . note that we could close @xmath45 from below , obtaining the charge @xmath79 , so the charge is indeed defined only modulo @xmath8 . the loop @xmath80 can be moved off @xmath29 ( having no boundary ) , and then there is no obstruction from shrinking it to a point in the @xmath24 direction . however , it can not disappear , since it still winds around @xmath32 . instead , we have obtained a cylindrical m2 brane , winding around @xmath32 and such an m2 brane is identified as the fundamental closed string . this is a good place to make an important remark . the process we have described demonstrates that the mqcd string is equivalent _ topologically _ to a closed string . this means that there is a _ kinematical _ possibility that one will transform to the other , however , such a process may be highly suppressed by the dynamics . in particular , as explained in @xcite , to be able to investigate the mqcd string while ignoring stringy effects , we should be in a range of parameters where an mqcd string _ can not _ transform to a closed string . this range of parameters was explored in @xcite . in this work we are mainly concerned with kinematics , as manifested by the topology of world - manifolds . so when we say that two curves are equivalent , it should be understood in this kinematical sense . as explained previously , the charge of the mqcd string protects it from breaking , so in the absence of @xmath42 charged particles ( quarks ) , it is infinite in physical space @xmath81 and is stable ( infinite in time ) . this is interpreted as a signal of confinement in the gauge theory . to investigate confinement further , one introduces heavy fundamental matter ( quarks ) that , at this stage , is used to provide external probes of the gauge force @xcite . this is done , as described in the beginning of this section , by adding semi - infinite d4 branes tubes of m5 brane in m theory . the large mass means that the d4 branes are very far ( in the @xmath23 direction ) from the finite d4 branes . this implies that they have negligible influence on the geometry of @xmath29 at this region and the analysis of the previous subsection continues to apply . in particular , the mqcd string still exists and , at low enough energies ( at which it is restricted to be kept far from the semi - infinite d4 branes ) it has the same stability properties . to identify the quarks , one considers the @xmath9 model in the coulomb phase , where the unbroken gauge symmetry is abelian and the quarks are not confined . the corresponding bps states are identified with m2 branes with a boundary that surrounds ( in the @xmath23 plane ) one finite and one semi - infinite d4 brane ; in principle , it may be either a disk ( with one boundary ) or a cylinder ( with two boundaries , each winding one d4 brane ) . these two types of surfaces are illustrated in figure [ f - qrk ] . they are ` continuously ' connected , , they are homologically equivalent . in the following , we will refer only to the cylindrical topology , but we do not loose generality by this , since one can deform ` continuously ' ( in the above homological sense ) between the two topologies . in particular , the unique boundary loop @xmath82 of a disk can be seen as a limiting case of a two - loop boundary @xmath83 of a cylinder . the winding numbers of the boundary ( in the @xmath23 plane ) around the d4 branes represent the color and flavor quantum numbers of the quark ( the two possible orientations correspond to a quark and anti - quark ) . next one rotates the left ns5 brane . in field theory , this corresponds to a mass for the adjoint hypermultiplet . as a result , monopoles condense and this is expected to trigger confinement of electric charge . this indeed can be seen in the brane configuration @xcite : upon rotation , all the @xmath8 finite tubes ( d4 branes ) join to a single tube , winding @xmath8 times around @xmath32 . as a result , the m2 brane configurations described in figure [ f - qrk ] are not possible any more . to understand this , note that @xmath84 winds once around @xmath32 , so @xmath85 should also wind once , but there is no such loop ( with color quantum numbers ) . this is seen as a manifestation of confinement : there are no isolated quarks . to see what _ is _ possible , one may check what happens to @xmath85 , as the left ns5 brane is rotated . this is described ( for @xmath86 ) in figure [ f - cnf ] . this is a schematic and @xmath32 directions . more details can be found in @xcite . ] illustration of a cross - section of the m5 brane between the two ns5 branes . the shaded circles represent the d4 brane tubes . as the tubes join , @xmath85 is forced to have a part that _ is not _ on @xmath29 . since the m2 brane can not end there , it must connect to another piece of an m2 brane . comparing with figure [ f - sus ] , one can see that this other piece can be the mqcd string with a unit charge . indeed , the curve @xmath80 in figure [ f - sus ] winds once around @xmath32 , exactly as is required from @xmath85 . to summarize , we have the following evidence for confinement in @xmath0 mqcd : * in the absence of external quarks , the mqcd string is infinite and stable ; * m2 branes representing isolated quarks do not exist ; * there is an m2 brane representing a meson : quark and anti - quark with an mqcd string extending between them . to investigate screening , we consider higher energy scales , which means that we allow more energetic excitations and , in particular , a longer string . in the m theoretic description it means that we allow m2 branes with a larger area , so the region of the semi - infinite d4 branes becomes accessible to the m2 brane . therefore , the stability of the mqcd string should be re - examined now , with the full curve ( [ sg - su ] ) replacing ( [ sg - su0 ] ) as @xmath29 . to use relation ( [ hr ] ) , all we have to do is to find which closed curves in @xmath30 are equivalent ( in @xmath65 ) to curves in @xmath29 . @xmath30 is unchanged by adding quarks ( only the m5 brane is modified ) , thus @xmath65 is still generated by a loop around the compact direction . the semi - infinite d4 brane which represents a quark winds _ once _ around the compact dimension , as can be easily verified simply connected , leading to the same result for @xmath60 . ] , therefore , _ all _ closed curves are equivalent to curves in @xmath29 , so @xmath60 is the trivial group . this means that in this case @xmath29 leaves no residual conserved charge from the winding number in @xmath30 and the mqcd string is not protected anymore by topological considerations it can break . moreover , it is clear that the semi - infinite d4 brane should be involved in this process . in the following we demonstrate how the mqcd string can be broken . we consider a world - manifold that will describe such a process or , equivalently , a family of m2 branes parameterized by time . this is described in figure [ f - scr ] . the physical and internal spaces are represented each by one direction . we also suppress one of the dimensions of the m2 brane , so each point represents a line ( either open or closed ) . we start with the left m2 brane ( figure [ f - scr]a ) , which represents a meson . a more detailed description of this m2 brane is given in figure [ f - mes ] . the vertical line in figure [ f - scr]a , extending in physical space , is the mqcd string ; each point on it is an open line ( see figure [ f - mes ] ) , like @xmath45 in figure [ f - sus ] . recall , however that it can be complemented by the curve @xmath56 to a closed curve , this corresponds to a transition from an mqcd string to a fundamental string and , therefore , should be highly suppressed dynamically . however , the kinematical possibility is still important , since we need to realize it only at _ discrete points _ the ends of the mqcd string . the corresponding dynamical suppression is related to the threshold for the production of a quark - anti - quark pair . ] the horizontal lines , extending only in the internal space , are the quarks ; each point on them represents a closed loop @xmath82 ( see figure [ f - mes ] ) ; these loops interpolate between the loops @xmath85 and @xmath84 , where @xmath84 is on a semi - infinite d4 brane . in fact , the two quarks can have different @xmath84 , being on different d4 branes ( which means a different flavor , like in a @xmath87 meson ) . the total m2 brane is , therefore , a surface with three boundaries : @xmath88 , @xmath89 and the third is composed of the two boundaries of the mqcd string connected by two copies of @xmath56 . the next m2 brane ( figure [ f - scr]b ) describes an intermediate configuration . the additional horizontal line is a cylinder in internal space which starts at @xmath85 ( at some point in physical space along the mqcd string ) , passes through some closed loop @xmath82 and goes back to @xmath85 . at the initial state , @xmath90 ; at an intermediate state , @xmath82 is arbitrary ( but homologic to @xmath85 ) ; at the final state ( figure [ f - scr]c ) , @xmath82 arrives to @xmath91 , which is on a semi - infinite d4 brane and , therefore , can be a boundary for the m2 brane . so at this stage , the two parts are free to separate in space ( as described in figure [ f - scr]d ) , each representing a full meson . so , the introduction of dynamical quarks , makes the mqcd string breakable . this was shown both by topological considerations and by an explicit construction . so far we mostly considered a model obtained from the @xmath9 supersymmetric one , by decoupling the adjoint field @xmath92 giving it an infinite mass . however , the above analysis equally applies to a more general situation , as we will now show . we return to the @xmath9 supersymmetric model and break the @xmath9 supersymmetry by the following superpotential : w = f ( ) f()=_k=2^n-1_k ^k . such a perturbation has the same effect as a mass for the adjoint : the coulomb branch is lifted and in the surviving vacua monopoles condense , causing confinement . the brane configuration corresponding to this model was found in @xcite . the superpotential ( [ supot - phi ] ) is realized by a `` curved '' left ns5 brane direction , can be seen as a set of parallel ns5 branes extending in this direction . this configuration was considered in @xcite . ] , described asymptotically by the relation @xmath93 . in semi - classical m theory there is , as before , a single m5 brane @xmath27 . we are interested in curves @xmath29 with vanishing genus corresponding to vacua with no massless photons . imposing the appropriate asymptotic conditions , one obtains v = z+ w=[f(v(z ) ) ] _ > t = z^n - f(z+z ) . here @xmath94 is a global coordinate of the curve , @xmath95 is an @xmath96 matrix related to the masses of the quarks and @xmath97 , where @xmath98 is the instanton factor of the @xmath9 model ( @xmath99 ) . @xmath100_>$ ] means to expand the function @xmath101 in powers of @xmath94 and to take only the _ positive _ powers . to relate this curve to the previous one ( eq . ( [ sg - su ] ) ) , one should choose @xmath102 . this implies @xmath103 and , therefore , leads to the curve ^n t = w^n - f(w+m ) v=+ , where @xmath104^\frac{1}{n - f } \hsc m:=\mu z \hsc \lm^b:=\mu^n\lm_2^{b_2 } \hsc \hs.\ ] ] taking the limit @xmath105 with @xmath106 fixed ( and with @xmath107 ) , one obtains the curve ( [ sg - su ] ) with @xmath108 . with this background , one can turn to the analysis of the mqcd string and one finds that nothing essential is changed . the space @xmath30 is the same and the curve @xmath29 , although more distorted in the @xmath26 direction , can still be identified with a punctured plane this time the @xmath94 plane . moreover , the behavior of @xmath70 is completely unaffected by the details of @xmath109 and , since this is the only element relevant to the topological analysis , the results are the same : for @xmath41 there is an mqcd string stabilized by a conserved @xmath110 charge . the quark is realized in the same way ( note that the semi - infinite d4 brane realizing it is connected to the right ns5 brane , which is not affected by @xmath109 ) , so the description of confinement and screening is also the same . this is all in agreement with the expectations from field theory . one can generalize the models of the previous section by adding more ns5 branes . we consider a series of @xmath112 ns5 branes , each having fixed @xmath19 and two dimensions extended in the @xmath113 direction . there are also @xmath114 d4 branes connecting the @xmath115th and the @xmath116th ns5 brane , @xmath117 semi - infinite d4 branes extending to the left of the first ( left ) ns5 brane and @xmath118 semi infinite d4 branes extending to the right of the last ( right ) ns5 brane . when the ns5 branes are parallel , the low energy effective field theory on the finite d4 branes is an @xmath9 4d gauge theory with a gauge group @xmath119 . for each @xmath120 , there is a hypermultiplet in the @xmath121 ( bi - fundamental ) representation . in addition , the semi - infinite d4 branes contribute each a hypermultiplet in the fundamental representation of either @xmath122 ( the left ones ) or @xmath123 ( the right ones ) . rotations of the ns5 branes correspond , as before , to masses for the scalars in the adjoint representations and we will consider rotations that respect @xmath124 supersymmetry . in the semi - classical m - theoretical description , this is a single m5 brane @xmath27 . for generic angles between the ns5 branes , one expects @xmath29 to be of genus 0 , and by imposing the appropriate asymptotic condition , one obtains the following curve v=_=1^l w=_=1^l , t= _ = 1^l(z - z_)^n_-1-n_. here @xmath94 is a global coordinate of the curve , @xmath125 determine the asymptotic position of the ns5 branes ( which are related to the masses of the adjoint scalars and the scales of the gauge factors ) , and @xmath126 are related to the masses of the bi - fundamentals , left fundamentals and right fundamentals , respectively . we start with the model with no fundamental matter ( @xmath127 ) . for generic adjoint masses , the model is expected to be in the confining phase . the elements of @xmath128 acting trivially on the bi - fundamental representation are of the form @xmath129 , so the kernel of this representation is @xmath42 , where @xmath130 is the greatest common divisor of @xmath131 and @xmath114 . therefore , for the above model , the kernel is @xmath42 with @xmath132 , embedded diagonally in all the group factors . when @xmath133 , one expects the existence of strings ( flux tubes ) with a conserved @xmath42 charge , representing a confining force , while when @xmath124 , there should be no conserved charge , which means that the confining force is screened . as before , we will identify the @xmath42 group with the relative homology group @xmath60 . the manifold @xmath30 is the same , leading to @xmath54 . to find @xmath60 using eq . ( [ hr ] ) , we look for classes in @xmath65 that can be represented by loops in @xmath29 ( and are , therefore , trivial in @xmath60 ) . the surface @xmath29 , as described in eqs . ( [ sg - lsu1],[sg - lsu2 ] ) ( with @xmath127 ) , can be identified with the @xmath94 plane , complemented by @xmath134 ( which is a finite point on @xmath29 ! ) and punctured at the points @xmath135 , where at least one of the space - time coordinates diverges . loops in @xmath29 are classified by their winding number around these punctures . from eq . ( [ sg - lsu2 ] ) we learn that a loop around @xmath135 winds the compact direction @xmath136 times . all these loops together generate in @xmath65 the subgroup @xmath137 , where @xmath138 and using eq . ( [ hr ] ) , we obtain @xmath139 which is the desired result . figure [ f - lsu ] describes @xmath29 for @xmath140 gauge group . it is the @xmath94 plane , with which @xmath29 can be identified . as in figure [ f - sus ] , we concentrate on the relevant aspect , which is the @xmath70 behavior and , in particular , its argument , which parameterizes the winding of @xmath29 around the compact dimension . the points @xmath135 are the asymptotic locations of the three ns5 branes . the dashed lines are lines of constant @xmath141 ( constant @xmath19 ) . each is a closed curve ( possibly passing through @xmath134 ) . near some of these curves , there are arrows that indicate the positive direction of @xmath19 and @xmath32 and a number @xmath142 which is the winding number of this circle around @xmath32 . there are @xmath142 points on this circle with the same @xmath32 and , therefore , the same @xmath70 . the mqcd string @xmath45 may be chosen to connect two such points . for definiteness , one can choose constant @xmath70 along it and interpolate linearly between the end values of @xmath23 and @xmath26 . two such strings are described : @xmath143 between the left d4 branes and @xmath144 between the right ones . as for the @xmath7 gauge group , each of these strings can be complemented by a line @xmath56 on @xmath29 to a closed curve @xmath85 winding once around @xmath32 . the point @xmath145 is the asymptotic location of a semi - infinite d4 brane connected to the right of the right ns5 brane and representing a quark in the fundamental representation of @xmath146 . a curve @xmath84 around this point winds once around @xmath32 . using this figure , we can demonstrate confinement and screening in this model , along the lines of the previous section . an m2 brane describing a quark should be a cylinder with a curve like @xmath84 as one of its two boundaries . the other boundary @xmath85 should wind around @xmath32 once . when @xmath147 and @xmath148 are relatively prime , one can construct @xmath85 by combining loops around @xmath149 , @xmath150 and @xmath151 ( for example , when @xmath152 , one can take for @xmath85 a loop around @xmath150 ) . so in this case , there are isolated quarks , which means that the confining force is completely screened ( loops around @xmath149 , @xmath150 and @xmath151 correspond to screening by an adjoint of @xmath122 , bi - fundamental and adjoint of @xmath146 respectively ) . when @xmath153 , there is no curve on @xmath29 that winds once around @xmath85 ( except at the region near @xmath154 , which is identified as a _ flavor _ d4 brane and not a _ color _ one ) . a possible boundary is @xmath155 , which corresponds to a quark attached to an mqcd string . so we conclude that in this case quarks can not exist in isolation and they are bound to ends of mqcd strings . we chose @xmath154 to be close to @xmath156 , which means that it is heavy and , therefore , decouples at low enough energies ( an infinite mass for this quark would correspond to @xmath157 ) . it is manifested geometrically by the fact that away from @xmath154 and @xmath156 , the curve is as for the model without fundamentals and an mqcd string is stable there quarks are confined . at higher energies , the region of @xmath154 and @xmath156 becomes accessible to the mqcd string and it can break . this is demonstrated by figure [ f - scr ] also for the present case , where this time @xmath84 and @xmath85 are as in figure [ f - lsu ] . to summarize , also for this model , we obtained results in agreement with the field theoretical expectations . we now move to the analysis of confinement and screening in models with @xmath14 or @xmath13 gauge groups . these models can be realized in type iia string theory @xcite by performing an orientifold projection , with a 4d orientifold plane ( o4 plane ) parallel to the d4 branes . the brane configuration is almost as in figure [ f - cfg ] , the only difference being that it is restricted to be symmetric around the o4 plane , which is the @xmath19 axis ( multiplied , as everything else , by @xmath28 ) . this means that the d4 branes can move off the @xmath19 axis only in symmetric pairs . there are two possible types of projections , one leading to @xmath14 gauge theory ( @xmath8 even ) with fundamental matter fields and the other , to @xmath13 gauge theory with matter in the vector representation . in both cases @xmath20 is even and there are @xmath158 @xmath9 hypermultiplets . the realization of fields and the identification of parameters and moduli is as in the @xmath7 model , described in subsection [ su - brane ] . so is the supersymmetry . these configurations also have a ( semi - classical ) m - theoretical description , as a single smooth m5 brane @xcite . as the type iia configuration , the m5 brane is also symmetric around the orientifold plane ( modulo a shift in the compact dimension , which is invisible in the type iia description ) . in addition , it was found that the o4 plane contributes additional tubes to the m5 brane , as would do d4 branes with the same rr charge . the charge of an o4 plane depends on the projection type . for an @xmath14 projection it is @xmath159 ( in units of d4 brane charge ) and for the @xmath13 projection it is @xmath160 . when an orientifold plane crosses a ns5 brane , the sign of its charge changes @xcite , so the ns5 brane feels a net charge of two units . this implies that in the @xmath14 model there are two extra tubes between the ns5 branes and in the @xmath13 model there are two extra tubes in the semi - infinite regions . these tubes can not move off the @xmath19 axis . when the ns5 branes are not parallel , all the finite tubes merge to a single tube , as in the @xmath7 model ; for orthogonal ns5 branes , corresponding to an @xmath0 model without an adjoint , @xmath29 assumes the following form @xcite t = v^-k(m - v)/m vw=:=[^3k-2fm]^ , where @xmath161 for @xmath14 and @xmath162 for @xmath13 . the corresponding m5 brane is invariant under the transformation v - v w - w x^7-x^7 t(-1)^n t. these models are also expected to be confining , so we would like to find the corresponding mqcd string , following the considerations in subsection [ sec - hom ] . since this should be an m2 brane , we have to determine first what is the effect of an orientifold projection on such branes . this will be considered in the next subsection . in the m theoretical description of the o4 orientifold , the transformation @xmath163 which is gauged involves not only the space reflection ( [ reflect - m ] ) , but also an orientation reversal of the m2 brane . one way to see this is from 11d supergravity the low energy approximation of m theory . to gauge @xmath163 , it has to be a symmetry of the corresponding action . this action includes a term adada where @xmath164 is a 3-form field . the transformation ( [ reflect - m ] ) reverses the orientation of the 11d space - time ( inverts an odd number of coordinates ) , so to keep the term ( [ sugra ] ) @xmath163-invariant , @xmath164 should change sign ( as already observed in @xcite ) . the m2 brane couples to @xmath164 through a term @xmath165 , and to keep this term invariant too , the orientation of the m2 brane should be inverted . the same conclusion is obtained by considering the brane realization of @xmath166 susy quantum gauge theory ( corresponding to parallel ns5 branes ) . the central charge of a bps state is given by @xmath167 , where @xmath168 is the seiberg - witten differential and @xmath82 is a homology cycle in @xmath29 . in m theory , a bps state is realized by an m2 brane @xmath55 with boundaries on the m5 brane ( on @xmath29 ) , the homology cycle @xmath82 is the boundary of @xmath55 and the integral gives the ( signed ) surface of this brane . in an orientifold model , the @xmath163-mirror @xmath169 of @xmath55 represents the same state and thus should lead to the same central charge . now one observes that @xmath170 is odd under the transformation ( [ reflect - m ] ) , while @xmath29 is symmetric for the @xmath9 models @xcite is not needed here . ] , so the integral @xmath167 changes sign when @xmath82 is transformed by ( [ reflect - m ] ) and , to obtain the same central charge , the transformation ( [ reflect - m ] ) must be accompanied by a flip of orientation . finally , the flip in orientation is also consistent with the limit of ( perturbative ) type iia string theory . there , by definition , @xmath163 flips the orientation of the fundamental string . in m theory , a fundamental string corresponds to a tube - like m2 brane wrapped around the compact dimension , parameterized by @xmath32 . the transformation ( [ reflect - m ] ) does not change the @xmath32 direction , so the direction of the string is correlated with the orientation of the corresponding ( tube - like ) m2 brane . for the analysis of the mqcd string , we need a characterization of the space of m2 brane configurations in a model with an o4 orientifold . if this was an `` orbifold model '' , obtained by gauging the reflection ( [ reflect - m ] ) , the answer would be clear : one would consider general m2 brane configurations in the _ quotient space _ @xmath171 , where = /_2 = ^6s^1 and @xmath172 is the reflection ( [ reflect - m ] ) . in the present case , as explained above , the gauge transformation @xmath163 involves also an orientation reversal of the m2 brane , leading to an undetermined orientation in the quotient space , so to keep the information about the orientation , one has to consider also the _ covering space _ @xmath173 . in the covering space , one should consider general configurations , with the understanding that configurations related by @xmath163 should be identified . one should observe that there are closed configurations in the quotient space @xmath174 that are represented in the covering space @xmath173 by configurations that are _ not _ closed , , `` jump '' between points identified by @xmath163 ( the `` twisted '' sector ; leading to unoriented surfaces in @xmath174 ) so , in general , one should consider also such configurations . we now specialize to a configuration representing an mqcd string , looking for the conserved charge realizing the flux ( recall the discussion in subsection [ sec - hom ] ) . these are m2 branes ending on the m5 brane but , as in the previous models , one can start by classifying m2 branes in the absence of the m5 brane . to find a candidate for a conserved charge , we consider first the case of a closed m2 brane in the covering space ( corresponding to an orientable m2 brane in the quotient space ) . the internal space is @xmath175 , the same as in the @xmath7 case , so these m2 branes seem to be characterized , as before , by the winding number of the cross - section @xmath176 around the compact dimension , leading to @xmath177 . however , here the situation is different . this is because we look for a physical property , which must be @xmath163-invariant ( since this is a gauge symmetry ) , while the winding number is not . to analyze this , it is worthwhile to consider the quotient space . denoting by @xmath45 the curve in @xmath30 that is the image of @xmath176 , the winding number is _ dx^10= _ cdx^10 , where the integration over @xmath45 is performed using the orientation induced on it by that of @xmath176 . different curves @xmath176 in @xmath178 which are related by @xmath163 , map to the same @xmath45 in @xmath30 , but induce on it a different ( continuous ) orientation , leading to a different winding number . for example , replacing @xmath176 by its @xmath163 image leads to the opposite orientation on @xmath45 . since @xmath179 is @xmath163 invariant , the orientation flip implies an opposite winding number . this means that only the winding number modulo 2 _ can _ be gauge invariant and we now show that _ it is _ indeed gauge invariant . to show this , we should compare two different orientations . in general , the orientations can be different only on a part of @xmath45 . this part is , however , necessarily closed ( since both orientations are continuous ) , so it contributes an integer to the winding number . therefore , when its orientation is flipped , the change in the winding number is even , so the winding number modulo 2 is invariant . the above argument also demonstrates that the winding number modulo 2 can be calculated directly in the quotient space @xmath30 , by choosing an _ arbitrary _ continuous orientation on @xmath45 . we return now to the general m2 brane , which can be also unorientable . the lack of orientability is a global property of the surface and will not be reflected in the classification of cross - sections , so these will still be classified by the winding number modulo 2 . the results obtained so far can be summarized as follows : * the gauge invariant property that can be used to classify the cross - sections of an mqcd strings is a @xmath172 charge : the winding number modulo 2 of the cross - section around the compact dimension . * this property can be calculated directly in the quotient space @xmath30 , using eq . ( [ wind ] ) with an arbitrary ( continuous ) orientation on the cross - section @xmath45 . next we analyze the dependence of the @xmath172 charge on the cross - section . for this , let @xmath143 and @xmath144 be two cross - sections ( in the quotient space ) and @xmath55 the part of the m2 brane between them ( so @xmath180 is the boundary of @xmath55 ) . if @xmath55 is orientable , one can choose a global orientation and this leads , as in the @xmath7 case , to the same winding number for any two cross - sections , so the @xmath172 charge is trivially independent of the cross - section . to check the unorientable case , let us represent it as an orientable surface with insertions of cross - cups . assuming even @xmath8 in the transformation ( [ reflect - m ] ) ( odd @xmath8 will be considered in subsection [ odd - n ] ) , @xmath32 is a gauge invariant function on @xmath55 . a difference in the winding numbers of @xmath143 and @xmath144 can be traced to a non - trivial winding for a loop around a cross - cup . however , since opposite points on a cross - cup are identified and , therefore have the same @xmath32 value , the winding number for a loop around the cross - cup is always even , so @xmath143 and @xmath144 have the same @xmath172 charge also in this case . we , therefore , have a well defined , gauge invariant @xmath172 charge for the mqcd string . to proceed , we consider each gauge group separately . as for @xmath7 , we first consider the model without hypermultiplets ( @xmath41 ) . all the fields are in the adjoint representation , which is invariant under the center of the group . the center of @xmath181 is @xmath172 , so we expect an mqcd string with such a charge . this is indeed what we find . as in the previous models , one can consider only closed cross - sections , by supplementing open ones with curves in @xmath29 . these were analyzed in the previous subsection and we obtained a @xmath172 charge that , in the absence of an m5 brane , was conserved . the m5 brane can lead to non - conservation by adding a loop in @xmath29 to a cross - section @xmath45 . the curve @xmath29 is given ( in the covering space ) by eq . ( [ sg - sop ] ) ( with @xmath41 ) , leading to identification modulo @xmath142 in the winding number . @xmath182 is even so the @xmath172 charge is still conserved . this @xmath172 charge can , therefore , be identified with the center of the @xmath14 gauge group . figure [ f - sps ] describes two examples of an mqcd string ( or , more precisely , their cross - sections in the covering space ) , with a unit charge . this looks exactly as for @xmath7 ( compare to figure [ f - sus ] ) , but in that case @xmath176 and @xmath183 are distinct and carry different ( opposite ) @xmath42 charge ( winding number ) , while in the present case , they are @xmath163-equivalent , carrying _ the same _ ( @xmath172 ) charge . it is easy to see how two such strings can annihilate each other : @xmath183 can be rotated until it coincides with @xmath176 and they can join to a loop with a vanishing winding number , which can contract to a point . confinement and screening are analyzed in the same way as in @xmath7 , by introducing semi - infinite d4 branes , representing heavy quarks . also here m2 branes representing isolated quarks look as in figure [ f - qrk ] and are possible only when the ns5 branes are parallel . when the ns5 branes are rotated , the color d4 branes join to a single tube which winds around @xmath32 more then once , so a quark tube can not end on it . instead , it can connect to an mqcd string , whose cross section in the internal space is described in figure [ f - sps ] . so quarks are connected in pairs by a flux tube , forming singlets . note that unlike @xmath7 , in @xmath14 there is no distinction between the fundamental and anti - fundamental representations and any two quarks can combine to form a singlet ( being the analog of both mesons and baryons in @xmath7 ) . this is reflected correctly in the m theory description . in @xmath7 , the two representations were distinguished by the orientation of the corresponding m2 branes and an mqcd string could connect only quark m2 branes of opposite orientation . here , the two orientations are related by @xmath163 so the mqcd string can connect any two quark m2 branes . as in previous models , at low energies , compared to the mass of the quarks , the mqcd string is stable and there is confinement . at higher energies , pairs of quarks can be produced dynamically , screening the confining force . this process is described by figure [ f - scr ] also for the present case . to summarize , we have reproduced all the results described in section 2 , also for the symplectic groups . we now turn to the orthogonal groups . the group @xmath13 is not simply connected : it is covered twice by its universal covering group @xmath186 , so @xmath187 . @xmath188 has a center of four elements - either @xmath189 ( for odd @xmath190 ) or @xmath191 ( for even @xmath190 ) , so this should be the charge of a general string in the model without hypermultiplets ( @xmath41 ) . the search for the mqcd string is identical to the one for @xmath181 , with the same results : mqcd strings with a @xmath172 charge . this means that we do not obtain the full spectrum of possible mqcd strings , but actually this could be expected . a string ( flux tube ) with a unit charge would have to be connected to @xmath13 spinors . these are not realized in the present brane configuration , so it is reasonable that the corresponding strings are not realized either . we conclude that we have found an mqcd string carrying the charge of the _ vectorial _ representation of @xmath184 . this is verified when we add semi - infinite d4 branes , which realize quarks in this representation : the m2 branes realizing the quark states are bound in pairs by such a string , forming a meson , as in the @xmath181 model . the demonstration of confinement and screening is also identical in the two models . note , however , that not all charges are screened : the dynamical matter can screen ( for @xmath192 ) any external probe which is in a representation of @xmath184 ( rather then @xmath188 ) . when spinorial external probes are available , they should feel a confining force , but we do not have a description of those in the present formulation , so this can not be checked . the center of @xmath194 is @xmath172 and the vectorial representation is uncharged under it ( being a representation of @xmath195 ) . therefore , in view of the discussion of the @xmath184 models , we expect here no mqcd string . to check this , we return to the discussion at the end of subsection [ s - m2 ] . the present case corresponds to odd @xmath8 in the transformation ( [ reflect - m ] ) , meaning that an orientation flip in @xmath55 is accompanied by a jump of @xmath196 in @xmath32 . a loop around a cross - cup will , therefore , have an _ odd _ winding number and this implies that there will always be cross - sections of the mqcd string with an even , trivial , winding number . thus , in the present case , although the @xmath172 charge of a specific cross - section of the string is well defined , it does not protect the string from breaking , since it is not the same for all cross - sections . this means that there is no stable mqcd string , in agreement with the field theoretical expectations . note that we obtained this result even before considering the m5 brane ! for completeness , let us nevertheless consider the influence of @xmath29 . it leads , as before , to identification modulo @xmath142 in the winding number and , since this time @xmath197 is odd , the @xmath172 charge of a cross - section is no longer conserved . as for @xmath184 , the absence on a stable mqcd string means that matter in representations of @xmath193 ( and , in particular , the vectorial one ) is screened . spinorial charge is expected to be confined . in this work we used the realization of supersymmetric gauge theories in m theory to investigate confinement and screening . we considered the following ( @xmath0 supersymmetric ) models : * @xmath9 susy @xmath7 gauge theory with @xmath20 fundamental flavors , perturbed by a polynomial superpotential for the adjoint hypermultiplet ( @xmath0 sqcd can be seen as a special case of this family , for which the adjoint hypermultiplet is infinitely heavy ) ; * @xmath9 susy @xmath111 gauge theory with matter in fundamental and bi - fundamental representations , perturbed by masses for the adjoint hypermultiplets ; * @xmath14 gauge theory with @xmath20 fundamental hypermultiplets ; * @xmath13 gauge theory with @xmath20 vectorial hypermultiplets . for all the above models , we found results in agreement with the field theoretical expectations . starting without fundamental matter ( vectorial , for @xmath13 ) , the mqcd string ( , a candidate for the field theoretical electric flux tube ) was identified as an m2 brane ending on the m5 brane ( following @xcite ) . this string was shown to carry a topologically - conserved charge under a group isomorphic to the group @xmath11 of gauge transformations that act trivially on the fields in the field theory . this is in accord with the identification of this charge as the gauge charge that is _ not screened _ by the dynamical matter and , therefore , confined . we then introduced heavy quarks carrying unscreened charge and showed that they can not exist in isolation , but can be connected to an mqcd string , forming neutral mesons . this is a demonstration of confinement . finally , we considered higher energy scales , at which these quarks become dynamical and found that this destabilizes the mqcd string , demonstrating the screening of the charge that these quarks carry . in the @xmath13 models , we did not obtain strings carrying spinorial charge : for example , for @xmath184 , the mqcd string carries a @xmath172 charge ( which was shown to correspond to the vector representation ) , while the field theory expectation is for a charge group @xmath11 with 4 elements . the fact that the charge group was at most @xmath172 , followed from the topology of the internal space @xmath30 ( as implied by the o4 projection ) , and was independent of the m5 brane . this can serve as guidance in the search for a realization of models with spinors : the presence of heavy charge , since the vector screens the rest . ] spinors and vectors would imply the existence of a @xmath189 or @xmath198 charge group and to obtain such a charge , a different background is needed ( , larger orientifold group ) . a different configuration of ns5 branes and d4 branes would correspond to a different m5 brane in _ the same background _ , so it will not lead to the desired result . the above considerations demonstrate that the properties of the mqcd string can be used in the identification of the field theoretical model realized by an m5 configuration , or serve as a consistency check . it is , therefore , worthwhile to perform this analysis in other such realizations . * acknowledgment : * we are grateful to a. giveon , a. hanany , m. henningson , k. landsteiner , r. livne , e. lopez , y. oz and a. zaffaroni for helpful discussions . o.p . is grateful to the theory division at cern , where part of this work was performed , for hospitality . this work is supported in part by bsf american - israel bi - national science foundation , and by the israel science foundation founded by the israel academy of sciences and humanities centers of excellence program .

electrical pacing is a common procedure that is used in both experimental and clinical settings for studying and/or annihilating anatomical reentry . in a recent study [ comtois and vinet , chaos * 12 * , 903 ( 2002 ) ] , new ways to terminate the one - dimensional reentry using a simple protocol consisting of only two stimulations were discovered . the probability of annihilating the reentrant activity is much more probable by these new scenarios than by the usual local unidirectional block . this paper is an extension of the previous study in which the sensitivity of the new scenarios of annihilation to the pathway length is studied . it follows that reentry can be stopped over a limited interval of the pathway length and that increasing the length beyond the upper limit of this interval yields to a transition to sustained double - wave reentry . a similar dynamical mechanism , labeled alternans amplification , is found to be responsible for both behaviors .

physics0310031

the picture of a fixed waveform traveling at constant speed around a ring of excitable tissue , is still a common representation of functional reentry in the clinical setting , particularly in reference to common atrial flutter @xcite . however , the findings that complex reentries are possible even in a simple homogeneous one - dimensional ionic loop model and that their occurrence is dependent on the steepness of the restitution curve of the action potential duration has altered the current understanding of the phenomena @xcite , in which reentry was postulated to remain stable and periodic as long as there was a minimal excitable gap ahead of the wavefront . these findings have also altered the thinking about the effect of antiarrhythmic drugs . @xcite . overdrive pacing using a transvenously inserted catheter in the right atrium is a standard clinical procedure to interrupt atrial flutter . it is very successful , particularly when it is applied in conjunction with the administration of class i or iii antiarrhythmic drug @xcite the use of rapid pacing is likely to increase with the implant of permanent single or dual site stimulator for the prevention of atrial tachycardias@xcite . however , the mechanism by which overdrive pacing interrupts reentry and the electrophysiological parameters of the reentry circuit that may determined an optimal choice of parameters for the pacing protocol are not understood . as a first step to improve the pacing algorithm , we have previously studied of a simple protocol of stimulation consisting of two electrical stimuli applied in the pathway of a periodic reentry @xcite . new scenarios of reentry annihilation were identified , different from the classical unidirectional block @xcite , which is still considered to be the most important mode of termination . these alternative scenarios of reentry annihilation follow from a spatiotemporal process that we have called alternans amplification @xcite . the first objective of this paper is to understand the effect of length of the reentry pathway on these scenarios of annihilation . we also show that beyond a critical length of the reentry pathway , alternans amplification induces a transition double - wave reentry instead of annihilation . results obtained with two different models are presented . the first model ( ionic loop : il ) is a one - dimensional reaction diffusion system , using a cardiac ionic model to represent the transmembrane currents . the second model is an integral - delay equation ( i d ) based on the local properties of propagation and repolarization . the well - known monodomain cable equation for an 1d homogeneous excitable cardiac tissue embedded in an unbounded external medium of negligible resistivity is:@xmath0 where @xmath1 is the transmembrane potential ( mv ) , @xmath2 is the membrane capacitance ( @xmath3f@xmath4 ) , @xmath5 the surface - to - volume ratio ( @xmath6m@xmath7 , assuming cylindrical cells with radius of @xmath8 m ) and @xmath9 is the intracellular resistivity ( @xmath10 cm ) . @xmath11 is given by a modified version of the beeler - reuter model ( mbr ) of the cardiac cell membrane , whose details and space - clamp dynamics are given in @xcite . for each time step ( @xmath12 , the system becomes a second order ordinary differential equation that is computed with a galerkin finite element method projected on a linear basis function and a regular spatial mesh ( @xmath13)@xcite . the resulting tridiagonal linear system of equations is solved with a simplified lu decomposition method . the choice of @xmath14 and @xmath15 is motivated by the fact that depolarization is the stiffest part of the process . programs were written in fortran77 and ran on sgi workstations ( silicon graphics ) . following the generation of an action potential , two quantities are measured at each site to analyze the propagation : the activation time ( @xmath16 ) and repolarization time ( @xmath17 ) . @xmath16 corresponds to the onset of the action potential and is defined as the moment at which @xmath18 reaches its maximum during the upstroke of the action potential . @xmath17 is meant to indicate the time from which a new action potential can be generated by an incoming activation front or an external stimulus . a large set of simulations of sustained reentry in loops of different lengths has showed that an active propagating response was generated if stimulation was applied at least@xmath19 ms after the@xmath20 mv post - upstroke downcrossing in repolarization . accordingly , this instant ( i.e. 30 ms after @xmath21 mv crossing ) is taken as @xmath17 . the action potential ( @xmath22 ) is defined as the time interval from @xmath16 to @xmath17 , such that @xmath23 . the diastolic interval ( @xmath24 ) associated to an activation occurring at @xmath16 is defined as the time between the previous @xmath17 and current @xmath16 . with these definitions , a site is excitable if @xmath25 . the integral - delay model used in this study is an extension of a previous formulation that was developed to describe sustained unidirectional propagation on the loop @xcite . a first relation @xmath26 gives the duration of the action potential as a function of @xmath27 in the space - clamped configuration . if the nodes were disconnected from their neighbors , the repolarization time following an activation occurring at @xmath16 would be:@xmath28 provided that @xmath29 . if @xmath30 , the node is unexcitable . the actual repolarization time of a node at position @xmath31 is expressed as a weighted average of @xmath32 over a symmetric neighborhood of length @xmath33 , i.e.:@xmath34 where , @xmath35 with ( @xmath36 , @xmath37 @xmath4 , @xmath38 cm ) , is the weighting function representing the effect of resistive coupling on the repolarization phase . @xmath39 controls the spatial decay of the weighting function and @xmath40 is a normalization coefficient such that @xmath41 . the calculation of @xmath42 associated to one excitation is performed at each node at the next instance when it is stimulated by an incoming front or an external stimulus . at this moment , the @xmath43 associated to the previous excitation of each point of the neighborhood are collected and averaged to produce @xmath42 . in this way , a front whose propagation stops at some location still produces a continuous distribution of @xmath42 around the region of block since @xmath42 is an weighted average of the @xmath43 of the sites excited by the front that is blocked and of those that were not reached by that front and still have the @xmath32 associated to their previous excitation . it provides at once a representation of the acceleration of repolarization of the excited cells induced by the load of those that are not excited , and of the prolongation of repolarization in those that were not excited by the electrotonic depolarization induces by the proximal excited cells . once the @xmath42 associated to the last excitation that we label @xmath44 has been computed , the diastolic interval associated to the current stimulation , which take place at the time @xmath16 , is calculated as : @xmath45 if @xmath46the stimulus produces an action potential@xmath47 which propagates on both side with a conduction time @xmath48 and reached the neighbouring nodes at the time @xmath49 @xmath50 . if @xmath30 , the point is not activated , and its @xmath32 is not changed . the integral - delay model was originally developed to represent the propagation of a single activation front during reentry , without external stimulation . in this context , @xmath51 is always the repolarisation associated to the previous passage of the activation front and it can be written as @xmath52 . similarly , @xmath53 . with these relations , eq . [ mnide ] , becomes:@xmath54 if the conduction time is constant @xmath55 and eq . [ mnide ] is equivalent to the version of the i d model introduced in @xcite . in fact this version neglects the effect of the delay of propagation in calculating the effect of coupling on repolarization . if @xmath56 is taken as a @xmath57 function , which is equivalent to ignoring the effect of coupling , eq . [ mnide ] corresponds to the version of the integral - delay model of courtemanche and al.the simulation of the i d model were performed using @xmath58 where @xmath59 ms / cm , @xmath27 is in ms and @xmath60 is in ms / cm , and @xmath61 \frac{\left ( d+30\right ) ^{4.92}}{f^{4.92}+\left ( d+30\right ) ^{4.92 } } \label{eq_restitution}%\ ] ] with @xmath62 ms , @xmath63 ms , @xmath64 ms , and @xmath65 ms @xcite . these functions were obtained by fitting the data gathered from different regimes of propagation ( free running periodic and qp reentry obtained with the il model)@xcite . computations were performed with a spatial discretization of @xmath66 @xmath67 , as in the ionic model . for the il model , @xmath68 consists of a 2.5 ms current pulse applied over an interval of 450 @xmath67 with amplitude of 60 @xmath69a/@xmath70 . dual stimulations were applied on periodic reentry . the timing of stimuli was controlled by two parameters : @xmath71 , the time interval between the last activation at the center of the stimulated area and the onset of the first stimulus @xmath72 ; @xmath73 , the time interval between the onset of @xmath72 and that of the second stimulus @xmath74 for the i d model , the value of @xmath27 , defines the response of the nodes that are stimulated . if @xmath30 , the stimulus is applied in the refractory period and doest not have an effect . if @xmath29 , the stimulus depolarizes the tissue , defining @xmath26 , and induces bi - directional propagation . as for the il model the stimulation covers 450 @xmath67 . the post - stimuli dynamics are constrained by the steady states of the system . for loops longer than a minimum length @xmath75 , sustained reentries are stable attractor of the system . these sustained reentries can be either periodic ( period-1 ) or quasiperiodic ( qp ) , and hold a single ( sw ) , two ( dw ) or more traveling activation fronts . table [ table_attract ] lists the stable solutions of both the i d and il models for @xmath76 . the number and nature of the sustained reentries change with @xmath77 and condition the outcomes of the stimulations . [ c]|l|ll|interval & reentry type & + @xmath77 ( cm ) & sw ( single wave ) & + @xmath78 & @xmath79 & + @xmath80 & qp , mode-0 & + @xmath81 & @xmath82{c}% \text{qp},\text{mode-0}\\ \text{qp},\text{mode-1}% \end{array } $ ] & + @xmath83 & period-1 & + @xmath84 & @xmath85 & + @xmath86 & period-1 & + @xmath87 & @xmath85 & + qp reentries are characterized by a spatial oscillation of @xmath27 , with a wavelength @xmath88 that is an irrational fraction of @xmath77 ( fig . [ quasi_sust]a and c ) . for both sw and dw reentries , there is an interval of @xmath77 in which two different qp solutions coexist . these solutions , labeled mode-0 and mode-1 , have a similar structure but @xmath89 . in sw reentries , the passage of each activation front is associated with a profile of @xmath90 and @xmath27 holding one ( mode-0 , panel a ) or multiple ( mode-1 , panel c ) maxima and minima over two turns . for dw qp reentries , @xmath77 is twice the value for sw qp reentries , such that mode-0 has one maximum and one minimum over one turn . [ quasi_sust]a ) . successive activations at each site alternate between long and short @xmath90 and @xmath27 values , except at a number of nodes corresponding to the boundaries from which the phase of the alternation is inverted . as illustrated by the time - course of @xmath16 and @xmath17(fig . [ quasi_sust ] , right panels ) , the quasi - periodic nature of the propagation makes the position of the extrema and of the nodes to drift slowly in the direction inverse to the propagation of the activation fronts . qp reentry is thus constituted by discordant alternans @xcite with boundaries between short and long apd moving around the loop . [ ptb ] fig_di_quasisust.eps period-1 sw reentry can be annihilated by an isolated stimulus applied in the narrow vulnerable window in which the stimulation produces only a retrograde front , corresponding to the well - known mechanism of unidirectional block @xcite . in a previous paper , we have also described other modes of annihilation as well as different transient dynamics that were induced by two successive stimuli @xcite . these new modes of annihilation were compatible with experimental observations , and relevant to antiarrhythmic pacing therapy @xcite . however , that study was restricted to a specific length of the loop ( @xmath91 cm @xmath92 ) . in the following , we present a systematic investigation of the outcomes of double stimuli applied on period-1 sw reentry as a function of the timing of the pulses and the length of the loop . complex dynamics can be induced by a second stimulus @xmath74 thanks to the asymmetric profile in @xmath17 left by the interaction of first stimulus @xmath72 with the reentry activation front @xmath93 . when @xmath71 , the time between the last passage of @xmath93 and the onset of @xmath72 , is beyond the vulnerable window , @xmath72 produces both a retrograde ( @xmath94 ) and an antegrade ( @xmath95 ) activation front . as illustrated in fig . [ fig_gradtrep]a , the key factor determining the dynamics that can be induced by @xmath74 is the region located between the site of stimulation ( @xmath96 ) and the site of the collision between @xmath93 and @xmath94 ( @xmath97 , identified by the arrow in [ fig_gradtrep]a ) . @xmath42 ( thin line in fig . [ fig_gradtrep]a ) is minimum near @xmath96 and reaches its maximum at @xmath97 . the il and i d models produce the same profile of @xmath17 ( thin lines and triangles , respectively , in [ fig_gradtrep]a ) , showing that the i d model that was initially developed to describe sustained reentry also provides an appropriate low - dimensional representation of the dynamics when stimulations are applied . the location of @xmath97 as well as the profile of @xmath17 around @xmath96 depend on both @xmath71 and @xmath77 . figure [ fig_gradtrep]b shows the profile of @xmath17 obtained from loops of different lengths stimulated at the same diastolic interval @xmath98 , where @xmath99 is the action potential duration of the stable reentry for each @xmath77 . the position of @xmath97 is shifted to the left ( arrows in [ fig_gradtrep]b ) because the collision is delayed on longer loops . however all the loops have the same invariant profile of @xmath42 in the time and space interval that they share before the collision . if @xmath72 is applied at larger @xmath100 value , the distance between @xmath96 and @xmath97 is shortened , @xmath101 and @xmath102 are increased , such that the extent and depth of the cusp in @xmath42 around @xmath96 are diminished . [ ptb ] fig_gtrep.eps the spatial profile of @xmath42 for short @xmath71 is asymmetrical , with a sharp gradient between @xmath96 and @xmath103 , and a more gradual increase at the right of @xmath96 . owing to this asymmetry , the outcome of @xmath74 depends on @xmath73 , the time interval between the onset of the two stimuli . figure [ blockr2 ] illustrates a case in which @xmath74 is applied after the collision of @xmath93 and @xmath94 , at an instant where @xmath95 , the antegrade front created by @xmath72 , still has not reached @xmath97 . @xmath74 creates both an antegrade ( @xmath104 ) and a retrograde ( @xmath105 ) front , but @xmath105 is blocked between @xmath96 and @xmath97 . thereupon , the system is left with two antegrade fronts ( @xmath95 and @xmath104 ) . this occurs as long as @xmath105 does not propagate beyond @xmath97 to collide with @xmath95 , in which case @xmath104 is left alone to perpetuate the reentry . this is an alternate scenario of unidirectional block that creates a propagating wave in the same direction as @xmath93 and @xmath95 . [ ptb ] fig_blr2.eps all the complex dynamics occur in the range of @xmath73 for which @xmath105 is blocked between @xmath96 and @xmath97 whereas @xmath104 propagates . this depends on the profile of @xmath106 left by @xmath72 , which was shown to be an invariant function of @xmath100 in fig . [ fig_gradtrep]b . figure [ fig_map_super ] shows the global characteristics of the dynamics in the [ @xmath107 ] parameters plane for two values of @xmath77 ( @xmath108 and @xmath109 cm @xmath110 ) . the parameter plane can be divided in three areas . in the region labeled @xmath111 at low @xmath73 values , @xmath112is applied during the refractory period and does not produce a response . in the upper region , labeled `` @xmath113 '' , @xmath105 propagates beyond @xmath97 , collides with @xmath95 , and @xmath104 is left alone to maintain the reentry . in the middle area , @xmath105 is blocked between @xmath97 and @xmath114and complex dynamics may occur . the phase plane area in which @xmath105 is blocked , for @xmath100 , between @xmath115 and @xmath116 ms ( dotted vertical line ) , is the same for the two @xmath77 . the specific subsets in which complex dynamics occurs ( represented by different shaded areas in fig . [ fig_map_super ] ) change with @xmath77 , and are discussed later . the article is focused on the area where @xmath105 is blocked , between @xmath117 and @xmath118 ms . the lower limit of the area coincides with @xmath119 and is close to the action potential duration restitution curve @xmath120 . the upper limit is nearly a constant , around @xmath121 ms . appendix [ appendix_a ] shows that this @xmath73 upper limit is set by the locus between @xmath114 and @xmath97 where @xmath122 is equal to the maximum conduction time . this locus , indicated by a circle for the specific @xmath100 illustrated in [ fig_gradtrep]b , does not depend on @xmath77 , which explains why the upper limit for the block of @xmath105 is identical in the two loops . @xmath122 remains everywhere below the maximum conduction time if @xmath100 is too long , explaining why the area of complex dynamics disappears beyond @xmath100 @xmath118 ms . it is also demonstrated in appendix [ appendix_a ] that the maximum slope of the @xmath123 function must be greater than @xmath124 to allow the block of @xmath105 . the same condition that controls the stability of the period-1 reentry @xcite thus determines if complex dynamics can be induced by @xmath74 . [ ptb ] fig_map_super.eps once @xmath104 has started to propagate and @xmath105 has been blocked , there are 4 possible outcomes : \1 ) @xmath95 is blocked and @xmath104 perpetuates the sw reentry , \2 ) @xmath104 is blocked , and @xmath95 maintains the sw reentry , \3 ) @xmath95 and @xmath104 are blocked , and reentry is annihilated \4 ) neither @xmath95 nor @xmath104 are blocked , and there is a transition to dw reentry . as seen in table [ table_attract ] , this last option can only occur if @xmath125 . the next section discusses the cases 3 ) and 4 ) in which the system does not return to the original sw period-1 reentry . the three new scenarios of reentry annihilation reported in @xcite are shown in fig . [ fig_bloc_aa ] . these scenarios of termination differ in the number of revolution made by the @xmath95 and @xmath126activation fronts before they are blocked . accordingly , we introduce the notation @xmath127 , meaning that the front @xmath128 is blocked ( @xmath129 ) after @xmath130 turns around the loop . let consider the simplest case ( [ @xmath131 , fig . [ fig_bloc_aa]a ) in which both @xmath95 and @xmath104 are blocked after one rotation . @xmath95 is blocked first near @xmath96 , when it reaches the refractory tail left by @xmath104 and @xmath105 . this occurs because @xmath95 has already completed a fraction of its rotation when @xmath74 is applied . as a consequence , @xmath95 comes back to reactivate the sites near @xmath96 after a time interval much shorter than the period of rotation @xmath132 of the stable reentry . the block of @xmath104 takes place between @xmath97 and the site where @xmath105 was blocked . when @xmath95 travels in this zone , the region has last been excited by @xmath94 , and a time interval longer than @xmath132 has elapsed since this last excitation . besides , both the action potential and @xmath42 associated to @xmath94 were short since @xmath72 was premature . as a consequence , @xmath95 produces action potentials that are longer than those of the stable reentry . since the time between the passage of @xmath95 and the return of @xmath104 is also shorter than @xmath132 , @xmath104 is blocked . the block of @xmath104 thus results from a process of amplified alternation in a region between @xmath96 and @xmath97 . the premature @xmath94 that creates short action potential is followed by the late @xmath95 generating long action potential . in the two other scenarios , @xmath95 nor @xmath104 are blocked after either 2 or 3 turns ( [ @xmath133 , [ @xmath134 in fig . [ fig_bloc_aa]b - c ) . each passage of @xmath95 leaves around @xmath96 a convex profile of @xmath42 , which is turned to a concave profile by the subsequent passage of @xmath104 . the panel b and c illustrates the process of alternans amplification . alternans amplification may end up by the blockade of one of the front , in which case propagation reverts to sw reentry . it may saturated , leading to dw reentry , or may lead to annihilation . the next section explores the conditions for annihilation . [ ptb ] fig_bloc_aa.eps [ [ the - scenarios - of - annihilation - by - alternans - amplification - occurs - over - limited - range - of - l ] ] the scenarios of annihilation by alternans amplification occurs over limited range of @xmath77 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the conditions leading to each type of alternans amplification annihilation vary with @xmath77 , and none exists for @xmath77 @xmath135@xmath136 cm . the four panels of fig . [ fig_map_single ] picture the extent of the different zones of annihilation for @xmath77 ranging from @xmath137 , the minimum @xmath77 with stable period-1 sw reentry , to the limiting @xmath138 cm value . at short @xmath77 ( @xmath139 @xmath140 , fig . [ fig_map_single]a ) , there is a large [ @xmath100 , @xmath73 ] area with [ @xmath131 block located at low @xmath100 values , and a small adjacent area of [ @xmath133 block . the zone of [ @xmath133 always remains minimal ; and it is the first to disappear at @xmath142 cm . the zone of [ @xmath143 appears at intermediate @xmath77 , expands , and is the last to disappear . [ h ] fig_map1.eps [ [ disappearance - of - thea_11dashva_21dashv - annihilation ] ] disappearance of the [ @xmath131 annihilation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + taken together , the maps of fig . [ fig_map_super ] and fig . [ fig_map_single ] a - c shows that the area of [ @xmath144 is embedded in the larger region , invariant with respect to @xmath77 , in which @xmath105 is blocked between @xmath97 and @xmath96 . the zone of [ @xmath131 is located at intermediate @xmath145 , just over the area , labeled [ @xmath146 ( fig . [ fig_map_super ] ) , in which @xmath95can propagated beyond @xmath96 , but @xmath104 is stopped by the refractory tail left by @xmath95 . as @xmath77 is increased , the lower boundary of [ @xmath131 is lifted , thus diminishing the [ @xmath131 area until it disappears completely . the lost of [ @xmath131 annihilation is thus caused by the inability of the system to block @xmath95 at its first return . appendix [ appendix_b ] provides the conditions needed for @xmath95 to block in the tail of @xmath104 and proves that there is a limiting @xmath77 beyond which this can not happen . to summarize : 1 ) increasing @xmath73 produces longer action potential for @xmath74 , delays @xmath102 , and thus augments the likelihood of @xmath95 to be blocked . 2 ) however , the increase of @xmath73 and @xmath102 are bounded by the condition of @xmath105 being blocked between @xmath96 and @xmath97 , and these limiting values are independent of @xmath77 . 3 ) since increasing @xmath77 delays the return of @xmath95 , there is a length @xmath77 from which the return of @xmath95 always occurs after the limiting @xmath102 value . [ [ opening - the - way - to - more - than - one - rotation - for - both - fronts ] ] opening the way to more than one rotation for both fronts + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + figure [ fig_map_single ] shows that , for each value of @xmath77 , the zone of [ @xmath131 and [ @xmath147 blocks are disjoint , being respectively located at low and high @xmath100 . as seen in fig [ fig_map_single]d , the zones of [ @xmath148 remain located at large @xmath100 even at values of @xmath77 for which [ @xmath131 does not exists anymore . however , enlarging @xmath77 also leads to the appearance and extension of areas in which @xmath95 and @xmath104 persist together for an increasing number of turns before one of the front is blocked ( up to 7 activations in [ fig_map_single]d ) . as @xmath77 is increased , the areas of the parameters plane associated to these other forms of transient complex dynamics extend toward low @xmath100 , to finally cover all the range from @xmath117 to @xmath149 ms when @xmath150 cm , a value close to @xmath151 from which dw reentry becomes possible . the areas of the parameters plane associated to these behaviors with coexistence of @xmath95 and @xmath104 for multiple turns form a sequence of contiguous parallel tongues . all the higher modes of block and complex dynamics require that both @xmath95 and @xmath104 travel beyond @xmath114 , at least at their first return . conditions for the block of @xmath104 are derived in appendix [ appendix_c ] based on an approximation using the @xmath120 equation and rules for @xmath95 and @xmath104 return cycles . it shows that the block of @xmath104 depends on the balance between the return cycle of @xmath95 and @xmath104 . in both the i d and il model , annihilation by alternans amplification is impossible from @xmath77 @xmath152 30.5 cm . in fact , for 31 @xmath153 30.5 cm , only transient complex dynamics with a final return to period-1 sw reentry are observed . however , the maximum number of turns with @xmath95 and @xmath104 co - traveling grows , just as the number of tongues in the [ @xmath154 ] plane associated to different numbers of turns during which the two fronts coexist . each new tongue appears at low @xmath100 and @xmath73 values , and expands as @xmath77 is further increased . finally , a first transition to qp mode-0 dw reentry is detected at @xmath155 cm ( fig . [ map_6200]a ) . transition to dw reentry thus appears as the asymptotic limit of the prolongation of the transient propagation with two fronts . however , transition to dw reentry begins much beyond @xmath156 @xmath157 cm , the value at which sustained mode-0 qp dw reentry starts to exist . in fact , at @xmath155 cm , the system is rather in the range of @xmath77 for which both dw mode-0 and mode-1 solutions coexists . [ tbh ] fig_map_6200.eps for longer @xmath77 , ( fig . [ map_6200]b ) the region with transition to mode-0 dw reentry expands , as it was the case for all the zones associated to two fronts transient propagation created at shorter @xmath77 . transition to mode-1 dw reentry also appears . however , the transition to mode-1 appears at low @xmath100 , but at two disjoint intermediate @xmath73 values , giving rise to two separated areas . for still longer @xmath77 , the areas of mode-0 and mode-1 transition enlarge until covering , from @xmath150 cm , all the area where @xmath105 is blocked . hence , from this value of @xmath158transition to dw reentry occurs in all instances where @xmath105 is blocked . finally , beyond @xmath159 cm , all transitions go to period-1 dw reentry , which is the only dw solution that remains . increasing the length of the pathway correlates with an increase in the complexity of the dynamics and to the transition to sustained dw reentry . the results presented in fig . [ map_6200 ] show that changing the timing of the stimuli when @xmath160 can induce the transition either to the mode-0 or mode-1 dw qp reentry . the il model was simulated with @xmath150 cm ( @xmath161 but near @xmath151 ) to circumscribe the basin of attraction in the parameter space associated to each dw qp reentry . the large area in which two antegrade fronts are created ( @xmath162 ms ) is separated between two regions , with lower @xmath73 converging to mode-0 , and higher @xmath73 to mode-1 . in this last section , we compare the transient dynamics leading to each of two modes of dw qp reentry . figure [ fig_di_quasi ] shows the spatial profile of @xmath27 associated to @xmath95 and @xmath104 for two cases converging respectively to mode-0 ( panel a ) and mode-1(panel b ) . in this representation,@xmath163the passage of @xmath104 ( @xmath164 , dotted line ) at each location is followed by that of @xmath95 ( @xmath165full line ) . at first , @xmath104 propagates with a short @xmath27 , producing brief action potentials everywhere along the loop . as a consequence , @xmath95 travelling afterwards meets long @xmath166 . @xmath167 has its first maximum ( p@xmath168 ) when it travels between @xmath97 and the point where @xmath105was blocked , such that @xmath169 has a minimum at the same location from which it increases slowly until its next return in the same region . from there , @xmath164 starts to alternate between long and short values with a spatial period close to 2 turns , and @xmath167 follows a complementary profile . this first phase of the propagation , lasting for approximately 8 turns , can be labeled as concordant alternans since @xmath169 and @xmath170 have alternating values and that each remains short or long for at least one complete rotation . this pattern does not correspond neither to mode-0 nor mode-1 , since both stabilized dw qp solutions have a wavelength less than @xmath77 . however , from the beginning ( @xmath171 ) , @xmath170 already shows a second spatial oscillation in @xmath27 that is superimposed to the concordant alternans . the structure of this oscillation , which has a wavelength close to one turn , makes the difference between the cases converging to mode-0 and mode-1 . in the former case , the oscillation embeds two peaks \{p@xmath168 , p@xmath172 } ( @xmath170 for mode-0 , continuous line in fig . [ fig_di_quasi]a ) , while in the later case , it has three peaks \{p@xmath168 , p@xmath172 , p@xmath173 } ( @xmath170 for mode-1 , continuous line in fig . [ fig_di_quasi]b ) . these superimposed spatial variations persist while the concordant alternans dissipate . during this process , the position of the peaks does not change much . as the amplitude of the concordant alternans decreases , the respective height of the discordant alternans increases up to a point where the boundary with large gradient in @xmath27 begins to move around the loop due to the quasiperiodic nature of the propagation . this travelling mechanism is akin to the propagation of paced discordant alternans on a cable of cardiac tissue@xcite . [ tbh ] fig_di_quasi.eps the main difference between the transition to mode-0 and mode-1 is the presence of the third peak p@xmath173 in @xmath170 . the p@xmath173 peak of @xmath170 is induced by the increase of @xmath73 . on one hand , a larger @xmath73 produces a longer @xmath42 such that @xmath95 meets a lower @xmath27 and the amplitude of p@xmath168 is reduced . but a larger @xmath73 also implies that @xmath104 travels faster , comes back sooner to the stimulation sites , and set the stage for a new maximum in the @xmath170 profile . of course the transition from two to three peaks is continuous process since a similar variation of @xmath27 with less amplitude already exists with the transition to mode-0 in fig . [ fig_di_quasi]a . it means that there must be a minimal spatial profile that corresponds to the boundary between the two basins of attraction ( the transition to either mode-0 or mode-1 ) . alternans amplification , leading either to reentry annihilation or transition to dw reentry , can exist if the slope of the apd restitution curve is larger than one . hence , the condition on the slope of the apd restitution curve that is mandatory for the existence of sustained qp reentry@xcite also enables double - pulse stimulation to produce a new mode of unidirectional block , in which only an antegrade front propagates away from the stimulation site . on any closed circuit with two activation fronts travelling in opposite direction , this opens a large spatio - temporal window in which an ectopic focus or an external source firing twice can start a reentry . this is consistent with the use of burst pacing as a standard experimental and clinical procedure to start tachycardia @xcite . in a loop already holding a sw reentry , the @xmath174 protocol can induced the simultaneous propagation of two antegrade fronts whose final outcome depends on the timing of the stimuli and the length of the loop . the transient or persistent coexistence of two antegrade fronts is a new type of dynamics in which the effects of the stimuli can not be represent as perturbations of a limit - cycle , as it has been done for models in which the steep slope criteria was not fulfill @xcite . in clinical and experimental investigations , stimulations are currently used to either study the characteristics of reentry circuits through resetting or to stop the tachycardias @xcite . most often , stimulations are applied at one site and propagation is assessed through one or a few recording electrodes . we may consider the dynamics that will be observed with this setting for each of the three scenarios of annihilation by alternans amplification . since the distance traveled by @xmath94 from @xmath114 to @xmath97 usually covers a small portion of the reentry pathway , electrodes are much more likely to be positioned in the complementary portion of the circuit . for [ @xmath131 , such electrodes will detect the last passage of @xmath93 before the collision with @xmath94 , then @xmath175and @xmath104 . the detection of @xmath175and @xmath104 would clearly exclude classical unidirectional block being responsible for the annihilation . in fact , this modeling study was initiated after a set of experimental and clinical studies on flutter using multichannel ( 4 to 8 channels ) recordings @xcite . in these , cases of annihilation were reported in which @xmath93 , @xmath175 , @xmath104 were detected , and in which the propagation of @xmath175and @xmath104 was blocked in the segment of the circuit were the collision of @xmath93 and @xmath175was presumed to have occurred . this scenario , that was called collision block , is consistent with the [ @xmath131 block by alternans amplification @xcite . in the case of [ @xmath133 block , @xmath95 and @xmath104 would be detected twice ( see [ fig_bloc_aa]b ) . since the time intervals from each @xmath95 to the next @xmath104 are longer than that between @xmath104 and the following @xmath95 , the sequence of the differences between activation times alternates around a value shorter than the period of the original reentry , the block occurring after the longer interval of the time series . for [ @xmath143 , each front is seen thrice , with similar oscillation in the time series of the difference . however , the structure of oscillation of the cycle as well as the last value before annihilation depend on the position of the electrode in the circuit . in a protocol for annihilation by unidirectional block , the detection of @xmath175and @xmath104 would indicate that the stimuli were beyond the vulnerable window and would trigger the application of a new stimulus at shorter coupling interval . however , this could rather reinitiate a reentry bound to stop by alternans amplification . to allow the unidirectional block of @xmath105 , @xmath72 must be applied early beyond the vulnerable window , in the portion of the excitable gap referred as partially refractory by electrophysiologists @xcite . the prematurity of @xmath72 , in conjunction with the high slope of the apd restitution curve and of the @xmath176 dispersion curve , create a concave asymmetrical profile of @xmath17 around @xmath114 . for a given @xmath72 , the @xmath73 intervals for which @xmath105 is blocked depends on the local dynamics around @xmath114 that is again determined by the apd restitution and the dispersion of @xmath176 . in fact , when the parameters @xmath100 is used to describe the timing of the first stimulus , the [ @xmath177 range to get a block of @xmath105 becomes independent of @xmath77 . the pivotal role of the @xmath26 and @xmath178functions is further confirmed by the capacity of the i d model to reproduce the dynamics of the ionic model . however , in order to avoid discontinuity in @xmath17 in the region where @xmath105 is blocked and get a correct representation of the dynamics , spatial averaging must be included in the computation of the spatial profile of @xmath17 . originally , spatial averaging was added to the i d model to reproduce the modulation of the repolarization by resistive coupling in order to correct the shortcomings of the model regarding the details of the bifurcation from periodic to qp propagation @xcite . including spatial averaging becomes even more essential when stimulations and blocks produce steep gradients in @xmath17 @xcite . while the wide range of [ @xmath177 for which @xmath105 is blocked does not depends on @xmath77 , this is not the case for the subset of this interval leading to reentry annihilation by alternans amplification . as shown in fig . [ fig_map_single ] , the interval of block is very large for @xmath179 and decreases gradually until it disappears at @xmath180 . for the mbr model used herein , annihilation by alternans amplification exists for @xmath181 cm , and it occurs on a range of [ @xmath177 that remains much wider than the @xmath124 ms standard vulnerable window for most of this range of @xmath77 . this is consistent with the result of a clinical study in which dual pulses stimulation was found to be four times more effective than single stimulus to stop monomorphic ventricular tachycardia in man @xcite . as @xmath77 is increased , the [ @xmath177 area of @xmath105 block also encloses a growing number of tongues with increasingly prolong coexistence of @xmath95 and @xmath104 , a process that culminates in the appearance of zones of transition to dw reentry . then the zones with transition to dw reentry extend until they cover completely the area in which @xmath105 is blocked . from @xmath182 blocking @xmath105 yields automatically to sustained dw reentry . hence , transition to dw reentry does not necessarily imply a heterogeneous substrate as it is been proposed elsewhere @xcite , but can also be achieved in a homogeneous medium through the creation of a functional heterogeneity by a limited number of electrical stimulations . annihilation and transition to dw occur on separated ranges of loop lengths . on shorter loop , reentry annihilation is produced when alternans amplification reaches an amplitude high enough to block @xmath95and @xmath104 . on longer loop , the distributed alternans saturate at an amplitude that still permits sustained propagation . in our version of the br model , the transition from sw qp to period-1 reentry at @xmath183 is supercritical . however , we have shown that , for other sets of parameters , the bifurcation is subcritical , with bistability between qp and period-1 reentry near @xmath183 @xcite . in these cases , it is possible that the two - stimulations protocol applied to the sw period-1 reentry near @xmath183 could induced a transition to sw qp propagation , a phenomenon that was not possible with the instance of the mbr model used in this paper . our results are consistent with different clinical and experimental observations , and open the possibility to design more effective anti - arrhythmic pacing strategies . however , the modeling studies must be extended to more realistic representations to evaluate properly potential applications . preliminary results from an ongoing work on a two - dimensional annulus show that block by alternans amplification can still be obtained on this setting , but that additional scenarios are possible , including termination through transient fibrillation that has also been observed in real cardiac tissue . tissue heterogeneity , either at the level of the ionic properties or of the cells coupling , could also be important since termination of reentry has been obtained in a one - dimensional loop model embedding a small area of slower conduction but using an ionic model with minimal apd restitution properties @xcite . it remains to be seen if annihilation based on alternans amplification would be amplified or reduced by the inclusion of spatial inhomogeneity . investigation will also have to be extended to bidomain model in order to get a more proper representation of the stimulus . the simplified representation used herein can be an acceptable approximation for low amplitude stimuli or to mimic the effect spontaneous firing of a group of cells . however it is known that current spread of the stimulus depend on the properties of the external and internal medium @xcite . the mbr representation of the ionic properties is also oversimplified . however , since most phenomena described in this paper occurs in the few first beat after the stimulations and can be explained from the apd restitution and speed dispersion , it is unlikely that slow memory effects appearing in the dynamics of more complex model would change the behaviour . characterizing the apd restitution and speed dispersion of the more complex model in the range of frequency of repetitive activity associated to reentry should allow a prediction of the possible dynamics . this work is a further illustration of the richness and diversity of the dynamics that can results from the restitution of apd and dispersion of the speed even in a simplified model of the tissue . it has revealed some unexpected behaviours , like block by alternans amplification , which can be much more prevalent than the mechanism of unidirectional block that are usually assumed to be dominant . the low - dimensional model whose behaviour is equivalent to the ionic model , provides a generic understanding of the dynamics that can used a guideline to investigate the effects of future complexification of the model . this work was supported by grants from the natural sciences and engineering research council of canada ( av ) , the _ fonds qubcois de la recherche sur la nature et les technologies_(pc ) , as well as by the technical and computer resources of the rseau qubcois de calcul de haute performance . 99 g. r. mines , trans roy soc can 4 , 43 ( 1914 ) . frame and m. b. simson , circulation * 78 * , 1277 ( 1988 ) . and e. k. rhee , circ res * 68 * , 493 ( 1991 ) . , j. n. graziano , and p. a. boyden , j cardiovasc electrophysiol * 4 * , 672 ( 1993 ) . , b. mensour , a. vinet , and t. kus , can j cardiol * 19 * , 244 ( 2003 ) . , l. glass , and j. p. keener , phys rev lett * 70 * , 2182 ( 1993 ) . and f. a. roberge , ann biomed eng * 22 * , 568 ( 1994 ) . , journal of biological systems * 7 * , 451 ( 1999 ) . , ann biomed eng * 28 * , 704 ( 2000 ) . , y. h. kim , o. voroshilovsky , z. qu , j. r. kil , m. h. lee , h. s. karagueuzian , j. n. weiss , and p. s. chen , proc natl acad sci * 97 * , 6061 ( 2000 ) . , a. garfinkel , p. s. chen , and j. n. weiss , circulation * 102 * , 1664 ( 2000 ) . , g. marenzi , c. tondo , d. cardinale , f. giraldi , g. lauri , and m. guazzi , am j cardiol * 68 * , 492 ( 1991 ) . and o. m. orning , eur heart j * 14 * , 421 ( 1993 ) . , s. saksena , m. hill , r. b. krol , a. n. munsif , i. giorgberidze , p. mathew , and r. mehra , j am coll cardiol * 29 * , 1007 ( 1997 ) . and a. vinet , chaos * 12 * , 903 ( 2002 ) . and y. rudy , pacing clin electrophysiol * 14 * , 1700 ( 1991 ) . and y. rudy , j cardiovasc electrophysiol * 6 * , 115 ( 1995 ) . , m. s. hanna , and l. h. frame , circulation * 94 * , 2268 ( 1996 ) . and f. a. roberge , j theor biol * 170 * , 183 ( 1994b ) . , j. p. keener , and l. glass , siam journal on applied mathematics * 56 * , 119 ( 1996 ) . , f. h. fenton , s. j. evans , h. m. hastings , and a. karma , j cardiovasc electrophysiol * 12 * , 196 ( 2001 ) . , j. l. mcharg , and r. f. . j. gilmour , am j physiol heart circ physiol * 282 * , h516 ( 2002 ) . and a. karma , phys rev lett * 88 * , 208101 ( 2002 ) . , y. i. zilberter , and c. f. starmer , physica d * 70 * , 321 ( 1994 ) . , e. jalil , a. vinet , and t. kus , pacing clin electrophysiol * 23 * , 1200 ( 2000 ) . and j. p. keener , chaos * 12 * , 788 ( 2002 ) . , r. cardinal , p. lefranc , f. helie , p. rocque , t. kus , and p. page , circulation * 93 * , 1845 ( 1996 ) . , z. qu , y. h. kim , t. j. wu , a. garfinkel , j. n. weiss , h. s. karagueuzian , and p. s. chen , circ res * 84 * , 1318 ( 1999 ) . , a. vinet , and r. cardinal , j cardiovasc electrophysiol * 11 * , 531 ( 2000 ) . , d. li , and l. yue , annu rev physiol * 62 * , 51 ( 2000 ) . and l. glass , physical review e * 53 * , 6353 ( 1996 ) . , y. nagai , k. hall , m. talajic , and s. nattel , phys rev e * 65 * , 021908 ( 2002 ) . , j. jung , m. stopp , and h. schieffer , eur heart j * 18 * , 866 ( 1997 ) . , m. e. rosenthal , n. j. stamato , f. e. marchlinski , a. e. buxton , l. h. frame , j. m. miller , and m. e. josephson , j am coll cardiol * 8 * , 294 ( 1986 ) . , p. brugada , l. boersma , l. mont , c. kirchhof , h. j. wellens , and m. a. allessie , circulation * 83 * , 1621 ( 1991 ) . and m. m. scheinman , circulation * 97 * , 1589 ( 1998 ) . , k. m. stein , and d. j. christini , chaos * 12 * , 893 ( 2002 ) . and d. j. christini , phys rev e * 66 * , 061903 ( 2002 ) . , b. j. roth , and n. a. trayanova , j cardiovasc electrophysiol * 11 * , 274 ( 2000 ) . and e. cytrynbaum , j theor biol * 223 * , 233 ( 2003 ) . three conditions must be fulfilled for the block to occur : 1 ) @xmath105 , the retrograde front created by @xmath74 , must be blocked between the stimulation site @xmath114and @xmath184 the locus of the collision between the reentry front @xmath93 and retrograde front @xmath94 created by @xmath72 ; 2 ) afterward , @xmath95 , the antegrade front produced by @xmath72 , must be blocked when it returns near @xmath114 ; 3 ) finally , @xmath104 , the antegrade front produced by @xmath74 , must also be blocked when it travels between @xmath114 and @xmath97 the following three appendixes formulate the constraints associated with each of these conditions . in all three appendixes , we consider that @xmath185and @xmath104 travels toward increasing value of @xmath31 , we use @xmath186 , the repolarization time associated to the passage of the last activation front of the reentry @xmath93 at @xmath96 , as the reference time @xmath187 , and introduce a spatial coordinate @xmath188 to follow the retrograde fronts @xmath94 and @xmath105 . @xmath105 can propagate as long as its activation time @xmath189 is larger than @xmath190 , the repolarization time associated to @xmath94 . in the limiting case @xmath191 , @xmath192 , and @xmath105 travels with the maximum conduction time @xmath193 = @xmath194 . when @xmath105 reaches a point @xmath195 with @xmath196 it can continue to propagate if @xmath197 @xmath198 . the limiting case for the propagation of @xmath105 is thus @xmath199 we neglect the effect of coupling in the calculation of @xmath200 to obtain @xmath201 where @xmath202 @xmath203 is the diastolic interval associated to the propagation of @xmath94 . from our choice of reference time and the definition of @xmath204 , @xmath205 . in the mbr model , the @xmath60 dispersion curve is very steep such that the conduction time is minimal , except for a short interval of @xmath27 close to @xmath115 . as a consequence , we approximate that both @xmath94 and @xmath93 have been propagating with the minimum conduction time ( i.e. maximum speed ) @xmath206 , such that @xmath207 substituting these relations in eq . [ cond_r2s ] yields@xmath208 taking the spatial derivatives of eq . [ eq_trep_r1 ] yields@xmath209 which , thanks to eq . [ eq_trep_r1 ] , gives @xmath210 the existence and location of the critical point depends on the slope of the restitution curve , and on the relative difference between the maximum and minimum conduction time . from the @xmath211 given by eq . [ eq_dispersion ] , @xmath212 , and@xmath213 using eq . [ eq_restitution ] to solve this equation , we obtain that @xmath214 ms . this result means that , if @xmath215 ms , @xmath216 everywhere between @xmath97 and @xmath114 , and @xmath105 can not be blocked . it means also that if @xmath217 , there is a critical point @xmath195 whose position depends on @xmath100 but is independent of @xmath77 . the last step is to determine the maximum @xmath73 ( @xmath218 ) to get a block of @xmath219 at the @xmath220 . assuming again that @xmath105 travels with the minimum conduction time , the first condition of eq . [ cond_r2 ] becomes@xmath221 reducing to the limit@xmath222 the maximum @xmath73 for the block of @xmath219is independent of @xmath77 and of @xmath100 , provided that .@xmath217 . for @xmath223 to induce propagation @xmath73 , must also be @xmath224 , which is the refractory period at @xmath114 . in summary , for all @xmath225 ms , @xmath105 blocks if @xmath226 . the main approximation used herein is that both @xmath94 and @xmath105 propagate everywhere with the minimal propagation time . with regard to @xmath94 , the error introduces by the approximation is minimal unless @xmath227 . for @xmath105 , the conduction time is obviously underestimated when its activation time comes close to @xmath200 . the approximation thus overestimates @xmath228 , whose value is around 50 ms for the i d and il model , compared to the @xmath229 ms provided by the approximation . appendix [ appendix_a ] shows that @xmath230 $ ] , @xmath105 exist and is blocked if @xmath231,$ ] a set of conditions that do not depend on @xmath77 . the next event is the annihilation @xmath95 when it returns near @xmath114 and hits the refractory tail left @xmath104 . using @xmath186 as a reference time @xmath95 comes back to @xmath114 at the time @xmath232 , in which @xmath233 is the time taken by @xmath95 to propagates over the loop on its first turn . it is blocked if the system is still refractory , which means @xmath234 neglecting the effect of coupling on repolarization , @xmath235 is approximated as : @xmath236 in which @xmath237 , the diastolic interval associated to @xmath74 , is estimated by @xmath238 substituting eq . [ trep_s2_blocka1 ] and [ eq_ds2ds1 ] in eq . [ cond_block_a1 ] yields@xmath239 in which @xmath233 is the only non- local term . for most ionic models and experimental preparations , @xmath240 is a monotonic decreasing function of @xmath100 ( as in fig . 4 of ref . the effect of the prematurity on the return cycle comes from the steepness of the @xmath60 dispersion curve ( eq . [ eq_dispersion ] ) which is close to @xmath206 as soon as @xmath27 is greater than a few tenths of ms . therefore , the prolongation of the return cycle depends on the limited region beyond @xmath114 over which @xmath95 does not propagate at the maximum speed . hence , we write @xmath241 where @xmath242 is maximum for @xmath117 and @xmath243 the condition [ cond_block_a1_rr ] for the block of @xmath95 becomes @xmath244 consider @xmath245 as the minimum value of @xmath73 to get a block of @xmath95 . lets consider next the case where @xmath217 and @xmath246 , for which @xmath247 . the condition for the block of @xmath95 becomes @xmath248 which can not be fulfilled since both @xmath100 and @xmath249 are smaller than the diastolic interval of the free reentry . the block of @xmath95 will occur from the minimum @xmath250 . this explains why the dynamical regime [ @xmath251 is found in the lower portion of the [ @xmath177 area in which @xmath105 blocked , as seen in figure [ fig_map_super ] . the condition for the block of @xmath105 are that @xmath252 $ ] and @xmath253 $ ] . since the right hand side of eq . [ eq_condf ] is a growing function of @xmath77 , @xmath254 for any fixed value of @xmath100 must also increase with @xmath77 until reaching @xmath255 . hence , there is a limit value of @xmath77 from which @xmath95cannot be blocked on its first return . for @xmath256 , the value of @xmath100 for which @xmath257 , eq . [ eq_condf ] becomes@xmath258 shortening @xmath256 increases both the left and right side of the equation , such that the value of @xmath256 depends on the balance between the slope of @xmath259 ( i.e. the change in the return cycle ) and @xmath260 ( the change in repolarisation time at @xmath114 ) . the derivative of the equation with respect to @xmath77 gives:@xmath261 where @xmath262 . in the mbr model , @xmath263 whereas @xmath264 at low @xmath27 value , such that @xmath265 as observed in the numerical simulations . appendix [ appendix_b ] shows that @xmath95 is blocked only over a subset of the [ @xmath266 area for which @xmath105 exist and is blocked ( i.e. @xmath252 $ ] , @xmath267 $ ] ) . with respect to @xmath268 , the lower bound of this subset is greater than @xmath269 and increases with @xmath77 , while the upper bound remains constant at @xmath255 . with respect to @xmath270the variation of the limits of this subset is given by a complex expression that depends on the slope of both the restitution and dispersion curves . for the mbr model , the upper @xmath100 of the subset decreases toward 0 as @xmath77 is increased . the last step is to obtain the conditions for the block of @xmath271 @xmath104 blocks between @xmath97 and the locus where @xmath219has stopped when it hits the refractory tail left by @xmath95 . using the @xmath204 coordinate , the condition for the block of @xmath104 is that there is a point where @xmath272 @xmath273 is approximated as @xmath274where @xmath275 since @xmath105 has not propagated in this region . using eq . [ eq_trep_r1 ] of appendix [ appendix_a ] for @xmath276 and eq . [ eqtau1 ] of appendix [ appendix_b ] for @xmath277 yields @xmath278 @xmath279 @xmath280 where @xmath281 is the time needed for @xmath104 to travel from the stimulation site to @xmath204 . for @xmath95 , we have assumed in [ appendix_b ] that the prolongation of the return cycle was occurring mainly in a short region around @xmath114 beyond which @xmath95 was travelling at maximal speed . the situation is different for @xmath104 . as it could be seen in fig . [ fig_bloc_aa ] a , the @xmath27 associated to @xmath104 as it propagates away form @xmath282meaning that it speed of propagation diminishes . nevertheless , we write @xmath283 with eq . [ eq_trepola1 ] and [ eq_tact2 ] , the condition [ conda2_1 ] for the block @xmath104 after one rotation becomes @xmath284 to get the [ @xmath131 block , the conditions given by eq . [ lim_a1_bl_final ] and eq . [ lim_a2_bl_final1 ] must both be fulfilled , leading to the final condition : @xmath285 with the supplementary constraint that @xmath73 and @xmath100 must remain in the interval where @xmath105 is block . since both @xmath73 and @xmath100 are bound , there is a limiting @xmath77 value beyond which [ @xmath286 block can not occur . for the il and i d model , @xmath163we found that @xmath104 was always stop when @xmath95 was blocked , meaning that eq . [ lim_a2_bl_final1 ] was fulfilled whenever eq . [ lim_a1_bl_final ] was satisfied . however , the condition depends on the restitution and dispersion curve and , through @xmath287 , the interaction of @xmath104 with the spatial profile of @xmath288 left by @xmath95 .